dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-07-26 10:50:06 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

Convert VP8LFastSLog2 to fixed point

Browse Source 
Speedups: 1% with '-lossless', 2% with '-lossless -q 100 -m6'

Change-Id: I1d79ea8e3e9e4bac7bcea4d7cbcc1bd56273988e
This commit is contained in:
Vincent Rabaud
2024-07-09 15:44:18 +02:00
parent 66408c2c7c
commit fb444b692b
7 changed files with 128 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/dsp/lossless_enc_mips32.c
View File

@ -23,12 +23,12 @@
#include <stdlib.h>
#include <string.h>
static float FastSLog2Slow_MIPS32(uint32_t v) {
static uint64_t FastSLog2Slow_MIPS32(uint32_t v) {
assert(v >= LOG_LOOKUP_IDX_MAX);
if (v < APPROX_LOG_WITH_CORRECTION_MAX) {
uint32_t log_cnt, y, correction;
uint32_t log_cnt, y;
uint64_t correction;
const int c24 = 24;
const float v_f = (float)v;
uint32_t temp;
// Xf = 256 = 2^8
@ -49,13 +49,14 @@ static float FastSLog2Slow_MIPS32(uint32_t v) {
// log2(Xf) = log2(floor(Xf)) + log2(1 + (v % y) / v)
// The correction factor: log(1 + d) ~ d; for very small d values, so
// log2(1 + (v % y) / v) ~ LOG_2_RECIPROCAL * (v % y)/v
// LOG_2_RECIPROCAL ~ 23/16
// (v % y) = (v % 2^log_cnt) = v & (2^log_cnt - 1)
correction = (23 * (v & (y - 1))) >> 4;
return v_f * (kLog2fTable[temp] + log_cnt) + correction;
correction = LOG_2_RECIPROCAL_FIXED * (v & (y - 1));
return (uint64_t)v * (kLog2Table[temp] +
((uint64_t)log_cnt << LOG_2_PRECISION_BITS)) +
correction;
} else {
return (float)(LOG_2_RECIPROCAL * v * log((double)v));
return (uint64_t)(LOG_2_RECIPROCAL_FIXED_DOUBLE * v * log((double)v) + .5);
}
}
@ -82,8 +83,7 @@ static uint32_t FastLog2Slow_MIPS32(uint32_t v) {
if (v >= APPROX_LOG_MAX) {
// Since the division is still expensive, add this correction factor only
// for large values of 'v'.
const uint64_t correction =
(uint64_t)LOG_2_RECIPROCAL_FIXED * (v & (y - 1));
const uint64_t correction = LOG_2_RECIPROCAL_FIXED * (v & (y - 1));
log_2 += (uint32_t)DivRound(correction, v);
}
return log_2;
@ -227,7 +227,7 @@ static WEBP_INLINE void GetEntropyUnrefinedHelper(
bit_entropy->sum += (*val_prev) * streak;
bit_entropy->nonzeros += streak;
bit_entropy->nonzero_code = *i_prev;
bit_entropy->entropy -= VP8LFastSLog2(*val_prev) * streak;
bit_entropy->entropy += VP8LFastSLog2(*val_prev) * streak;
if (bit_entropy->max_val < *val_prev) {
bit_entropy->max_val = *val_prev;
}
@ -259,7 +259,7 @@ static void GetEntropyUnrefined_MIPS32(const uint32_t X[], int length,
}
GetEntropyUnrefinedHelper(0, i, &x_prev, &i_prev, bit_entropy, stats);
bit_entropy->entropy += VP8LFastSLog2(bit_entropy->sum);
bit_entropy->entropy = VP8LFastSLog2(bit_entropy->sum) - bit_entropy->entropy;
}
static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
@ -282,7 +282,7 @@ static void GetCombinedEntropyUnrefined_MIPS32(const uint32_t X[],
}
GetEntropyUnrefinedHelper(0, i, &xy_prev, &i_prev, entropy, stats);
entropy->entropy += VP8LFastSLog2(entropy->sum);
entropy->entropy = VP8LFastSLog2(entropy->sum) - entropy->entropy;
}
#define ASM_START \