2015-02-13 14:40:36 -05:00
|
|
|
// 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 (
|
2016-10-03 12:26:25 -07:00
|
|
|
"cmd/compile/internal/ssa"
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
"cmd/compile/internal/types"
|
2017-03-06 07:32:37 -08:00
|
|
|
"cmd/internal/dwarf"
|
2015-02-13 14:40:36 -05:00
|
|
|
"cmd/internal/obj"
|
2017-04-18 12:53:25 -07:00
|
|
|
"cmd/internal/objabi"
|
2017-03-22 20:27:54 -07:00
|
|
|
"cmd/internal/src"
|
2016-04-06 12:01:40 -07:00
|
|
|
"cmd/internal/sys"
|
2015-02-13 14:40:36 -05:00
|
|
|
"fmt"
|
2019-05-20 15:07:01 +07:00
|
|
|
"internal/race"
|
2017-04-27 07:37:40 -07:00
|
|
|
"math/rand"
|
2016-02-25 10:35:19 -08:00
|
|
|
"sort"
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
"sync"
|
2017-05-17 08:29:18 -07:00
|
|
|
"time"
|
2015-02-13 14:40:36 -05:00
|
|
|
)
|
|
|
|
|
|
|
|
|
|
// "Portable" code generation.
|
|
|
|
|
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
var (
|
|
|
|
|
nBackendWorkers int // number of concurrent backend workers, set by a compiler flag
|
|
|
|
|
compilequeue []*Node // functions waiting to be compiled
|
|
|
|
|
)
|
|
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
func emitptrargsmap(fn *Node) {
|
|
|
|
|
if fn.funcname() == "_" {
|
2015-11-04 15:54:41 -05:00
|
|
|
return
|
|
|
|
|
}
|
2018-03-06 16:20:10 -08:00
|
|
|
sym := lookup(fmt.Sprintf("%s.args_stackmap", fn.funcname()))
|
2017-04-21 07:51:41 -07:00
|
|
|
lsym := sym.Linksym()
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
nptr := int(fn.Type.ArgWidth() / int64(Widthptr))
|
2015-02-23 16:07:24 -05:00
|
|
|
bv := bvalloc(int32(nptr) * 2)
|
|
|
|
|
nbitmap := 1
|
2018-03-06 16:20:10 -08:00
|
|
|
if fn.Type.NumResults() > 0 {
|
2015-02-13 14:40:36 -05:00
|
|
|
nbitmap = 2
|
|
|
|
|
}
|
2017-04-21 13:59:51 -07:00
|
|
|
off := duint32(lsym, 0, uint32(nbitmap))
|
|
|
|
|
off = duint32(lsym, off, uint32(bv.n))
|
2017-09-18 23:29:46 -07:00
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
if fn.IsMethod() {
|
|
|
|
|
onebitwalktype1(fn.Type.Recvs(), 0, bv)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2018-03-06 16:20:10 -08:00
|
|
|
if fn.Type.NumParams() > 0 {
|
|
|
|
|
onebitwalktype1(fn.Type.Params(), 0, bv)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-04-21 13:59:51 -07:00
|
|
|
off = dbvec(lsym, off, bv)
|
2017-09-18 23:29:46 -07:00
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
if fn.Type.NumResults() > 0 {
|
|
|
|
|
onebitwalktype1(fn.Type.Results(), 0, bv)
|
2017-04-21 13:59:51 -07:00
|
|
|
off = dbvec(lsym, off, bv)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-21 14:20:54 -07:00
|
|
|
ggloblsym(lsym, int32(off), obj.RODATA|obj.LOCAL)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-09-15 21:43:53 +02:00
|
|
|
// cmpstackvarlt reports whether the stack variable a sorts before b.
|
2015-09-19 23:55:27 +02:00
|
|
|
//
|
2015-02-13 14:40:36 -05:00
|
|
|
// Sort the list of stack variables. Autos after anything else,
|
|
|
|
|
// within autos, unused after used, within used, things with
|
|
|
|
|
// pointers first, zeroed things first, and then decreasing size.
|
|
|
|
|
// Because autos are laid out in decreasing addresses
|
|
|
|
|
// on the stack, pointers first, zeroed things first and decreasing size
|
|
|
|
|
// really means, in memory, things with pointers needing zeroing at
|
|
|
|
|
// the top of the stack and increasing in size.
|
|
|
|
|
// Non-autos sort on offset.
|
2015-09-19 23:55:27 +02:00
|
|
|
func cmpstackvarlt(a, b *Node) bool {
|
2017-04-25 18:14:12 -07:00
|
|
|
if (a.Class() == PAUTO) != (b.Class() == PAUTO) {
|
|
|
|
|
return b.Class() == PAUTO
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-25 18:14:12 -07:00
|
|
|
if a.Class() != PAUTO {
|
2016-02-20 21:36:12 -08:00
|
|
|
return a.Xoffset < b.Xoffset
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-04-27 15:17:57 -07:00
|
|
|
if a.Name.Used() != b.Name.Used() {
|
|
|
|
|
return a.Name.Used()
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
ap := types.Haspointers(a.Type)
|
|
|
|
|
bp := types.Haspointers(b.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
if ap != bp {
|
2015-09-19 23:55:27 +02:00
|
|
|
return ap
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-02-27 19:56:38 +02:00
|
|
|
ap = a.Name.Needzero()
|
|
|
|
|
bp = b.Name.Needzero()
|
2015-02-13 14:40:36 -05:00
|
|
|
if ap != bp {
|
2015-09-19 23:55:27 +02:00
|
|
|
return ap
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-02-20 21:36:12 -08:00
|
|
|
if a.Type.Width != b.Type.Width {
|
|
|
|
|
return a.Type.Width > b.Type.Width
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2015-09-19 23:55:27 +02:00
|
|
|
return a.Sym.Name < b.Sym.Name
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2016-02-25 10:35:19 -08:00
|
|
|
// byStackvar implements sort.Interface for []*Node using cmpstackvarlt.
|
|
|
|
|
type byStackVar []*Node
|
|
|
|
|
|
|
|
|
|
func (s byStackVar) Len() int { return len(s) }
|
|
|
|
|
func (s byStackVar) Less(i, j int) bool { return cmpstackvarlt(s[i], s[j]) }
|
|
|
|
|
func (s byStackVar) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
|
|
|
|
|
|
2017-03-17 07:49:22 -07:00
|
|
|
func (s *ssafn) AllocFrame(f *ssa.Func) {
|
2017-03-17 09:10:57 -07:00
|
|
|
s.stksize = 0
|
2017-03-17 09:19:56 -07:00
|
|
|
s.stkptrsize = 0
|
2017-03-17 07:49:22 -07:00
|
|
|
fn := s.curfn.Func
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
// Mark the PAUTO's unused.
|
2017-03-17 07:49:22 -07:00
|
|
|
for _, ln := range fn.Dcl {
|
2017-04-25 18:14:12 -07:00
|
|
|
if ln.Class() == PAUTO {
|
2017-04-27 15:17:57 -07:00
|
|
|
ln.Name.SetUsed(false)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2016-10-03 12:26:25 -07:00
|
|
|
for _, l := range f.RegAlloc {
|
|
|
|
|
if ls, ok := l.(ssa.LocalSlot); ok {
|
2017-04-27 15:17:57 -07:00
|
|
|
ls.N.(*Node).Name.SetUsed(true)
|
2016-10-03 12:26:25 -07:00
|
|
|
}
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-10-03 12:26:25 -07:00
|
|
|
scratchUsed := false
|
|
|
|
|
for _, b := range f.Blocks {
|
|
|
|
|
for _, v := range b.Values {
|
2017-09-18 14:53:56 -07:00
|
|
|
if n, ok := v.Aux.(*Node); ok {
|
|
|
|
|
switch n.Class() {
|
2018-03-08 21:00:36 +00:00
|
|
|
case PPARAM, PPARAMOUT:
|
|
|
|
|
// Don't modify nodfp; it is a global.
|
|
|
|
|
if n != nodfp {
|
|
|
|
|
n.Name.SetUsed(true)
|
|
|
|
|
}
|
|
|
|
|
case PAUTO:
|
2017-04-27 15:17:57 -07:00
|
|
|
n.Name.SetUsed(true)
|
2017-03-31 16:52:50 -07:00
|
|
|
}
|
2016-10-03 12:26:25 -07:00
|
|
|
}
|
2016-10-04 13:00:21 -07:00
|
|
|
if !scratchUsed {
|
|
|
|
|
scratchUsed = v.Op.UsesScratch()
|
2016-10-03 12:26:25 -07:00
|
|
|
}
|
2017-09-18 14:53:56 -07:00
|
|
|
|
2016-10-03 12:26:25 -07:00
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-30 14:34:16 -07:00
|
|
|
if f.Config.NeedsFpScratch && scratchUsed {
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
s.scratchFpMem = tempAt(src.NoXPos, s.curfn, types.Types[TUINT64])
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-17 07:49:22 -07:00
|
|
|
sort.Sort(byStackVar(fn.Dcl))
|
2016-10-03 12:26:25 -07:00
|
|
|
|
|
|
|
|
// Reassign stack offsets of the locals that are used.
