go/src/cmd/link/internal/ld/pcln.go

526 lines
13 KiB
Go
Raw Normal View History

// Copyright 2013 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 ld
import (
"cmd/internal/objabi"
cmd/internal/src: cache prefixed filenames CL 37234 introduced string concatenation into some hot code. This CL does that work earlier and caches the result. Updates #19386 Performance impact vs master: name old time/op new time/op delta Template 223ms ± 5% 216ms ± 5% -2.98% (p=0.001 n=20+20) Unicode 98.7ms ± 4% 99.0ms ± 4% ~ (p=0.749 n=20+19) GoTypes 631ms ± 4% 626ms ± 4% ~ (p=0.253 n=20+20) Compiler 2.91s ± 1% 2.87s ± 3% -1.11% (p=0.005 n=18+20) SSA 4.48s ± 2% 4.36s ± 2% -2.77% (p=0.000 n=20+20) Flate 130ms ± 2% 129ms ± 6% ~ (p=0.428 n=19+20) GoParser 160ms ± 4% 157ms ± 3% -1.62% (p=0.005 n=20+18) Reflect 395ms ± 2% 394ms ± 4% ~ (p=0.445 n=20+20) Tar 120ms ± 5% 118ms ± 6% ~ (p=0.101 n=19+20) XML 224ms ± 3% 223ms ± 3% ~ (p=0.544 n=19+19) name old user-ns/op new user-ns/op delta Template 291user-ms ± 5% 265user-ms ± 5% -9.02% (p=0.000 n=20+19) Unicode 140user-ms ± 3% 139user-ms ± 8% ~ (p=0.904 n=20+20) GoTypes 844user-ms ± 3% 849user-ms ± 3% ~ (p=0.251 n=20+18) Compiler 4.06user-s ± 5% 3.98user-s ± 2% ~ (p=0.056 n=20+20) SSA 6.89user-s ± 5% 6.50user-s ± 3% -5.61% (p=0.000 n=20+20) Flate 164user-ms ± 5% 163user-ms ± 4% ~ (p=0.365 n=20+19) GoParser 206user-ms ± 6% 204user-ms ± 4% ~ (p=0.534 n=20+18) Reflect 501user-ms ± 4% 505user-ms ± 5% ~ (p=0.383 n=20+20) Tar 151user-ms ± 3% 152user-ms ± 7% ~ (p=0.798 n=17+20) XML 283user-ms ± 7% 280user-ms ± 5% ~ (p=0.301 n=20+20) name old alloc/op new alloc/op delta Template 42.5MB ± 0% 40.2MB ± 0% -5.59% (p=0.000 n=20+20) Unicode 31.7MB ± 0% 31.0MB ± 0% -2.19% (p=0.000 n=20+18) GoTypes 124MB ± 0% 117MB ± 0% -5.90% (p=0.000 n=20+20) Compiler 533MB ± 0% 490MB ± 0% -8.07% (p=0.000 n=20+20) SSA 989MB ± 0% 893MB ± 0% -9.74% (p=0.000 n=20+20) Flate 27.8MB ± 0% 26.1MB ± 0% -5.92% (p=0.000 n=20+20) GoParser 34.3MB ± 0% 32.1MB ± 0% -6.43% (p=0.000 n=19+20) Reflect 84.6MB ± 0% 81.4MB ± 0% -3.84% (p=0.000 n=20+20) Tar 28.8MB ± 0% 27.7MB ± 0% -3.89% (p=0.000 n=20+20) XML 47.2MB ± 0% 44.2MB ± 0% -6.45% (p=0.000 n=20+19) name old allocs/op new allocs/op delta Template 420k ± 1% 381k ± 1% -9.35% (p=0.000 n=20+20) Unicode 338k ± 1% 324k ± 1% -4.29% (p=0.000 n=20+19) GoTypes 1.28M ± 0% 1.15M ± 0% -10.30% (p=0.000 n=20+20) Compiler 5.06M ± 0% 4.41M ± 0% -12.92% (p=0.000 n=20+20) SSA 9.14M ± 0% 7.91M ± 0% -13.46% (p=0.000 n=19+20) Flate 267k ± 0% 241k ± 1% -9.53% (p=0.000 n=20+20) GoParser 347k ± 1% 312k ± 0% -10.15% (p=0.000 n=19+20) Reflect 1.07M ± 0% 1.00M ± 0% -6.86% (p=0.000 n=20+20) Tar 274k ± 1% 256k ± 1% -6.73% (p=0.000 n=20+20) XML 448k ± 0% 398k ± 0% -11.17% (p=0.000 