go/src/cmd/compile/internal/gc/esc.go

2273 lines
64 KiB
Go
Raw Normal View History

// Copyright 2011 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package gc
import (
"cmd/compile/internal/types"
"fmt"
"strconv"
"strings"
)
// Escape analysis.
// An escape analysis pass for a set of functions. The
// analysis assumes that closures and the functions in which
// they appear are analyzed together, so that the aliasing
// between their variables can be modeled more precisely.
//
// First escfunc, esc and escassign recurse over the ast of
// each function to dig out flow(dst,src) edges between any
// pointer-containing nodes and store those edges in
// e.nodeEscState(dst).Flowsrc. For values assigned to a
// variable in an outer scope or used as a return value,
// they store a flow(theSink, src) edge to a fake node 'the
// Sink'. For variables referenced in closures, an edge
// flow(closure, &var) is recorded and the flow of a closure
// itself to an outer scope is tracked the same way as other
// variables.
//
// Then escflood walks the graph in destination-to-source
// order, starting at theSink, propagating a computed
// "escape level", and tags as escaping values it can
// reach that are either & (address-taken) nodes or new(T),
// and tags pointer-typed or pointer-containing function
// parameters it can reach as leaking.
//
// If a value's address is taken but the address does not escape,
// then the value can stay on the stack. If the value new(T) does
// not escape, then new(T) can be rewritten into a stack allocation.
// The same is true of slice literals.
func escapes(all []*Node) {
visitBottomUp(all, escAnalyze)
}
const (
EscFuncUnknown = 0 + iota
EscFuncPlanned
EscFuncStarted
EscFuncTagged
)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// A Level encodes the reference state and context applied to
// (stack, heap) allocated memory.
//
// value is the overall sum of *(1) and &(-1) operations encountered
// along a path from a destination (sink, return value) to a source
// (allocation, parameter).
//
// suffixValue is the maximum-copy-started-suffix-level on
// a flow path from a sink/destination. That is, a value
// with suffixValue N is guaranteed to be dereferenced at least
// N deep (chained applications of DOTPTR or IND or INDEX)
// before the result is assigned to a sink.
//
// For example, suppose x is a pointer to T, declared type T struct { left, right *T }
// sink = x.left.left --> level(x)=2, x is reached via two dereferences (DOTPTR) and does not escape to sink.
// sink = &T{right:x} --> level(x)=-1, x is accessible from sink via one "address of"
// sink = &T{right:&T{right:x}} --> level(x)=-2, x is accessible from sink via two "address of"
//
// However, in the next example x's level value and suffixValue differ:
// sink = &T{right:&T{right:x.left}} --> level(x).value=-1, level(x).suffixValue=1
// The positive suffixValue indicates that x is NOT accessible
// from sink. Without a separate suffixValue to capture this, x would
// appear to escape because its "value" would be -1. (The copy
// operations are sometimes implicit in the source code; in this case,
// the value of x.left was copied into a field of an newly allocated T).
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
//
// Each node's level (value and suffixValue) is the maximum for
// all flow paths from (any) sink to that node.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// There's one of these for each Node, and the integer values
// rarely exceed even what can be stored in 4 bits, never mind 8.
type Level struct {
value, suffixValue int8
}
// There are loops in the escape graph,
// causing arbitrary recursion into deeper and deeper
// levels. Cut this off safely by making minLevel sticky:
// once you get that deep, you cannot go down any further
// but you also cannot go up any further. This is a
// conservative fix. Making minLevel smaller (more negative)
// would handle more complex chains of indirections followed
// by address-of operations, at the cost of repeating the
// traversal once for each additional allowed level when a
// loop is encountered. Using -2 suffices to pass all the
// tests we have written so far, which we assume matches the
// level of complexity we want the escape analysis code to
// handle.
const MinLevel = -2
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func (l Level) int() int {
return int(l.value)
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func levelFrom(i int) Level {
if i <= MinLevel {
return Level{value: MinLevel}
}
return Level{value: int8(i)}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func satInc8(x int8) int8 {
if x == 127 {
return 127
}
return x + 1
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func min8(a, b int8) int8 {
if a < b {
return a
}
return b
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func max8(a, b int8) int8 {
if a > b {
return a
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
return b
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// inc returns the level l + 1, representing the effect of an indirect (*) operation.
func (l Level) inc() Level {
if l.value <= MinLevel {
return Level{value: MinLevel}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
return Level{value: satInc8(l.value), suffixValue: satInc8(l.suffixValue)}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// dec returns the level l - 1, representing the effect of an address-of (&) operation.
func (l Level) dec() Level {
if l.value <= MinLevel {
return Level{value: MinLevel}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
return Level{value: l.value - 1, suffixValue: l.suffixValue - 1}
}
// copy returns the level for a copy of a value with level l.
// The resulting suffixValue is at least zero, or larger if it was already larger.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
func (l Level) copy() Level {
return Level{value: l.value, suffixValue: max8(l.suffixValue, 0)}
}
func (l1 Level) min(l2 Level) Level {
return Level{
value: min8(l1.value, l2.value),
suffixValue: min8(l1.suffixValue, l2.suffixValue)}
}
// guaranteedDereference returns the number of dereferences
// applied to a pointer before addresses are taken/generated.
// This is the maximum level computed from path suffixes starting
// with copies where paths flow from destination to source.
func (l Level) guaranteedDereference() int {
return int(l.suffixValue)
}
// An EscStep documents one step in the path from memory
// that is heap allocated to the (alleged) reason for the
// heap allocation.
type EscStep struct {
src, dst *Node // the endpoints of this edge in the escape-to-heap chain.
where *Node // sometimes the endpoints don't match source locations; set 'where' to make that right
parent *EscStep // used in flood to record path
why string // explanation for this step in the escape-to-heap chain
busy bool // used in prevent to snip cycles.
}
type NodeEscState struct {
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
Curfn *Node
Flowsrc []EscStep // flow(this, src)
Retval Nodes // on OCALLxxx, list of dummy return values
Loopdepth int32 // -1: global, 0: return variables, 1:function top level, increased inside function for every loop or label to mark scopes
Level Level
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
Walkgen uint32
Maxextraloopdepth int32
}
func (e *EscState) nodeEscState(n *Node) *NodeEscState {
if nE, ok := n.Opt().(*NodeEscState); ok {
return nE
}
if n.Opt() != nil {
Fatalf("nodeEscState: opt in use (%T)", n.Opt())
}
nE := &NodeEscState{
Curfn: Curfn,
}
n.SetOpt(nE)
e.opts = append(e.opts, n)
return nE
}
func (e *EscState) track(n *Node) {
if Curfn == nil {
Fatalf("EscState.track: Curfn nil")
}
n.Esc = EscNone // until proven otherwise
nE := e.nodeEscState(n)
nE.Loopdepth = e.loopdepth
e.noesc = append(e.noesc, n)
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Escape constants are numbered in order of increasing "escapiness"
// to help make inferences be monotonic. With the exception of
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// EscNever which is sticky, eX < eY means that eY is more exposed
// than eX, and hence replaces it in a conservative analysis.
const (
EscUnknown = iota
EscNone // Does not escape to heap, result, or parameters.
EscReturn // Is returned or reachable from returned.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
EscHeap // Reachable from the heap
EscNever // By construction will not escape.
EscBits = 3
EscMask = (1 << EscBits) - 1
EscContentEscapes = 1 << EscBits // value obtained by indirect of parameter escapes to heap
EscReturnBits = EscBits + 1
// Node.esc encoding = | escapeReturnEncoding:(width-4) | contentEscapes:1 | escEnum:3
)
// escMax returns the maximum of an existing escape value
// (and its additional parameter flow flags) and a new escape type.
func escMax(e, etype uint16) uint16 {
if e&EscMask >= EscHeap {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// normalize
if e&^EscMask != 0 {
Fatalf("Escape information had unexpected return encoding bits (w/ EscHeap, EscNever), e&EscMask=%v", e&EscMask)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
if e&EscMask > etype {
return e
}
if etype == EscNone || etype == EscReturn {
return (e &^ EscMask) | etype
}
return etype
}
// For each input parameter to a function, the escapeReturnEncoding describes
// how the parameter may leak to the function's outputs. This is currently the
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// "level" of the leak where level is 0 or larger (negative level means stored into
// something whose address is returned -- but that implies stored into the heap,
// hence EscHeap, which means that the details are not currently relevant. )
const (
bitsPerOutputInTag = 3 // For each output, the number of bits for a tag
bitsMaskForTag = uint16(1<<bitsPerOutputInTag) - 1 // The bit mask to extract a single tag.
maxEncodedLevel = int(bitsMaskForTag - 1) // The largest level that can be stored in a tag.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
)
type EscState struct {
// Fake node that all
// - return values and output variables
// - parameters on imported functions not marked 'safe'
// - assignments to global variables
// flow to.
theSink Node
dsts []*Node // all dst nodes
loopdepth int32 // for detecting nested loop scopes
pdepth int // for debug printing in recursions.
dstcount int // diagnostic
edgecount int // diagnostic
noesc []*Node // list of possible non-escaping nodes, for printing
recursive bool // recursive function or group of mutually recursive functions.
opts []*Node // nodes with .Opt initialized
walkgen uint32
}
func newEscState(recursive bool) *EscState {
e := new(EscState)
e.theSink.Op = ONAME
e.theSink.Orig = &e.theSink
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
e.theSink.SetClass(PEXTERN)
e.theSink.Sym = lookup(".sink")
e.nodeEscState(&e.theSink).Loopdepth = -1
e.recursive = recursive
return e
}
func (e *EscState) stepWalk(dst, src *Node, why string, parent *EscStep) *EscStep {
// TODO: keep a cache of these, mark entry/exit in escwalk to avoid allocation
// Or perhaps never mind, since it is disabled unless printing is on.
// We may want to revisit this, since the EscStep nodes would make
// an excellent replacement for the poorly-separated graph-build/graph-flood
// stages.
if Debug['m'] == 0 {
return nil
}
return &EscStep{src: src, dst: dst, why: why, parent: parent}
}
func (e *EscState) stepAssign(step *EscStep, dst, src *Node, why string) *EscStep {
if Debug['m'] == 0 {
return nil
}
if step != nil { // Caller may have known better.
if step.why == "" {
step.why = why
}
if step.dst == nil {
step.dst = dst
}
if step.src == nil {
step.src = src
}
return step
}
return &EscStep{src: src, dst: dst, why: why}
}
func (e *EscState) stepAssignWhere(dst, src *Node, why string, where *Node) *EscStep {
if Debug['m'] == 0 {
return nil
}
return &EscStep{src: src, dst: dst, why: why, where: where}
}
// funcSym returns fn.Func.Nname.Sym if no nils are encountered along the way.
func funcSym(fn *Node) *types.Sym {
if fn == nil || fn.Func.Nname == nil {
return nil
}
return fn.Func.Nname.Sym
}
// curfnSym returns n.Curfn.Nname.Sym if no nils are encountered along the way.
func (e *EscState) curfnSym(n *Node) *types.Sym {
nE := e.nodeEscState(n)
return funcSym(nE.Curfn)
}
func escAnalyze(all []*Node, recursive bool) {
e := newEscState(recursive)
for _, n := range all {
if n.Op == ODCLFUNC {
n.Esc = EscFuncPlanned
if Debug['m'] > 3 {
Dump("escAnalyze", n)
}
}
}
// flow-analyze functions
for _, n := range all {
if n.Op == ODCLFUNC {
e.escfunc(n)
}
}
// visit the upstream of each dst, mark address nodes with
// addrescapes, mark parameters unsafe
escapes := make([]uint16, len(e.dsts))
for i, n := range e.dsts {
escapes[i] = n.Esc
}
for _, n := range e.dsts {
e.escflood(n)
}
for {
done := true
for i, n := range e.dsts {
if n.Esc != escapes[i] {
done = false
if Debug['m'] > 2 {
Warnl(n.Pos, "Reflooding %v %S", e.curfnSym(n), n)
}
escapes[i] = n.Esc
e.escflood(n)
}
}
if done {
break
}
}
// for all top level functions, tag the typenodes corresponding to the param nodes
for _, n := range all {
if n.Op == ODCLFUNC {
e.esctag(n)
}
}
if Debug['m'] != 0 {
for _, n := range e.noesc {
if n.Esc == EscNone && n.Op != OADDR {
Warnl(n.Pos, "%v %S does not escape", e.curfnSym(n), n)
}
}
}
for _, x := range e.opts {
x.SetOpt(nil)
}
}
func (e *EscState) escfunc(fn *Node) {
if fn.Esc != EscFuncPlanned {
Fatalf("repeat escfunc %v", fn.Func.Nname)
}
fn.Esc = EscFuncStarted
saveld := e.loopdepth
e.loopdepth = 1
savefn := Curfn
Curfn = fn
for _, ln := range Curfn.Func.Dcl {
if ln.Op != ONAME {
continue
}
lnE := e.nodeEscState(ln)
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
switch ln.Class() {
// out params are in a loopdepth between the sink and all local variables
case PPARAMOUT:
lnE.Loopdepth = 0
case PPARAM:
lnE.Loopdepth = 1
if ln.Type != nil && !types.Haspointers(ln.Type) {
break
}
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if Curfn.Nbody.Len() == 0 && !Curfn.Noescape() {
ln.Esc = EscHeap
} else {
ln.Esc = EscNone // prime for escflood later
}
e.noesc = append(e.noesc, ln)
}
}
// in a mutually recursive group we lose track of the return values
if e.recursive {
for _, ln := range Curfn.Func.Dcl {
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
if ln.Op == ONAME && ln.Class() == PPARAMOUT {
e.escflows(&e.theSink, ln, e.stepAssign(nil, ln, ln, "returned from recursive function"))
}
}
}
e.escloopdepthlist(Curfn.Nbody)
e.esclist(Curfn.Nbody, Curfn)
Curfn = savefn
e.loopdepth = saveld
}
// Mark labels that have no backjumps to them as not increasing e.loopdepth.
