rt/snapraid
1
0
Fork 0
mirror of https://github.com/amadvance/snapraid.git synced 2026-10-26 10:51:54 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Use an implicit clobber list for asm snipsets

Browse source 
There is no speed gain in a custom clobber list.
This commit is contained in:
Andrea Mazzoleni 2015-03-09 22:25:29 +01:00
parent 54df8acd06
commit 763dccaaed
3 changed files with 36 additions and 126 deletions
Download patch file Download diff file Expand all files Collapse all files

58
raid/internal.h
View file

@ -195,20 +195,15 @@ static __always_inline void raid_asm_begin(void)
{
}
static __always_inline void raid_asm_end(void)
#ifdef CONFIG_SSE2
static __always_inline void raid_asm_sse_end(void)
{
/* SSE2 and AVX2 code uses non-temporal writes, like MOVNTDQ, */
/* SSE and AVX code uses non-temporal writes, like MOVNTDQ, */
/* that use a weak memory model. To ensure that other processors */
/* see correctly the data written, we use a store-store memory */
/* barrier at the end of the asm code */
#ifdef CONFIG_SSE2
asm volatile ("sfence" : : : "memory");
#endif
}
#ifdef CONFIG_SSE2
static __always_inline void raid_asm_clobber_xmm4(void)
{
/* clobbers registers used in the asm code */
/* this is required because in the Windows ABI, */
/* registers xmm6-xmm15 should be kept by the callee. */
@ -218,60 +213,27 @@ static __always_inline void raid_asm_clobber_xmm4(void)
/* compiler supports SSE2 registers in the clobber list */
#ifdef __SSE2__
asm volatile ("" : : : "%xmm0", "%xmm1", "%xmm2", "%xmm3");
#endif
}
#endif
#ifdef CONFIG_SSE2
static __always_inline void raid_asm_clobber_xmm8(void)
{
raid_asm_clobber_xmm4();
#ifdef __SSE2__
asm volatile ("" : : : "%xmm4", "%xmm5", "%xmm6", "%xmm7");
#ifdef CONFIG_X86_64
asm volatile ("" : : : "%xmm8", "%xmm9", "%xmm10", "%xmm11");
asm volatile ("" : : : "%xmm12", "%xmm13", "%xmm14", "%xmm15");
#endif
#endif
}
#endif
#ifdef CONFIG_AVX2
static __always_inline void raid_asm_clobber_ymm4(void)
static __always_inline void raid_asm_avx_end(void)
{
raid_asm_clobber_xmm4();
raid_asm_sse_end();
/* reset the upper part of the ymm registers */
/* to avoid the 70 clocks penality on the next */
/* xmm register use */
asm volatile ("vzeroupper" : : : "memory");
}
#endif
#ifdef CONFIG_AVX2
static __always_inline void raid_asm_clobber_ymm8(void)
{
raid_asm_clobber_xmm8();
asm volatile ("vzeroupper" : : : "memory");
}
#endif
#endif /* CONFIG_X86 */
#ifdef CONFIG_X86_64
#ifdef CONFIG_SSE2
static __always_inline void raid_asm_clobber_xmm16(void)
{
raid_asm_clobber_xmm8();
#ifdef __SSE2__
asm volatile ("" : : : "%xmm8", "%xmm9", "%xmm10", "%xmm11");
asm volatile ("" : : : "%xmm12", "%xmm13", "%xmm14", "%xmm15");
#endif
}
#endif
#ifdef CONFIG_AVX2
static __always_inline void raid_asm_clobber_ymm16(void)
{
raid_asm_clobber_xmm16();
asm volatile ("vzeroupper" : : : "memory");
}
#endif
#endif /* CONFIG_X86_64 */
#endif