diff options
| author | Wang Xuerui <git@xen0n.name> | 2019-11-04 13:29:20 +0000 |
|---|---|---|
| committer | Brad Fitzpatrick <bradfitz@golang.org> | 2019-11-04 18:53:43 +0000 |
| commit | 210e3677f997fcce48e749437853203358f1b7b4 (patch) | |
| tree | 0a36c2bd7c03f557a99e2dfc54f48b970bf63ffb /src/runtime/sys_linux_mips64x.s | |
| parent | bf7e55b618ff4ec8e823b77c3c775d0fb4dba1ba (diff) | |
| download | go-210e3677f997fcce48e749437853203358f1b7b4.tar.xz | |
runtime: use vDSO clock_gettime on linux/mips64x
Speed up nanotime1 and walltime1 on MIPS64 with vDSO, just like the
other vDSO-enabled targets.
Benchmark numbers on Loongson 3A3000 (GOARCH=mips64le, 1.4GHz) against
current master:
benchmark old ns/op new ns/op delta
BenchmarkNow 868 293 -66.24%
BenchmarkNowUnixNano 851 296 -65.22%
Performance hit on fallback case, tested by using a wrong vDSO symbol name:
benchmark old ns/op new ns/op delta
BenchmarkNow 868 889 +2.42%
BenchmarkNowUnixNano 851 893 +4.94%
Change-Id: Ibfb48893cd060536359863ffee7624c00def646b
GitHub-Last-Rev: 03a58ac2e4e036a4f61227cfd013082871e92863
GitHub-Pull-Request: golang/go#35181
Reviewed-on: https://go-review.googlesource.com/c/go/+/203578
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Diffstat (limited to 'src/runtime/sys_linux_mips64x.s')
| -rw-r--r-- | src/runtime/sys_linux_mips64x.s | 78 |
1 files changed, 74 insertions, 4 deletions
diff --git a/src/runtime/sys_linux_mips64x.s b/src/runtime/sys_linux_mips64x.s index e4d02a3953..723cfe43d9 100644 --- a/src/runtime/sys_linux_mips64x.s +++ b/src/runtime/sys_linux_mips64x.s @@ -211,23 +211,88 @@ TEXT runtime·mincore(SB),NOSPLIT|NOFRAME,$0-28 // func walltime1() (sec int64, nsec int32) TEXT runtime·walltime1(SB),NOSPLIT,$16 + MOVV R29, R16 // R16 is unchanged by C code + MOVV R29, R1 + + MOVV g_m(g), R17 // R17 = m + + // Set vdsoPC and vdsoSP for SIGPROF traceback. + MOVV R31, m_vdsoPC(R17) + MOVV R29, m_vdsoSP(R17) + + MOVV m_curg(R17), R4 + MOVV g, R5 + BNE R4, R5, noswitch + + MOVV m_g0(R17), R4 + MOVV (g_sched+gobuf_sp)(R4), R1 // Set SP to g0 stack + +noswitch: + SUBV $16, R1 + AND $~15, R1 // Align for C code + MOVV R1, R29 + MOVW $0, R4 // CLOCK_REALTIME MOVV $0(R29), R5 - MOVV $SYS_clock_gettime, R2 - SYSCALL + + MOVV runtime·vdsoClockgettimeSym(SB), R25 + BEQ R25, fallback + + JAL (R25) + +finish: MOVV 0(R29), R3 // sec MOVV 8(R29), R5 // nsec + + MOVV R16, R29 // restore SP + MOVV R0, m_vdsoSP(R17) // clear vdsoSP + MOVV R3, sec+0(FP) MOVW R5, nsec+8(FP) RET +fallback: + MOVV $SYS_clock_gettime, R2 + SYSCALL + JMP finish + TEXT runtime·nanotime1(SB),NOSPLIT,$16 + MOVV R29, R16 // R16 is unchanged by C code + MOVV R29, R1 + + MOVV g_m(g), R17 // R17 = m + + // Set vdsoPC and vdsoSP for SIGPROF traceback. + MOVV R31, m_vdsoPC(R17) + MOVV R29, m_vdsoSP(R17) + + MOVV m_curg(R17), R4 + MOVV g, R5 + BNE R4, R5, noswitch + + MOVV m_g0(R17), R4 + MOVV (g_sched+gobuf_sp)(R4), R1 // Set SP to g0 stack + +noswitch: + SUBV $16, R1 + AND $~15, R1 // Align for C code + MOVV R1, R29 + MOVW $1, R4 // CLOCK_MONOTONIC MOVV $0(R29), R5 - MOVV $SYS_clock_gettime, R2 - SYSCALL + + MOVV runtime·vdsoClockgettimeSym(SB), R25 + BEQ R25, fallback + + JAL (R25) + +finish: MOVV 0(R29), R3 // sec MOVV 8(R29), R5 // nsec + + MOVV R16, R29 // restore SP + MOVV R0, m_vdsoSP(R17) // clear vdsoSP + // sec is in R3, nsec in R5 // return nsec in R3 MOVV $1000000000, R4 @@ -237,6 +302,11 @@ TEXT runtime·nanotime1(SB),NOSPLIT,$16 MOVV R3, ret+0(FP) RET +fallback: + MOVV $SYS_clock_gettime, R2 + SYSCALL + JMP finish + TEXT runtime·rtsigprocmask(SB),NOSPLIT|NOFRAME,$0-28 MOVW how+0(FP), R4 MOVV new+8(FP), R5 |