|
2018-12-21 16:36:45 -08:00
|
|
|
lastHasPtr := false
|
2017-03-17 07:49:22 -07:00
|
|
|
for i, n := range fn.Dcl {
|
2017-04-25 18:14:12 -07:00
|
|
|
if n.Op != ONAME || n.Class() != PAUTO {
|
2015-02-13 14:40:36 -05:00
|
|
|
continue
|
|
|
|
|
}
|
2017-04-27 15:17:57 -07:00
|
|
|
if !n.Name.Used() {
|
2017-03-17 07:49:22 -07:00
|
|
|
fn.Dcl = fn.Dcl[:i]
|
2016-10-03 12:26:25 -07:00
|
|
|
break
|
|
|
|
|
}
|
2015-02-13 14:40:36 -05:00
|
|
|
|
|
|
|
|
dowidth(n.Type)
|
2016-10-03 12:26:25 -07:00
|
|
|
w := n.Type.Width
|
2017-03-17 13:35:36 -07:00
|
|
|
if w >= thearch.MAXWIDTH || w < 0 {
|
2015-08-30 23:10:03 +02:00
|
|
|
Fatalf("bad width")
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2018-12-21 16:36:45 -08:00
|
|
|
if w == 0 && lastHasPtr {
|
|
|
|
|
// Pad between a pointer-containing object and a zero-sized object.
|
|
|
|
|
// This prevents a pointer to the zero-sized object from being interpreted
|
|
|
|
|
// as a pointer to the pointer-containing object (and causing it
|
|
|
|
|
// to be scanned when it shouldn't be). See issue 24993.
|
|
|
|
|
w = 1
|
|
|
|
|
}
|
2017-03-17 09:10:57 -07:00
|
|
|
s.stksize += w
|
|
|
|
|
s.stksize = Rnd(s.stksize, int64(n.Type.Align))
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
if types.Haspointers(n.Type) {
|
2017-03-17 09:19:56 -07:00
|
|
|
s.stkptrsize = s.stksize
|
2018-12-21 16:36:45 -08:00
|
|
|
lastHasPtr = true
|
|
|
|
|
} else {
|
|
|
|
|
lastHasPtr = false
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-03-17 13:35:36 -07:00
|
|
|
if thearch.LinkArch.InFamily(sys.MIPS, sys.MIPS64, sys.ARM, sys.ARM64, sys.PPC64, sys.S390X) {
|
2017-03-17 09:10:57 -07:00
|
|
|
s.stksize = Rnd(s.stksize, int64(Widthptr))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2017-03-17 09:10:57 -07:00
|
|
|
n.Xoffset = -s.stksize
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-17 09:10:57 -07:00
|
|
|
s.stksize = Rnd(s.stksize, int64(Widthreg))
|
2017-03-17 09:19:56 -07:00
|
|
|
s.stkptrsize = Rnd(s.stkptrsize, int64(Widthreg))
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
func funccompile(fn *Node) {
|
|
|
|
|
if Curfn != nil {
|
|
|
|
|
Fatalf("funccompile %v inside %v", fn.Func.Nname.Sym, Curfn.Func.Nname.Sym)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
if fn.Type == nil {
|
|
|
|
|
if nerrors == 0 {
|
|
|
|
|
Fatalf("funccompile missing type")
|
|
|
|
|
}
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// assign parameter offsets
|
2017-03-14 09:12:55 -07:00
|
|
|
dowidth(fn.Type)
|
2015-02-13 14:40:36 -05:00
|
|
|
|
2016-04-24 13:50:26 -07:00
|
|
|
if fn.Nbody.Len() == 0 {
|
2018-11-01 12:30:23 -04:00
|
|
|
// Initialize ABI wrappers if necessary.
|
|
|
|
|
fn.Func.initLSym(false)
|
2018-03-06 16:20:10 -08:00
|
|
|
emitptrargsmap(fn)
|
2016-03-07 22:05:49 -08:00
|
|
|
return
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2018-03-06 16:20:10 -08:00
|
|
|
dclcontext = PAUTO
|
|
|
|
|
Curfn = fn
|
|
|
|
|
|
|
|
|
|
compile(fn)
|
|
|
|
|
|
|
|
|
|
Curfn = nil
|
|
|
|
|
dclcontext = PEXTERN
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
func compile(fn *Node) {
|
2015-02-13 14:40:36 -05:00
|
|
|
saveerrors()
|
|
|
|
|
|
2017-03-14 09:12:55 -07:00
|
|
|
order(fn)
|
2015-02-13 14:40:36 -05:00
|
|
|
if nerrors != 0 {
|
2016-03-07 22:05:49 -08:00
|
|
|
return
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-14 09:12:55 -07:00
|
|
|
walk(fn)
|
2015-02-13 14:40:36 -05:00
|
|
|
if nerrors != 0 {
|
2016-03-07 22:05:49 -08:00
|
|
|
return
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2015-10-20 10:00:07 -07:00
|
|
|
if instrumenting {
|
2017-03-14 09:12:55 -07:00
|
|
|
instrument(fn)
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
|
|
|
|
|
2017-03-23 16:39:11 -07:00
|
|
|
// From this point, there should be no uses of Curfn. Enforce that.
|
|
|
|
|
Curfn = nil
|
|
|
|
|
|
2019-01-22 13:49:47 -08:00
|
|
|
if fn.funcname() == "_" {
|
|
|
|
|
// We don't need to generate code for this function, just report errors in its body.
|
|
|
|
|
// At this point we've generated any errors needed.
|
|
|
|
|
// (Beyond here we generate only non-spec errors, like "stack frame too large".)
|
|
|
|
|
// See issue 29870.
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
2017-04-12 13:23:07 -07:00
|
|
|
// Set up the function's LSym early to avoid data races with the assemblers.
|
2018-11-01 12:30:23 -04:00
|
|
|
fn.Func.initLSym(true)
|
2017-04-12 13:23:07 -07:00
|
|
|
|
2018-09-01 20:16:39 -07:00
|
|
|
// Make sure type syms are declared for all types that might
|
|
|
|
|
// be types of stack objects. We need to do this here
|
|
|
|
|
// because symbols must be allocated before the parallel
|
|
|
|
|
// phase of the compiler.
|
2019-01-22 10:08:10 -08:00
|
|
|
for _, n := range fn.Func.Dcl {
|
|
|
|
|
switch n.Class() {
|
|
|
|
|
case PPARAM, PPARAMOUT, PAUTO:
|
|
|
|
|
if livenessShouldTrack(n) && n.Addrtaken() {
|
|
|
|
|
dtypesym(n.Type)
|
|
|
|
|
// Also make sure we allocate a linker symbol
|
|
|
|
|
// for the stack object data, for the same reason.
|
|
|
|
|
if fn.Func.lsym.Func.StackObjects == nil {
|
2019-04-10 16:44:46 -07:00
|
|
|
fn.Func.lsym.Func.StackObjects = Ctxt.Lookup(fn.Func.lsym.Name + ".stkobj")
|
2018-09-01 20:16:39 -07:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
if compilenow() {
|
|
|
|
|
compileSSA(fn, 0)
|
|
|
|
|
} else {
|
|
|
|
|
compilequeue = append(compilequeue, fn)
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// compilenow reports whether to compile immediately.
|
|
|
|
|
// If functions are not compiled immediately,
|
|
|
|
|
// they are enqueued in compilequeue,
|
|
|
|
|
// which is drained by compileFunctions.
|
|
|
|
|
func compilenow() bool {
|
cmd/compile: don't update outer variables after capturevars is complete
When compiling concurrently, we walk all functions before compiling
any of them. Walking functions can cause variables to switch from
being non-addrtaken to addrtaken, e.g. to prepare for a runtime call.
Typechecking propagates addrtaken-ness of closure variables to
their outer variables, so that capturevars can decide whether to
pass the variable's value or a pointer to it.
When all functions are compiled immediately, as long as the containing
function is compiled prior to the closure, this propagation has no effect.
When compilation is deferred, though, in rare cases, this results in
a change in the addrtaken-ness of a variable in the outer function,
which in turn changes the compiler's output.
(This is rare because in a great many cases, a temporary has been
introduced, insulating the outer variable from modification.)
But concurrent compilation must generate identical results.
To fix this, track whether capturevars has run.
If it has, there is no need to update outer variables
when closure variables change.
Capturevars always runs before any functions are walked or compiled.
The remainder of the changes in this CL are to support the test.
In particular, -d=compilelater forces the compiler to walk all
functions before compiling any of them, despite being non-concurrent.
This is useful because -live is fundamentally incompatible with
concurrent compilation, but we want -c=1 to have no behavior changes.
Fixes #20250
Change-Id: I89bcb54268a41e8588af1ac8cc37fbef856a90c2
Reviewed-on: https://go-review.googlesource.com/42853
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Robert Griesemer <gri@golang.org>
2017-05-05 15:22:59 -07:00
|
|
|
return nBackendWorkers == 1 && Debug_compilelater == 0
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
}
|
|
|
|
|
|
2017-10-10 17:43:41 -07:00
|
|
|
const maxStackSize = 1 << 30
|
2017-05-30 15:05:42 -07:00
|
|
|
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
// compileSSA builds an SSA backend function,
|
|
|
|
|
// uses it to generate a plist,
|
|
|
|
|
// and flushes that plist to machine code.
|
|
|
|
|
// worker indicates which of the backend workers is doing the processing.
|
|
|
|
|
func compileSSA(fn *Node, worker int) {
|
2017-10-10 17:43:41 -07:00
|
|
|
f := buildssa(fn, worker)
|
2018-05-29 09:30:12 -07:00
|
|
|
// Note: check arg size to fix issue 25507.
|
|
|
|
|
if f.Frontend().(*ssafn).stksize >= maxStackSize || fn.Type.ArgWidth() >= maxStackSize {
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
largeStackFramesMu.Lock()
|
2018-10-25 06:23:54 -07:00
|
|
|
largeStackFrames = append(largeStackFrames, largeStack{locals: f.Frontend().(*ssafn).stksize, args: fn.Type.ArgWidth(), pos: fn.Pos})
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
largeStackFramesMu.Unlock()
|
2017-10-10 17:43:41 -07:00
|
|
|
return
|
cmd/compile: rework reporting of oversized stack frames
We don't support stack frames over 2GB.