n=20+18) Performance impact when applied together with CL 37234 atop CL 37234's parent commit (i.e. as if it were a part of CL 37234), to show that this commit makes CL 37234 completely performance-neutral: name old time/op new time/op delta Template 222ms ±14% 222ms ±14% ~ (p=1.000 n=14+15) Unicode 104ms ±18% 106ms ±18% ~ (p=0.650 n=13+14) GoTypes 653ms ± 7% 638ms ± 5% ~ (p=0.145 n=14+12) Compiler 3.10s ± 1% 3.13s ±10% ~ (p=1.000 n=2+2) SSA 4.73s ±11% 4.68s ±11% ~ (p=0.567 n=15+15) Flate 136ms ± 4% 133ms ± 7% ~ (p=0.231 n=12+14) GoParser 163ms ±11% 169ms ±10% ~ (p=0.352 n=14+14) Reflect 415ms ±15% 423ms ±20% ~ (p=0.715 n=15+14) Tar 133ms ±17% 130ms ±23% ~ (p=0.252 n=14+15) XML 236ms ±16% 235ms ±14% ~ (p=0.874 n=14+14) name old user-ns/op new user-ns/op delta Template 271user-ms ±10% 271user-ms ±10% ~ (p=0.780 n=14+15) Unicode 143user-ms ± 5% 146user-ms ±11% ~ (p=0.432 n=12+14) GoTypes 864user-ms ± 5% 866user-ms ± 9% ~ (p=0.905 n=14+13) Compiler 4.17user-s ± 1% 4.26user-s ± 7% ~ (p=1.000 n=2+2) SSA 6.79user-s ± 8% 6.79user-s ± 6% ~ (p=0.902 n=15+15) Flate 169user-ms ± 8% 164user-ms ± 5% -3.13% (p=0.014 n=14+14) GoParser 212user-ms ± 7% 217user-ms ±22% ~ (p=1.000 n=13+15) Reflect 521user-ms ± 7% 533user-ms ±15% ~ (p=0.511 n=14+14) Tar 165user-ms ±17% 161user-ms ±15% ~ (p=0.345 n=15+15) XML 294user-ms ±11% 292user-ms ±10% ~ (p=0.839 n=14+14) name old alloc/op new alloc/op delta Template 39.9MB ± 0% 39.9MB ± 0% ~ (p=0.621 n=15+14) Unicode 31.0MB ± 0% 31.0MB ± 0% ~ (p=0.098 n=13+15) GoTypes 117MB ± 0% 117MB ± 0% ~ (p=0.775 n=15+15) Compiler 488MB ± 0% 488MB ± 0% ~ (p=0.333 n=2+2) SSA 892MB ± 0% 892MB ± 0% +0.03% (p=0.000 n=15+15) Flate 26.1MB ± 0% 26.1MB ± 0% ~ (p=0.098 n=15+15) GoParser 31.8MB ± 0% 31.8MB ± 0% ~ (p=0.525 n=15+13) Reflect 81.2MB ± 0% 81.2MB ± 0% +0.06% (p=0.001 n=12+14) Tar 27.5MB ± 0% 27.5MB ± 0% ~ (p=0.595 n=15+15) XML 44.1MB ± 0% 44.1MB ± 0% ~ (p=0.486 n=15+15) name old allocs/op new allocs/op delta Template 378k ± 1% 378k ± 0% ~ (p=0.949 n=15+14) Unicode 324k ± 0% 324k ± 1% ~ (p=0.057 n=14+15) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.461 n=15+15) Compiler 4.39M ± 0% 4.39M ± 0% ~ (p=0.333 n=2+2) SSA 7.90M ± 0% 7.90M ± 0% +0.06% (p=0.008 n=15+15) Flate 240k ± 1% 241k ± 0% ~ (p=0.233 n=15+15) GoParser 309k ± 1% 309k ± 0% ~ (p=0.867 n=15+12) Reflect 1.00M ± 0% 1.00M ± 0% ~ (p=0.139 n=12+15) Tar 254k ± 1% 253k ± 1% ~ (p=0.345 n=15+15) XML 398k ± 0% 397k ± 1% ~ (p=0.267 n=15+15) Change-Id: Ic999a0f456a371c99eebba0f9747263a13836e33 Reviewed-on: https://go-review.googlesource.com/37766 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-03-04 07:09:33 -08:00
"cmd/internal/src"
"cmd/link/internal/sym"
"log"
"os"
"path/filepath"
)
// iteration over encoded pcdata tables.
