Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
cmd/internal/obj/x86: make function prologue more predictable
Static branch prediction guesses that forward branches aren't taken. Since stacks are rarely grown, make the forward branch mean grow. Sample disassembly for func f() { _ = [128]byte{} } Before: TEXT main.f(SB) x.go x.go:3 0x2000 65488b0c25a0080000 GS MOVQ GS:0x8a0, CX x.go:3 0x2009 483b6110 CMPQ 0x10(CX), SP x.go:3 0x200d 7707 JA 0x2016 x.go:3 0x200f e88c410400 CALL runtime.morestack_noctxt(SB) x.go:3 0x2014 ebea JMP main.f(SB) x.go:3 0x2016 4881ec80000000 SUBQ $0x80, SP x.go:4 0x201d 488d3c24 LEAQ 0(SP), DI x.go:4 0x2021 31c0 XORL AX, AX x.go:4 0x2023 e8cc640400 CALL 0x484f4 x.go:5 0x2028 4881c480000000 ADDQ $0x80, SP x.go:5 0x202f c3 RET After: TEXT main.f(SB) x.go x.go:3 0x2000 65488b0c25a0080000 GS MOVQ GS:0x8a0, CX x.go:3 0x2009 483b6110 CMPQ 0x10(CX), SP x.go:3 0x200d 761a JBE 0x2029 x.go:3 0x200f 4881ec80000000 SUBQ $0x80, SP x.go:4 0x2016 488d3c24 LEAQ 0(SP), DI x.go:4 0x201a 31c0 XORL AX, AX x.go:4 0x201c e813740400 CALL 0x49434 x.go:5 0x2021 4881c480000000 ADDQ $0x80, SP x.go:5 0x2028 c3 RET x.go:3 0x2029 e8224f0400 CALL runtime.morestack_noctxt(SB) x.go:3 0x202e ebd0 JMP main.f(SB) Updates golang#10587. Sample benchmarks on a 2.8 GHz Intel Core i7: package sort name old mean new mean delta SearchWrappers 134ns × (0.99,1.01) 132ns × (0.99,1.01) -1.73% (p=0.000 n=15+14) SortString1K 215µs × (0.99,1.01) 213µs × (0.99,1.01) -0.61% (p=0.020 n=14+15) StableString1K 311µs × (0.99,1.02) 309µs × (0.99,1.02) ~ (p=0.077 n=14+15) SortInt1K 103µs × (0.99,1.02) 100µs × (0.98,1.01) -3.34% (p=0.000 n=15+15) StableInt1K 102µs × (0.99,1.01) 98µs × (0.97,1.04) -3.53% (p=0.000 n=15+15) SortInt64K 10.1ms × (0.98,1.02) 9.7ms × (0.99,1.01) -3.86% (p=0.000 n=14+15) StableInt64K 8.70ms × (0.99,1.01) 8.44ms × (0.99,1.03) -2.93% (p=0.000 n=14+15) Sort1e2 51.2µs × (1.00,1.01) 48.9µs × (0.99,1.02) -4.48% (p=0.000 n=13+15) Stable1e2 100µs × (0.99,1.02) 99µs × (0.99,1.01) -1.15% (p=0.000 n=14+13) Sort1e4 11.1ms × (0.99,1.02) 10.4ms × (0.99,1.01) -6.02% (p=0.000 n=15+14) Stable1e4 30.6ms × (0.99,1.01) 30.3ms × (0.99,1.02) -1.02% (p=0.001 n=15+14) Sort1e6 1.75s × (0.99,1.02) 1.66s × (0.98,1.03) -4.95% (p=0.000 n=14+15) Stable1e6 6.31s × (0.99,1.01) 6.26s × (0.99,1.01) -0.79% (p=0.002 n=15+15) package regexp name old mean new mean delta Literal 131ns × (0.99,1.01) 130ns × (0.99,1.03) -1.07% (p=0.004 