// Walk hasn't generated (goto|label).Left.Sym.Label yet, so we'll cheat
// and set it to one of the following two. Then in esc we'll clear it again.
var (
looping Node
nonlooping Node
)
func (e *EscState) escloopdepthlist(l Nodes) {
for _, n := range l.Slice() {
e.escloopdepth(n)
}
}
func (e *EscState) escloopdepth(n *Node) {
if n == nil {
return
}
e.escloopdepthlist(n.Ninit)
switch n.Op {
case OLABEL:
if n.Sym == nil {
Fatalf("esc:label without label: %+v", n)
}
// Walk will complain about this label being already defined, but that's not until
// after escape analysis. in the future, maybe pull label & goto analysis out of walk and put before esc
n.Sym.Label = asTypesNode(&nonlooping)
case OGOTO:
if n.Sym == nil {
Fatalf("esc:goto without label: %+v", n)
}
// If we come past one that's uninitialized, this must be a (harmless) forward jump
// but if it's set to nonlooping the label must have preceded this goto.
if asNode(n.Sym.Label) == &nonlooping {
n.Sym.Label = asTypesNode(&looping)
}
}
e.escloopdepth(n.Left)
e.escloopdepth(n.Right)
e.escloopdepthlist(n.List)
e.escloopdepthlist(n.Nbody)
e.escloopdepthlist(n.Rlist)
}
func (e *EscState) esclist(l Nodes, parent *Node) {
for _, n := range l.Slice() {
e.esc(n, parent)
}
}
func (e *EscState) isSliceSelfAssign(dst, src *Node) bool {
// Detect the following special case.
//
// func (b *Buffer) Foo() {
// n, m := ...
// b.buf = b.buf[n:m]
// }
//
// This assignment is a no-op for escape analysis,
// it does not store any new pointers into b that were not already there.
// However, without this special case b will escape, because we assign to OIND/ODOTPTR.
// Here we assume that the statement will not contain calls,
// that is, that order will move any calls to init.
// Otherwise base ONAME value could change between the moments
// when we evaluate it for dst and for src.
// dst is ONAME dereference.
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if dst.Op != ODEREF && dst.Op != ODOTPTR || dst.Left.Op != ONAME {
return false
}
// src is a slice operation.
switch src.Op {
case OSLICE, OSLICE3, OSLICESTR:
// OK.
case OSLICEARR, OSLICE3ARR:
// Since arrays are embedded into containing object,
// slice of non-pointer array will introduce a new pointer into b that was not already there
// (pointer to b itself). After such assignment, if b contents escape,
// b escapes as well. If we ignore such OSLICEARR, we will conclude
// that b does not escape when b contents do.
//
// Pointer to an array is OK since it's not stored inside b directly.
// For slicing an array (not pointer to array), there is an implicit OADDR.
// We check that to determine non-pointer array slicing.
if src.Left.Op == OADDR {
return false
}
default:
return false
}
// slice is applied to ONAME dereference.
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if src.Left.Op != ODEREF && src.Left.Op != ODOTPTR || src.Left.Left.Op != ONAME {
return false
}
// dst and src reference the same base ONAME.
return dst.Left == src.Left.Left
}
// isSelfAssign reports whether assignment from src to dst can
// be ignored by the escape analysis as it's effectively a self-assignment.
func (e *EscState) isSelfAssign(dst, src *Node) bool {
if e.isSliceSelfAssign(dst, src) {
return true
}
// Detect trivial assignments that assign back to the same object.
//
// It covers these cases:
// val.x = val.y
// val.x[i] = val.y[j]
// val.x1.x2 = val.x1.y2
// ... etc
//
// These assignments do not change assigned object lifetime.
if dst == nil || src == nil || dst.Op != src.Op {
return false
}
switch dst.Op {
case ODOT, ODOTPTR:
// Safe trailing accessors that are permitted to differ.
case OINDEX:
if e.mayAffectMemory(dst.Right) || e.mayAffectMemory(src.Right) {
return false
}
default:
return false
}
// The expression prefix must be both "safe" and identical.
return samesafeexpr(dst.Left, src.Left)
}
// mayAffectMemory reports whether n evaluation may affect program memory state.
// If expression can't affect it, then it can be safely ignored by the escape analysis.
func (e *EscState) mayAffectMemory(n *Node) bool {
// We may want to use "memory safe" black list instead of general
// "side-effect free", which can include all calls and other ops
// that can affect allocate or change global state.
// It's safer to start from a whitelist for now.
//
// We're ignoring things like division by zero, index out of range,
// and nil pointer dereference here.
switch n.Op {
case ONAME, OCLOSUREVAR, OLITERAL:
return false
// Left+Right group.
case OINDEX, OADD, OSUB, OOR, OXOR, OMUL, OLSH, ORSH, OAND, OANDNOT, ODIV, OMOD:
return e.mayAffectMemory(n.Left) || e.mayAffectMemory(n.Right)
// Left group.
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
case ODOT, ODOTPTR, ODEREF, OCONVNOP, OCONV, OLEN, OCAP,
ONOT, OBITNOT, OPLUS, ONEG, OALIGNOF, OOFFSETOF, OSIZEOF:
return e.mayAffectMemory(n.Left)
default:
return true
}
}
func (e *EscState) esc(n *Node, parent *Node) {
if n == nil {
return
}
lno := setlineno(n)
// ninit logically runs at a different loopdepth than the rest of the for loop.
e.esclist(n.Ninit, n)
if n.Op == OFOR || n.Op == OFORUNTIL || n.Op == ORANGE {
e.loopdepth++
}
// type switch variables have no ODCL.
// process type switch as declaration.
// must happen before processing of switch body,
// so before recursion.
if n.Op == OSWITCH && n.Left != nil && n.Left.Op == OTYPESW {
for _, cas := range n.List.Slice() { // cases
// it.N().Rlist is the variable per case
if cas.Rlist.Len() != 0 {
e.nodeEscState(cas.Rlist.First()).Loopdepth = e.loopdepth
}
}
}
// Big stuff and non-constant-sized stuff escapes unconditionally.
// "Big" conditions that were scattered around in walk have been
// gathered here.
if n.Esc != EscHeap && n.Type != nil &&
(n.Type.Width > maxStackVarSize ||
(n.Op == ONEW || n.Op == OPTRLIT) && n.Type.Elem().Width >= maxImplicitStackVarSize ||
n.Op == OMAKESLICE && !isSmallMakeSlice(n)) {
// isSmallMakeSlice returns false for non-constant len/cap.
// If that's the case, print a more accurate escape reason.
var msgVerb, escapeMsg string
if n.Op == OMAKESLICE && (!Isconst(n.Left, CTINT) || !Isconst(n.Right, CTINT)) {
msgVerb, escapeMsg = "has ", "non-constant size"
} else {
msgVerb, escapeMsg = "is ", "too large for stack"
}
if Debug['m'] > 2 {
Warnl(n.Pos, "%v "+msgVerb+escapeMsg, n)
}
n.Esc = EscHeap
addrescapes(n)
e.escassignSinkWhy(n, n, escapeMsg) // TODO category: tooLarge
}
cmd/compile: eliminate dead code in if statements after typechecking This is a more thorough and cleaner fix than doing dead code elimination separately during inlining, escape analysis, and export. Unfortunately, it does add another full walk of the AST. The performance impact is very small, but not non-zero. If a label or goto is present in the dead code, it is not eliminated. This restriction can be removed once label/goto checking occurs much earlier in the compiler. In practice, it probably doesn't matter much. Updates #19699 Fixes #19705 name old alloc/op new alloc/op delta Template 39.2MB ± 0% 39.3MB ± 0% +0.28% (p=0.008 n=5+5) Unicode 29.8MB ± 0% 29.8MB ± 0% ~ (p=1.000 n=5+5) GoTypes 113MB ± 0% 113MB ± 0% -0.55% (p=0.008 n=5+5) SSA 1.25GB ± 0% 1.25GB ± 0% +0.02% (p=0.008 n=5+5) Flate 25.3MB ± 0% 25.3MB ± 0% -0.24% (p=0.032 n=5+5) GoParser 31.7MB ± 0% 31.8MB ± 0% +0.31% (p=0.008 n=5+5) Reflect 78.2MB ± 0% 78.3MB ± 0% ~ (p=0.421 n=5+5) Tar 26.6MB ± 0% 26.7MB ± 0% +0.21% (p=0.008 n=5+5) XML 42.2MB ± 0% 42.2MB ± 0% ~ (p=0.056 n=5+5) name old allocs/op new allocs/op delta Template 385k ± 0% 387k ± 0% +0.51% (p=0.016 n=5+5) Unicode 321k ± 0% 321k ± 0% ~ (p=1.000 n=5+5) GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=1.000 n=5+5) SSA 9.71M ± 0% 9.72M ± 0% +0.10% (p=0.008 n=5+5) Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5) GoParser 315k ± 0% 317k ± 0% +0.71% (p=0.008 n=5+5) Reflect 980k ± 0% 983k ± 0% +0.30% (p=0.032 n=5+5) Tar 251k ± 0% 252k ± 0% +0.55% (p=0.016 n=5+5) XML 392k ± 0% 393k ± 0% +0.30% (p=0.008 n=5+5) Change-Id: Ia10ff4bbf5c6eae782582cc9cbc9785494d4fb83 Reviewed-on: https://go-review.googlesource.com/38773 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-03-27 11:38:20 -07:00
e.esc(n.Left, n)
if n.Op == ORANGE {
// ORANGE node's Right is evaluated before the loop
e.loopdepth--
}
cmd/compile: eliminate dead code in if statements after typechecking This is a more thorough and cleaner fix than doing dead code elimination separately during inlining, escape analysis, and export. Unfortunately, it does add another full walk of the AST. The performance impact is very small, but not non-zero. If a label or goto is present in the dead code, it is not eliminated. This restriction can be removed once label/goto checking occurs much earlier in the compiler. In practice, it probably doesn't matter much. Updates #19699 Fixes #19705 name old alloc/op new alloc/op delta Template 39.2MB ± 0% 39.3MB ± 0% +0.28% (p=0.008 n=5+5) Unicode 29.8MB ± 0% 29.8MB ± 0% ~ (p=1.000 n=5+5) GoTypes 113MB ± 0% 113MB ± 0% -0.55% (p=0.008 n=5+5) SSA 1.25GB ± 0% 1.25GB ± 0% +0.02% (p=0.008 n=5+5) Flate 25.3MB ± 0% 25.3MB ± 0% -0.24% (p=0.032 n=5+5) GoParser 31.7MB ± 0% 31.8MB ± 0% +0.31% (p=0.008 n=5+5) Reflect 78.2MB ± 0% 78.3MB ± 0% ~ (p=0.421 n=5+5) Tar 26.6MB ± 0% 26.7MB ± 0% +0.21% (p=0.008 n=5+5) XML 42.2MB ± 0% 42.2MB ± 0% ~ (p=0.056 n=5+5) name old allocs/op new allocs/op delta Template 385k ± 0% 387k ± 0% +0.51% (p=0.016 n=5+5) Unicode 321k ± 0% 321k ± 0% ~ (p=1.000 n=5+5) GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=1.000 n=5+5) SSA 9.71M ± 0% 9.72M ± 0% +0.10% (p=0.008 n=5+5) Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5) GoParser 315k ± 0% 317k ± 0% +0.71% (p=0.008 n=5+5) Reflect 980k ± 0% 983k ± 0% +0.30% (p=0.032 n=5+5) Tar 251k ± 0% 252k ± 0% +0.55% (p=0.016 n=5+5) XML 392k ± 0% 393k ± 0% +0.30% (p=0.008 n=5+5) Change-Id: Ia10ff4bbf5c6eae782582cc9cbc9785494d4fb83 Reviewed-on: https://go-review.googlesource.com/38773 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-03-27 11:38:20 -07:00
e.esc(n.Right, n)
if n.Op == ORANGE {
e.loopdepth++
}
cmd/compile: eliminate dead code in if statements after typechecking This is a more thorough and cleaner fix than doing dead code elimination separately during inlining, escape analysis, and export. Unfortunately, it does add another full walk of the AST. The performance impact is very small, but not non-zero. If a label or goto is present in the dead code, it is not eliminated. This restriction can be removed once label/goto checking occurs much earlier in the compiler. In practice, it probably doesn't matter much. Updates #19699 Fixes #19705 name old alloc/op new alloc/op delta Template 39.2MB ± 0% 39.3MB ± 0% +0.28% (p=0.008 n=5+5) Unicode 29.8MB ± 0% 29.8MB ± 0% ~ (p=1.000 n=5+5) GoTypes 113MB ± 0% 113MB ± 0% -0.55% (p=0.008 n=5+5) SSA 1.25GB ± 0% 1.25GB ± 0% +0.02% (p=0.008 n=5+5) Flate 25.3MB ± 0% 25.3MB ± 0% -0.24% (p=0.032 n=5+5) GoParser 31.7MB ± 0% 31.8MB ± 0% +0.31% (p=0.008 n=5+5) Reflect 78.2MB ± 0% 78.3MB ± 0% ~ (p=0.421 n=5+5) Tar 26.6MB ± 0% 26.7MB ± 0% +0.21% (p=0.008 n=5+5) XML 42.2MB ± 0% 42.2MB ± 0% ~ (p=0.056 n=5+5) name old allocs/op new allocs/op delta Template 385k ± 0% 387k ± 0% +0.51% (p=0.016 n=5+5) Unicode 321k ± 0% 321k ± 0% ~ (p=1.000 n=5+5) GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=1.000 n=5+5) SSA 9.71M ± 0% 9.72M ± 0% +0.10% (p=0.008 n=5+5) Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5) GoParser 315k ± 0% 317k ± 0% +0.71% (p=0.008 n=5+5) Reflect 980k ± 0% 983k ± 0% +0.30% (p=0.032 n=5+5) Tar 251k ± 0% 252k ± 0% +0.55% (p=0.016 n=5+5) XML 392k ± 0% 393k ± 0% +0.30% (p=0.008 n=5+5) Change-Id: Ia10ff4bbf5c6eae782582cc9cbc9785494d4fb83 Reviewed-on: https://go-review.googlesource.com/38773 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-03-27 11:38:20 -07:00
e.esclist(n.Nbody, n)
e.esclist(n.List, n)
e.esclist(n.Rlist, n)
if n.Op == OFOR || n.Op == OFORUNTIL || n.Op == ORANGE {
e.loopdepth--
}
if Debug['m'] > 2 {
fmt.Printf("%v:[%d] %v esc: %v\n", linestr(lineno), e.loopdepth, funcSym(Curfn), n)
}
opSwitch:
switch n.Op {
// Record loop depth at declaration.
case ODCL:
if n.Left != nil {
e.nodeEscState(n.Left).Loopdepth = e.loopdepth
}
case OLABEL:
switch asNode(n.Sym.Label) {
case &nonlooping:
if Debug['m'] > 2 {
fmt.Printf("%v:%v non-looping label\n", linestr(lineno), n)
}
case &looping:
if Debug['m'] > 2 {
fmt.Printf("%v: %v looping label\n", linestr(lineno), n)
}
e.loopdepth++
}
n.Sym.Label = nil
case ORANGE:
if n.List.Len() >= 2 {
// Everything but fixed array is a dereference.
