dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-12-28 06:38:20 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

MIPS: dspr2: add some specific mips code to commit I2c3f2b12f8df15b785fad5a9c56316e954ae0c53

Browse Source 
added some C-code tuning also

Change-Id: I67ce70a063ef6b5821b9158a4defd6987eccbb9a
This commit is contained in:
Djordje Pesut 2014-09-04 13:38:15 +02:00
parent e564062522
commit 80b8099fd8
1 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/dec/vp8l.c
View File

@ -758,7 +758,6 @@ static WEBP_INLINE uint32_t Rotate8b(uint32_t V) {
static WEBP_INLINE void CopySmallPattern(const uint8_t* data_src, static WEBP_INLINE void CopySmallPattern(const uint8_t* data_src,
uint8_t* data_dst, uint8_t* data_dst,
int length, uint32_t pattern) { int length, uint32_t pattern) {
int ilength = length;
uint32_t* pdata; uint32_t* pdata;
int j = 0; int j = 0;
int i; int i;
@ -768,17 +767,16 @@ static WEBP_INLINE void CopySmallPattern(const uint8_t* data_src,
pattern = Rotate8b(pattern); pattern = Rotate8b(pattern);
++j; ++j;
} }
ilength -= j; length -= j;
data_src += j; data_src += j;
// Copy the pattern 4 bytes at a time. // Copy the pattern 4 bytes at a time.
pdata = (uint32_t*)data_dst; pdata = (uint32_t*)data_dst;
for (i = 0; i < (ilength >> 2); ++i) { for (i = 0; i < (length >> 2); ++i) {
pdata[i] = pattern; pdata[i] = pattern;
} }
// Finish with left-overs. 'pattern' is still correctly positioned, // Finish with left-overs. 'pattern' is still correctly positioned,
// so no Rotate8b() call is needed. // so no Rotate8b() call is needed.
data_dst = (uint8_t*)pdata; for (i <<= 2; i < length; ++i) {
for (i <<= 2; i < ilength; ++i) {
data_dst[i] = data_src[i]; data_dst[i] = data_src[i];
} }
} }
@ -793,6 +791,8 @@ static WEBP_INLINE void CopyBlock(uint8_t* data_dst, int dist, int length) {
#if defined(__arm__) || defined(_M_ARM) // arm doesn't like multiply that much #if defined(__arm__) || defined(_M_ARM) // arm doesn't like multiply that much
pattern |= pattern << 8; pattern |= pattern << 8;
pattern |= pattern << 16; pattern |= pattern << 16;
#elif defined(WEBP_USE_MIPS_DSP_R2)
__asm__ volatile ("replv.qb %0, %0" : "+r"(pattern));
#else #else
pattern = 0x01010101u * pattern; pattern = 0x01010101u * pattern;
#endif #endif
@ -801,6 +801,8 @@ static WEBP_INLINE void CopyBlock(uint8_t* data_dst, int dist, int length) {
pattern = *(const uint16_t*)data_src; pattern = *(const uint16_t*)data_src;
#if defined(__arm__) || defined(_M_ARM) #if defined(__arm__) || defined(_M_ARM)
pattern |= pattern << 16; pattern |= pattern << 16;
#elif defined(WEBP_USE_MIPS_DSP_R2)
__asm__ volatile ("replv.ph %0, %0" : "+r"(pattern));
#else #else
pattern = 0x00010001u * pattern; pattern = 0x00010001u * pattern;
#endif #endif