n=14+15) NotLiteral 2.13µs × (0.99,1.01) 2.01µs × (0.99,1.03) -5.71% (p=0.000 n=14+14) MatchClass 3.15µs × (0.99,1.01) 3.04µs × (0.99,1.02) -3.40% (p=0.000 n=15+15) MatchClass_InRange 2.92µs × (0.99,1.01) 2.77µs × (0.99,1.02) -5.05% (p=0.000 n=13+15) ReplaceAll 2.17µs × (0.99,1.02) 2.06µs × (0.99,1.01) -5.19% (p=0.000 n=15+13) AnchoredLiteralShortNonMatch 116ns × (0.99,1.02) 113ns × (0.99,1.01) -2.75% (p=0.000 n=15+14) AnchoredLiteralLongNonMatch 125ns × (0.99,1.01) 127ns × (0.98,1.02) +1.49% (p=0.000 n=15+15) AnchoredShortMatch 178ns × (0.99,1.02) 175ns × (0.99,1.01) -1.62% (p=0.000 n=15+13) AnchoredLongMatch 328ns × (0.99,1.00) 341ns × (0.99,1.01) +3.73% (p=0.000 n=12+15) OnePassShortA 773ns × (0.99,1.02) 752ns × (0.99,1.01) -2.78% (p=0.000 n=15+13) NotOnePassShortA 794ns × (0.99,1.03) 780ns × (0.99,1.02) -1.75% (p=0.001 n=15+15) OnePassShortB 608ns × (0.99,1.01) 591ns × (0.99,1.02) -2.86% (p=0.000 n=15+14) NotOnePassShortB 576ns × (0.99,1.01) 571ns × (0.99,1.02) -0.74% (p=0.035 n=15+15) OnePassLongPrefix 131ns × (0.99,1.02) 130ns × (0.99,1.02) -1.32% (p=0.003 n=15+15) OnePassLongNotPrefix 503ns × (0.99,1.02) 481ns × (0.99,1.01) -4.34% (p=0.000 n=15+13) MatchEasy0_32 102ns × (0.98,1.01) 101ns × (0.99,1.02) ~ (p=0.907 n=15+14) MatchEasy0_1K 617ns × (0.99,1.02) 634ns × (0.98,1.02) +2.77% (p=0.000 n=15+15) MatchEasy0_32K 10.9µs × (0.99,1.01) 11.1µs × (0.99,1.01) +1.59% (p=0.000 n=15+15) MatchEasy0_1M 406µs × (0.99,1.02) 410µs × (0.99,1.02) +1.01% (p=0.000 n=14+15) MatchEasy0_32M 13.4ms × (0.99,1.01) 13.7ms × (0.99,1.02) +1.64% (p=0.000 n=12+15) MatchEasy1_32 83.7ns × (0.98,1.02) 83.0ns × (0.98,1.02) ~ (p=0.190 n=15+15) MatchEasy1_1K 1.46µs × (0.99,1.02) 1.39µs × (0.99,1.02) -4.83% (p=0.000 n=15+15) MatchEasy1_32K 49.4µs × (0.99,1.01) 49.4µs × (0.99,1.01) ~ (p=0.205 n=15+15) MatchEasy1_1M 1.72ms × (0.99,1.02) 1.75ms × (0.99,1.01) +1.34% (p=0.000 n=15+15) MatchEasy1_32M 55.5ms × (0.99,1.01) 56.1ms × (0.99,1.02) +1.10% (p=0.002 n=15+15) MatchMedium_32 1.37µs × (0.99,1.04) 1.33µs × (0.99,1.01) -2.87% (p=0.000 n=15+15) MatchMedium_1K 41.1µs × (0.99,1.02) 40.4µs × (0.99,1.02) -1.59% (p=0.000 n=15+15) MatchMedium_32K 1.71ms × (0.99,1.01) 1.75ms × (0.99,1.02) +2.36% (p=0.000 n=14+15) MatchMedium_1M 54.5ms × (0.99,1.01) 56.1ms × (0.99,1.01) +2.94% (p=0.000 n=13+15) MatchMedium_32M 1.75s × (0.99,1.01) 1.80s × (0.99,1.01) +2.77% (p=0.000 n=15+15) MatchHard_32 2.12µs × (0.99,1.02) 2.06µs × (0.99,1.01) -2.60% (p=0.000 n=15+14) MatchHard_1K 64.4µs × (0.98,1.02) 62.2µs × (0.99,1.01) -3.33% (p=0.000 n=15+15) MatchHard_32K 2.74ms × (0.99,1.01) 2.75ms × (0.99,1.01) ~ (p=0.310 n=15+14) MatchHard_1M 87.1ms × (0.99,1.02) 88.2ms × (0.99,1.01) +1.36% (p=0.000 n=14+15) MatchHard_32M 2.79s × (0.99,1.02) 2.83s × (0.99,1.02) +1.26% (p=0.004 n=15+14) go1 benchmarks name old time/op new time/op delta BinaryTree17 3.34s ± 3% 3.28s ± 2% -1.86% (p=0.000 n=67+66) Fannkuch11 2.50s ± 1% 2.51s ± 1% +0.24% (p=0.016 n=63+66) FmtFprintfEmpty 50.3ns ± 1% 50.2ns ± 2% -0.30% (p=0.001 n=62+67) FmtFprintfString 178ns ± 1% 166ns ± 1% -7.10% (p=0.000 n=62+59) FmtFprintfInt 168ns ± 1% 161ns ± 2% -4.41% (p=0.000 n=66+64) FmtFprintfIntInt 292ns ± 1% 282ns ± 2% -3.55% (p=0.000 n=62+60) FmtFprintfPrefixedInt 245ns ± 2% 239ns ± 2% -2.24% (p=0.000 n=66+65) FmtFprintfFloat 338ns ± 2% 326ns ± 1% -3.42% (p=0.000 n=64+59) FmtManyArgs 1.14µs ± 1% 1.10µs ± 2% -3.55% (p=0.000 n=62+62) GobDecode 8.88ms ± 2% 8.74ms ± 1% -1.55% (p=0.000 n=66+62) GobEncode 6.84ms ± 2% 6.61ms ± 2% -3.32% (p=0.000 n=61+67) Gzip 356ms ± 2% 352ms ± 2% -1.07% (p=0.000 n=67+66) Gunzip 90.6ms ± 2% 89.8ms ± 1% -0.83% (p=0.000 n=65+64) HTTPClientServer 82.6µs ± 2% 82.5µs ± 2% ~ (p=0.832 n=65+63) JSONEncode 17.5ms ± 2% 16.8ms ± 2% -3.77% (p=0.000 n=63+63) JSONDecode 63.3ms ± 2% 59.0ms ± 2% -6.85% (p=0.000 n=64+63) Mandelbrot200 3.85ms ± 1% 3.85ms ± 1% ~ (p=0.127 n=65+62) GoParse 3.75ms ± 2% 3.66ms ± 2% -2.39% (p=0.000 n=66+64) RegexpMatchEasy0_32 100ns ± 2% 100ns ± 1% -0.65% (p=0.000 n=62+64) RegexpMatchEasy0_1K 342ns ± 1% 341ns ± 1% -0.43% (p=0.000 n=65+64) RegexpMatchEasy1_32 82.8ns ± 2% 82.8ns ± 2% ~ (p=0.977 n=63+64) RegexpMatchEasy1_1K 511ns ± 2% 506ns ± 2% -1.01% (p=0.000 n=63+64) RegexpMatchMedium_32 139ns ± 1% 134ns ± 3% -3.27% (p=0.000 n=59+60) RegexpMatchMedium_1K 41.8µs ± 2% 40.5µs ± 2% -3.05% (p=0.000 n=62+64) RegexpMatchHard_32 2.13µs ± 1% 2.09µs ± 1% -2.22% (p=0.000 n=60+65) RegexpMatchHard_1K 64.4µs ± 3% 62.8µs ± 2% -2.58% (p=0.000 n=65+59) Revcomp 531ms ± 2% 529ms ± 1% -0.28% (p=0.022 n=61+61) Template 73.2ms ± 1% 73.1ms ± 1% ~ (p=0.794 n=66+63) TimeParse 369ns ± 1% 352ns ± 1% -4.68% (p=0.000 n=65+66) TimeFormat 374ns ± 2% 348ns ± 2% -7.01% (p=0.000 n=66+64) Change-Id: Ib190b5bb48a3e9087711d9e3383621d3103dd342 Reviewed-on: https://go-review.googlesource.com/10367 Reviewed-by: Russ Cox <[email protected]>
- Loading branch information