Rather than detect this during backend compilation,
check for it at the end of compilation.
This is arguably a more accurate check anyway,
since it takes into account the full frame,
including local stack, arguments, and arch-specific
rounding, although it's unlikely anyone would ever notice.
Also, rather than reporting the error right away,
take note of it and report it later, at the top level.
This is not relevant now, but it will help with making
the backend concurrent, as the append to the list of
oversized functions can be cheaply protected by a plain mutex.
Updates #15756
Updates #19250
Change-Id: Id3fa21906616d62e9dc66e27a17fd5f83304e96e
Reviewed-on: https://go-review.googlesource.com/38972
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
2017-03-30 11:46:45 -07:00
|
|
|
}
|
2017-10-10 17:43:41 -07:00
|
|
|
pp := newProgs(fn, worker)
|
2018-05-29 16:44:36 -07:00
|
|
|
defer pp.Free()
|
2017-10-10 17:43:41 -07:00
|
|
|
genssa(f, pp)
|
2018-05-29 16:44:36 -07:00
|
|
|
// Check frame size again.
|
|
|
|
|
// The check above included only the space needed for local variables.
|
|
|
|
|
// After genssa, the space needed includes local variables and the callee arg region.
|
|
|
|
|
// We must do this check prior to calling pp.Flush.
|
|
|
|
|
// If there are any oversized stack frames,
|
|
|
|
|
// the assembler may emit inscrutable complaints about invalid instructions.
|
|
|
|
|
if pp.Text.To.Offset >= maxStackSize {
|
|
|
|
|
largeStackFramesMu.Lock()
|
2018-10-25 06:23:54 -07:00
|
|
|
locals := f.Frontend().(*ssafn).stksize
|
|
|
|
|
largeStackFrames = append(largeStackFrames, largeStack{locals: locals, args: fn.Type.ArgWidth(), callee: pp.Text.To.Offset - locals, pos: fn.Pos})
|
2018-05-29 16:44:36 -07:00
|
|
|
largeStackFramesMu.Unlock()
|
|
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
2018-02-13 17:39:38 -05:00
|
|
|
pp.Flush() // assemble, fill in boilerplate, etc.
|
2017-04-19 14:20:52 -07:00
|
|
|
// fieldtrack must be called after pp.Flush. See issue 20014.
|
|
|
|
|
fieldtrack(pp.Text.From.Sym, fn.Func.FieldTrack)
|
2017-02-06 17:06:02 -08:00
|
|
|
}
|
|
|
|
|
|
2017-05-17 08:29:18 -07:00
|
|
|
func init() {
|
2019-05-20 15:07:01 +07:00
|
|
|
if race.Enabled {
|
2017-05-17 08:29:18 -07:00
|
|
|
rand.Seed(time.Now().UnixNano())
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
// compileFunctions compiles all functions in compilequeue.
|
|
|
|
|
// It fans out nBackendWorkers to do the work
|
|
|
|
|
// and waits for them to complete.
|
|
|
|
|
func compileFunctions() {
|
|
|
|
|
if len(compilequeue) != 0 {
|
2017-04-27 16:37:25 -07:00
|
|
|
sizeCalculationDisabled = true // not safe to calculate sizes concurrently
|
2019-05-20 15:07:01 +07:00
|
|
|
if race.Enabled {
|
2017-04-27 07:37:40 -07:00
|
|
|
// Randomize compilation order to try to shake out races.
|
|
|
|
|
tmp := make([]*Node, len(compilequeue))
|
|
|
|
|
perm := rand.Perm(len(compilequeue))
|
|
|
|
|
for i, v := range perm {
|
|
|
|
|
tmp[v] = compilequeue[i]
|
|
|
|
|
}
|
|
|
|
|
copy(compilequeue, tmp)
|
|
|
|
|
} else {
|
|
|
|
|
// Compile the longest functions first,
|
|
|
|
|
// since they're most likely to be the slowest.
|
|
|
|
|
// This helps avoid stragglers.
|
2019-04-30 15:23:14 -04:00
|
|
|
sort.Slice(compilequeue, func(i, j int) bool {
|
2017-04-27 07:37:40 -07:00
|
|
|
return compilequeue[i].Nbody.Len() > compilequeue[j].Nbody.Len()
|
|
|
|
|
})
|
|
|
|
|
}
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
var wg sync.WaitGroup
|
2017-10-06 11:32:28 -04:00
|
|
|
Ctxt.InParallel = true
|
2017-05-09 17:36:42 -07:00
|
|
|
c := make(chan *Node, nBackendWorkers)
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
for i := 0; i < nBackendWorkers; i++ {
|
|
|
|
|
wg.Add(1)
|
|
|
|
|
go func(worker int) {
|
|
|
|
|
for fn := range c {
|
|
|
|
|
compileSSA(fn, worker)
|
|
|
|
|
}
|
|
|
|
|
wg.Done()
|
|
|
|
|
}(i)
|
|
|
|
|
}
|
|
|
|
|
for _, fn := range compilequeue {
|
|
|
|
|
c <- fn
|
|
|
|
|
}
|
|
|
|
|
close(c)
|
|
|
|
|
compilequeue = nil
|
|
|
|
|
wg.Wait()
|
2017-10-06 11:32:28 -04:00
|
|
|
Ctxt.InParallel = false
|
2017-04-27 16:37:25 -07:00
|
|
|
sizeCalculationDisabled = false
|
cmd/compile: add initial backend concurrency support
This CL adds initial support for concurrent backend compilation.
BACKGROUND
The compiler currently consists (very roughly) of the following phases:
1. Initialization.
2. Lexing and parsing into the cmd/compile/internal/syntax AST.
3. Translation into the cmd/compile/internal/gc AST.
4. Some gc AST passes: typechecking, escape analysis, inlining,
closure handling, expression evaluation ordering (order.go),
and some lowering and optimization (walk.go).
5. Translation into the cmd/compile/internal/ssa SSA form.
6. Optimization and lowering of SSA form.
7. Translation from SSA form to assembler instructions.
8. Translation from assembler instructions to machine code.
9. Writing lots of output: machine code, DWARF symbols,
type and reflection info, export data.
Phase 2 was already concurrent as of Go 1.8.
Phase 3 is planned for eventual removal;
we hope to go straight from syntax AST to SSA.
Phases 5–8 are per-function; this CL adds support for
processing multiple functions concurrently.
The slowest phases in the compiler are 5 and 6,
so this offers the opportunity for some good speed-ups.
Unfortunately, it's not quite that straightforward.
In the current compiler, the latter parts of phase 4
(order, walk) are done function-at-a-time as needed.
Making order and walk concurrency-safe proved hard,
and they're not particularly slow, so there wasn't much reward.
To enable phases 5–8 to be done concurrently,
when concurrent backend compilation is requested,
we complete phase 4 for all functions
before starting later phases for any functions.
Also, in reality, we automatically generate new
functions in phase 9, such as method wrappers
and equality and has routines.
Those new functions then go through phases 4–8.
This CL disables concurrent backend compilation
after the first, big, user-provided batch of
functions has been compiled.
This is done to keep things simple,
and because the autogenerated functions
tend to be small, few, simple, and fast to compile.
USAGE
Concurrent backend compilation still defaults to off.
To set the number of functions that may be backend-compiled
concurrently, use the compiler flag -c.
In future work, cmd/go will automatically set -c.
Furthermore, this CL has been intentionally written
so that the c=1 path has no backend concurrency whatsoever,
not even spawning any goroutines.
This helps ensure that, should problems arise
late in the development cycle,
we can simply have cmd/go set c=1 always,
and revert to the original compiler behavior.
MUTEXES
Most of the work required to make concurrent backend
compilation safe has occurred over the past month.
This CL adds a handful of mutexes to get the rest of the way there;
they are the mutexes that I didn't see a clean way to avoid.
Some of them may still be eliminable in future work.
In no particular order:
* gc.funcsymsmu. The global funcsyms slice is populated
lazily when we need function symbols for closures.
This occurs during gc AST to SSA translation.
The function funcsym also does a package lookup,
which is a source of races on types.Pkg.Syms;
funcsymsmu also covers that package lookup.
This mutex is low priority: it adds a single global,
it is in an infrequently used code path, and it is low contention.
Since funcsyms may now be added in any order,
we must sort them to preserve reproducible builds.
* gc.largeStackFramesMu. We don't discover until after SSA compilation
that a function's stack frame is gigantic.
Recording that error happens basically never,
but it does happen concurrently.
Fix with a low priority mutex and sorting.
* obj.Link.hashmu. ctxt.hash stores the mapping from
types.Syms (compiler symbols) to obj.LSyms (linker symbols).
It is accessed fairly heavily through all the phases.
This is the only heavily contended mutex.
* gc.signatlistmu. The global signatlist map is
populated with types through several of the concurrent phases,
including notably via ngotype during DWARF generation.
It is low priority for removal.
* gc.typepkgmu. Looking up symbols in the types package
happens a fair amount during backend compilation
and DWARF generation, particularly via ngotype.
This mutex helps us to avoid a broader mutex on types.Pkg.Syms.
It has low-to-moderate contention.
* types.internedStringsmu. gc AST to SSA conversion and
some SSA work introduce new autotmps.
Those autotmps have their names interned to reduce allocations.