func getvarint(pp *[]byte) uint32 {
v := uint32(0)
p := *pp
for shift := 0; ; shift += 7 {
v |= uint32(p[0]&0x7F) << uint(shift)
tmp4 := p
p = p[1:]
if tmp4[0]&0x80 == 0 {
break
}
}
*pp = p
return v
}
func pciternext(it *Pciter) {
it.pc = it.nextpc
if it.done != 0 {
return
}
if -cap(it.p) >= -cap(it.d.P[len(it.d.P):]) {
it.done = 1
return
}
// value delta
v := getvarint(&it.p)
if v == 0 && it.start == 0 {
it.done = 1
return
}
it.start = 0
dv := int32(v>>1) ^ (int32(v<<31) >> 31)
it.value += dv
// pc delta
v = getvarint(&it.p)
it.nextpc = it.pc + v*it.pcscale
}
func pciterinit(ctxt *Link, it *Pciter, d *sym.Pcdata) {
it.d = *d
it.p = it.d.P
it.pc = 0
it.nextpc = 0
it.value = -1
it.start = 1
it.done = 0
it.pcscale = uint32(ctxt.Arch.MinLC)
pciternext(it)
}
func addvarint(d *sym.Pcdata, val uint32) {
n := int32(0)
for v := val; v >= 0x80; v >>= 7 {
n++
}
n++
old := len(d.P)
for cap(d.P) < len(d.P)+int(n) {
d.P = append(d.P[:cap(d.P)], 0)
}
d.P = d.P[:old+int(n)]
p := d.P[old:]
var v uint32
for v = val; v >= 0x80; v >>= 7 {
p[0] = byte(v | 0x80)
p = p[1:]
}
p[0] = byte(v)
}
func addpctab(ctxt *Link, ftab *sym.Symbol, off int32, d *sym.Pcdata) int32 {
var start int32
if len(d.P) > 0 {
start = int32(len(ftab.P))
ftab.AddBytes(d.P)
}
return int32(ftab.SetUint32(ctxt.Arch, int64(off), uint32(start)))
}
func ftabaddstring(ctxt *Link, ftab *sym.Symbol, s string) int32 {
n := int32(len(s)) + 1
start := int32(len(ftab.P))
ftab.Grow(int64(start) + int64(n) + 1)
copy(ftab.P[start:], s)
return start
}
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
// numberfile assigns a file number to the file if it hasn't been assigned already.
func numberfile(ctxt *Link, file *sym.Symbol) {
if file.Type != sym.SFILEPATH {
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
ctxt.Filesyms = append(ctxt.Filesyms, file)
file.Value = int64(len(ctxt.Filesyms))
file.Type = sym.SFILEPATH
cmd/internal/src: cache prefixed filenames CL 37234 introduced string concatenation into some hot code. This CL does that work earlier and caches the result. Updates #19386 Performance impact vs master: name old time/op new time/op delta Template 223ms ± 5% 216ms ± 5% -2.98% (p=0.001 n=20+20) Unicode 98.7ms ± 4% 99.0ms ± 4% ~ (p=0.749 n=20+19) GoTypes 631ms ± 4% 626ms ± 4% ~ (p=0.253 n=20+20) Compiler 2.91s ± 1% 2.87s ± 3% -1.11% (p=0.005 n=18+20) SSA 4.48s ± 2% 4.36s ± 2% -2.77% (p=0.000 n=20+20) Flate 130ms ± 2% 129ms ± 6% ~ (p=0.428 n=19+20) GoParser 160ms ± 4% 157ms ± 3% -1.62% (p=0.005 n=20+18) Reflect 395ms ± 2% 394ms ± 4% ~ (p=0.445 n=20+20) Tar 120ms ± 5% 118ms ± 6% ~ (p=0.101 n=19+20) XML 224ms ± 3% 223ms ± 3% ~ (p=0.544 n=19+19) name old user-ns/op new user-ns/op delta Template 291user-ms ± 5% 265user-ms ± 5% -9.02% (p=0.000 n=20+19) Unicode 140user-ms ± 3% 139user-ms ± 8% ~ (p=0.904 n=20+20) GoTypes 844user-ms ± 3% 849user-ms ± 3% ~ (p=0.251 n=20+18) Compiler 4.06user-s ± 5% 3.98user-s ± 2% ~ (p=0.056 n=20+20) SSA 6.89user-s ± 5% 6.50user-s ± 3% -5.61% (p=0.000 n=20+20) Flate 164user-ms ± 5% 163user-ms ± 4% ~ (p=0.365 n=20+19) GoParser 206user-ms ± 6% 204user-ms ± 4% ~ (p=0.534 n=20+18) Reflect 501user-ms ± 4% 505user-ms ± 5% ~ (p=0.383 n=20+20) Tar 151user-ms ± 3% 152user-ms ± 7% ~ (p=0.798 n=17+20) XML 283user-ms ± 7% 280user-ms ± 5% ~ (p=0.301 n=20+20) name old alloc/op new alloc/op delta Template 42.5MB ± 0% 40.2MB ± 0% -5.59% (p=0.000 n=20+20) Unicode 31.7MB ± 0% 31.0MB ± 0% -2.19% (p=0.000 n=20+18) GoTypes 124MB ± 0% 117MB ± 0% -5.90% (p=0.000 n=20+20) Compiler 533MB ± 0% 490MB ± 0% -8.07% (p=0.000 n=20+20) SSA 