github_mirrors/vectorscan
3
0
Fork 1
mirror of https://github.com/VectorCamp/vectorscan.git synced 2025-06-28 16:41:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

simd: simplify the set-all-ones util funcs

Browse Source 
Modern compilers (gcc, icc) get this right, with the benefit of
removing our last use of inline asm in this file.
This commit is contained in:
Matthew Barr 2016-06-06 16:24:55 +10:00
parent 4d6934fc77
commit adf820bbba
1 changed files with 7 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
src/util/simd_utils.h
View File

@ -88,10 +88,6 @@
# endif # endif
#endif #endif
#ifdef _WIN32
#define NO_ASM
#endif
// Fallback to identity case. // Fallback to identity case.
#ifndef assume_aligned #ifndef assume_aligned
#define assume_aligned(x, y) (x) #define assume_aligned(x, y) (x)
@ -106,13 +102,12 @@ extern const char vbs_mask_data[];
#endif #endif
static really_inline m128 ones128(void) { static really_inline m128 ones128(void) {
#if !defined(NO_ASM) #if defined(__GNUC__) || defined(__INTEL_COMPILER)
// trick from Intel's optimization guide to generate all-ones. We have to /* gcc gets this right */
// use inline asm, as there's no intrinsic for this yet. return _mm_set1_epi8(0xFF);
m128 ret;
__asm__ ("pcmpeqb %0,%0" : "=x"(ret));
return ret;
#else #else
/* trick from Intel's optimization guide to generate all-ones.
* ICC converts this to the single cmpeq instruction */
return _mm_cmpeq_epi8(_mm_setzero_si128(), _mm_setzero_si128()); return _mm_cmpeq_epi8(_mm_setzero_si128(), _mm_setzero_si128());
#endif #endif
} }
@ -172,19 +167,8 @@ static really_inline unsigned short cmpmsk8(m128 a, m128 b) {
#define eq128(a, b) _mm_cmpeq_epi8((a), (b)) #define eq128(a, b) _mm_cmpeq_epi8((a), (b))
#define movemask128(a) ((u32)_mm_movemask_epi8((a))) #define movemask128(a) ((u32)_mm_movemask_epi8((a)))
// We found that this generated better code with gcc-4.1 and with the default
// tuning settings on gcc-4.4 than just using the _mm_set1_epi8() instrinsic.
static really_inline m128 set16x8(u8 c) { static really_inline m128 set16x8(u8 c) {
#if !defined(__AVX2__)
m128 a = _mm_cvtsi32_si128((int)c);
a = _mm_unpacklo_epi8(a, a);
a = _mm_unpacklo_epi8(a, a);
a = _mm_shuffle_epi32(a, 0);
return a;
#else
// uses a broadcast for much win
return _mm_set1_epi8(c); return _mm_set1_epi8(c);
#endif
} }
static really_inline u32 movd(const m128 in) { static really_inline u32 movd(const m128 in) {
@ -369,8 +353,7 @@ m128 variable_byte_shift_m128(m128 in, s32 amount) {
static really_inline static really_inline
m256 set32x8(u32 in) { m256 set32x8(u32 in) {
m128 a = _mm_cvtsi32_si128(in); return _mm256_set1_epi8(in);
return _mm256_broadcastb_epi8(a);
} }
#define eq256(a, b) _mm256_cmpeq_epi8((a), (b)) #define eq256(a, b) _mm256_cmpeq_epi8((a), (b))
@ -423,12 +406,7 @@ static really_inline m256 zeroes256(void) {
static really_inline m256 ones256(void) { static really_inline m256 ones256(void) {
#if defined(__AVX2__) #if defined(__AVX2__)
m256 rv; m256 rv = _mm256_set1_epi8(0xFF);
#if defined(NO_ASM)
rv = eq256(zeroes256(), zeroes256());
#else
__asm__ ("vpcmpeqb %0,%0,%0" : "=x"(rv));
#endif
#else #else
m256 rv = {ones128(), ones128()}; m256 rv = {ones128(), ones128()};
#endif #endif