// If fixed array is really the address of fixed array,
// it is also a dereference, because it is implicitly
// dereferenced (see #12588)
if n.Type.IsArray() &&
!(n.Right.Type.IsPtr() && types.Identical(n.Right.Type.Elem(), n.Type)) {
e.escassignWhyWhere(n.List.Second(), n.Right, "range", n)
} else {
e.escassignDereference(n.List.Second(), n.Right, e.stepAssignWhere(n.List.Second(), n.Right, "range-deref", n))
}
}
case OSWITCH:
if n.Left != nil && n.Left.Op == OTYPESW {
for _, cas := range n.List.Slice() {
// cases
// n.Left.Right is the argument of the .(type),
// it.N().Rlist is the variable per case
if cas.Rlist.Len() != 0 {
e.escassignWhyWhere(cas.Rlist.First(), n.Left.Right, "switch case", n)
}
}
}
case OAS, OASOP:
// Filter out some no-op assignments for escape analysis.
if e.isSelfAssign(n.Left, n.Right) {
if Debug['m'] != 0 {
Warnl(n.Pos, "%v ignoring self-assignment in %S", e.curfnSym(n), n)
}
break
}
e.escassign(n.Left, n.Right, e.stepAssignWhere(nil, nil, "", n))
case OAS2: // x,y = a,b
if n.List.Len() == n.Rlist.Len() {
rs := n.Rlist.Slice()
where := n
for i, n := range n.List.Slice() {
e.escassignWhyWhere(n, rs[i], "assign-pair", where)
}
}
case OAS2RECV: // v, ok = <-ch
e.escassignWhyWhere(n.List.First(), n.Rlist.First(), "assign-pair-receive", n)
case OAS2MAPR: // v, ok = m[k]
e.escassignWhyWhere(n.List.First(), n.Rlist.First(), "assign-pair-mapr", n)
case OAS2DOTTYPE: // v, ok = x.(type)
e.escassignWhyWhere(n.List.First(), n.Rlist.First(), "assign-pair-dot-type", n)
case OSEND: // ch <- x
e.escassignSinkWhy(n, n.Right, "send")
case ODEFER:
if e.loopdepth == 1 { // top level
break
}
// arguments leak out of scope
// TODO: leak to a dummy node instead
// defer f(x) - f and x escape
e.escassignSinkWhy(n, n.Left.Left, "defer func")
e.escassignSinkWhy(n, n.Left.Right, "defer func ...") // ODDDARG for call
for _, arg := range n.Left.List.Slice() {
e.escassignSinkWhy(n, arg, "defer func arg")
}
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
case OGO:
// go f(x) - f and x escape
e.escassignSinkWhy(n, n.Left.Left, "go func")
e.escassignSinkWhy(n, n.Left.Right, "go func ...") // ODDDARG for call
for _, arg := range n.Left.List.Slice() {
e.escassignSinkWhy(n, arg, "go func arg")
}
case OCALLMETH, OCALLFUNC, OCALLINTER:
e.esccall(n, parent)
// esccall already done on n.Rlist.First()
// tie its Retval to n.List
case OAS2FUNC: // x,y = f()
rs := e.nodeEscState(n.Rlist.First()).Retval.Slice()
where := n
for i, n := range n.List.Slice() {
if i >= len(rs) {
break
}
e.escassignWhyWhere(n, rs[i], "assign-pair-func-call", where)
}
if n.List.Len() != len(rs) {
Fatalf("esc oas2func")
}
case ORETURN:
retList := n.List
if retList.Len() == 1 && Curfn.Type.NumResults() > 1 {
// OAS2FUNC in disguise
// esccall already done on n.List.First()
// tie e.nodeEscState(n.List.First()).Retval to Curfn.Func.Dcl PPARAMOUT's
retList = e.nodeEscState(n.List.First()).Retval
}
i := 0
for _, lrn := range Curfn.Func.Dcl {
if i >= retList.Len() {
break
}
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
if lrn.Op != ONAME || lrn.Class() != PPARAMOUT {
continue
}
e.escassignWhyWhere(lrn, retList.Index(i), "return", n)
i++
}
if i < retList.Len() {
Fatalf("esc return list")
}
// Argument could leak through recover.
case OPANIC:
e.escassignSinkWhy(n, n.Left, "panic")
case OAPPEND:
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if !n.IsDDD() {
for _, nn := range n.List.Slice()[1:] {
e.escassignSinkWhy(n, nn, "appended to slice") // lose track of assign to dereference
}
} else {
// append(slice1, slice2...) -- slice2 itself does not escape, but contents do.
slice2 := n.List.Second()
e.escassignDereference(&e.theSink, slice2, e.stepAssignWhere(n, slice2, "appended slice...", n)) // lose track of assign of dereference
if Debug['m'] > 3 {
Warnl(n.Pos, "%v special treatment of append(slice1, slice2...) %S", e.curfnSym(n), n)
}
}
e.escassignDereference(&e.theSink, n.List.First(), e.stepAssignWhere(n, n.List.First(), "appendee slice", n)) // The original elements are now leaked, too
case OCOPY:
e.escassignDereference(&e.theSink, n.Right, e.stepAssignWhere(n, n.Right, "copied slice", n)) // lose track of assign of dereference
case OCONV, OCONVNOP:
e.escassignWhyWhere(n, n.Left, "converted", n)
case OCONVIFACE:
e.track(n)
e.escassignWhyWhere(n, n.Left, "interface-converted", n)
case OARRAYLIT:
// Link values to array
for _, elt := range n.List.Slice() {
if elt.Op == OKEY {
elt = elt.Right
cmd/compile: stop adding implicit OKEY nodes Keys are uncommon in array and slice literals, and normalizing OARRAYLIT and OSLICELIT nodes to always use OKEY ends up not reducing complexity much. Instead, only create OKEY nodes to represent explicit keys, and recalculate implicit keys when/where necessary. Fixes #15350. name old time/op new time/op delta Template 299ms ± 9% 299ms ±12% ~ (p=0.694 n=28+30) Unicode 165ms ± 7% 162ms ± 9% ~ (p=0.084 n=27+27) GoTypes 950ms ± 9% 963ms ± 5% ~ (p=0.301 n=30+29) Compiler 4.23s ± 7% 4.17s ± 7% ~ (p=0.057 n=29+27) name old user-ns/op new user-ns/op delta Template 389M ±15% 400M ±12% ~ (p=0.202 n=30+29) Unicode 246M ±21% 232M ±22% -5.76% (p=0.006 n=28+29) GoTypes 1.34G ± 8% 1.34G ± 7% ~ (p=0.775 n=28+30) Compiler 5.91G ± 6% 5.87G ± 7% ~ (p=0.298 n=28+29) name old alloc/op new alloc/op delta Template 41.2MB ± 0% 41.2MB ± 0% ~ (p=0.085 n=30+30) Unicode 34.0MB ± 0% 31.5MB ± 0% -7.28% (p=0.000 n=30+29) GoTypes 121MB ± 0% 121MB ± 0% ~ (p=0.657 n=30+30) Compiler 511MB ± 0% 511MB ± 0% -0.01% (p=0.001 n=29+29) name old allocs/op new allocs/op delta Template 390k ± 0% 390k ± 0% ~ (p=0.225 n=30+29) Unicode 318k ± 0% 293k ± 0% -8.03% (p=0.000 n=30+29) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.745 n=30+30) Compiler 4.35M ± 0% 4.35M ± 0% ~ (p=0.105 n=30+30) Change-Id: I6310739a0bfdb54f1ab8a460b2c03615ad1ff5bc Reviewed-on: https://go-review.googlesource.com/32221 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-27 02:02:30 -07:00
}
e.escassign(n, elt, e.stepAssignWhere(n, elt, "array literal element", n))
}
case OSLICELIT:
// Slice is not leaked until proven otherwise
e.track(n)
// Link values to slice
for _, elt := range n.List.Slice() {
if elt.Op == OKEY {
elt = elt.Right
cmd/compile: stop adding implicit OKEY nodes Keys are uncommon in array and slice literals, and normalizing OARRAYLIT and OSLICELIT nodes to always use OKEY ends up not reducing complexity much. Instead, only create OKEY nodes to represent explicit keys, and recalculate implicit keys when/where necessary. Fixes #15350. name old time/op new time/op delta Template 299ms ± 9% 299ms ±12% ~ (p=0.694 n=28+30) Unicode 165ms ± 7% 162ms ± 9% ~ (p=0.084 n=27+27) GoTypes 950ms ± 9% 963ms ± 5% ~ (p=0.301 n=30+29) Compiler 4.23s ± 7% 4.17s ± 7% ~ (p=0.057 n=29+27) name old user-ns/op new user-ns/op delta Template 389M ±15% 400M ±12% ~ (p=0.202 n=30+29) Unicode 246M ±21% 232M ±22% -5.76% (p=0.006 n=28+29) GoTypes 1.34G ± 8% 1.34G ± 7% ~ (p=0.775 n=28+30) Compiler 5.91G ± 6% 5.87G ± 7% ~ (p=0.298 n=28+29) name old alloc/op new alloc/op delta Template 41.2MB ± 0% 41.2MB ± 0% ~ (p=0.085 n=30+30) Unicode 34.0MB ± 0% 31.5MB ± 0% -7.28% (p=0.000 n=30+29) GoTypes 121MB ± 0% 121MB ± 0% ~ (p=0.657 n=30+30) Compiler 511MB ± 0% 511MB ± 0% -0.01% (p=0.001 n=29+29) name old allocs/op new allocs/op delta Template 390k ± 0% 390k ± 0% ~ (p=0.225 n=30+29) Unicode 318k ± 0% 293k ± 0% -8.03% (p=0.000 n=30+29) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.745 n=30+30) Compiler 4.35M ± 0% 4.35M ± 0% ~ (p=0.105 n=30+30) Change-Id: I6310739a0bfdb54f1ab8a460b2c03615ad1ff5bc Reviewed-on: https://go-review.googlesource.com/32221 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-27 02:02:30 -07:00
}
e.escassign(n, elt, e.stepAssignWhere(n, elt, "slice literal element", n))
}
// Link values to struct.
case OSTRUCTLIT:
for _, elt := range n.List.Slice() {
e.escassignWhyWhere(n, elt.Left, "struct literal element", n)
}
case OPTRLIT:
e.track(n)
// Link OSTRUCTLIT to OPTRLIT; if OPTRLIT escapes, OSTRUCTLIT elements do too.
e.escassignWhyWhere(n, n.Left, "pointer literal [assign]", n)
case OCALLPART:
e.track(n)
// Contents make it to memory, lose track.
e.escassignSinkWhy(n, n.Left, "call part")
case OMAPLIT:
e.track(n)
// Keys and values make it to memory, lose track.
for _, elt := range n.List.Slice() {
e.escassignSinkWhy(n, elt.Left, "map literal key")
e.escassignSinkWhy(n, elt.Right, "map literal value")
}
case OCLOSURE:
// Link addresses of captured variables to closure.
for _, v := range n.Func.Closure.Func.Cvars.Slice() {
if v.Op == OXXX { // unnamed out argument; see dcl.go:/^funcargs
continue
}
a := v.Name.Defn
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if !v.Name.Byval() {
a = nod(OADDR, a, nil)
a.Pos = v.Pos
e.nodeEscState(a).Loopdepth = e.loopdepth
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
a = typecheck(a, ctxExpr)
}
e.escassignWhyWhere(n, a, "captured by a closure", n)
}
fallthrough
case OMAKECHAN,
OMAKEMAP,
OMAKESLICE,
ONEW,
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ORUNES2STR,
OBYTES2STR,
OSTR2RUNES,
OSTR2BYTES,
ORUNESTR:
e.track(n)
case OADDSTR:
e.track(n)
// Arguments of OADDSTR do not escape.
case OADDR:
// current loop depth is an upper bound on actual loop depth
// of addressed value.
e.track(n)
// for &x, use loop depth of x if known.
// it should always be known, but if not, be conservative
// and keep the current loop depth.
if n.Left.Op == ONAME {
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
switch n.Left.Class() {
// PPARAM is loop depth 1 always.