989MB ± 0% 893MB ± 0% -9.74% (p=0.000 n=20+20) Flate 27.8MB ± 0% 26.1MB ± 0% -5.92% (p=0.000 n=20+20) GoParser 34.3MB ± 0% 32.1MB ± 0% -6.43% (p=0.000 n=19+20) Reflect 84.6MB ± 0% 81.4MB ± 0% -3.84% (p=0.000 n=20+20) Tar 28.8MB ± 0% 27.7MB ± 0% -3.89% (p=0.000 n=20+20) XML 47.2MB ± 0% 44.2MB ± 0% -6.45% (p=0.000 n=20+19) name old allocs/op new allocs/op delta Template 420k ± 1% 381k ± 1% -9.35% (p=0.000 n=20+20) Unicode 338k ± 1% 324k ± 1% -4.29% (p=0.000 n=20+19) GoTypes 1.28M ± 0% 1.15M ± 0% -10.30% (p=0.000 n=20+20) Compiler 5.06M ± 0% 4.41M ± 0% -12.92% (p=0.000 n=20+20) SSA 9.14M ± 0% 7.91M ± 0% -13.46% (p=0.000 n=19+20) Flate 267k ± 0% 241k ± 1% -9.53% (p=0.000 n=20+20) GoParser 347k ± 1% 312k ± 0% -10.15% (p=0.000 n=19+20) Reflect 1.07M ± 0% 1.00M ± 0% -6.86% (p=0.000 n=20+20) Tar 274k ± 1% 256k ± 1% -6.73% (p=0.000 n=20+20) XML 448k ± 0% 398k ± 0% -11.17% (p=0.000 n=20+18) Performance impact when applied together with CL 37234 atop CL 37234's parent commit (i.e. as if it were a part of CL 37234), to show that this commit makes CL 37234 completely performance-neutral: name old time/op new time/op delta Template 222ms ±14% 222ms ±14% ~ (p=1.000 n=14+15) Unicode 104ms ±18% 106ms ±18% ~ (p=0.650 n=13+14) GoTypes 653ms ± 7% 638ms ± 5% ~ (p=0.145 n=14+12) Compiler 3.10s ± 1% 3.13s ±10% ~ (p=1.000 n=2+2) SSA 4.73s ±11% 4.68s ±11% ~ (p=0.567 n=15+15) Flate 136ms ± 4% 133ms ± 7% ~ (p=0.231 n=12+14) GoParser 163ms ±11% 169ms ±10% ~ (p=0.352 n=14+14) Reflect 415ms ±15% 423ms ±20% ~ (p=0.715 n=15+14) Tar 133ms ±17% 130ms ±23% ~ (p=0.252 n=14+15) XML 236ms ±16% 235ms ±14% ~ (p=0.874 n=14+14) name old user-ns/op new user-ns/op delta Template 271user-ms ±10% 271user-ms ±10% ~ (p=0.780 n=14+15) Unicode 143user-ms ± 5% 146user-ms ±11% ~ (p=0.432 n=12+14) GoTypes 864user-ms ± 5% 866user-ms ± 9% ~ (p=0.905 n=14+13) Compiler 4.17user-s ± 1% 4.26user-s ± 7% ~ (p=1.000 n=2+2) SSA 6.79user-s ± 8% 6.79user-s ± 6% ~ (p=0.902 n=15+15) Flate 169user-ms ± 8% 164user-ms ± 5% -3.13% (p=0.014 n=14+14) GoParser 212user-ms ± 7% 217user-ms ±22% ~ (p=1.000 n=13+15) Reflect 521user-ms ± 7% 533user-ms ±15% ~ (p=0.511 n=14+14) Tar 165user-ms ±17% 161user-ms ±15% ~ (p=0.345 n=15+15) XML 294user-ms ±11% 292user-ms ±10% ~ (p=0.839 n=14+14) name old alloc/op new alloc/op delta Template 39.9MB ± 0% 39.9MB ± 0% ~ (p=0.621 n=15+14) Unicode 31.0MB ± 0% 31.0MB ± 0% ~ (p=0.098 n=13+15) GoTypes 117MB ± 0% 117MB ± 0% ~ (p=0.775 n=15+15) Compiler 488MB ± 0% 488MB ± 0% ~ (p=0.333 n=2+2) SSA 892MB ± 0% 892MB ± 0% +0.03% (p=0.000 n=15+15) Flate 26.1MB ± 0% 26.1MB ± 0% ~ (p=0.098 n=15+15) GoParser 31.8MB ± 0% 31.8MB ± 0% ~ (p=0.525 n=15+13) Reflect 81.2MB ± 0% 81.2MB ± 0% +0.06% (p=0.001 n=12+14) Tar 27.5MB ± 0% 27.5MB ± 0% ~ (p=0.595 n=15+15) XML 44.1MB ± 0% 44.1MB ± 0% ~ (p=0.486 n=15+15) name old allocs/op new allocs/op delta Template 378k ± 1% 378k ± 0% ~ (p=0.949 n=15+14) Unicode 324k ± 0% 324k ± 1% ~ (p=0.057 n=14+15) GoTypes 1.15M ± 0% 1.15M ± 0% ~ (p=0.461 n=15+15) Compiler 4.39M ± 0% 4.39M ± 0% ~ (p=0.333 n=2+2) SSA 7.90M ± 0% 7.90M ± 0% +0.06% (p=0.008 n=15+15) Flate 240k ± 1% 241k ± 0% ~ (p=0.233 n=15+15) GoParser 309k ± 1% 309k ± 0% ~ (p=0.867 n=15+12) Reflect 1.00M ± 0% 1.00M ± 0% ~ (p=0.139 n=12+15) Tar 254k ± 1% 253k ± 1% ~ (p=0.345 n=15+15) XML 398k ± 0% 397k ± 1% ~ (p=0.267 n=15+15) Change-Id: Ic999a0f456a371c99eebba0f9747263a13836e33 Reviewed-on: https://go-review.googlesource.com/37766 Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2017-03-04 07:09:33 -08:00
path := file.Name[len(src.FileSymPrefix):]