That interning requires protecting types.internedStrings.
The autotmp names are heavily re-used, and the mutex
overhead and contention here are low, so it is probably
a worthwhile performance optimization to keep this mutex.
TESTING
I have been testing this code locally by running
'go install -race cmd/compile'
and then doing
'go build -a -gcflags=-c=128 std cmd'
for all architectures and a variety of compiler flags.
This obviously needs to be made part of the builders,
but it is too expensive to make part of all.bash.
I have filed #19962 for this.
REPRODUCIBLE BUILDS
This version of the compiler generates reproducible builds.
Testing reproducible builds also needs automation, however,
and is also too expensive for all.bash.
This is #19961.
Also of note is that some of the compiler flags used by 'toolstash -cmp'
are currently incompatible with concurrent backend compilation.
They still work fine with c=1.
Time will tell whether this is a problem.
NEXT STEPS
* Continue to find and fix races and bugs,
using a combination of code inspection, fuzzing,
and hopefully some community experimentation.
I do not know of any outstanding races,
but there probably are some.
* Improve testing.
* Improve performance, for many values of c.
* Integrate with cmd/go and fine tune.
* Support concurrent compilation with the -race flag.
It is a sad irony that it does not yet work.
* Minor code cleanup that has been deferred during
the last month due to uncertainty about the
ultimate shape of this CL.
PERFORMANCE
Here's the buried lede, at last. :)
All benchmarks are from my 8 core 2.9 GHz Intel Core i7 darwin/amd64 laptop.
First, going from tip to this CL with c=1 has almost no impact.
name old time/op new time/op delta
Template 195ms ± 3% 194ms ± 5% ~ (p=0.370 n=30+29)
Unicode 86.6ms ± 3% 87.0ms ± 7% ~ (p=0.958 n=29+30)
GoTypes 548ms ± 3% 555ms ± 4% +1.35% (p=0.001 n=30+28)
Compiler 2.51s ± 2% 2.54s ± 2% +1.17% (p=0.000 n=28+30)
SSA 5.16s ± 3% 5.16s ± 2% ~ (p=0.910 n=30+29)
Flate 124ms ± 5% 124ms ± 4% ~ (p=0.947 n=30+30)
GoParser 146ms ± 3% 146ms ± 3% ~ (p=0.150 n=29+28)
Reflect 354ms ± 3% 352ms ± 4% ~ (p=0.096 n=29+29)
Tar 107ms ± 5% 106ms ± 3% ~ (p=0.370 n=30+29)
XML 200ms ± 4% 201ms ± 4% ~ (p=0.313 n=29+28)
[Geo mean] 332ms 333ms +0.10%
name old user-time/op new user-time/op delta
Template 227ms ± 5% 225ms ± 5% ~ (p=0.457 n=28+27)
Unicode 109ms ± 4% 109ms ± 5% ~ (p=0.758 n=29+29)
GoTypes 713ms ± 4% 721ms ± 5% ~ (p=0.051 n=30+29)
Compiler 3.36s ± 2% 3.38s ± 3% ~ (p=0.146 n=30+30)
SSA 7.46s ± 3% 7.47s ± 3% ~ (p=0.804 n=30+29)
Flate 146ms ± 7% 147ms ± 3% ~ (p=0.833 n=29+27)
GoParser 179ms ± 5% 179ms ± 5% ~ (p=0.866 n=30+30)
Reflect 431ms ± 4% 429ms ± 4% ~ (p=0.593 n=29+30)
Tar 124ms ± 5% 123ms ± 5% ~ (p=0.140 n=29+29)
XML 243ms ± 4% 242ms ± 7% ~ (p=0.404 n=29+29)
[Geo mean] 415ms 415ms +0.02%
name old obj-bytes new obj-bytes delta
Template 382k ± 0% 382k ± 0% ~ (all equal)
Unicode 203k ± 0% 203k ± 0% ~ (all equal)
GoTypes 1.18M ± 0% 1.18M ± 0% ~ (all equal)
Compiler 3.98M ± 0% 3.98M ± 0% ~ (all equal)
SSA 8.28M ± 0% 8.28M ± 0% ~ (all equal)
Flate 230k ± 0% 230k ± 0% ~ (all equal)
GoParser 287k ± 0% 287k ± 0% ~ (all equal)
Reflect 1.00M ± 0% 1.00M ± 0% ~ (all equal)
Tar 190k ± 0% 190k ± 0% ~ (all equal)
XML 416k ± 0% 416k ± 0% ~ (all equal)
[Geo mean] 660k 660k +0.00%
Comparing this CL to itself, from c=1 to c=2
improves real times 20-30%, costs 5-10% more CPU time,
and adds about 2% alloc.
The allocation increase comes from allocating more ssa.Caches.
name old time/op new time/op delta
Template 202ms ± 3% 149ms ± 3% -26.15% (p=0.000 n=49+49)
Unicode 87.4ms ± 4% 84.2ms ± 3% -3.68% (p=0.000 n=48+48)
GoTypes 560ms ± 2% 398ms ± 2% -28.96% (p=0.000 n=49+49)
Compiler 2.46s ± 3% 1.76s ± 2% -28.61% (p=0.000 n=48+46)
SSA 6.17s ± 2% 4.04s ± 1% -34.52% (p=0.000 n=49+49)
Flate 126ms ± 3% 92ms ± 2% -26.81% (p=0.000 n=49+48)
GoParser 148ms ± 4% 107ms ± 2% -27.78% (p=0.000 n=49+48)
Reflect 361ms ± 3% 281ms ± 3% -22.10% (p=0.000 n=49+49)
Tar 109ms ± 4% 86ms ± 3% -20.81% (p=0.000 n=49+47)
XML 204ms ± 3% 144ms ± 2% -29.53% (p=0.000 n=48+45)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 246ms ± 4% ~ (p=0.401 n=50+48)
Unicode 109ms ± 4% 111ms ± 4% +1.47% (p=0.000 n=44+50)
GoTypes 728ms ± 3% 765ms ± 3% +5.04% (p=0.000 n=46+50)
Compiler 3.33s ± 3% 3.41s ± 2% +2.31% (p=0.000 n=49+48)
SSA 8.52s ± 2% 9.11s ± 2% +6.93% (p=0.000 n=49+47)
Flate 149ms ± 4% 161ms ± 3% +8.13% (p=0.000 n=50+47)
GoParser 181ms ± 5% 192ms ± 2% +6.40% (p=0.000 n=49+46)
Reflect 452ms ± 9% 474ms ± 2% +4.99% (p=0.000 n=50+48)
Tar 126ms ± 6% 136ms ± 4% +7.95% (p=0.000 n=50+49)
XML 247ms ± 5% 264ms ± 3% +6.94% (p=0.000 n=48+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 39.3MB ± 0% +1.48% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.2MB ± 0% +1.19% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 114MB ± 0% +0.69% (p=0.008 n=5+5)
Compiler 443MB ± 0% 447MB ± 0% +0.95% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.26GB ± 0% +0.89% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 25.9MB ± 1% +2.35% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 32.2MB ± 0% +1.59% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 78.9MB ± 0% +0.91% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.0MB ± 0% +1.80% (p=0.008 n=5+5)
XML 42.4MB ± 0% 43.4MB ± 0% +2.35% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 379k ± 0% 378k ± 0% ~ (p=0.421 n=5+5)
Unicode 322k ± 0% 321k ± 0% ~ (p=0.222 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.548 n=5+5)
Compiler 4.12M ± 0% 4.11M ± 0% -0.14% (p=0.032 n=5+5)
SSA 9.72M ± 0% 9.72M ± 0% ~ (p=0.421 n=5+5)
Flate 234k ± 1% 234k ± 0% ~ (p=0.421 n=5+5)
GoParser 316k ± 1% 315k ± 0% ~ (p=0.222 n=5+5)
Reflect 980k ± 0% 979k ± 0% ~ (p=0.095 n=5+5)
Tar 249k ± 1% 249k ± 1% ~ (p=0.841 n=5+5)
XML 392k ± 0% 391k ± 0% ~ (p=0.095 n=5+5)
From c=1 to c=4, real time is down ~40%, CPU usage up 10-20%, alloc up ~5%:
name old time/op new time/op delta
Template 203ms ± 3% 131ms ± 5% -35.45% (p=0.000 n=50+50)
Unicode 87.2ms ± 4% 84.1ms ± 2% -3.61% (p=0.000 n=48+47)
GoTypes 560ms ± 4% 310ms ± 2% -44.65% (p=0.000 n=50+49)
Compiler 2.47s ± 3% 1.41s ± 2% -43.10% (p=0.000 n=50+46)
SSA 6.17s ± 2% 3.20s ± 2% -48.06% (p=0.000 n=49+49)
Flate 126ms ± 4% 74ms ± 2% -41.06% (p=0.000 n=49+48)
GoParser 148ms ± 4% 89ms ± 3% -39.97% (p=0.000 n=49+50)
Reflect 360ms ± 3% 242ms ± 3% -32.81% (p=0.000 n=49+49)
Tar 108ms ± 4% 73ms ± 4% -32.48% (p=0.000 n=50+49)
XML 203ms ± 3% 119ms ± 3% -41.56% (p=0.000 n=49+48)
name old user-time/op new user-time/op delta
Template 246ms ± 9% 287ms ± 9% +16.98% (p=0.000 n=50+50)