// PPARAMOUT is loop depth 0 for writes
// but considered loop depth 1 for address-of,
// so that writing the address of one result
// to another (or the same) result makes the
// first result move to the heap.
case PPARAM, PPARAMOUT:
nE := e.nodeEscState(n)
nE.Loopdepth = 1
break opSwitch
}
}
nE := e.nodeEscState(n)
leftE := e.nodeEscState(n.Left)
if leftE.Loopdepth != 0 {
nE.Loopdepth = leftE.Loopdepth
}
case ODOT,
ODOTPTR,
OINDEX:
// Propagate the loopdepth of t to t.field.
if n.Left.Op != OLITERAL { // OLITERAL node doesn't have esc state
e.nodeEscState(n).Loopdepth = e.nodeEscState(n.Left).Loopdepth
}
}
lineno = lno
}
// escassignWhyWhere bundles a common case of
// escassign(e, dst, src, e.stepAssignWhere(dst, src, reason, where))
func (e *EscState) escassignWhyWhere(dst, src *Node, reason string, where *Node) {
var step *EscStep
if Debug['m'] != 0 {
step = e.stepAssignWhere(dst, src, reason, where)
}
e.escassign(dst, src, step)
}
// escassignSinkWhy bundles a common case of
// escassign(e, &e.theSink, src, e.stepAssign(nil, dst, src, reason))
func (e *EscState) escassignSinkWhy(dst, src *Node, reason string) {
var step *EscStep
if Debug['m'] != 0 {
step = e.stepAssign(nil, dst, src, reason)
}
e.escassign(&e.theSink, src, step)
}
// escassignSinkWhyWhere is escassignSinkWhy but includes a call site
// for accurate location reporting.
func (e *EscState) escassignSinkWhyWhere(dst, src *Node, reason string, call *Node) {
var step *EscStep
if Debug['m'] != 0 {
step = e.stepAssignWhere(dst, src, reason, call)
}
e.escassign(&e.theSink, src, step)
}
// Assert that expr somehow gets assigned to dst, if non nil. for
// dst==nil, any name node expr still must be marked as being
// evaluated in curfn. For expr==nil, dst must still be examined for
// evaluations inside it (e.g *f(x) = y)
func (e *EscState) escassign(dst, src *Node, step *EscStep) {
if dst.isBlank() || dst == nil || src == nil || src.Op == ONONAME || src.Op == OXXX {
return
}
if Debug['m'] > 2 {
fmt.Printf("%v:[%d] %v escassign: %S(%0j)[%v] = %S(%0j)[%v]\n",
linestr(lineno), e.loopdepth, funcSym(Curfn),
dst, dst, dst.Op,
src, src, src.Op)
}
setlineno(dst)
originalDst := dst
dstwhy := "assigned"
// Analyze lhs of assignment.
// Replace dst with &e.theSink if we can't track it.
switch dst.Op {
default:
Dump("dst", dst)
Fatalf("escassign: unexpected dst")
case OARRAYLIT,
OSLICELIT,
OCLOSURE,
OCONV,
OCONVIFACE,
OCONVNOP,
OMAPLIT,
OSTRUCTLIT,
OPTRLIT,
ODDDARG,
OCALLPART:
case ONAME:
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
if dst.Class() == PEXTERN {
dstwhy = "assigned to top level variable"
dst = &e.theSink
}
case ODOT: // treat "dst.x = src" as "dst = src"
e.escassign(dst.Left, src, e.stepAssign(step, originalDst, src, "dot-equals"))
return
case OINDEX:
if dst.Left.Type.IsArray() {
e.escassign(dst.Left, src, e.stepAssign(step, originalDst, src, "array-element-equals"))
return
}
dstwhy = "slice-element-equals"
dst = &e.theSink // lose track of dereference
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
case ODEREF:
dstwhy = "star-equals"
dst = &e.theSink // lose track of dereference
case ODOTPTR:
dstwhy = "star-dot-equals"
dst = &e.theSink // lose track of dereference
// lose track of key and value
case OINDEXMAP:
e.escassign(&e.theSink, dst.Right, e.stepAssign(nil, originalDst, src, "key of map put"))
dstwhy = "value of map put"
dst = &e.theSink
}
lno := setlineno(src)
e.pdepth++
switch src.Op {
case OADDR, // dst = &x
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ODEREF, // dst = *x
ODOTPTR, // dst = (*x).f
ONAME,
ODDDARG,
OPTRLIT,
OARRAYLIT,
OSLICELIT,
OMAPLIT,
OSTRUCTLIT,
OMAKECHAN,
OMAKEMAP,
OMAKESLICE,
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ORUNES2STR,
OBYTES2STR,
OSTR2RUNES,
OSTR2BYTES,
OADDSTR,
ONEW,
OCALLPART,
ORUNESTR,
OCONVIFACE:
e.escflows(dst, src, e.stepAssign(step, originalDst, src, dstwhy))
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 18:17:20 +03:00
case OCLOSURE:
// OCLOSURE is lowered to OPTRLIT,
// insert OADDR to account for the additional indirection.
a := nod(OADDR, src, nil)
a.Pos = src.Pos
e.nodeEscState(a).Loopdepth = e.nodeEscState(src).Loopdepth
a.Type = types.NewPtr(src.Type)
e.escflows(dst, a, e.stepAssign(nil, originalDst, src, dstwhy))
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 18:17:20 +03:00
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Flowing multiple returns to a single dst happens when
// analyzing "go f(g())": here g() flows to sink (issue 4529).
case OCALLMETH, OCALLFUNC, OCALLINTER:
for _, n := range e.nodeEscState(src).Retval.Slice() {
e.escflows(dst, n, e.stepAssign(nil, originalDst, n, dstwhy))
}
// A non-pointer escaping from a struct does not concern us.
case ODOT:
if src.Type != nil && !types.Haspointers(src.Type) {
break
}
fallthrough
// Conversions, field access, slice all preserve the input value.
case OCONV,
OCONVNOP,
ODOTMETH,
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
// treat recv.meth as a value with recv in it, only happens in ODEFER and OGO
// iface.method already leaks iface in esccall, no need to put in extra ODOTINTER edge here
OSLICE,
OSLICE3,
OSLICEARR,
OSLICE3ARR,
OSLICESTR:
// Conversions, field access, slice all preserve the input value.
e.escassign(dst, src.Left, e.stepAssign(step, originalDst, src, dstwhy))
case ODOTTYPE,
ODOTTYPE2:
if src.Type != nil && !types.Haspointers(src.Type) {
break
}
e.escassign(dst, src.Left, e.stepAssign(step, originalDst, src, dstwhy))
case OAPPEND:
// Append returns first argument.
// Subsequent arguments are already leaked because they are operands to append.
e.escassign(dst, src.List.First(), e.stepAssign(step, dst, src.List.First(), dstwhy))
case OINDEX:
// Index of array preserves input value.
if src.Left.Type.IsArray() {
e.escassign(dst, src.Left, e.stepAssign(step, originalDst, src, dstwhy))
} else {
e.escflows(dst, src, e.stepAssign(step, originalDst, src, dstwhy))
}
// Might be pointer arithmetic, in which case
// the operands flow into the result.
// TODO(rsc): Decide what the story is here. This is unsettling.
case OADD,
OSUB,
OOR,
OXOR,
OMUL,
ODIV,
OMOD,
OLSH,
ORSH,
OAND,
OANDNOT,
OPLUS,
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ONEG,
OBITNOT:
e.escassign(dst, src.Left, e.stepAssign(step, originalDst, src, dstwhy))
e.escassign(dst, src.Right, e.stepAssign(step, originalDst, src, dstwhy))
}
e.pdepth--
lineno = lno
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Common case for escapes is 16 bits 000000000xxxEEEE
// where commonest cases for xxx encoding in-to-out pointer
// flow are 000, 001, 010, 011 and EEEE is computed Esc bits.
// Note width of xxx depends on value of constant
// bitsPerOutputInTag -- expect 2 or 3, so in practice the
// tag cache array is 64 or 128 long. Some entries will
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// never be populated.
var tags [1 << (bitsPerOutputInTag + EscReturnBits)]string
// mktag returns the string representation for an escape analysis tag.
func mktag(mask int) string {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
switch mask & EscMask {
case EscNone, EscReturn:
default:
Fatalf("escape mktag")
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
if mask < len(tags) && tags[mask] != "" {
return tags[mask]
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
s := fmt.Sprintf("esc:0x%x", mask)
if mask < len(tags) {
tags[mask] = s
}
return s
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
// parsetag decodes an escape analysis tag and returns the esc value.
func parsetag(note string) uint16 {
if !strings.HasPrefix(note, "esc:") {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
return EscUnknown
}
n, _ := strconv.ParseInt(note[4:], 0, 0)
em := uint16(n)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
if em == 0 {
return EscNone
}
return em
}
// describeEscape returns a string describing the escape tag.
// The result is either one of {EscUnknown, EscNone, EscHeap} which all have no further annotation
// or a description of parameter flow, which takes the form of an optional "contentToHeap"
// indicating that the content of this parameter is leaked to the heap, followed by a sequence
// of level encodings separated by spaces, one for each parameter, where _ means no flow,
// = means direct flow, and N asterisks (*) encodes content (obtained by indirection) flow.
// e.g., "contentToHeap _ =" means that a parameter's content (one or more dereferences)
// escapes to the heap, the parameter does not leak to the first output, but does leak directly
// to the second output (and if there are more than two outputs, there is no flow to those.)
func describeEscape(em uint16) string {
var s string
switch em & EscMask {
case EscUnknown:
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
s = "EscUnknown"
case EscNone:
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
s = "EscNone"
case EscHeap:
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
s = "EscHeap"
case EscReturn:
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
s = "EscReturn"
}
if em&EscContentEscapes != 0 {
if s != "" {
s += " "
}
s += "contentToHeap"
}
for em >>= EscReturnBits; em != 0; em >>= bitsPerOutputInTag {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// See encoding description above
if s != "" {
s += " "
}
switch embits := em & bitsMaskForTag; embits {
case 0:
s += "_"
case 1:
s += "="
default:
for i := uint16(0); i < embits-1; i++ {
s += "*"
}
}
}
return s
}
// escassignfromtag models the input-to-output assignment flow of one of a function
// calls arguments, where the flow is encoded in "note".
func (e *EscState) escassignfromtag(note string, dsts Nodes, src, call *Node) uint16 {
em := parsetag(note)
if src.Op == OLITERAL {
return em
}
if Debug['m'] > 3 {
fmt.Printf("%v::assignfromtag:: src=%S, em=%s\n",
linestr(lineno), src, describeEscape(em))
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
if em == EscUnknown {
e.escassignSinkWhyWhere(src, src, "passed to call[argument escapes]", call)
return em
}
if em == EscNone {
return em
}
// If content inside parameter (reached via indirection)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// escapes to heap, mark as such.
if em&EscContentEscapes != 0 {
e.escassign(&e.theSink, e.addDereference(src), e.stepAssignWhere(src, src, "passed to call[argument content escapes]", call))
}
em0 := em
dstsi := 0
for em >>= EscReturnBits; em != 0 && dstsi < dsts.Len(); em >>= bitsPerOutputInTag {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Prefer the lowest-level path to the reference (for escape purposes).
// Two-bit encoding (for example. 1, 3, and 4 bits are other options)
// 01 = 0-level
// 10 = 1-level, (content escapes),
// 11 = 2-level, (content of content escapes),
embits := em & bitsMaskForTag
if embits > 0 {
n := src
for i := uint16(0); i < embits-1; i++ {
n = e.addDereference(n) // encode level>0 as indirections
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
e.escassign(dsts.Index(dstsi), n, e.stepAssignWhere(dsts.Index(dstsi), src, "passed-to-and-returned-from-call", call))
}
dstsi++
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// If there are too many outputs to fit in the tag,
// that is handled at the encoding end as EscHeap,
// so there is no need to check here.
if em != 0 && dstsi >= dsts.Len() {
Fatalf("corrupt esc tag %q or messed up escretval list\n", note)
}
return em0
}
func (e *EscState) escassignDereference(dst *Node, src *Node, step *EscStep) {
if src.Op == OLITERAL {
return
}
e.escassign(dst, e.addDereference(src), step)
}
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
// addDereference constructs a suitable ODEREF note applied to src.
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Because this is for purposes of escape accounting, not execution,
// some semantically dubious node combinations are (currently) possible.
func (e *EscState) addDereference(n *Node) *Node {
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ind := nod(ODEREF, n, nil)
e.nodeEscState(ind).Loopdepth = e.nodeEscState(n).Loopdepth
ind.Pos = n.Pos
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
t := n.Type
if t.IsPtr() || t.IsSlice() {
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
// This should model our own sloppy use of ODEREF to encode
// decreasing levels of indirection; i.e., "indirecting" a slice
// yields the type of an element.
t = t.Elem()
} else if t.IsString() {
t = types.Types[TUINT8]
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
ind.Type = t
return ind
}
// escNoteOutputParamFlow encodes maxEncodedLevel/.../1/0-level flow to the vargen'th parameter.
// Levels greater than maxEncodedLevel are replaced with maxEncodedLevel.
// If the encoding cannot describe the modified input level and output number, then EscHeap is returned.
func escNoteOutputParamFlow(e uint16, vargen int32, level Level) uint16 {
// Flow+level is encoded in two bits.
// 00 = not flow, xx = level+1 for 0 <= level <= maxEncodedLevel
// 16 bits for Esc allows 6x2bits or 4x3bits or 3x4bits if additional information would be useful.
if level.int() <= 0 && level.guaranteedDereference() > 0 {
return escMax(e|EscContentEscapes, EscNone) // At least one deref, thus only content.