file.Name = expandGoroot(path)
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
}
}
func renumberfiles(ctxt *Link, files []*sym.Symbol, d *sym.Pcdata) {
var f *sym.Symbol
// Give files numbers.
for i := 0; i < len(files); i++ {
f = files[i]
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
numberfile(ctxt, f)
}
newval := int32(-1)
var out sym.Pcdata
var it Pciter
for pciterinit(ctxt, &it, d); it.done == 0; pciternext(&it) {
// value delta
oldval := it.value
var val int32
if oldval == -1 {
val = -1
} else {
if oldval < 0 || oldval >= int32(len(files)) {
log.Fatalf("bad pcdata %d", oldval)
}
val = int32(files[oldval].Value)
}
dv := val - newval
newval = val
v := (uint32(dv) << 1) ^ uint32(dv>>31)
addvarint(&out, v)
// pc delta
addvarint(&out, (it.nextpc-it.pc)/it.pcscale)
}
// terminating value delta
addvarint(&out, 0)
*d = out
}
// onlycsymbol reports whether this is a cgo symbol provided by the
// runtime and only used from C code.
func onlycsymbol(s *sym.Symbol) bool {
switch s.Name {
case "_cgo_topofstack", "_cgo_panic", "crosscall2":
return true
}
return false
}
func emitPcln(ctxt *Link, s *sym.Symbol) bool {
if s == nil {
return true
}
if ctxt.BuildMode == BuildModePlugin && Headtype == objabi.Hdarwin && onlycsymbol(s) {
return false
}
// We want to generate func table entries only for the "lowest level" symbols,
// not containers of subsymbols.
if s.Type&sym.SCONTAINER != 0 {
return false
}
return true
}
// pclntab initializes the pclntab symbol with
// runtime function and file name information.
var pclntabZpcln sym.FuncInfo
// These variables are used to initialize runtime.firstmoduledata, see symtab.go:symtab.
var pclntabNfunc int32
var pclntabFiletabOffset int32
var pclntabPclntabOffset int32
var pclntabFirstFunc *sym.Symbol
var pclntabLastFunc *sym.Symbol
func (ctxt *Link) pclntab() {
funcdataBytes := int64(0)
ftab := ctxt.Syms.Lookup("runtime.pclntab", 0)
ftab.Type = sym.SPCLNTAB
ftab.Attr |= sym.AttrReachable
// See golang.org/s/go12symtab for the format. Briefly:
// 8-byte header
// nfunc [thearch.ptrsize bytes]
// function table, alternating PC and offset to func struct [each entry thearch.ptrsize bytes]
// end PC [thearch.ptrsize bytes]
// offset to file table [4 bytes]
nfunc := int32(0)
// Find container symbols, mark them with sym.SCONTAINER
for _, s := range ctxt.Textp {
if s.Outer != nil {
s.Outer.Type |= sym.SCONTAINER
}
}
for _, s := range ctxt.Textp {
if emitPcln(ctxt, s) {
nfunc++
}
}
pclntabNfunc = nfunc
ftab.Grow(8 + int64(ctxt.Arch.PtrSize) + int64(nfunc)*2*int64(ctxt.Arch.PtrSize) + int64(ctxt.Arch.PtrSize) + 4)
ftab.SetUint32(ctxt.Arch, 0, 0xfffffffb)
ftab.SetUint8(ctxt.Arch, 6, uint8(ctxt.Arch.MinLC))
ftab.SetUint8(ctxt.Arch, 7, uint8(ctxt.Arch.PtrSize))
ftab.SetUint(ctxt.Arch, 8, uint64(nfunc))
pclntabPclntabOffset = int32(8 + ctxt.Arch.PtrSize)
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
funcnameoff := make(map[string]int32)
nameToOffset := func(name string) int32 {
nameoff, ok := funcnameoff[name]
if !ok {
nameoff = ftabaddstring(ctxt, ftab, name)
funcnameoff[name] = nameoff
}
return nameoff
}
nfunc = 0
var last *sym.Symbol
for _, s := range ctxt.Textp {
last = s
if !emitPcln(ctxt, s) {
continue
}
pcln := s.FuncInfo
if pcln == nil {
pcln = &pclntabZpcln
}
if pclntabFirstFunc == nil {
pclntabFirstFunc = s
}
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
if len(pcln.InlTree) > 0 {
if len(pcln.Pcdata) <= objabi.PCDATA_InlTreeIndex {
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
// Create inlining pcdata table.
