diff options
author | mtklein <mtklein@chromium.org> | 2015-09-21 10:50:56 -0700 |
---|---|---|
committer | Commit bot <commit-bot@chromium.org> | 2015-09-21 10:50:56 -0700 |
commit | 46243a7c02a1d5116e55a27ff59218f9c320df97 (patch) | |
tree | 1e18dc3cbab5663c61b8e9465d322166f015face /include/core | |
parent | 4ae26258ef1705faa303eb9eb6b779695b9143f8 (diff) |
Combined approach.
This combines some ideas from these two CLs:
- try stosd/w
- update memset16/32 inlining heuristics
BUG=skia:4316
Blinking in and out for perf.skia.org.
TBR=reed@google.com
Review URL: https://codereview.chromium.org/1356133002
Diffstat (limited to 'include/core')
-rw-r--r-- | include/core/SkUtils.h | 42 |
1 files changed, 29 insertions, 13 deletions
diff --git a/include/core/SkUtils.h b/include/core/SkUtils.h index 4e24bd0883..652cdae36e 100644 --- a/include/core/SkUtils.h +++ b/include/core/SkUtils.h @@ -9,6 +9,27 @@ #define SkUtils_DEFINED #include "SkTypes.h" +#if defined(SK_BUILD_FOR_WIN) + #include <intrin.h> +#endif + +#if defined(SK_CPU_X86) + static inline void rep_stosw(uint16_t buffer[], uint16_t value, int count) { + #if defined(SK_BUILD_FOR_WIN) + __stosw(buffer, value, count); + #else + __asm__ __volatile__ ( "rep stosw" : "+D"(buffer), "+c"(count) : "a"(value) ); + #endif + } + + static inline void rep_stosd(uint32_t buffer[], uint32_t value, int count) { + #if defined(SK_BUILD_FOR_WIN) + __stosd((PDWORD)buffer, value, count); + #else + __asm__ __volatile__ ( "rep stosl" : "+D"(buffer), "+c"(count) : "a"(value) ); + #endif + } +#endif namespace SkOpts { extern void (*memset16)(uint16_t[], uint16_t, int); @@ -17,13 +38,8 @@ namespace SkOpts { /////////////////////////////////////////////////////////////////////////////// -// The inlining heuristics below were determined using bench/MemsetBench.cpp -// on a x86 desktop, a Nexus 7 with and without NEON, and a Nexus 9: -// - on x86, inlining was never faster, -// - on ARMv7, inlining was faster for N<=10. Putting this check inside the NEON -// code was not helpful; it's got to be here outside. -// - NEON code generation for ARMv8 with GCC 4.9 is terrible, -// making the NEON code ~8x slower that just a serial loop. +// The stosw/d and inlining heuristics below were determined using +// bench/MemsetBench.cpp and perf.skia.org. /** Similar to memset(), but it assigns a 16bit value into the buffer. @param buffer The memory to have value copied into it @@ -31,10 +47,10 @@ namespace SkOpts { @param count The number of times value should be copied into the buffer. */ static inline void sk_memset16(uint16_t buffer[], uint16_t value, int count) { -#if defined(SK_CPU_ARM64) +#if defined(SK_CPU_X86) + if (count > 30) { rep_stosw(buffer, value, count); return; } +#elif defined(SK_ARM_HAS_NEON) while (count --> 0) { *buffer++ = value; } return; -#elif defined(SK_CPU_ARM32) - if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; } #endif SkOpts::memset16(buffer, value, count); } @@ -45,10 +61,10 @@ static inline void sk_memset16(uint16_t buffer[], uint16_t value, int count) { @param count The number of times value should be copied into the buffer. */ static inline void sk_memset32(uint32_t buffer[], uint32_t value, int count) { -#if defined(SK_CPU_ARM64) +#if defined(SK_CPU_X86) + if (count > 30) { rep_stosd(buffer, value, count); return; } +#elif defined(SK_ARM_HAS_NEON) while (count --> 0) { *buffer++ = value; } return; -#elif defined(SK_CPU_ARM32) - if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; } #endif SkOpts::memset32(buffer, value, count); } |