}
if level.int() < 0 {
return EscHeap
}
if level.int() > maxEncodedLevel {
// Cannot encode larger values than maxEncodedLevel.
level = levelFrom(maxEncodedLevel)
}
encoded := uint16(level.int() + 1)
shift := uint(bitsPerOutputInTag*(vargen-1) + EscReturnBits)
old := (e >> shift) & bitsMaskForTag
if old == 0 || encoded != 0 && encoded < old {
old = encoded
}
encodedFlow := old << shift
if (encodedFlow>>shift)&bitsMaskForTag != old {
// Encoding failure defaults to heap.
return EscHeap
}
return (e &^ (bitsMaskForTag << shift)) | encodedFlow
}
func (e *EscState) initEscRetval(call *Node, fntype *types.Type) {
cE := e.nodeEscState(call)
cE.Retval.Set(nil) // Suspect this is not nil for indirect calls.
for i, f := range fntype.Results().Fields().Slice() {
buf := fmt.Sprintf(".out%d", i)
ret := newname(lookup(buf))
ret.SetAddable(false) // TODO(mdempsky): Seems suspicious.
ret.Type = f.Type
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
ret.SetClass(PAUTO)
ret.Name.Curfn = Curfn
e.nodeEscState(ret).Loopdepth = e.loopdepth
ret.Name.SetUsed(true)
ret.Pos = call.Pos
cE.Retval.Append(ret)
}
}
// This is a bit messier than fortunate, pulled out of esc's big
// switch for clarity. We either have the paramnodes, which may be
// connected to other things through flows or we have the parameter type
// nodes, which may be marked "noescape". Navigating the ast is slightly
// different for methods vs plain functions and for imported vs
// this-package
func (e *EscState) esccall(call *Node, parent *Node) {
var fntype *types.Type
var indirect bool
var fn *Node
switch call.Op {
default:
Fatalf("esccall")
case OCALLFUNC:
fn = call.Left
fntype = fn.Type
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
indirect = fn.Op != ONAME || fn.Class() != PFUNC
case OCALLMETH:
fn = asNode(call.Left.Sym.Def)
if fn != nil {
fntype = fn.Type
} else {
fntype = call.Left.Type
}
case OCALLINTER:
fntype = call.Left.Type
indirect = true
}
argList := call.List
args := argList.Slice()
if indirect {
// We know nothing!
// Leak all the parameters
for _, arg := range args {
e.escassignSinkWhy(call, arg, "parameter to indirect call")
if Debug['m'] > 3 {
fmt.Printf("%v::esccall:: indirect call <- %S, untracked\n", linestr(lineno), arg)
}
}
// Set up bogus outputs
e.initEscRetval(call, fntype)
// If there is a receiver, it also leaks to heap.
if call.Op != OCALLFUNC {
rf := fntype.Recv()
r := call.Left.Left
if types.Haspointers(rf.Type) {
e.escassignSinkWhy(call, r, "receiver in indirect call")
}
} else { // indirect and OCALLFUNC = could be captured variables, too. (#14409)
rets := e.nodeEscState(call).Retval.Slice()
for _, ret := range rets {
e.escassignDereference(ret, fn, e.stepAssignWhere(ret, fn, "captured by called closure", call))
}
}
return
}
cE := e.nodeEscState(call)
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
if fn != nil && fn.Op == ONAME && fn.Class() == PFUNC &&
fn.Name.Defn != nil && fn.Name.Defn.Nbody.Len() != 0 && fn.Name.Param.Ntype != nil && fn.Name.Defn.Esc < EscFuncTagged {
// function in same mutually recursive group. Incorporate into flow graph.
if Debug['m'] > 3 {
fmt.Printf("%v::esccall:: %S in recursive group\n", linestr(lineno), call)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
if fn.Name.Defn.Esc == EscFuncUnknown || cE.Retval.Len() != 0 {
Fatalf("graph inconsistency")
}
i := 0
// Receiver.
if call.Op != OCALLFUNC {
rf := fntype.Recv()
if rf.Sym != nil && !rf.Sym.IsBlank() {
n := fn.Name.Defn.Func.Dcl[0]
i++
if n.Class() != PPARAM {
Fatalf("esccall: not a parameter %+v", n)
}
e.escassignWhyWhere(n, call.Left.Left, "recursive call receiver", call)
}
}
// Parameters.
for _, param := range fntype.Params().FieldSlice() {
if param.Sym == nil || param.Sym.IsBlank() {
// Unnamed parameter is not listed in Func.Dcl.
// But we need to consume the arg.
if param.IsDDD() && !call.IsDDD() {
args = nil
} else {
args = args[1:]
}
continue
}
n := fn.Name.Defn.Func.Dcl[i]
i++
if n.Class() != PPARAM {
Fatalf("esccall: not a parameter %+v", n)
}
if len(args) == 0 {
continue
}
arg := args[0]
if n.IsDDD() && !call.IsDDD() {
// Introduce ODDDARG node to represent ... allocation.
arg = nod(ODDDARG, nil, nil)
arr := types.NewArray(n.Type.Elem(), int64(len(args)))
arg.Type = types.NewPtr(arr) // make pointer so it will be tracked
arg.Pos = call.Pos
e.track(arg)
call.Right = arg
}
e.escassignWhyWhere(n, arg, "arg to recursive call", call) // TODO this message needs help.
if arg == args[0] {
args = args[1:]
continue
}
// "..." arguments are untracked
for _, a := range args {
if Debug['m'] > 3 {
fmt.Printf("%v::esccall:: ... <- %S, untracked\n", linestr(lineno), a)
}
e.escassignSinkWhyWhere(arg, a, "... arg to recursive call", call)
}
// ... arg consumes all remaining arguments
args = nil
}
// Results.
for _, n := range fn.Name.Defn.Func.Dcl[i:] {
if n.Class() == PPARAMOUT {
cE.Retval.Append(n)
}
}
// Sanity check: all arguments must be consumed.
if len(args) != 0 {
Fatalf("esccall not consumed all args %+v\n", call)
}
return
}
// Imported or completely analyzed function. Use the escape tags.
if cE.Retval.Len() != 0 {
Fatalf("esc already decorated call %+v\n", call)
}
if Debug['m'] > 3 {
fmt.Printf("%v::esccall:: %S not recursive\n", linestr(lineno), call)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
// set up out list on this call node with dummy auto ONAMES in the current (calling) function.
e.initEscRetval(call, fntype)
// Receiver.
if call.Op != OCALLFUNC {
rf := fntype.Recv()
r := call.Left.Left
if types.Haspointers(rf.Type) {
e.escassignfromtag(rf.Note, cE.Retval, r, call)
}
}
for i, param := range fntype.Params().FieldSlice() {
note := param.Note
var arg *Node
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if param.IsDDD() && !call.IsDDD() {
rest := args[i:]
if len(rest) == 0 {
break
}
// Introduce ODDDARG node to represent ... allocation.
arg = nod(ODDDARG, nil, nil)
arg.Pos = call.Pos
arr := types.NewArray(param.Type.Elem(), int64(len(rest)))
arg.Type = types.NewPtr(arr) // make pointer so it will be tracked
e.track(arg)
call.Right = arg
// Store arguments into slice for ... arg.
for _, a := range rest {
if Debug['m'] > 3 {
fmt.Printf("%v::esccall:: ... <- %S\n", linestr(lineno), a)
}
if note == uintptrEscapesTag {
e.escassignSinkWhyWhere(arg, a, "arg to uintptrescapes ...", call)
} else {
e.escassignWhyWhere(arg, a, "arg to ...", call)
}
}
} else {
arg = args[i]
if note == uintptrEscapesTag {
e.escassignSinkWhy(arg, arg, "escaping uintptr")
}
}
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if types.Haspointers(param.Type) && e.escassignfromtag(note, cE.Retval, arg, call)&EscMask == EscNone && parent.Op != ODEFER && parent.Op != OGO {
a := arg
for a.Op == OCONVNOP {
a = a.Left
}
switch a.Op {
// The callee has already been analyzed, so its arguments have esc tags.
// The argument is marked as not escaping at all.
// Record that fact so that any temporary used for
// synthesizing this expression can be reclaimed when
// the function returns.
// This 'noescape' is even stronger than the usual esc == EscNone.
// arg.Esc == EscNone means that arg does not escape the current function.
// arg.SetNoescape(true) here means that arg does not escape this statement
// in the current function.
case OCALLPART, OCLOSURE, ODDDARG, OARRAYLIT, OSLICELIT, OPTRLIT, OSTRUCTLIT:
a.SetNoescape(true)
}
}
}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// escflows records the link src->dst in dst, throwing out some quick wins,
// and also ensuring that dst is noted as a flow destination.
func (e *EscState) escflows(dst, src *Node, why *EscStep) {
if dst == nil || src == nil || dst == src {
return
}
// Don't bother building a graph for scalars.
if src.Type != nil && !types.Haspointers(src.Type) && !isReflectHeaderDataField(src) {
if Debug['m'] > 3 {
fmt.Printf("%v::NOT flows:: %S <- %S\n", linestr(lineno), dst, src)
}
return
}
if Debug['m'] > 3 {
fmt.Printf("%v::flows:: %S <- %S\n", linestr(lineno), dst, src)
}
dstE := e.nodeEscState(dst)
if len(dstE.Flowsrc) == 0 {
e.dsts = append(e.dsts, dst)
e.dstcount++
}
e.edgecount++
if why == nil {
dstE.Flowsrc = append(dstE.Flowsrc, EscStep{src: src})
} else {
starwhy := *why
starwhy.src = src // TODO: need to reconcile this w/ needs of explanations.
dstE.Flowsrc = append(dstE.Flowsrc, starwhy)
}
}
// Whenever we hit a reference node, the level goes up by one, and whenever
// we hit an OADDR, the level goes down by one. as long as we're on a level > 0
// finding an OADDR just means we're following the upstream of a dereference,
// so this address doesn't leak (yet).
// If level == 0, it means the /value/ of this node can reach the root of this flood.
// so if this node is an OADDR, its argument should be marked as escaping iff
// its currfn/e.loopdepth are different from the flood's root.
// Once an object has been moved to the heap, all of its upstream should be considered
// escaping to the global scope.
func (e *EscState) escflood(dst *Node) {
switch dst.Op {
case ONAME, OCLOSURE:
default:
return
}
dstE := e.nodeEscState(dst)
if Debug['m'] > 2 {
fmt.Printf("\nescflood:%d: dst %S scope:%v[%d]\n", e.walkgen, dst, e.curfnSym(dst), dstE.Loopdepth)
}
for i := range dstE.Flowsrc {
e.walkgen++
s := &dstE.Flowsrc[i]
s.parent = nil
e.escwalk(levelFrom(0), dst, s.src, s)
}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// funcOutputAndInput reports whether dst and src correspond to output and input parameters of the same function.
func funcOutputAndInput(dst, src *Node) bool {
// Note if dst is marked as escaping, then "returned" is too weak.
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
return dst.Op == ONAME && dst.Class() == PPARAMOUT &&
src.Op == ONAME && src.Class() == PPARAM && src.Name.Curfn == dst.Name.Curfn
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
func (es *EscStep) describe(src *Node) {
if Debug['m'] < 2 {
return
}
step0 := es
for step := step0; step != nil && !step.busy; step = step.parent {
// TODO: We get cycles. Trigger is i = &i (where var i interface{})
step.busy = true
// The trail is a little odd because of how the
// graph is constructed. The link to the current
// Node is parent.src unless parent is nil in which
// case it is step.dst.
nextDest := step.parent
dst := step.dst
where := step.where
if nextDest != nil {
dst = nextDest.src
}
if where == nil {
where = dst
}
Warnl(src.Pos, "\tfrom %v (%s) at %s", dst, step.why, where.Line())
}
for step := step0; step != nil && step.busy; step = step.parent {
step.busy = false
}
}
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
const NOTALOOPDEPTH = -1
func (e *EscState) escwalk(level Level, dst *Node, src *Node, step *EscStep) {
e.escwalkBody(level, dst, src, step, NOTALOOPDEPTH)
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
}
func (e *EscState) escwalkBody(level Level, dst *Node, src *Node, step *EscStep, extraloopdepth int32) {
if src.Op == OLITERAL {
return
}
srcE := e.nodeEscState(src)
if srcE.Walkgen == e.walkgen {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// Esclevels are vectors, do not compare as integers,
// and must use "min" of old and new to guarantee
// convergence.
level = level.min(srcE.Level)
if level == srcE.Level {
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
// Have we been here already with an extraloopdepth,
// or is the extraloopdepth provided no improvement on
// what's already been seen?
if srcE.Maxextraloopdepth >= extraloopdepth || srcE.Loopdepth >= extraloopdepth {
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
return
}
srcE.Maxextraloopdepth = extraloopdepth
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
} else { // srcE.Walkgen < e.walkgen -- first time, reset this.
srcE.Maxextraloopdepth = NOTALOOPDEPTH
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
srcE.Walkgen = e.walkgen
srcE.Level = level
modSrcLoopdepth := srcE.Loopdepth
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
if extraloopdepth > modSrcLoopdepth {
modSrcLoopdepth = extraloopdepth
}
if Debug['m'] > 2 {
fmt.Printf("escwalk: level:%d depth:%d %.*s op=%v %S(%0j) scope:%v[%d] extraloopdepth=%v\n",
level, e.pdepth, e.pdepth, "\t\t\t\t\t\t\t\t\t\t", src.Op, src, src, e.curfnSym(src), srcE.Loopdepth, extraloopdepth)
}
e.pdepth++
// Input parameter flowing to output parameter?