Unicode 109ms ± 4% 118ms ± 5% +7.56% (p=0.000 n=46+50)
GoTypes 735ms ± 4% 806ms ± 2% +9.62% (p=0.000 n=50+50)
Compiler 3.34s ± 4% 3.56s ± 2% +6.78% (p=0.000 n=49+49)
SSA 8.54s ± 3% 10.04s ± 3% +17.55% (p=0.000 n=50+50)
Flate 149ms ± 6% 176ms ± 3% +17.82% (p=0.000 n=50+48)
GoParser 181ms ± 5% 213ms ± 3% +17.47% (p=0.000 n=50+50)
Reflect 453ms ± 6% 499ms ± 2% +10.11% (p=0.000 n=50+48)
Tar 126ms ± 5% 149ms ±11% +18.76% (p=0.000 n=50+50)
XML 246ms ± 5% 287ms ± 4% +16.53% (p=0.000 n=49+50)
name old alloc/op new alloc/op delta
Template 38.8MB ± 0% 40.4MB ± 0% +4.21% (p=0.008 n=5+5)
Unicode 29.8MB ± 0% 30.9MB ± 0% +3.68% (p=0.008 n=5+5)
GoTypes 113MB ± 0% 116MB ± 0% +2.71% (p=0.008 n=5+5)
Compiler 443MB ± 0% 455MB ± 0% +2.75% (p=0.008 n=5+5)
SSA 1.25GB ± 0% 1.27GB ± 0% +1.84% (p=0.008 n=5+5)
Flate 25.3MB ± 0% 26.9MB ± 1% +6.31% (p=0.008 n=5+5)
GoParser 31.7MB ± 0% 33.2MB ± 0% +4.61% (p=0.008 n=5+5)
Reflect 78.2MB ± 0% 80.2MB ± 0% +2.53% (p=0.008 n=5+5)
Tar 26.6MB ± 0% 27.9MB ± 0% +5.19% (p=0.008 n=5+5)
XML 42.4MB ± 0% 44.6MB ± 0% +5.20% (p=0.008 n=5+5)
name old allocs/op new allocs/op delta
Template 380k ± 0% 379k ± 0% -0.39% (p=0.032 n=5+5)
Unicode 321k ± 0% 321k ± 0% ~ (p=0.841 n=5+5)
GoTypes 1.14M ± 0% 1.14M ± 0% ~ (p=0.421 n=5+5)
Compiler 4.12M ± 0% 4.14M ± 0% +0.52% (p=0.008 n=5+5)
SSA 9.72M ± 0% 9.76M ± 0% +0.37% (p=0.008 n=5+5)
Flate 234k ± 1% 234k ± 1% ~ (p=0.690 n=5+5)
GoParser 316k ± 0% 317k ± 1% ~ (p=0.841 n=5+5)
Reflect 981k ± 0% 981k ± 0% ~ (p=1.000 n=5+5)
Tar 250k ± 0% 249k ± 1% ~ (p=0.151 n=5+5)
XML 393k ± 0% 392k ± 0% ~ (p=0.056 n=5+5)
Going beyond c=4 on my machine tends to increase CPU time and allocs
without impacting real time.
The CPU time numbers matter, because when there are many concurrent
compilation processes, that will impact the overall throughput.
The numbers above are in many ways the best case scenario;
we can take full advantage of all cores.
Fortunately, the most common compilation scenario is incremental
re-compilation of a single package during a build/test cycle.
Updates #15756
Change-Id: I6725558ca2069edec0ac5b0d1683105a9fff6bea
Reviewed-on: https://go-review.googlesource.com/40693
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Robert Griesemer <gri@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-03-19 08:27:26 -07:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
2019-09-26 08:38:33 -04:00
|
|
|
func debuginfo(fnsym *obj.LSym, infosym *obj.LSym, curfn interface{}) ([]dwarf.Scope, dwarf.InlCalls) {
|
2017-03-23 16:39:04 -07:00
|
|
|
fn := curfn.(*Node)
|
2017-10-06 11:32:28 -04:00
|
|
|
if fn.Func.Nname != nil {
|
|
|
|
|
if expect := fn.Func.Nname.Sym.Linksym(); fnsym != expect {
|
|
|
|
|
Fatalf("unexpected fnsym: %v != %v", fnsym, expect)
|
|
|
|
|
}
|
2017-03-03 16:45:21 -08:00
|
|
|
}
|
|
|
|
|
|
2019-09-26 09:53:37 -04:00
|
|
|
var apdecls []*Node
|
|
|
|
|
// Populate decls for fn.
|
2017-03-23 16:39:04 -07:00
|
|
|
for _, n := range fn.Func.Dcl {
|
2015-02-13 14:40:36 -05:00
|
|
|
if n.Op != ONAME { // might be OTYPE or OLITERAL
|
|
|
|
|
continue
|
|
|
|
|
}
|
2017-04-25 18:14:12 -07:00
|
|
|
switch n.Class() {
|
2016-10-03 12:26:25 -07:00
|
|
|
case PAUTO:
|
2017-04-27 15:17:57 -07:00
|
|
|
if !n.Name.Used() {
|
2017-12-01 15:23:30 -05:00
|
|
|
// Text == nil -> generating abstract function
|
|
|
|
|
if fnsym.Func.Text != nil {
|
|
|
|
|
Fatalf("debuginfo unused node (AllocFrame should truncate fn.Func.Dcl)")
|
|
|
|
|
}
|
|
|
|
|
continue
|
2016-10-03 12:26:25 -07:00
|
|
|
}
|
|
|
|
|
case PPARAM, PPARAMOUT:
|
2017-02-06 17:06:02 -08:00
|
|
|
default:
|
|
|
|
|
continue
|
2015-02-13 14:40:36 -05:00
|
|
|
}
|
2019-09-26 09:53:37 -04:00
|
|
|
apdecls = append(apdecls, n)
|
|
|
|
|
fnsym.Func.RecordAutoType(ngotype(n).Linksym())
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
|
|
|
|
|
2019-09-26 09:53:37 -04:00
|
|
|
decls, dwarfVars := createDwarfVars(fnsym, fn.Func, apdecls)
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
|
2019-09-26 08:38:33 -04:00
|
|
|
// For each type referenced by the functions auto vars, attach a
|
|
|
|
|
// dummy relocation to the function symbol to insure that the type
|
|
|
|
|
// included in DWARF processing during linking.
|
|
|
|
|
typesyms := []*obj.LSym{}
|
|
|
|
|
for t, _ := range fnsym.Func.Autot {
|
|
|
|
|
typesyms = append(typesyms, t)
|
|
|
|
|
}
|
|
|
|
|
sort.Sort(obj.BySymName(typesyms))
|
|
|
|
|
for _, sym := range typesyms {
|
|
|
|
|
r := obj.Addrel(infosym)
|
|
|
|
|
r.Sym = sym
|
|
|
|
|
r.Type = objabi.R_USETYPE
|
|
|
|
|
}
|
|
|
|
|
fnsym.Func.Autot = nil
|
|
|
|
|
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
var varScopes []ScopeID
|
|
|
|
|
for _, decl := range decls {
|
2017-08-18 10:22:19 +02:00
|
|
|
pos := decl.Pos
|
|
|
|
|
if decl.Name.Defn != nil && (decl.Name.Captured() || decl.Name.Byval()) {
|
|
|
|
|
// It's not clear which position is correct for captured variables here:
|
|
|
|
|
// * decl.Pos is the wrong position for captured variables, in the inner
|
|
|
|
|
// function, but it is the right position in the outer function.
|
|
|
|
|
// * decl.Name.Defn is nil for captured variables that were arguments
|
|
|
|
|
// on the outer function, however the decl.Pos for those seems to be
|
|
|
|
|
// correct.
|
|
|
|
|
// * decl.Name.Defn is the "wrong" thing for variables declared in the
|
|
|
|
|
// header of a type switch, it's their position in the header, rather
|
|
|
|
|
// than the position of the case statement. In principle this is the
|
|
|
|
|
// right thing, but here we prefer the latter because it makes each
|
|
|
|
|
// instance of the header variable local to the lexical block of its
|
|
|
|
|
// case statement.
|
|
|
|
|
// This code is probably wrong for type switch variables that are also
|
|
|
|
|
// captured.
|
|
|
|
|
pos = decl.Name.Defn.Pos
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2017-08-18 10:22:19 +02:00
|
|
|
varScopes = append(varScopes, findScope(fn.Func.Marks, pos))
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2017-10-06 11:32:28 -04:00
|
|
|
|
|
|
|
|
scopes := assembleScopes(fnsym, fn, dwarfVars, varScopes)
|
|
|
|
|
var inlcalls dwarf.InlCalls
|
|
|
|
|
if genDwarfInline > 0 {
|
2018-03-01 15:36:04 +00:00
|
|
|
inlcalls = assembleInlines(fnsym, dwarfVars)
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
return scopes, inlcalls
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2017-02-06 17:06:02 -08:00
|
|
|
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
// createSimpleVars creates a DWARF entry for every variable declared in the
|
|
|
|
|
// function, claiming that they are permanently on the stack.
