dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-07-17 06:19:54 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

rescaler: add some SSE2 code

Browse Source 
The rounding and arithmetic is not the same as previously, to prevent overflow cases for large upscale factors.

We still rely on 32b x 32b -> 64b multiplies. Raised the fixed-point precision to 32b
so that we have some nice shifts from epi64 to epi32.
Changed rescaler_t type to 'uint32_t' in order to squeeze in all the precision required.

The MIPS code has been disabled because it's now out-of-sync. Will be fixed in
a subsequent CL when the dust settles.
~30-35% faster

Change-Id: I32e4ddc00933f1b1aa3463403086199fd5dad07b
This commit is contained in:
Pascal Massimino
2015-09-25 14:34:02 +02:00
parent 1df1d0eedb
commit 76a7dc39e5
10 changed files with 320 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
src/dsp/rescaler_mips_dsp_r2.c
View File

@ -30,7 +30,7 @@ static void ImportRowShrink(WebPRescaler* const wrk, const uint8_t* src) {
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
int* frow = wrk->frow + channel;
rescaler_t* frow = wrk->frow + channel;
const uint8_t* src1 = src + channel;
int temp3;
int base, frac, sum;
@ -84,7 +84,7 @@ static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
assert(!WebPRescalerInputDone(wrk));
for (channel = 0; channel < x_stride; ++channel) {
int* frow = wrk->frow + channel;
rescaler_t* frow = wrk->frow + channel;
const uint8_t* src1 = src + channel;
int temp1, temp2, temp3, temp4;
int frac;
@ -133,17 +133,15 @@ static void ImportRowExpand(WebPRescaler* const wrk, const uint8_t* src) {
}
static void ExportRowShrink(WebPRescaler* const wrk) {
uint8_t* dst = wrk->dst;
rescaler_t* irow = wrk->irow;
const int x_out_max = wrk->dst_width * wrk->num_channels;
assert(!WebPRescalerOutputDone(wrk));
assert(wrk->y_accum <= 0);
assert(!wrk->y_expand);
// if wrk->fxy_scale can fit into 32 bits use optimized code,
// otherwise use C code
if ((wrk->fxy_scale >> 32) == 0) {
uint8_t* dst = wrk->dst;
rescaler_t* irow = wrk->irow;
if (wrk->fxy_scale) {
const rescaler_t* frow = wrk->frow;
const int yscale = wrk->fy_scale * (-wrk->y_accum);
const int x_out_max = wrk->dst_width * wrk->num_channels;
int temp0, temp1, temp3, temp4, temp5, temp6, temp7;
const int temp2 = (int)wrk->fxy_scale;
@ -212,8 +210,12 @@ static void ExportRowShrink(WebPRescaler* const wrk) {
[rest]"r"(rest)
: "memory", "hi", "lo"
);
} else {
WebPRescalerExportRowShrinkC(wrk);
} else { // very special case for src = dst = 1x1
int x_out;
for (x_out = 0; x_out < x_out_max; ++x_out) {
dst[x_out] = irow[x_out];
irow[x_out] = 0;
}
}
}
@ -225,9 +227,17 @@ static void ExportRowShrink(WebPRescaler* const wrk) {
extern void WebPRescalerDspInitMIPSdspR2(void);
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInitMIPSdspR2(void) {
#if 0
// The assembly code is currently out-of-sync wrt the C-implementation.
// Disabled for now.
WebPRescalerImportRowExpand = ImportRowExpand;
WebPRescalerImportRowShrink = ImportRowShrink;
WebPRescalerExportRowShrink = ExportRowShrink;
#else
(void)ImportRowExpand;
(void)ImportRowShrink;
(void)ExportRowShrink;
#endif
}
#else // !WEBP_USE_MIPS_DSP_R2