pcdata := make([]sym.Pcdata, objabi.PCDATA_InlTreeIndex+1)
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
copy(pcdata, pcln.Pcdata)
pcln.Pcdata = pcdata
}
if len(pcln.Funcdataoff) <= objabi.FUNCDATA_InlTree {
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
// Create inline tree funcdata.
funcdata := make([]*sym.Symbol, objabi.FUNCDATA_InlTree+1)
funcdataoff := make([]int64, objabi.FUNCDATA_InlTree+1)
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
copy(funcdata, pcln.Funcdata)
copy(funcdataoff, pcln.Funcdataoff)
pcln.Funcdata = funcdata
pcln.Funcdataoff = funcdataoff
}
}
funcstart := int32(len(ftab.P))
funcstart += int32(-len(ftab.P)) & (int32(ctxt.Arch.PtrSize) - 1)
ftab.SetAddr(ctxt.Arch, 8+int64(ctxt.Arch.PtrSize)+int64(nfunc)*2*int64(ctxt.Arch.PtrSize), s)
ftab.SetUint(ctxt.Arch, 8+int64(ctxt.Arch.PtrSize)+int64(nfunc)*2*int64(ctxt.Arch.PtrSize)+int64(ctxt.Arch.PtrSize), uint64(funcstart))
// Write runtime._func. Keep in sync with ../../../../runtime/runtime2.go:/_func
// and package debug/gosym.
// fixed size of struct, checked below
off := funcstart
end := funcstart + int32(ctxt.Arch.PtrSize) + 3*4 + 5*4 + int32(len(pcln.Pcdata))*4 + int32(len(pcln.Funcdata))*int32(ctxt.Arch.PtrSize)
if len(pcln.Funcdata) > 0 && (end&int32(ctxt.Arch.PtrSize-1) != 0) {
end += 4
}
ftab.Grow(int64(end))
// entry uintptr
off = int32(ftab.SetAddr(ctxt.Arch, int64(off), s))
// name int32
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
nameoff := nameToOffset(s.Name)
off = int32(ftab.SetUint32(ctxt.Arch, int64(off), uint32(nameoff)))
// args int32
// TODO: Move into funcinfo.
args := uint32(0)
if s.FuncInfo != nil {
args = uint32(s.FuncInfo.Args)
}
off = int32(ftab.SetUint32(ctxt.Arch, int64(off), args))
// frame int32
// This has been removed (it was never set quite correctly anyway).
// Nothing should use it.
// Leave an obviously incorrect value.
// TODO: Remove entirely.
off = int32(ftab.SetUint32(ctxt.Arch, int64(off), 0x1234567))
if pcln != &pclntabZpcln {
renumberfiles(ctxt, pcln.File, &pcln.Pcfile)
if false {
// Sanity check the new numbering
var it Pciter
for pciterinit(ctxt, &it, &pcln.Pcfile); it.done == 0; pciternext(&it) {
if it.value < 1 || it.value > int32(len(ctxt.Filesyms)) {
Errorf(s, "bad file number in pcfile: %d not in range [1, %d]\n", it.value, len(ctxt.Filesyms))
errorexit()
}
}
}
}
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
if len(pcln.InlTree) > 0 {
inlTreeSym := ctxt.Syms.Lookup("inltree."+s.Name, 0)
inlTreeSym.Type = sym.SRODATA
inlTreeSym.Attr |= sym.AttrReachable | sym.AttrDuplicateOK
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
for i, call := range pcln.InlTree {
// Usually, call.File is already numbered since the file
// shows up in the Pcfile table. However, two inlined calls
// might overlap exactly so that only the innermost file
// appears in the Pcfile table. In that case, this assigns
// the outer file a number.
