generic-library/webp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

MIPS: MIPS32r1: Added optimization for HuffmanCost functions.

Browse Source 
HuffmanCost and HuffmanCostCombined optimized and added
'const' to some variables from ExtraCost functions.

Change-Id: I28b2b357a06766bee78bdab294b5fc8c05ac120d
This commit is contained in:
Djordje Pesut 2014-04-22 14:55:14 +02:00
parent c0220460e9
commit 1d62acf6af
1 changed files with 91 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

97
src/dsp/lossless_mips32.c
View File

@ -17,9 +17,10 @@
#if defined(WEBP_USE_MIPS32)
#include <assert.h>
#include <math.h>
#include <stdlib.h>
#include <assert.h>
#include <string.h>
#define APPROX_LOG_WITH_CORRECTION_MAX 65536
#define APPROX_LOG_MAX 4096
@ -108,8 +109,8 @@ static float FastLog2Slow(int v) {
// return (double)cost;
static double ExtraCost(const int* const population, int length) {
int i, temp0, temp1;
int* pop = (int*)&population[4];
const int* LoopEnd = (int*)&population[length];
const int* pop = &population[4];
const int* const LoopEnd = &population[length];
__asm__ volatile(
"mult $zero, $zero \n\t"
@ -154,9 +155,9 @@ static double ExtraCost(const int* const population, int length) {
static double ExtraCostCombined(const int* const X, const int* const Y,
int length) {
int i, temp0, temp1, temp2, temp3;
int* pX = (int*)&X[4];
int* pY = (int*)&Y[4];
const int* LoopEnd = (int*)&X[length];
const int* pX = &X[4];
const int* pY = &Y[4];
const int* const LoopEnd = &X[length];
__asm__ volatile(
"mult $zero, $zero \n\t"
@ -188,6 +189,88 @@ static double ExtraCostCombined(const int* const X, const int* const Y,
return (double)((int64_t)temp0 << 32 | temp1);
}
#define HUFFMAN_COST_PASS \
__asm__ volatile( \
"sll %[temp1], %[temp0], 3 \n\t" \
"addiu %[temp3], %[streak], -3 \n\t" \
"addu %[temp2], %[pstreaks], %[temp1] \n\t" \
"blez %[temp3], 1f \n\t" \
"srl %[temp1], %[temp1], 1 \n\t" \
"addu %[temp3], %[pcnts], %[temp1] \n\t" \
"lw %[temp0], 4(%[temp2]) \n\t" \
"lw %[temp1], 0(%[temp3]) \n\t" \
"addu %[temp0], %[temp0], %[streak] \n\t" \
"addiu %[temp1], %[temp1], 1 \n\t" \
"sw %[temp0], 4(%[temp2]) \n\t" \
"sw %[temp1], 0(%[temp3]) \n\t" \
"b 2f \n\t" \
"1: \n\t" \
"lw %[temp0], 0(%[temp2]) \n\t" \
"addu %[temp0], %[temp0], %[streak] \n\t" \
"sw %[temp0], 0(%[temp2]) \n\t" \
"2: \n\t" \
: [temp1]"=&r"(temp1), [temp2]"=&r"(temp2), \
[temp3]"=&r"(temp3), [temp0]"+r"(temp0) \
: [pstreaks]"r"(pstreaks), [pcnts]"r"(pcnts), \
[streak]"r"(streak) \
: "memory" \
);
// Returns the various RLE counts
static VP8LStreaks HuffmanCostCount(const int* population, int length) {
int i;
int streak = 0;
VP8LStreaks stats;
int* const pstreaks = &stats.streaks[0][0];
int* const pcnts = &stats.counts[0];
int temp0, temp1, temp2, temp3;
memset(&stats, 0, sizeof(stats));
for (i = 0; i < length - 1; ++i) {
++streak;
if (population[i] == population[i + 1]) {
continue;
}
temp0 = population[i] != 0;
HUFFMAN_COST_PASS
streak = 0;
}
++streak;
temp0 = population[i] != 0;
HUFFMAN_COST_PASS
return stats;
}
static VP8LStreaks HuffmanCostCombinedCount(const int* X, const int* Y,
int length) {
int i;
int streak = 0;
VP8LStreaks stats;
int* const pstreaks = &stats.streaks[0][0];
int* const pcnts = &stats.counts[0];
int temp0, temp1, temp2, temp3;
memset(&stats, 0, sizeof(stats));
for (i = 0; i < length - 1; ++i) {
const int xy = X[i] + Y[i];
const int xy_next = X[i + 1] + Y[i + 1];
++streak;
if (xy == xy_next) {
continue;
}
temp0 = xy != 0;
HUFFMAN_COST_PASS
streak = 0;
}
{
const int xy = X[i] + Y[i];
++streak;
temp0 = xy != 0;
HUFFMAN_COST_PASS
}
return stats;
}
#endif // WEBP_USE_MIPS32
//------------------------------------------------------------------------------
@ -201,5 +284,7 @@ void VP8LDspInitMIPS32(void) {
VP8LFastLog2Slow = FastLog2Slow;
VP8LExtraCost = ExtraCost;
VP8LExtraCostCombined = ExtraCostCombined;
VP8LHuffmanCostCount = HuffmanCostCount;
VP8LHuffmanCostCombinedCount = HuffmanCostCombinedCount;
#endif // WEBP_USE_MIPS32
}