|
2019-09-26 09:53:37 -04:00
|
|
|
func createSimpleVars(apDecls []*Node) ([]*Node, []*dwarf.Var, map[*Node]bool) {
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
var vars []*dwarf.Var
|
|
|
|
|
var decls []*Node
|
2017-10-06 11:32:28 -04:00
|
|
|
selected := make(map[*Node]bool)
|
2019-09-26 09:53:37 -04:00
|
|
|
for _, n := range apDecls {
|
2017-03-07 10:57:55 -08:00
|
|
|
if n.IsAutoTmp() {
|
|
|
|
|
continue
|
|
|
|
|
}
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
|
2018-11-20 16:33:33 -05:00
|
|
|
decls = append(decls, n)
|
|
|
|
|
vars = append(vars, createSimpleVar(n))
|
|
|
|
|
selected[n] = true
|
|
|
|
|
}
|
|
|
|
|
return decls, vars, selected
|
|
|
|
|
}
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
|
2018-11-20 16:33:33 -05:00
|
|
|
func createSimpleVar(n *Node) *dwarf.Var {
|
|
|
|
|
var abbrev int
|
|
|
|
|
offs := n.Xoffset
|
|
|
|
|
|
|
|
|
|
switch n.Class() {
|
|
|
|
|
case PAUTO:
|
|
|
|
|
abbrev = dwarf.DW_ABRV_AUTO
|
|
|
|
|
if Ctxt.FixedFrameSize() == 0 {
|
|
|
|
|
offs -= int64(Widthptr)
|
|
|
|
|
}
|
|
|
|
|
if objabi.Framepointer_enabled(objabi.GOOS, objabi.GOARCH) || objabi.GOARCH == "arm64" {
|
|
|
|
|
// There is a word space for FP on ARM64 even if the frame pointer is disabled
|
|
|
|
|
offs -= int64(Widthptr)
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2017-03-07 10:57:55 -08:00
|
|
|
|
2018-11-20 16:33:33 -05:00
|
|
|
case PPARAM, PPARAMOUT:
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM
|
|
|
|
|
offs += Ctxt.FixedFrameSize()
|
|
|
|
|
default:
|
|
|
|
|
Fatalf("createSimpleVar unexpected class %v for node %v", n.Class(), n)
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typename := dwarf.InfoPrefix + typesymname(n.Type)
|
|
|
|
|
inlIndex := 0
|
|
|
|
|
if genDwarfInline > 1 {
|
|
|
|
|
if n.InlFormal() || n.InlLocal() {
|
|
|
|
|
inlIndex = posInlIndex(n.Pos) + 1
|
|
|
|
|
if n.InlFormal() {
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
}
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2018-11-20 16:33:33 -05:00
|
|
|
declpos := Ctxt.InnermostPos(n.Pos)
|
|
|
|
|
return &dwarf.Var{
|
|
|
|
|
Name: n.Sym.Name,
|
|
|
|
|
IsReturnValue: n.Class() == PPARAMOUT,
|
|
|
|
|
IsInlFormal: n.InlFormal(),
|
|
|
|
|
Abbrev: abbrev,
|
|
|
|
|
StackOffset: int32(offs),
|
|
|
|
|
Type: Ctxt.Lookup(typename),
|
|
|
|
|
DeclFile: declpos.RelFilename(),
|
|
|
|
|
DeclLine: declpos.RelLine(),
|
|
|
|
|
DeclCol: declpos.Col(),
|
|
|
|
|
InlIndex: int32(inlIndex),
|
|
|
|
|
ChildIndex: -1,
|
|
|
|
|
}
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
2017-03-06 07:32:37 -08:00
|
|
|
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
// createComplexVars creates recomposed DWARF vars with location lists,
|
|
|
|
|
// suitable for describing optimized code.
|
2018-03-01 15:36:04 +00:00
|
|
|
func createComplexVars(fn *Func) ([]*Node, []*dwarf.Var, map[*Node]bool) {
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
debugInfo := fn.DebugInfo
|
2017-03-06 07:32:37 -08:00
|
|
|
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
// Produce a DWARF variable entry for each user variable.
|
|
|
|
|
var decls []*Node
|
|
|
|
|
var vars []*dwarf.Var
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
ssaVars := make(map[*Node]bool)
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
|
2018-03-23 20:20:50 +00:00
|
|
|
for varID, dvar := range debugInfo.Vars {
|
|
|
|
|
n := dvar.(*Node)
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
ssaVars[n] = true
|
|
|
|
|
for _, slot := range debugInfo.VarSlots[varID] {
|
|
|
|
|
ssaVars[debugInfo.Slots[slot].N.(*Node)] = true
|
|
|
|
|
}
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
if dvar := createComplexVar(fn, ssa.VarID(varID)); dvar != nil {
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
decls = append(decls, n)
|
|
|
|
|
vars = append(vars, dvar)
|
|
|
|
|
}
|
|
|
|
|
}
|
2017-10-23 11:41:47 -04:00
|
|
|
|
2017-10-06 11:32:28 -04:00
|
|
|
return decls, vars, ssaVars
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
// createDwarfVars process fn, returning a list of DWARF variables and the
|
|
|
|
|
// Nodes they represent.
|
2019-09-26 09:53:37 -04:00
|
|
|
func createDwarfVars(fnsym *obj.LSym, fn *Func, apDecls []*Node) ([]*Node, []*dwarf.Var) {
|
2017-10-06 11:32:28 -04:00
|
|
|
// Collect a raw list of DWARF vars.
|
|
|
|
|
var vars []*dwarf.Var
|
|
|
|
|
var decls []*Node
|
|
|
|
|
var selected map[*Node]bool
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
if Ctxt.Flag_locationlists && Ctxt.Flag_optimize && fn.DebugInfo != nil {
|
2018-03-01 15:36:04 +00:00
|
|
|
decls, vars, selected = createComplexVars(fn)
|
2017-10-06 11:32:28 -04:00
|
|
|
} else {
|
2019-09-26 09:53:37 -04:00
|
|
|
decls, vars, selected = createSimpleVars(apDecls)
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
|
2019-09-26 09:53:37 -04:00
|
|
|
dcl := apDecls
|
2017-10-06 11:32:28 -04:00
|
|
|
if fnsym.WasInlined() {
|
2017-12-11 15:53:31 -05:00
|
|
|
dcl = preInliningDcls(fnsym)
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// If optimization is enabled, the list above will typically be
|
|
|
|
|
// missing some of the original pre-optimization variables in the
|
|
|
|
|
// function (they may have been promoted to registers, folded into
|
2018-11-20 16:33:33 -05:00
|
|
|
// constants, dead-coded away, etc). Input arguments not eligible
|
|
|
|
|
// for SSA optimization are also missing. Here we add back in entries
|
2017-10-06 11:32:28 -04:00
|
|
|
// for selected missing vars. Note that the recipe below creates a
|
|
|
|
|
// conservative location. The idea here is that we want to
|
|
|
|
|
// communicate to the user that "yes, there is a variable named X
|
|
|
|
|
// in this function, but no, I don't have enough information to
|
|
|
|
|
// reliably report its contents."
|
2018-11-20 16:33:33 -05:00
|
|
|
// For non-SSA-able arguments, however, the correct information
|
|
|
|
|
// is known -- they have a single home on the stack.
|
2017-10-06 11:32:28 -04:00
|
|
|
for _, n := range dcl {
|
|
|
|
|
if _, found := selected[n]; found {
|
2017-10-23 11:41:47 -04:00
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
c := n.Sym.Name[0]
|
2017-12-11 15:53:31 -05:00
|
|
|
if c == '.' || n.Type.IsUntyped() {
|
2017-10-23 11:41:47 -04:00
|
|
|
continue
|
|
|
|
|
}
|
2018-11-20 16:33:33 -05:00
|
|
|
if n.Class() == PPARAM && !canSSAType(n.Type) {
|
|
|
|
|
// SSA-able args get location lists, and may move in and
|
|
|
|
|
// out of registers, so those are handled elsewhere.
|
|
|
|
|
// Autos and named output params seem to get handled
|
|
|
|
|
// with VARDEF, which creates location lists.
|
|
|
|
|
// Args not of SSA-able type are treated here; they
|
|
|
|
|
// are homed on the stack in a single place for the
|
|
|
|
|
// entire call.
|
|
|
|
|
vars = append(vars, createSimpleVar(n))
|
|
|
|
|
decls = append(decls, n)
|
|
|
|
|
continue
|
|
|
|
|
}
|
2017-10-23 11:41:47 -04:00
|
|
|
typename := dwarf.InfoPrefix + typesymname(n.Type)
|
|
|
|
|
decls = append(decls, n)
|
|
|
|
|
abbrev := dwarf.DW_ABRV_AUTO_LOCLIST
|
2019-03-22 08:14:45 -04:00
|
|
|
isReturnValue := (n.Class() == PPARAMOUT)
|
2017-10-23 11:41:47 -04:00
|
|
|
if n.Class() == PPARAM || n.Class() == PPARAMOUT {
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM_LOCLIST
|
2019-03-22 08:14:45 -04:00
|
|
|
} else if n.Class() == PAUTOHEAP {
|
|
|
|
|
// If dcl in question has been promoted to heap, do a bit
|
|
|
|
|
// of extra work to recover original class (auto or param);
|
|
|
|
|
// see issue 30908. This insures that we get the proper
|
|
|
|
|
// signature in the abstract function DIE, but leaves a
|
|
|
|
|
// misleading location for the param (we want pointer-to-heap
|
|
|
|
|
// and not stack).