numberfile(ctxt, call.File)
nameoff := nameToOffset(call.Func.Name)
inlTreeSym.SetUint32(ctxt.Arch, int64(i*16+0), uint32(call.Parent))
inlTreeSym.SetUint32(ctxt.Arch, int64(i*16+4), uint32(call.File.Value))
inlTreeSym.SetUint32(ctxt.Arch, int64(i*16+8), uint32(call.Line))
inlTreeSym.SetUint32(ctxt.Arch, int64(i*16+12), uint32(nameoff))
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
}
pcln.Funcdata[objabi.FUNCDATA_InlTree] = inlTreeSym
pcln.Pcdata[objabi.PCDATA_InlTreeIndex] = pcln.Pcinline
cmd/compile,link: generate PC-value tables with inlining information In order to generate accurate tracebacks, the runtime needs to know the inlined call stack for a given PC. This creates two tables per function for this purpose. The first table is the inlining tree (stored in the function's funcdata), which has a node containing the file, line, and function name for every inlined call. The second table is a PC-value table that maps each PC to a node in the inlining tree (or -1 if the PC is not the result of inlining). To give the appearance that inlining hasn't happened, the runtime also needs the original source position information of inlined AST nodes. Previously the compiler plastered over the line numbers of inlined AST nodes with the line number of the call. This meant that the PC-line table mapped each PC to line number of the outermost call in its inlined call stack, with no way to access the innermost line number. Now the compiler retains line numbers of inlined AST nodes and writes the innermost source position information to the PC-line and PC-file tables. Some tools and tests expect to see outermost line numbers, so we provide the OutermostLine function for displaying line info. To keep track of the inlined call stack for an AST node, we extend the src.PosBase type with an index into a global inlining tree. Every time the compiler inlines a call, it creates a node in the global inlining tree for the call, and writes its index to the PosBase of every inlined AST node. The parent of this node is the inlining tree index of the call. -1 signifies no parent. For each function, the compiler creates a local inlining tree and a PC-value table mapping each PC to an index in the local tree. These are written to an object file, which is read by the linker. The linker re-encodes these tables compactly by deduplicating function names and file names. This change increases the size of binaries by 4-5%. For example, this is how the go1 benchmark binary is impacted by this change: section old bytes new bytes delta .text 3.49M ± 0% 3.49M ± 0% +0.06% .rodata 1.12M ± 0% 1.21M ± 0% +8.21% .gopclntab 1.50M ± 0% 1.68M ± 0% +11.89% .debug_line 338k ± 0% 435k ± 0% +28.78% Total 9.21M ± 0% 9.58M ± 0% +4.01% Updates #19348. Change-Id: Ic4f180c3b516018138236b0c35e0218270d957d3 Reviewed-on: https://go-review.googlesource.com/37231 Run-TryBot: David Lazar <lazard@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Austin Clements <austin@google.com>
2017-02-17 12:28:05 -05:00
}
// pcdata
off = addpctab(ctxt, ftab, off, &pcln.Pcsp)
off = addpctab(ctxt, ftab, off, &pcln.Pcfile)
off = addpctab(ctxt, ftab, off, &pcln.Pcline)
off = int32(ftab.SetUint32(ctxt.Arch, int64(off), uint32(len(pcln.Pcdata))))
off = int32(ftab.SetUint32(ctxt.Arch, int64(off), uint32(len(pcln.Funcdata))))
for i := 0; i < len(pcln.Pcdata); i++ {
off = addpctab(ctxt, ftab, off, &pcln.Pcdata[i])
}
// funcdata, must be pointer-aligned and we're only int32-aligned.
// Missing funcdata will be 0 (nil pointer).
if len(pcln.Funcdata) > 0 {
if off&int32(ctxt.Arch.PtrSize-1) != 0 {
off += 4
}
for i := 0; i < len(pcln.Funcdata); i++ {
if pcln.Funcdata[i] == nil {
ftab.SetUint(ctxt.Arch, int64(off)+int64(ctxt.Arch.PtrSize)*int64(i), uint64(pcln.Funcdataoff[i]))
} else {
// TODO: Dedup.
funcdataBytes += pcln.Funcdata[i].Size
ftab.SetAddrPlus(ctxt.Arch, int64(off)+int64(ctxt.Arch.PtrSize)*int64(i), pcln.Funcdata[i], pcln.Funcdataoff[i])
}
}
off += int32(len(pcln.Funcdata)) * int32(ctxt.Arch.PtrSize)
}
if off != end {
Errorf(s, "bad math in functab: funcstart=%d off=%d but end=%d (npcdata=%d nfuncdata=%d ptrsize=%d)", funcstart, off, end, len(pcln.Pcdata), len(pcln.Funcdata), ctxt.Arch.PtrSize)
errorexit()
}
nfunc++
}
pclntabLastFunc = last
// Final entry of table is just end pc.