var leaks bool
var osrcesc uint16 // used to prevent duplicate error messages
dstE := e.nodeEscState(dst)
if funcOutputAndInput(dst, src) && src.Esc&EscMask < EscHeap && dst.Esc != EscHeap {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// This case handles:
// 1. return in
// 2. return &in
// 3. tmp := in; return &tmp
// 4. return *in
if Debug['m'] != 0 {
if Debug['m'] <= 2 {
Warnl(src.Pos, "leaking param: %S to result %v level=%v", src, dst.Sym, level.int())
step.describe(src)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
} else {
Warnl(src.Pos, "leaking param: %S to result %v level=%v", src, dst.Sym, level)
}
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
if src.Esc&EscMask != EscReturn {
src.Esc = EscReturn | src.Esc&EscContentEscapes
}
src.Esc = escNoteOutputParamFlow(src.Esc, dst.Name.Vargen, level)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
goto recurse
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// If parameter content escapes to heap, set EscContentEscapes
// Note minor confusion around escape from pointer-to-struct vs escape from struct
if dst.Esc == EscHeap &&
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
src.Op == ONAME && src.Class() == PPARAM && src.Esc&EscMask < EscHeap &&
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
level.int() > 0 {
src.Esc = escMax(EscContentEscapes|src.Esc, EscNone)
}
leaks = level.int() <= 0 && level.guaranteedDereference() <= 0 && dstE.Loopdepth < modSrcLoopdepth
leaks = leaks || level.int() <= 0 && dst.Esc&EscMask == EscHeap
osrcesc = src.Esc
switch src.Op {
case ONAME:
cmd/compile: move Node.Class to flags Put it at position zero, since it is fairly hot. This shrinks gc.Node into a smaller size class on 64 bit systems. name old time/op new time/op delta Template 193ms ± 5% 192ms ± 3% ~ (p=0.353 n=94+93) Unicode 86.1ms ± 5% 85.0ms ± 4% -1.23% (p=0.000 n=95+98) GoTypes 546ms ± 3% 544ms ± 4% -0.40% (p=0.007 n=94+97) Compiler 2.56s ± 3% 2.54s ± 3% -0.67% (p=0.000 n=99+97) SSA 5.13s ± 2% 5.10s ± 3% -0.55% (p=0.000 n=94+98) Flate 122ms ± 6% 121ms ± 4% -0.75% (p=0.002 n=97+95) GoParser 144ms ± 5% 144ms ± 4% ~ (p=0.298 n=98+97) Reflect 348ms ± 4% 349ms ± 4% ~ (p=0.350 n=98+97) Tar 105ms ± 5% 104ms ± 5% ~ (p=0.154 n=96+98) XML 200ms ± 5% 198ms ± 4% -0.71% (p=0.015 n=97+98) [Geo mean] 330ms 328ms -0.52% name old user-time/op new user-time/op delta Template 229ms ±11% 224ms ± 7% -2.16% (p=0.001 n=100+87) Unicode 109ms ± 5% 109ms ± 6% ~ (p=0.897 n=96+91) GoTypes 712ms ± 4% 709ms ± 4% ~ (p=0.085 n=96+98) Compiler 3.41s ± 3% 3.36s ± 3% -1.43% (p=0.000 n=98+98) SSA 7.46s ± 3% 7.31s ± 3% -2.02% (p=0.000 n=100+99) Flate 145ms ± 6% 143ms ± 6% -1.11% (p=0.001 n=99+97) GoParser 177ms ± 5% 176ms ± 5% -0.78% (p=0.018 n=95+95) Reflect 432ms ± 7% 435ms ± 9% ~ (p=0.296 n=100+100) Tar 121ms ± 7% 121ms ± 5% ~ (p=0.072 n=100+95) XML 241ms ± 4% 239ms ± 5% ~ (p=0.085 n=97+99) [Geo mean] 413ms 410ms -0.73% name old alloc/op new alloc/op delta Template 38.4MB ± 0% 37.7MB ± 0% -1.85% (p=0.008 n=5+5) Unicode 30.1MB ± 0% 28.8MB ± 0% -4.09% (p=0.008 n=5+5) GoTypes 112MB ± 0% 110MB ± 0% -1.69% (p=0.008 n=5+5) Compiler 470MB ± 0% 461MB ± 0% -1.91% (p=0.008 n=5+5) SSA 1.13GB ± 0% 1.11GB ± 0% -1.70% (p=0.008 n=5+5) Flate 25.0MB ± 0% 24.6MB ± 0% -1.67% (p=0.008 n=5+5) GoParser 31.6MB ± 0% 31.1MB ± 0% -1.66% (p=0.008 n=5+5) Reflect 77.1MB ± 0% 75.8MB ± 0% -1.69% (p=0.008 n=5+5) Tar 26.3MB ± 0% 25.7MB ± 0% -2.06% (p=0.008 n=5+5) XML 41.9MB ± 0% 41.1MB ± 0% -1.93% (p=0.008 n=5+5) [Geo mean] 73.5MB 72.0MB -2.03% name old allocs/op new allocs/op delta Template 383k ± 0% 383k ± 0% ~ (p=0.690 n=5+5) Unicode 343k ± 0% 343k ± 0% ~ (p=0.841 n=5+5) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.310 n=5+5) Compiler 4.43M ± 0% 4.42M ± 0% -0.17% (p=0.008 n=5+5) SSA 9.85M ± 0% 9.85M ± 0% ~ (p=0.310 n=5+5) Flate 236k ± 0% 236k ± 1% ~ (p=0.841 n=5+5) GoParser 320k ± 0% 320k ± 0% ~ (p=0.421 n=5+5) Reflect 988k ± 0% 987k ± 0% ~ (p=0.690 n=5+5) Tar 252k ± 0% 251k ± 0% ~ (p=0.095 n=5+5) XML 399k ± 0% 399k ± 0% ~ (p=1.000 n=5+5) [Geo mean] 741k 740k -0.07% Change-Id: I9e952b58a98e30a12494304db9ce50d0a85e459c Reviewed-on: https://go-review.googlesource.com/41797 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Marvin Stenger <marvin.stenger94@gmail.com>
2017-04-25 18:14:12 -07:00
if src.Class() == PPARAM && (leaks || dstE.Loopdepth < 0) && src.Esc&EscMask < EscHeap {
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
if level.guaranteedDereference() > 0 {
src.Esc = escMax(EscContentEscapes|src.Esc, EscNone)
if Debug['m'] != 0 {
if Debug['m'] <= 2 {
if osrcesc != src.Esc {
Warnl(src.Pos, "leaking param content: %S", src)
step.describe(src)
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
} else {
Warnl(src.Pos, "leaking param content: %S level=%v dst.eld=%v src.eld=%v dst=%S",
src, level, dstE.Loopdepth, modSrcLoopdepth, dst)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
}
} else {
src.Esc = EscHeap
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
if Debug['m'] != 0 {
if Debug['m'] <= 2 {
Warnl(src.Pos, "leaking param: %S", src)
step.describe(src)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
} else {
Warnl(src.Pos, "leaking param: %S level=%v dst.eld=%v src.eld=%v dst=%S",
src, level, dstE.Loopdepth, modSrcLoopdepth, dst)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
}
}
}
// Treat a captured closure variable as equivalent to the
// original variable.
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if src.IsClosureVar() {
e.escwalk(level, dst, src.Name.Defn, e.stepWalk(dst, src.Name.Defn, "closure-var", step))
}
case OPTRLIT, OADDR:
why := "pointer literal"
if src.Op == OADDR {
why = "address-of"
}
if leaks {
src.Esc = EscHeap
if Debug['m'] != 0 && osrcesc != src.Esc && src.Op != OADDR {
cmd/gc: fix escape analysis of closures Fixes #10353 See test/escape2.go:issue10353. Previously new(int) did not escape to heap, and so heap-allcated closure was referencing a stack var. This breaks the invariant that heap must not contain pointers to stack. Look at the following program: package main func main() { foo(new(int)) bar(new(int)) } func foo(x *int) func() { return func() { println(*x) } } // Models what foo effectively does. func bar(x *int) *C { return &C{x} } type C struct { x *int } Without this patch escape analysis works as follows: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:0 depth:1 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x to result ~r1 escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:5: new(int) escapes to heap /tmp/live2.go:4: main new(int) does not escape new(int) does not escape while being captured by the closure. With this patch escape analysis of foo and bar works similarly: $ go build -gcflags="-m -m -m -l" esc.go escflood:1: dst ~r1 scope:foo[0] escwalk: level:0 depth:0 &(func literal)( l(9)) scope:foo[0] escwalk: level:-1 depth:1 func literal( l(9) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:9: func literal escapes to heap escwalk: level:-1 depth:2 x( l(8) class(PPARAM) f(1) esc(no) ld(1)) scope:foo[1] /tmp/live2.go:8: leaking param: x escflood:2: dst ~r1 scope:bar[0] escwalk: level:0 depth:0 &C literal( l(15) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:15: &C literal escapes to heap escwalk: level:-1 depth:1 &C literal( l(15)) scope:bar[0] escwalk: level:-1 depth:2 x( l(14) class(PPARAM) f(1) esc(no) ld(1)) scope:bar[1] /tmp/live2.go:14: leaking param: x /tmp/live2.go:4: new(int) escapes to heap /tmp/live2.go:5: new(int) escapes to heap Change-Id: Ifd14b7ae3fc11820e3b5eb31eb07f35a22ed0932 Reviewed-on: https://go-review.googlesource.com/8408 Reviewed-by: Russ Cox <rsc@golang.org> Run-TryBot: Dmitry Vyukov <dvyukov@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2015-04-06 18:17:20 +03:00
p := src
if p.Left.Op == OCLOSURE {
p = p.Left // merely to satisfy error messages in tests
}
if Debug['m'] > 2 {
Warnl(src.Pos, "%S escapes to heap, level=%v, dst=%v dst.eld=%v, src.eld=%v",
p, level, dst, dstE.Loopdepth, modSrcLoopdepth)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
} else {
Warnl(src.Pos, "%S escapes to heap", p)
step.describe(src)
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
}
cmd/compile: fix liveness computation for heap-escaped parameters The liveness computation of parameters generally was never correct, but forcing all parameters to be live throughout the function covered up that problem. The new SSA back end is too clever: even though it currently keeps the parameter values live throughout the function, it may find optimizations that mean the current values are not written back to the original parameter stack slots immediately or ever (for example if a parameter is set to nil, SSA constant propagation may replace all later uses of the parameter with a constant nil, eliminating the need to write the nil value back to the stack slot), so the liveness code must now track the actual operations on the stack slots, exposing these problems. One small problem in the handling of arguments is that nodarg can return ONAME PPARAM nodes with adjusted offsets, so that there are actually multiple *Node pointers for the same parameter in the instruction stream. This might be possible to correct, but not in this CL. For now, we fix this by using n.Orig instead of n when considering PPARAM and PPARAMOUT nodes. The major problem in the handling of arguments is general confusion in the liveness code about the meaning of PPARAM|PHEAP and PPARAMOUT|PHEAP nodes, especially as contrasted with PAUTO|PHEAP. The difference between these two is that when a local variable "moves" to the heap, it's really just allocated there to start with; in contrast, when an argument moves to the heap, the actual data has to be copied there from the stack at the beginning of the function, and when a result "moves" to the heap the value in the heap has to be copied back to the stack when the function returns This general confusion is also present in the SSA back end. The PHEAP bit worked decently when I first introduced it 7 years ago (!) in 391425ae. The back end did nothing sophisticated, and in particular there was no analysis at all: no escape analysis, no liveness analysis, and certainly no SSA back end. But the complications caused in the various downstream consumers suggest that this should be a detail kept mainly in the front end. This CL therefore eliminates both the PHEAP bit and even the idea of "heap variables" from the back ends. First, it replaces the PPARAM|PHEAP, PPARAMOUT|PHEAP, and PAUTO|PHEAP variable classes with the single PAUTOHEAP, a pseudo-class indicating a variable maintained on the heap and available by indirecting a local variable kept on the stack (a plain PAUTO). Second, walkexpr replaces all references to PAUTOHEAP variables with indirections of the corresponding PAUTO variable. The back ends and the liveness code now just see plain indirected variables. This may actually produce better code, but the real goal here is to eliminate these little-used and somewhat suspect code paths in the back end analyses. The OPARAM node type goes away too. A followup CL will do the same to PPARAMREF. I'm not sure that the back ends (SSA in particular) are handling those right either, and with the framework established in this CL that change is trivial and the result clearly more correct. Fixes #15747. Change-Id: I2770b1ce3cbc93981bfc7166be66a9da12013d74 Reviewed-on: https://go-review.googlesource.com/23393 Reviewed-by: Keith Randall <khr@golang.org> Run-TryBot: Russ Cox <rsc@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-05-25 01:33:24 -04:00
addrescapes(src.Left)
e.escwalkBody(level.dec(), dst, src.Left, e.stepWalk(dst, src.Left, why, step), modSrcLoopdepth)
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
extraloopdepth = modSrcLoopdepth // passes to recursive case, seems likely a no-op
} else {
e.escwalk(level.dec(), dst, src.Left, e.stepWalk(dst, src.Left, why, step))
}
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
case OAPPEND:
e.escwalk(level, dst, src.List.First(), e.stepWalk(dst, src.List.First(), "append-first-arg", step))
case ODDDARG:
if leaks {
src.Esc = EscHeap
if Debug['m'] != 0 && osrcesc != src.Esc {
Warnl(src.Pos, "%S escapes to heap", src)
step.describe(src)
}
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
extraloopdepth = modSrcLoopdepth
}
// similar to a slice arraylit and its args.