|
|
|
|
|
// TODO(thanm): generate a better location expression
|
|
|
|
|
stackcopy := n.Name.Param.Stackcopy
|
|
|
|
|
if stackcopy != nil && (stackcopy.Class() == PPARAM || stackcopy.Class() == PPARAMOUT) {
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM_LOCLIST
|
|
|
|
|
isReturnValue = (stackcopy.Class() == PPARAMOUT)
|
|
|
|
|
}
|
2017-10-23 11:41:47 -04:00
|
|
|
}
|
2017-10-06 11:32:28 -04:00
|
|
|
inlIndex := 0
|
|
|
|
|
if genDwarfInline > 1 {
|
|
|
|
|
if n.InlFormal() || n.InlLocal() {
|
|
|
|
|
inlIndex = posInlIndex(n.Pos) + 1
|
2017-12-11 15:53:31 -05:00
|
|
|
if n.InlFormal() {
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM_LOCLIST
|
|
|
|
|
}
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
declpos := Ctxt.InnermostPos(n.Pos)
|
2017-10-23 11:41:47 -04:00
|
|
|
vars = append(vars, &dwarf.Var{
|
|
|
|
|
Name: n.Sym.Name,
|
2019-03-22 08:14:45 -04:00
|
|
|
IsReturnValue: isReturnValue,
|
2017-10-23 11:41:47 -04:00
|
|
|
Abbrev: abbrev,
|
|
|
|
|
StackOffset: int32(n.Xoffset),
|
|
|
|
|
Type: Ctxt.Lookup(typename),
|
2018-02-06 09:36:13 -05:00
|
|
|
DeclFile: declpos.RelFilename(),
|
|
|
|
|
DeclLine: declpos.RelLine(),
|
2017-10-06 11:32:28 -04:00
|
|
|
DeclCol: declpos.Col(),
|
|
|
|
|
InlIndex: int32(inlIndex),
|
|
|
|
|
ChildIndex: -1,
|
2017-10-23 11:41:47 -04:00
|
|
|
})
|
2019-09-26 09:53:37 -04:00
|
|
|
// Record go type of to insure that it gets emitted by the linker.
|
|
|
|
|
fnsym.Func.RecordAutoType(ngotype(n).Linksym())
|
2017-10-23 11:41:47 -04:00
|
|
|
}
|
|
|
|
|
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
return decls, vars
|
|
|
|
|
}
|
|
|
|
|
|
2017-12-11 15:53:31 -05:00
|
|
|
// Given a function that was inlined at some point during the
|
|
|
|
|
// compilation, return a sorted list of nodes corresponding to the
|
|
|
|
|
// autos/locals in that function prior to inlining. If this is a
|
|
|
|
|
// function that is not local to the package being compiled, then the
|
|
|
|
|
// names of the variables may have been "versioned" to avoid conflicts
|
|
|
|
|
// with local vars; disregard this versioning when sorting.
|
|
|
|
|
func preInliningDcls(fnsym *obj.LSym) []*Node {
|
2017-10-06 11:32:28 -04:00
|
|
|
fn := Ctxt.DwFixups.GetPrecursorFunc(fnsym).(*Node)
|
2018-04-04 15:53:27 -07:00
|
|
|
var rdcl []*Node
|
|
|
|
|
for _, n := range fn.Func.Inl.Dcl {
|
2017-10-06 11:32:28 -04:00
|
|
|
c := n.Sym.Name[0]
|
2017-12-19 12:31:14 -08:00
|
|
|
// Avoid reporting "_" parameters, since if there are more than
|
2017-12-19 12:08:32 -05:00
|
|
|
// one, it can result in a collision later on, as in #23179.
|
|
|
|
|
if unversion(n.Sym.Name) == "_" || c == '.' || n.Type.IsUntyped() {
|
2017-10-06 11:32:28 -04:00
|
|
|
continue
|
|
|
|
|
}
|
|
|
|
|
rdcl = append(rdcl, n)
|
|
|
|
|
}
|
2017-12-11 15:53:31 -05:00
|
|
|
return rdcl
|
|
|
|
|
}
|
|
|
|
|
|
2017-09-29 14:14:03 -04:00
|
|
|
// stackOffset returns the stack location of a LocalSlot relative to the
|
|
|
|
|
// stack pointer, suitable for use in a DWARF location entry. This has nothing
|
|
|
|
|
// to do with its offset in the user variable.
|
2018-01-29 19:21:25 -05:00
|
|
|
func stackOffset(slot ssa.LocalSlot) int32 {
|
2017-09-29 14:14:03 -04:00
|
|
|
n := slot.N.(*Node)
|
|
|
|
|
var base int64
|
|
|
|
|
switch n.Class() {
|
|
|
|
|
case PAUTO:
|
|
|
|
|
if Ctxt.FixedFrameSize() == 0 {
|
|
|
|
|
base -= int64(Widthptr)
|
|
|
|
|
}
|
2018-08-29 14:55:03 +08:00
|
|
|
if objabi.Framepointer_enabled(objabi.GOOS, objabi.GOARCH) || objabi.GOARCH == "arm64" {
|
|
|
|
|
// There is a word space for FP on ARM64 even if the frame pointer is disabled
|
2017-09-29 14:14:03 -04:00
|
|
|
base -= int64(Widthptr)
|
|
|
|
|
}
|
|
|
|
|
case PPARAM, PPARAMOUT:
|
|
|
|
|
base += Ctxt.FixedFrameSize()
|
|
|
|
|
}
|
|
|
|
|
return int32(base + n.Xoffset + slot.Off)
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
// createComplexVar builds a single DWARF variable entry and location list.
|
|
|
|
|
func createComplexVar(fn *Func, varID ssa.VarID) *dwarf.Var {
|
|
|
|
|
debug := fn.DebugInfo
|
|
|
|
|
n := debug.Vars[varID].(*Node)
|
|
|
|
|
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
var abbrev int
|
|
|
|
|
switch n.Class() {
|
|
|
|
|
case PAUTO:
|
|
|
|
|
abbrev = dwarf.DW_ABRV_AUTO_LOCLIST
|
|
|
|
|
case PPARAM, PPARAMOUT:
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM_LOCLIST
|
|
|
|
|
default:
|
|
|
|
|
return nil
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
gotype := ngotype(n).Linksym()
|
|
|
|
|
typename := dwarf.InfoPrefix + gotype.Name[len("type."):]
|
2017-10-06 11:32:28 -04:00
|
|
|
inlIndex := 0
|
|
|
|
|
if genDwarfInline > 1 {
|
|
|
|
|
if n.InlFormal() || n.InlLocal() {
|
|
|
|
|
inlIndex = posInlIndex(n.Pos) + 1
|
2017-12-11 15:53:31 -05:00
|
|
|
if n.InlFormal() {
|
|
|
|
|
abbrev = dwarf.DW_ABRV_PARAM_LOCLIST
|
|
|
|
|
}
|
2017-10-06 11:32:28 -04:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
declpos := Ctxt.InnermostPos(n.Pos)
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
dvar := &dwarf.Var{
|
2017-10-06 11:32:28 -04:00
|
|
|
Name: n.Sym.Name,
|
|
|
|
|
IsReturnValue: n.Class() == PPARAMOUT,
|
|
|
|
|
IsInlFormal: n.InlFormal(),
|
|
|
|
|
Abbrev: abbrev,
|
|
|
|
|
Type: Ctxt.Lookup(typename),
|
2017-09-29 14:14:03 -04:00
|
|
|
// The stack offset is used as a sorting key, so for decomposed
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
// variables just give it the first one. It's not used otherwise.
|
2017-09-29 14:14:03 -04:00
|
|
|
// This won't work well if the first slot hasn't been assigned a stack
|
|
|
|
|
// location, but it's not obvious how to do better.
|
2018-02-05 17:04:44 -05:00
|
|
|
StackOffset: stackOffset(debug.Slots[debug.VarSlots[varID][0]]),
|
2018-02-06 09:36:13 -05:00
|
|
|
DeclFile: declpos.RelFilename(),
|
|
|
|
|
DeclLine: declpos.RelLine(),
|
2017-10-06 11:32:28 -04:00
|
|
|
DeclCol: declpos.Col(),
|
|
|
|
|
InlIndex: int32(inlIndex),
|
|
|
|
|
ChildIndex: -1,
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
cmd/compile: reimplement location list generation
Completely redesign and reimplement location list generation to be more
efficient, and hopefully not too hard to understand.
RegKills are gone. Instead of using the regalloc's liveness
calculations, redo them using the Ops' clobber information. Besides
saving a lot of Values, this avoids adding RegKills to blocks that would
be empty otherwise, which was messing up optimizations. This does mean
that it's much harder to tell whether the generation process is buggy
(there's nothing to cross-check it with), and there may be disagreements
with GC liveness. But the performance gain is significant, and it's nice
not to be messing with earlier compiler phases.
The intermediate representations are gone. Instead of producing
ssa.BlockDebugs, then dwarf.LocationLists, and then finally real
location lists, go directly from the SSA to a (mostly) real location
list. Because the SSA analysis happens before assembly, it stores
encoded block/value IDs where PCs would normally go. It would be easier
to do the SSA analysis after assembly, but I didn't want to retain the
SSA just for that.
Generation proceeds in two phases: first, it traverses the function in
CFG order, storing the state of the block at the beginning and end. End
states are used to produce the start states of the successor blocks. In
the second phase, it traverses in program text order and produces the
location lists. The processing in the second phase is redundant, but
much cheaper than storing the intermediate representation. It might be
possible to combine the two phases somewhat to take advantage of cases
where the CFG matches the block layout, but I haven't tried.
Location lists are finalized by adding a base address selection entry,
translating each encoded block/value ID to a real PC, and adding the
terminating zero entry. This probably won't work on OSX, where dsymutil
will choke on the base address selection. I tried emitting CU-relative
relocations for each address, and it was *very* bad for performance --
it uses more memory storing all the relocations than it does for the
actual location list bytes. I think I'm going to end up synthesizing the
relocations in the linker only on OSX, but TBD.
TestNexting needs updating: with more optimizations working, the
debugger doesn't stop on the continue (line 88) any more, and the test's
duplicate suppression kicks in. Also, dx and dy live a little longer
now, but they have the correct values.