ftab.SetAddrPlus(ctxt.Arch, 8+int64(ctxt.Arch.PtrSize)+int64(nfunc)*2*int64(ctxt.Arch.PtrSize), last, last.Size)
// Start file table.
start := int32(len(ftab.P))
start += int32(-len(ftab.P)) & (int32(ctxt.Arch.PtrSize) - 1)
pclntabFiletabOffset = start
ftab.SetUint32(ctxt.Arch, 8+int64(ctxt.Arch.PtrSize)+int64(nfunc)*2*int64(ctxt.Arch.PtrSize)+int64(ctxt.Arch.PtrSize), uint32(start))
ftab.Grow(int64(start) + (int64(len(ctxt.Filesyms))+1)*4)
ftab.SetUint32(ctxt.Arch, int64(start), uint32(len(ctxt.Filesyms)+1))
for i := len(ctxt.Filesyms) - 1; i >= 0; i-- {
s := ctxt.Filesyms[i]
ftab.SetUint32(ctxt.Arch, int64(start)+s.Value*4, uint32(ftabaddstring(ctxt, ftab, s.Name)))
}
ftab.Size = int64(len(ftab.P))
if ctxt.Debugvlog != 0 {
ctxt.Logf("%5.2f pclntab=%d bytes, funcdata total %d bytes\n", Cputime(), ftab.Size, funcdataBytes)
}
}
func expandGoroot(s string) string {
const n = len("$GOROOT")
if len(s) >= n+1 && s[:n] == "$GOROOT" && (s[n] == '/' || s[n] == '\\') {
root := objabi.GOROOT
if final := os.Getenv("GOROOT_FINAL"); final != "" {
root = final
}
return filepath.ToSlash(filepath.Join(root, s[n:]))
}
return s
}
const (
BUCKETSIZE = 256 * MINFUNC
SUBBUCKETS = 16
SUBBUCKETSIZE = BUCKETSIZE / SUBBUCKETS
NOIDX = 0x7fffffff
)
// findfunctab generates a lookup table to quickly find the containing
// function for a pc. See src/runtime/symtab.go:findfunc for details.
func (ctxt *Link) findfunctab() {
t := ctxt.Syms.Lookup("runtime.findfunctab", 0)
t.Type = sym.SRODATA
t.Attr |= sym.AttrReachable
t.Attr |= sym.AttrLocal
// find min and max address
min := ctxt.Textp[0].Value
max := int64(0)
for _, s := range ctxt.Textp {
max = s.Value + s.Size
}
// for each subbucket, compute the minimum of all symbol indexes
// that map to that subbucket.
n := int32((max - min + SUBBUCKETSIZE - 1) / SUBBUCKETSIZE)
indexes := make([]int32, n)
for i := int32(0); i < n; i++ {
indexes[i] = NOIDX
}
idx := int32(0)
for i, s := range ctxt.Textp {
if !emitPcln(ctxt, s) {
continue
}
p := s.Value
var e *sym.Symbol
i++
if i < len(ctxt.Textp) {
e = ctxt.Textp[i]
}
for !emitPcln(ctxt, e) && i < len(ctxt.Textp) {
e = ctxt.Textp[i]
i++
}
q := max
if e != nil {
q = e.Value
}
//print("%d: [%lld %lld] %s\n", idx, p, q, s->name);
for ; p < q; p += SUBBUCKETSIZE {
i = int((p - min) / SUBBUCKETSIZE)
if indexes[i] > idx {
indexes[i] = idx
}
}
i = int((q - 1 - min) / SUBBUCKETSIZE)
if indexes[i] > idx {
indexes[i] = idx
}
idx++
}
// allocate table
nbuckets := int32((max - min + BUCKETSIZE - 1) / BUCKETSIZE)
t.Grow(4*int64(nbuckets) + int64(n))
// fill in table
for i := int32(0); i < nbuckets; i++ {
base := indexes[i*SUBBUCKETS]
if base == NOIDX {
Errorf(nil, "hole in findfunctab")
}
t.SetUint32(ctxt.Arch, int64(i)*(4+SUBBUCKETS), uint32(base))
for j := int32(0); j < SUBBUCKETS && i*SUBBUCKETS+j < n; j++ {
idx = indexes[i*SUBBUCKETS+j]
if idx == NOIDX {
Errorf(nil, "hole in findfunctab")
}
if idx-base >= 256 {
Errorf(nil, "too many functions in a findfunc bucket! %d/%d %d %d", i, nbuckets, j, idx-base)
}
t.SetUint8(ctxt.Arch, int64(i)*(4+SUBBUCKETS)+4+int64(j), uint8(idx-base))
}
}
}