level = level.dec()
case OSLICELIT:
for _, elt := range src.List.Slice() {
if elt.Op == OKEY {
elt = elt.Right
cmd/compile: stop adding implicit OKEY nodes Keys are uncommon in array and slice literals, and normalizing OARRAYLIT and OSLICELIT nodes to always use OKEY ends up not reducing complexity much. Instead, only create OKEY nodes to represent explicit keys, and recalculate implicit keys when/where necessary. Fixes #15350. name old time/op new time/op delta Template 299ms ± 9% 299ms ±12% ~ (p=0.694 n=28+30) Unicode 165ms ± 7% 162ms ± 9% ~ (p=0.084 n=27+27) GoTypes 950ms ± 9% 963ms ± 5% ~ (p=0.301 n=30+29) Compiler 4.23s ± 7% 4.17s ± 7% ~ (p=0.057 n=29+27) name old user-ns/op new user-ns/op delta Template 389M ±15% 400M ±12% ~ (p=0.202 n=30+29) Unicode 246M ±21% 232M ±22% -5.76% (p=0.006 n=28+29) GoTypes 1.34G ± 8% 1.34G ± 7% ~ (p=0.775 n=28+30) Compiler 5.91G ± 6% 5.87G ± 7% ~ (p=0.298 n=28+29) name old alloc/op new alloc/op delta Template 41.2MB ± 0% 41.2MB ± 0% ~ (p=0.085 n=30+30) Unicode 34.0MB ± 0% 31.5MB ± 0% -7.28% (p=0.000 n=30+29) GoTypes 121MB ± 0% 121MB ± 0% ~ (p=0.657 n=30+30) Compiler 511MB ± 0% 511MB ± 0% -0.01% (p=0.001 n=29+29) name old allocs/op new allocs/op delta Template 390k ± 0% 390k ± 0% ~ (p=0.225 n=30+29) Unicode 318k ± 0% 293k ± 0% -8.03% (p=0.000 n=30+29) GoTypes 1.16M ± 0% 1.16M ± 0% ~ (p=0.745 n=30+30) Compiler 4.35M ± 0% 4.35M ± 0% ~ (p=0.105 n=30+30) Change-Id: I6310739a0bfdb54f1ab8a460b2c03615ad1ff5bc Reviewed-on: https://go-review.googlesource.com/32221 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2016-10-27 02:02:30 -07:00
}
e.escwalk(level.dec(), dst, elt, e.stepWalk(dst, elt, "slice-literal-element", step))
}
fallthrough
case OMAKECHAN,
OMAKEMAP,
OMAKESLICE,
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
ORUNES2STR,
OBYTES2STR,
OSTR2RUNES,
OSTR2BYTES,
OADDSTR,
OMAPLIT,
ONEW,
OCLOSURE,
OCALLPART,
ORUNESTR,
OCONVIFACE:
if leaks {
src.Esc = EscHeap
if Debug['m'] != 0 && osrcesc != src.Esc {
Warnl(src.Pos, "%S escapes to heap", src)
step.describe(src)
}
cmd/compile: better modeling of escape across loop levels Brief background on "why heap allocate". Things can be forced to the heap for the following reasons: 1) address published, hence lifetime unknown. 2) size unknown/too large, cannot be stack allocated 3) multiplicity unknown/too large, cannot be stack allocated 4) reachable from heap (not necessarily published) The bug here is a case of failing to enforce 4) when an object Y was reachable from a heap allocation X forced because of 3). It was found in the case of a closure allocated within a loop (X) and assigned to a variable outside the loop (multiplicity unknown) where the closure also captured a map (Y) declared outside the loop (reachable from heap). Note the variable declared outside the loop (Y) is not published, has known size, and known multiplicity (one). The only reason for heap allocation is that it was reached from a heap allocated item (X), but because that was not forced by publication, it has to be tracked by loop level, but escape-loop level was not tracked and thus a bug results. The fix is that when a heap allocation is newly discovered, use its looplevel as the minimum loop level for downstream escape flooding. Every attempt to generalize this bug to X-in-loop- references-Y-outside loop succeeded, so the fix was aimed to be general. Anywhere that loop level forces heap allocation, the loop level is tracked. This is not yet tested for all possible X and Y, but it is correctness- conservative and because it caused only one trivial regression in the escape tests, it is probably also performance-conservative. The new test checks the following: 1) in the map case, that if fn escapes, so does the map. 2) in the map case, if fn does not escape, neither does the map. 3) in the &x case, that if fn escapes, so does &x. 4) in the &x case, if fn does not escape, neither does &x. Fixes #13799. Change-Id: Ie280bef2bb86ec869c7c206789d0b68f080c3fdb Reviewed-on: https://go-review.googlesource.com/18234 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2016-01-04 16:44:20 -05:00
extraloopdepth = modSrcLoopdepth
cmd/compile: flow interface data to heap if CONVIFACE of a non-direct interface escapes Consider the following code: func f(x []*T) interface{} { return x } It returns an interface that holds a heap copy of x (by calling convT2I or friend), therefore x escape to heap. The current escape analysis only recognizes that x flows to the result. This is not sufficient, since if the result does not escape, x's content may be stack allocated and this will result a heap-to-stack pointer, which is bad. Fix this by realizing that if a CONVIFACE escapes and we're converting from a non-direct interface type, the data needs to escape to heap. Running "toolstash -cmp" on std & cmd, the generated machine code are identical for all packages. However, the export data (escape tags) differ in the following packages. It looks to me that all are similar to the "f" above, where the parameter should escape to heap. io/ioutil/ioutil.go:118 old: leaking param: r to result ~r1 level=0 new: leaking param: r image/image.go:943 old: leaking param: p to result ~r0 level=1 new: leaking param content: p net/url/url.go:200 old: leaking param: s to result ~r2 level=0 new: leaking param: s (as a consequence) net/url/url.go:183 old: leaking param: s to result ~r1 level=0 new: leaking param: s net/url/url.go:194 old: leaking param: s to result ~r1 level=0 new: leaking param: s net/url/url.go:699 old: leaking param: u to result ~r0 level=1 new: leaking param: u net/url/url.go:775 old: (*URL).String u does not escape new: leaking param content: u net/url/url.go:1038 old: leaking param: u to result ~r0 level=1 new: leaking param: u net/url/url.go:1099 old: (*URL).MarshalBinary u does not escape new: leaking param content: u flag/flag.go:235 old: leaking param: s to result ~r0 level=1 new: leaking param content: s go/scanner/errors.go:105 old: leaking param: p to result ~r0 level=0 new: leaking param: p database/sql/sql.go:204 old: leaking param: ns to result ~r0 level=0 new: leaking param: ns go/constant/value.go:303 old: leaking param: re to result ~r2 level=0, leaking param: im to result ~r2 level=0 new: leaking param: re, leaking param: im go/constant/value.go:846 old: leaking param: x to result ~r1 level=0 new: leaking param: x encoding/xml/xml.go:518 old: leaking param: d to result ~r1 level=2 new: leaking param content: d encoding/xml/xml.go:122 old: leaking param: leaking param: t to result ~r1 level=0 new: leaking param: t crypto/x509/verify.go:506 old: leaking param: c to result ~r8 level=0 new: leaking param: c crypto/x509/verify.go:563 old: leaking param: c to result ~r3 level=0, leaking param content: c new: leaking param: c crypto/x509/verify.go:615 old: (nothing) new: leaking closure reference c crypto/x509/verify.go:996 old: leaking param: c to result ~r1 level=0, leaking param content: c new: leaking param: c net/http/filetransport.go:30 old: leaking param: fs to result ~r1 level=0 new: leaking param: fs net/http/h2_bundle.go:2684 old: leaking param: mh to result ~r0 level=2 new: leaking param content: mh net/http/h2_bundle.go:7352 old: http2checkConnHeaders req does not escape new: leaking param content: req net/http/pprof/pprof.go:221 old: leaking param: name to result ~r1 level=0 new: leaking param: name cmd/internal/bio/must.go:21 old: leaking param: w to result ~r1 level=0 new: leaking param: w Fixes #29353. Change-Id: I7e7798ae773728028b0dcae5bccb3ada51189c68 Reviewed-on: https://go-review.googlesource.com/c/162829 Run-TryBot: Cherry Zhang <cherryyz@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Matthew Dempsky <mdempsky@google.com> Reviewed-by: David Chase <drchase@google.com>
2019-02-17 23:12:55 -05:00
if src.Op == OCONVIFACE {
lt := src.Left.Type
if !lt.IsInterface() && !isdirectiface(lt) && types.Haspointers(lt) {
// We're converting from a non-direct interface type.
// The interface will hold a heap copy of the data
// (by calling convT2I or friend). Flow the data to heap.
// See issue 29353.
e.escwalk(level, &e.theSink, src.Left, e.stepWalk(dst, src.Left, "interface-converted", step))
}
}
}
case ODOT,
ODOTTYPE:
e.escwalk(level, dst, src.Left, e.stepWalk(dst, src.Left, "dot", step))
case
OSLICE,
OSLICEARR,
OSLICE3,
OSLICE3ARR,
OSLICESTR:
e.escwalk(level, dst, src.Left, e.stepWalk(dst, src.Left, "slice", step))
case OINDEX:
if src.Left.Type.IsArray() {
e.escwalk(level, dst, src.Left, e.stepWalk(dst, src.Left, "fixed-array-index-of", step))
break
}
fallthrough
case ODOTPTR:
e.escwalk(level.inc(), dst, src.Left, e.stepWalk(dst, src.Left, "dot of pointer", step))
case OINDEXMAP:
e.escwalk(level.inc(), dst, src.Left, e.stepWalk(dst, src.Left, "map index", step))
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
case ODEREF:
e.escwalk(level.inc(), dst, src.Left, e.stepWalk(dst, src.Left, "indirection", step))
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
// In this case a link went directly to a call, but should really go
// to the dummy .outN outputs that were created for the call that
// themselves link to the inputs with levels adjusted.
// See e.g. #10466
// This can only happen with functions returning a single result.
case OCALLMETH, OCALLFUNC, OCALLINTER:
if srcE.Retval.Len() != 0 {
if Debug['m'] > 2 {
fmt.Printf("%v:[%d] dst %S escwalk replace src: %S with %S\n",
linestr(lineno), e.loopdepth,
dst, src, srcE.Retval.First())
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
}
src = srcE.Retval.First()
srcE = e.nodeEscState(src)
}
}
recurse:
cmd/internal/gc: improve flow of input params to output params This includes the following information in the per-function summary: outK = paramJ encoded in outK bits for paramJ outK = *paramJ encoded in outK bits for paramJ heap = paramJ EscHeap heap = *paramJ EscContentEscapes Note that (currently) if the address of a parameter is taken and returned, necessarily a heap allocation occurred to contain that reference, and the heap can never refer to stack, therefore the parameter and everything downstream from it escapes to the heap. The per-function summary information now has a tuneable number of bits (2 is probably noticeably better than 1, 3 is likely overkill, but it is now easy to check and the -m debugging output includes information that allows you to figure out if more would be better.) A new test was added to check pointer flow through struct-typed and *struct-typed parameters and returns; some of these are sensitive to the number of summary bits, and ought to yield better results with a more competent escape analysis algorithm. Another new test checks (some) correctness with array parameters, results, and operations. The old analysis inferred a piece of plan9 runtime was non-escaping by counteracting overconservative analysis with buggy analysis; with the bug fixed, the result was too conservative (and it's not easy to fix in this framework) so the source code was tweaked to get the desired result. A test was added against the discovered bug. The escape analysis was further improved splitting the "level" into 3 parts, one tracking the conventional "level" and the other two computing the highest-level-suffix-from-copy, which is used to generally model the cancelling effect of indirection applied to address-of. With the improved escape analysis enabled, it was necessary to modify one of the runtime tests because it now attempts to allocate too much on the (small, fixed-size) G0 (system) stack and this failed the test. Compiling src/std after touching src/runtime/*.go with -m logging turned on shows 420 fewer heap allocation sites (10538 vs 10968). Profiling allocations in src/html/template with for i in {1..5} ; do go tool 6g -memprofile=mastx.${i}.prof -memprofilerate=1 *.go; go tool pprof -alloc_objects -text mastx.${i}.prof ; done showed a 15% reduction in allocations performed by the compiler. Update #3753 Update #4720 Fixes #10466 Change-Id: I0fd97d5f5ac527b45f49e2218d158a6e89951432 Reviewed-on: https://go-review.googlesource.com/8202 Run-TryBot: David Chase <drchase@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-26 16:36:15 -04:00
level = level.copy()
for i := range srcE.Flowsrc {
s := &srcE.Flowsrc[i]
s.parent = step
e.escwalkBody(level, dst, s.src, s, extraloopdepth)
s.parent = nil
}
e.pdepth--
}
// addrescapes tags node n as having had its address taken
// by "increasing" the "value" of n.Esc to EscHeap.
// Storage is allocated as necessary to allow the address
// to be taken.
func addrescapes(n *Node) {
switch n.Op {
default:
// Unexpected Op, probably due to a previous type error. Ignore.
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
case ODEREF, ODOTPTR:
// Nothing to do.
case ONAME:
if n == nodfp {
break
}
// if this is a tmpname (PAUTO), it was tagged by tmpname as not escaping.
// on PPARAM it means something different.
if n.Class() == PAUTO && n.Esc == EscNever {
break
}
// If a closure reference escapes, mark the outer variable as escaping.
if n.IsClosureVar() {
addrescapes(n.Name.Defn)
break
}
if n.Class() != PPARAM && n.Class() != PPARAMOUT && n.Class() != PAUTO {
break
}
// This is a plain parameter or local variable that needs to move to the heap,
// but possibly for the function outside the one we're compiling.
// That is, if we have:
//
// func f(x int) {
// func() {
// global = &x
// }
// }
//
// then we're analyzing the inner closure but we need to move x to the
// heap in f, not in the inner closure. Flip over to f before calling moveToHeap.
oldfn := Curfn
Curfn = n.Name.Curfn
if Curfn.Func.Closure != nil && Curfn.Op == OCLOSURE {
Curfn = Curfn.Func.Closure
}
ln := lineno
lineno = Curfn.Pos
moveToHeap(n)
Curfn = oldfn
lineno = ln
// ODOTPTR has already been introduced,
// so these are the non-pointer ODOT and OINDEX.
// In &x[0], if x is a slice, then x does not
// escape--the pointer inside x does, but that
// is always a heap pointer anyway.
case ODOT, OINDEX, OPAREN, OCONVNOP:
if !n.Left.Type.IsSlice() {
addrescapes(n.Left)
}
}
}
// moveToHeap records the parameter or local variable n as moved to the heap.
func moveToHeap(n *Node) {
if Debug['r'] != 0 {
Dump("MOVE", n)
}
if compiling_runtime {
yyerror("%v escapes to heap, not allowed in runtime.", n)
}
if n.Class() == PAUTOHEAP {
Dump("n", n)
Fatalf("double move to heap")
}
// Allocate a local stack variable to hold the pointer to the heap copy.
