Optimize VP8SetResidualCoeffs.

Brings down WebP lossy encoding timings by 5%

Change-Id: Ia4a2fab0a887aaaf7841ce6d9ee16270d3e15489
This commit is contained in:
skal
2014-06-03 06:44:04 +02:00
parent ac591cf22e
commit 6679f8996f
6 changed files with 107 additions and 44 deletions

View File

@ -77,6 +77,41 @@ static WEBP_INLINE void PutLE32(uint8_t* const data, uint32_t val) {
PutLE16(data + 2, (int)(val >> 16));
}
// Returns (int)floor(log2(n)). n must be > 0.
// use GNU builtins where available.
#if defined(__GNUC__) && \
((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4)
static WEBP_INLINE int BitsLog2Floor(uint32_t n) {
return 31 ^ __builtin_clz(n);
}
#elif defined(_MSC_VER) && _MSC_VER > 1310 && \
(defined(_M_X64) || defined(_M_IX86))
#include <intrin.h>
#pragma intrinsic(_BitScanReverse)
static WEBP_INLINE int BitsLog2Floor(uint32_t n) {
uint32_t first_set_bit;
_BitScanReverse(&first_set_bit, n);
return first_set_bit;
}
#else
static WEBP_INLINE int BitsLog2Floor(uint32_t n) {
int log = 0;
uint32_t value = n;
int i;
for (i = 4; i >= 0; --i) {
const int shift = (1 << i);
const uint32_t x = value >> shift;
if (x != 0) {
value = x;
log += shift;
}
}
return log;
}
#endif
//------------------------------------------------------------------------------
#ifdef __cplusplus