Change-Id: Ie772dfe23a4e389ca573624fac4d05401ae32307
Reviewed-on: https://go-review.googlesource.com/89356
Run-TryBot: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
2017-10-26 15:40:17 -04:00
|
|
|
list := debug.LocationLists[varID]
|
|
|
|
|
if len(list) != 0 {
|
|
|
|
|
dvar.PutLocationList = func(listSym, startPC dwarf.Sym) {
|
|
|
|
|
debug.PutLocationList(list, Ctxt, listSym.(*obj.LSym), startPC.(*obj.LSym))
|
[dev.debug] cmd/compile: better DWARF with optimizations on
Debuggers use DWARF information to find local variables on the
stack and in registers. Prior to this CL, the DWARF information for
functions claimed that all variables were on the stack at all times.
That's incorrect when optimizations are enabled, and results in
debuggers showing data that is out of date or complete gibberish.
After this CL, the compiler is capable of representing variable
locations more accurately, and attempts to do so. Due to limitations of
the SSA backend, it's not possible to be completely correct.
There are a number of problems in the current design. One of the easier
to understand is that variable names currently must be attached to an
SSA value, but not all assignments in the source code actually result
in machine code. For example:
type myint int
var a int
b := myint(int)
and
b := (*uint64)(unsafe.Pointer(a))
don't generate machine code because the underlying representation is the
same, so the correct value of b will not be set when the user would
expect.
Generating the more precise debug information is behind a flag,
dwarflocationlists. Because of the issues described above, setting the
flag may not make the debugging experience much better, and may actually
make it worse in cases where the variable actually is on the stack and
the more complicated analysis doesn't realize it.
A number of changes are included:
- Add a new pseudo-instruction, RegKill, which indicates that the value
in the register has been clobbered.
- Adjust regalloc to emit RegKills in the right places. Significantly,
this means that phis are mixed with StoreReg and RegKills after
regalloc.
- Track variable decomposition in ssa.LocalSlots.
- After the SSA backend is done, analyze the result and build location
lists for each LocalSlot.
- After assembly is done, update the location lists with the assembled
PC offsets, recompose variables, and build DWARF location lists. Emit the
list as a new linker symbol, one per function.
- In the linker, aggregate the location lists into a .debug_loc section.
TODO:
- currently disabled for non-X86/AMD64 because there are no data tables.
go build -toolexec 'toolstash -cmp' -a std succeeds.
With -dwarflocationlists false:
before: f02812195637909ff675782c0b46836a8ff01976
after: 06f61e8112a42ac34fb80e0c818b3cdb84a5e7ec
benchstat -geomean /tmp/220352263 /tmp/621364410
completed 15 of 15, estimated time remaining 0s (eta 3:52PM)
name old time/op new time/op delta
Template 199ms ± 3% 198ms ± 2% ~ (p=0.400 n=15+14)
Unicode 96.6ms ± 5% 96.4ms ± 5% ~ (p=0.838 n=15+15)
GoTypes 653ms ± 2% 647ms ± 2% ~ (p=0.102 n=15+14)
Flate 133ms ± 6% 129ms ± 3% -2.62% (p=0.041 n=15+15)
GoParser 164ms ± 5% 159ms ± 3% -3.05% (p=0.000 n=15+15)
Reflect 428ms ± 4% 422ms ± 3% ~ (p=0.156 n=15+13)
Tar 123ms ±10% 124ms ± 8% ~ (p=0.461 n=15+15)
XML 228ms ± 3% 224ms ± 3% -1.57% (p=0.045 n=15+15)
[Geo mean] 206ms 377ms +82.86%
name old user-time/op new user-time/op delta
Template 292ms ±10% 301ms ±12% ~ (p=0.189 n=15+15)
Unicode 166ms ±37% 158ms ±14% ~ (p=0.418 n=15+14)
GoTypes 962ms ± 6% 963ms ± 7% ~ (p=0.976 n=15+15)
Flate 207ms ±19% 200ms ±14% ~ (p=0.345 n=14+15)
GoParser 246ms ±22% 240ms ±15% ~ (p=0.587 n=15+15)
Reflect 611ms ±13% 587ms ±14% ~ (p=0.085 n=15+13)
Tar 211ms ±12% 217ms ±14% ~ (p=0.355 n=14+15)
XML 335ms ±15% 320ms ±18% ~ (p=0.169 n=15+15)
[Geo mean] 317ms 583ms +83.72%
name old alloc/op new alloc/op delta
Template 40.2MB ± 0% 40.2MB ± 0% -0.15% (p=0.000 n=14+15)
Unicode 29.2MB ± 0% 29.3MB ± 0% ~ (p=0.624 n=15+15)
GoTypes 114MB ± 0% 114MB ± 0% -0.15% (p=0.000 n=15+14)
Flate 25.7MB ± 0% 25.6MB ± 0% -0.18% (p=0.000 n=13+15)
GoParser 32.2MB ± 0% 32.2MB ± 0% -0.14% (p=0.003 n=15+15)
Reflect 77.8MB ± 0% 77.9MB ± 0% ~ (p=0.061 n=15+15)
Tar 27.1MB ± 0% 27.0MB ± 0% -0.11% (p=0.029 n=15+15)
XML 42.7MB ± 0% 42.5MB ± 0% -0.29% (p=0.000 n=15+15)
[Geo mean] 42.1MB 75.0MB +78.05%
name old allocs/op new allocs/op delta
Template 402k ± 1% 398k ± 0% -0.91% (p=0.000 n=15+15)
Unicode 344k ± 1% 344k ± 0% ~ (p=0.715 n=15+14)
GoTypes 1.18M ± 0% 1.17M ± 0% -0.91% (p=0.000 n=15+14)
Flate 243k ± 0% 240k ± 1% -1.05% (p=0.000 n=13+15)
GoParser 327k ± 1% 324k ± 1% -0.96% (p=0.000 n=15+15)
Reflect 984k ± 1% 982k ± 0% ~ (p=0.050 n=15+15)
Tar 261k ± 1% 259k ± 1% -0.77% (p=0.000 n=15+15)
XML 411k ± 0% 404k ± 1% -1.55% (p=0.000 n=15+15)
[Geo mean] 439k 755k +72.01%
name old text-bytes new text-bytes delta
HelloSize 694kB ± 0% 694kB ± 0% -0.00% (p=0.000 n=15+15)
name old data-bytes new data-bytes delta
HelloSize 5.55kB ± 0% 5.55kB ± 0% ~ (all equal)
name old bss-bytes new bss-bytes delta
HelloSize 133kB ± 0% 133kB ± 0% ~ (all equal)
name old exe-bytes new exe-bytes delta
HelloSize 1.04MB ± 0% 1.04MB ± 0% ~ (all equal)
Change-Id: I991fc553ef175db46bb23b2128317bbd48de70d8
Reviewed-on: https://go-review.googlesource.com/41770
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2017-07-21 18:30:19 -04:00
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return dvar
|
2017-03-03 14:27:24 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// fieldtrack adds R_USEFIELD relocations to fnsym to record any
|
|
|
|
|
// struct fields that it used.
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
func fieldtrack(fnsym *obj.LSym, tracked map[*types.Sym]struct{}) {
|
2017-03-03 14:27:24 -08:00
|
|
|
if fnsym == nil {
|
|
|
|
|
return
|
|
|
|
|
}
|
2017-04-18 12:53:25 -07:00
|
|
|
if objabi.Fieldtrack_enabled == 0 || len(tracked) == 0 {
|
2017-03-03 14:27:24 -08:00
|
|
|
return
|
|
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
trackSyms := make([]*types.Sym, 0, len(tracked))
|
2017-03-03 14:27:24 -08:00
|
|
|
for sym := range tracked {
|
|
|
|
|
trackSyms = append(trackSyms, sym)
|
|
|
|
|
}
|
|
|
|
|
sort.Sort(symByName(trackSyms))
|
|
|
|
|
for _, sym := range trackSyms {
|
|
|
|
|
r := obj.Addrel(fnsym)
|
2017-04-21 07:51:41 -07:00
|
|
|
r.Sym = sym.Linksym()
|
2017-04-18 12:53:25 -07:00
|
|
|
r.Type = objabi.R_USEFIELD
|
2017-02-06 17:06:02 -08:00
|
|
|
}
|
2016-03-07 22:05:49 -08:00
|
|
|
}
|
|
|
|
|
|
cmd/compile: factor out Pkg, Sym, and Type into package types
- created new package cmd/compile/internal/types
- moved Pkg, Sym, Type to new package
- to break cycles, for now we need the (ugly) types/utils.go
file which contains a handful of functions that must be installed
early by the gc frontend
- to break cycles, for now we need two functions to convert between
*gc.Node and *types.Node (the latter is a dummy type)
- adjusted the gc's code to use the new package and the conversion
functions as needed
- made several Pkg, Sym, and Type methods functions as needed
- renamed constructors typ, typPtr, typArray, etc. to types.New,
types.NewPtr, types.NewArray, etc.
Passes toolstash-check -all.
Change-Id: I8adfa5e85c731645d0a7fd2030375ed6ebf54b72
Reviewed-on: https://go-review.googlesource.com/39855
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
2017-04-04 17:54:02 -07:00
|
|
|
type symByName []*types.Sym
|
2016-03-18 17:21:32 -07:00
|
|
|
|
|
|
|
|
func (a symByName) Len() int { return len(a) }
|
|
|
|
|
func (a symByName) Less(i, j int) bool { return a[i].Name < a[j].Name }
|
|
|
|
|
func (a symByName) Swap(i, j int) { a[i], a[j] = a[j], a[i] }
|