// temp will add it to the function declaration list automatically.
heapaddr := temp(types.NewPtr(n.Type))
heapaddr.Sym = lookup("&" + n.Sym.Name)
heapaddr.Orig.Sym = heapaddr.Sym
heapaddr.Pos = n.Pos
// Unset AutoTemp to persist the &foo variable name through SSA to
// liveness analysis.
// TODO(mdempsky/drchase): Cleaner solution?
heapaddr.Name.SetAutoTemp(false)
// Parameters have a local stack copy used at function start/end
// in addition to the copy in the heap that may live longer than
// the function.
if n.Class() == PPARAM || n.Class() == PPARAMOUT {
if n.Xoffset == BADWIDTH {
Fatalf("addrescapes before param assignment")
}
// We rewrite n below to be a heap variable (indirection of heapaddr).
// Preserve a copy so we can still write code referring to the original,
// and substitute that copy into the function declaration list
// so that analyses of the local (on-stack) variables use it.
stackcopy := newname(n.Sym)
stackcopy.SetAddable(false)
stackcopy.Type = n.Type
stackcopy.Xoffset = n.Xoffset
stackcopy.SetClass(n.Class())
stackcopy.Name.Param.Heapaddr = heapaddr
if n.Class() == PPARAMOUT {
// Make sure the pointer to the heap copy is kept live throughout the function.
// The function could panic at any point, and then a defer could recover.
// Thus, we need the pointer to the heap copy always available so the
// post-deferreturn code can copy the return value back to the stack.
// See issue 16095.
heapaddr.SetIsOutputParamHeapAddr(true)
}
n.Name.Param.Stackcopy = stackcopy
// Substitute the stackcopy into the function variable list so that
// liveness and other analyses use the underlying stack slot
// and not the now-pseudo-variable n.
found := false
for i, d := range Curfn.Func.Dcl {
if d == n {
Curfn.Func.Dcl[i] = stackcopy
found = true
break
}
// Parameters are before locals, so can stop early.
// This limits the search even in functions with many local variables.
if d.Class() == PAUTO {
break
}
}
if !found {
Fatalf("cannot find %v in local variable list", n)
}
Curfn.Func.Dcl = append(Curfn.Func.Dcl, n)
}
// Modify n in place so that uses of n now mean indirection of the heapaddr.
n.SetClass(PAUTOHEAP)
n.Xoffset = 0
n.Name.Param.Heapaddr = heapaddr
n.Esc = EscHeap
if Debug['m'] != 0 {
fmt.Printf("%v: moved to heap: %v\n", n.Line(), n)
}
}
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
// This special tag is applied to uintptr variables
// that we believe may hold unsafe.Pointers for
// calls into assembly functions.
const unsafeUintptrTag = "unsafe-uintptr"
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
// This special tag is applied to uintptr parameters of functions
// marked go:uintptrescapes.
const uintptrEscapesTag = "uintptr-escapes"
func (e *EscState) esctag(fn *Node) {
fn.Esc = EscFuncTagged
name := func(s *types.Sym, narg int) string {
if s != nil {
return s.Name
}
return fmt.Sprintf("arg#%d", narg)
}
// External functions are assumed unsafe,
// unless //go:noescape is given before the declaration.
if fn.Nbody.Len() == 0 {
cmd/compile: pack bool fields in Node, Name, Func and Type structs to bitsets This reduces compiler memory usage by up to 4% - see compilebench results below. name old time/op new time/op delta Template 245ms ± 4% 241ms ± 2% -1.88% (p=0.029 n=10+10) Unicode 126ms ± 3% 124ms ± 3% ~ (p=0.105 n=10+10) GoTypes 805ms ± 2% 813ms ± 3% ~ (p=0.515 n=8+10) Compiler 3.95s ± 2% 3.83s ± 1% -2.96% (p=0.000 n=9+10) MakeBash 47.4s ± 4% 46.6s ± 1% -1.59% (p=0.028 n=9+10) name old user-ns/op new user-ns/op delta Template 324M ± 5% 326M ± 3% ~ (p=0.935 n=10+10) Unicode 186M ± 5% 178M ±10% ~ (p=0.067 n=9+10) GoTypes 1.08G ± 7% 1.09G ± 4% ~ (p=0.956 n=10+10) Compiler 5.34G ± 4% 5.31G ± 1% ~ (p=0.501 n=10+8) name old alloc/op new alloc/op delta Template 41.0MB ± 0% 39.8MB ± 0% -3.03% (p=0.000 n=10+10) Unicode 32.3MB ± 0% 31.0MB ± 0% -4.13% (p=0.000 n=10+10) GoTypes 119MB ± 0% 116MB ± 0% -2.39% (p=0.000 n=10+10) Compiler 499MB ± 0% 487MB ± 0% -2.48% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Template 380k ± 1% 379k ± 1% ~ (p=0.436 n=10+10) Unicode 324k ± 1% 324k ± 0% ~ (p=0.853 n=10+10) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.481 n=10+10) Compiler 4.41M ± 0% 4.41M ± 0% -0.12% (p=0.007 n=10+10) name old text-bytes new text-bytes delta HelloSize 623k ± 0% 623k ± 0% ~ (all equal) CmdGoSize 6.64M ± 0% 6.64M ± 0% ~ (all equal) name old data-bytes new data-bytes delta HelloSize 5.81k ± 0% 5.81k ± 0% ~ (all equal) CmdGoSize 238k ± 0% 238k ± 0% ~ (all equal) name old bss-bytes new bss-bytes delta HelloSize 134k ± 0% 134k ± 0% ~ (all equal) CmdGoSize 152k ± 0% 152k ± 0% ~ (all equal) name old exe-bytes new exe-bytes delta HelloSize 967k ± 0% 967k ± 0% ~ (all equal) CmdGoSize 10.2M ± 0% 10.2M ± 0% ~ (all equal) Change-Id: I1f40af738254892bd6c8ba2eb43390b175753d52 Reviewed-on: https://go-review.googlesource.com/37445 Reviewed-by: Matthew Dempsky <mdempsky@google.com> Run-TryBot: Matthew Dempsky <mdempsky@google.com> TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-02-27 19:56:38 +02:00
if fn.Noescape() {
for _, f := range fn.Type.Params().Fields().Slice() {
if types.Haspointers(f.Type) {
f.Note = mktag(EscNone)
}
}
}
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
// Assume that uintptr arguments must be held live across the call.
// This is most important for syscall.Syscall.
// See golang.org/issue/13372.
// This really doesn't have much to do with escape analysis per se,
// but we are reusing the ability to annotate an individual function
// argument and pass those annotations along to importing code.
narg := 0
for _, f := range fn.Type.Params().Fields().Slice() {
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
narg++
if f.Type.Etype == TUINTPTR {
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
if Debug['m'] != 0 {
Warnl(fn.Pos, "%v assuming %v is unsafe uintptr", funcSym(fn), name(f.Sym, narg))
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
}
f.Note = unsafeUintptrTag
cmd/compile: recognize Syscall-like functions for liveness analysis Consider this code: func f(*int) func g() { p := new(int) f(p) } where f is an assembly function. In general liveness analysis assumes that during the call to f, p is dead in this frame. If f has retained p, p will be found alive in f's frame and keep the new(int) from being garbage collected. This is all correct and works. We use the Go func declaration for f to give the assembly function liveness information (the arguments are assumed live for the entire call). Now consider this code: func h1() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) } Here syscall.Syscall is taking the place of f, but because its arguments are uintptr, the liveness analysis and the garbage collector ignore them. Since p is no longer live in h once the call starts, if the garbage collector scans the stack while the system call is blocked, it will find no reference to the new(int) and reclaim it. If the kernel is going to write to *p once the call finishes, reclaiming the memory is a mistake. We can't change the arguments or the liveness information for syscall.Syscall itself, both for compatibility and because sometimes the arguments really are integers, and the garbage collector will get quite upset if it finds an integer where it expects a pointer. The problem is that these arguments are fundamentally untyped. The solution we have taken in the syscall package's wrappers in past releases is to insert a call to a dummy function named "use", to make it look like the argument is live during the call to syscall.Syscall: func h2() { p := new(int) syscall.Syscall(1, 2, 3, uintptr(unsafe.Pointer(p))) use(unsafe.Pointer(p)) } Keeping p alive during the call means that if the garbage collector scans the stack during the system call now, it will find the reference to p. Unfortunately, this approach is not available to users outside syscall, because 'use' is unexported, and people also have to realize they need to use it and do so. There is much existing code using syscall.Syscall without a 'use'-like function. That code will fail very occasionally in mysterious ways (see #13372). This CL fixes all that existing code by making the compiler do the right thing automatically, without any code modifications. That is, it takes h1 above, which is incorrect code today, and makes it correct code. Specifically, if the compiler sees a foreign func definition (one without a body) that has uintptr arguments, it marks those arguments as "unsafe uintptrs". If it later sees the function being called with uintptr(unsafe.Pointer(x)) as an argument, it arranges to mark x as having escaped, and it makes sure to hold x in a live temporary variable until the call returns, so that the garbage collector cannot reclaim whatever heap memory x points to. For now I am leaving the explicit calls to use in package syscall, but they can be removed early in a future cycle (likely Go 1.7). The rule has no effect on escape analysis, only on liveness analysis. Fixes #13372. Change-Id: I2addb83f70d08db08c64d394f9d06ff0a063c500 Reviewed-on: https://go-review.googlesource.com/18584 Reviewed-by: Ian Lance Taylor <iant@golang.org>
2016-01-13 00:46:28 -05:00
}
}
return
}
if fn.Func.Pragma&UintptrEscapes != 0 {
narg := 0
for _, f := range fn.Type.Params().Fields().Slice() {
narg++
if f.Type.Etype == TUINTPTR {
if Debug['m'] != 0 {
Warnl(fn.Pos, "%v marking %v as escaping uintptr", funcSym(fn), name(f.Sym, narg))
}
f.Note = uintptrEscapesTag
}
cmd/compile: bulk rename This change does a bulk rename of several identifiers in the compiler. See #27167 and https://docs.google.com/document/d/19_ExiylD9MRfeAjKIfEsMU1_RGhuxB9sA0b5Zv7byVI/ for context and for discussion of these particular renames. Commands run to generate this change: gorename -from '"cmd/compile/internal/gc".OPROC' -to OGO gorename -from '"cmd/compile/internal/gc".OCOM' -to OBITNOT gorename -from '"cmd/compile/internal/gc".OMINUS' -to ONEG gorename -from '"cmd/compile/internal/gc".OIND' -to ODEREF gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTR' -to OBYTES2STR gorename -from '"cmd/compile/internal/gc".OARRAYBYTESTRTMP' -to OBYTES2STRTMP gorename -from '"cmd/compile/internal/gc".OARRAYRUNESTR' -to ORUNES2STR gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTE' -to OSTR2BYTES gorename -from '"cmd/compile/internal/gc".OSTRARRAYBYTETMP' -to OSTR2BYTESTMP gorename -from '"cmd/compile/internal/gc".OSTRARRAYRUNE' -to OSTR2RUNES gorename -from '"cmd/compile/internal/gc".Etop' -to ctxStmt gorename -from '"cmd/compile/internal/gc".Erv' -to ctxExpr gorename -from '"cmd/compile/internal/gc".Ecall' -to ctxCallee gorename -from '"cmd/compile/internal/gc".Efnstruct' -to ctxMultiOK gorename -from '"cmd/compile/internal/gc".Easgn' -to ctxAssign gorename -from '"cmd/compile/internal/gc".Ecomplit' -to ctxCompLit Not altered: parameters and local variables (mostly in typecheck.go) named top, which should probably now be called ctx (and which should probably have a named type). Also not altered: Field called Top in gc.Func. gorename -from '"cmd/compile/internal/gc".Node.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/gc".Node.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/gc".nodeIsddd' -to nodeIsDDD gorename -from '"cmd/compile/internal/types".Field.Isddd' -to IsDDD gorename -from '"cmd/compile/internal/types".Field.SetIsddd' -to SetIsDDD gorename -from '"cmd/compile/internal/types".fieldIsddd' -to fieldIsDDD Not altered: function gc.hasddd, params and local variables called isddd Also not altered: fmt.go prints nodes using "isddd(%v)". cd cmd/compile/internal/gc; go generate I then manually found impacted comments using exact string match and fixed them up by hand. The comment changes were trivial. Passes toolstash-check. Fixes #27167. If this experiment is deemed a success, we will open a new tracking issue for renames to do at the end of the 1.13 cycles. Change-Id: I2dc541533d2ab0d06cb3d31d65df205ecfb151e8 Reviewed-on: https://go-review.googlesource.com/c/150140 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2018-11-18 08:34:38 -08:00
if f.IsDDD() && f.Type.Elem().Etype == TUINTPTR {
// final argument is ...uintptr.
if Debug['m'] != 0 {
Warnl(fn.Pos, "%v marking %v as escaping ...uintptr", funcSym(fn), name(f.Sym, narg))
}
f.Note = uintptrEscapesTag
}
}
}
for _, fs := range types.RecvsParams {
for _, f := range fs(fn.Type).Fields().Slice() {
if !types.Haspointers(f.Type) { // don't bother tagging for scalars
continue
}
if f.Note == uintptrEscapesTag {
// Note is already set in the loop above.
continue
}
// Unnamed parameters are unused and therefore do not escape.
if f.Sym == nil || f.Sym.IsBlank() {
f.Note = mktag(EscNone)
continue
}
switch esc := asNode(f.Nname).Esc; esc & EscMask {
case EscNone, // not touched by escflood
EscReturn:
f.Note = mktag(int(esc))
case EscHeap: // touched by escflood, moved to heap
}
}
}
}