diff options
| author | Johann Koenig <johannkoenig@google.com> | 2021-11-18 07:13:09 +0000 |
|---|---|---|
| committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2021-11-18 07:13:09 +0000 |
| commit | 6627a7c35b53e676fe89da0ea22b47b125342e98 (patch) | |
| tree | 1dd63279b10fd174c5e47857fbb08f4cb87bf86a /vp8/encoder | |
| parent | 52cfa1ca3546eec937bb4aa9d5fb157c58b907ad (diff) | |
| parent | 87ce2bc3e3266e670e0da71d7915c3c40e948c15 (diff) | |
| download | libvpx-6627a7c35b53e676fe89da0ea22b47b125342e98.tar libvpx-6627a7c35b53e676fe89da0ea22b47b125342e98.tar.gz libvpx-6627a7c35b53e676fe89da0ea22b47b125342e98.tar.bz2 libvpx-6627a7c35b53e676fe89da0ea22b47b125342e98.zip | |
Merge "replaced bsr() with get_msb() from bitops.h" into main
Diffstat (limited to 'vp8/encoder')
| -rw-r--r-- | vp8/encoder/x86/vp8_quantize_ssse3.c | 33 |
1 files changed, 6 insertions, 27 deletions
diff --git a/vp8/encoder/x86/vp8_quantize_ssse3.c b/vp8/encoder/x86/vp8_quantize_ssse3.c index 147c30cc3..f6df146f0 100644 --- a/vp8/encoder/x86/vp8_quantize_ssse3.c +++ b/vp8/encoder/x86/vp8_quantize_ssse3.c @@ -12,31 +12,7 @@ #include "./vp8_rtcd.h" #include "vp8/encoder/block.h" - -/* bitscan reverse (bsr) */ -#if defined(_MSC_VER) -#include <intrin.h> -#pragma intrinsic(_BitScanReverse) -static int bsr(int mask) { - unsigned long eob; - _BitScanReverse(&eob, mask); - eob++; - if (mask == 0) eob = 0; - return eob; -} -#else -static int bsr(int mask) { - int eob; -#if defined(__GNUC__) && __GNUC__ - __asm__ __volatile__("bsr %1, %0" : "=r"(eob) : "r"(mask) : "flags"); -#elif defined(__SUNPRO_C) || defined(__SUNPRO_CC) - asm volatile("bsr %1, %0" : "=r"(eob) : "r"(mask) : "flags"); -#endif - eob++; - if (mask == 0) eob = 0; - return eob; -} -#endif +#include "vpx_ports/bitops.h" /* get_msb */ void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d) { int eob, mask; @@ -108,7 +84,10 @@ void vp8_fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d) { mask = _mm_movemask_epi8(x); - eob = bsr(mask); + /* x2 is needed to increase the result from non-zero masks by 1, + * +1 is needed to mask undefined behavior for a null argument, + * the result of get_msb(1) is 0 */ + eob = get_msb(mask * 2 + 1); - *d->eob = 0xFF & eob; + *d->eob = eob; } |