dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2025-07-15 21:39:59 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

fix rescaler vertical interpolation

Browse Source 
* vertical expansion now uses bilinear interpolation
  * heavily assumes that the alpha plane is decoded in full, not row-by-row
  * split the RescalerExportRow and RescalerImportRow methods into Shrink
    and Expand variants.
  * MIPS implementation of ExportRowExpand is missing.

There's room for extra speed optim and code re-org, but let's keep that for later patches.

addresses https://code.google.com/p/webp/issues/detail?id=254

Change-Id: I8f12b855342bf07dd467fe85e4fde5fd814effdb
This commit is contained in:
Pascal Massimino
2015-09-18 10:45:03 +02:00
committed by James Zern
parent cd82440ec7
commit 5ff0079ece
10 changed files with 553 additions and 388 deletions
Download Patch File Download Diff File Expand all files Collapse all files

199
src/dsp/rescaler.c
View File

@ -17,78 +17,125 @@
//------------------------------------------------------------------------------
// Implementations of critical functions ImportRow / ExportRow
#define ROUNDER (1 << (WEBP_RESCALER_RFIX - 1))
#define MULT_FIX(x, y) (((int64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
#define ROUNDER (WEBP_RESCALER_ONE >> 1)
#define MULT_FIX(x, y) (((uint64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
static void RescalerImportRowC(WebPRescaler* const wrk,
const uint8_t* const src, int channel) {
//------------------------------------------------------------------------------
// Row import
void WebPRescalerImportRowExpandC(WebPRescaler* const wrk,
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels;
int x_in = channel;
int x_out;
if (!wrk->x_expand) {
uint32_t sum = 0;
int accum = 0;
for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
uint32_t base = 0;
// simple bilinear interpolation
int accum = wrk->x_add;
int left = src[x_in];
int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
x_in += x_stride;
x_out = channel;
assert(!WebPRescalerInputDone(wrk));
assert(wrk->x_expand);
while (1) {
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
x_out += x_stride;
if (x_out >= x_out_max) break;
accum -= wrk->x_sub;
if (accum < 0) {
left = right;
x_in += x_stride;
assert(x_in < wrk->src_width * x_stride);
right = src[x_in];
accum += wrk->x_add;
while (accum > 0) {
accum -= wrk->x_sub;
assert(x_in < wrk->src_width * x_stride);
base = src[x_in];
sum += base;
x_in += x_stride;
}
{ // Emit next horizontal pixel.
const int32_t frac = base * (-accum);
wrk->frow[x_out] = sum * wrk->x_sub - frac;
// fresh fractional start for next pixel
sum = (int)MULT_FIX(frac, wrk->fx_scale);
}
}
assert(accum == 0);
} else { // simple bilinear interpolation
int accum = wrk->x_add;
int left = src[x_in];
int right = (wrk->src_width > 1) ? src[x_in + x_stride] : left;
x_in += x_stride;
x_out = channel;
while (1) {
wrk->frow[x_out] = right * wrk->x_add + (left - right) * accum;
x_out += x_stride;
if (x_out >= x_out_max) break;
accum -= wrk->x_sub;
if (accum < 0) {
left = right;
x_in += x_stride;
assert(x_in < wrk->src_width * x_stride);
right = src[x_in];
accum += wrk->x_add;
}
}
assert(wrk->x_sub == 0 /* <- special case for src_width=1 */ || accum == 0);
}
// Accumulate the contribution of the new row.
assert(wrk->x_sub == 0 /* <- special case for src_width=1 */ || accum == 0);
}
void WebPRescalerImportRowShrinkC(WebPRescaler* const wrk,
const uint8_t* const src, int channel) {
const int x_stride = wrk->num_channels;
const int x_out_max = wrk->dst_width * wrk->num_channels;
int x_in = channel;
int x_out;
uint32_t sum = 0;
int accum = 0;
assert(!WebPRescalerInputDone(wrk));
assert(!wrk->x_expand);
for (x_out = channel; x_out < x_out_max; x_out += x_stride) {
wrk->irow[x_out] += wrk->frow[x_out];
uint32_t base = 0;
accum += wrk->x_add;
while (accum > 0) {
accum -= wrk->x_sub;
assert(x_in < wrk->src_width * x_stride);
base = src[x_in];
sum += base;
x_in += x_stride;
}
{ // Emit next horizontal pixel.
const rescaler_t frac = base * (-accum);
wrk->frow[x_out] = sum * wrk->x_sub - frac;
// fresh fractional start for next pixel
sum = (int)MULT_FIX(frac, wrk->fx_scale);
}
}
assert(accum == 0);
}
//------------------------------------------------------------------------------
// Row export
void WebPRescalerExportRowExpandC(WebPRescaler* const wrk) {
int x_out;
uint8_t* const dst = wrk->dst;
rescaler_t* const irow = wrk->irow;
const int x_out_max = wrk->dst_width * wrk->num_channels;
const rescaler_t* const frow = wrk->frow;
assert(!WebPRescalerOutputDone(wrk));
assert(wrk->y_accum <= 0);
assert(wrk->y_expand);
if (wrk->y_accum == 0) {
for (x_out = 0; x_out < x_out_max; ++x_out) {
const int v = (int)MULT_FIX(frow[x_out], wrk->fy_scale);
dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
}
} else {
const int64_t A = wrk->y_sub + wrk->y_accum;
const int64_t B = -wrk->y_accum;
for (x_out = 0; x_out < x_out_max; ++x_out) {
const int64_t I = A * frow[x_out] + B * irow[x_out];
const int v = (int)MULT_FIX(I, wrk->fxy_scale);
dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
}
}
}
void WebPRescalerExportRowC(WebPRescaler* const wrk, int x_out) {
if (wrk->y_accum <= 0) {
uint8_t* const dst = wrk->dst;
int32_t* const irow = wrk->irow;
const int32_t* const frow = wrk->frow;
const int yscale = wrk->fy_scale * (-wrk->y_accum);
const int x_out_max = wrk->dst_width * wrk->num_channels;
for (; x_out < x_out_max; ++x_out) {
void WebPRescalerExportRowShrinkC(WebPRescaler* const wrk) {
int x_out;
uint8_t* const dst = wrk->dst;
rescaler_t* const irow = wrk->irow;
const int x_out_max = wrk->dst_width * wrk->num_channels;
const rescaler_t* const frow = wrk->frow;
const int yscale = wrk->fy_scale * (-wrk->y_accum);
assert(!WebPRescalerOutputDone(wrk));
assert(wrk->y_accum <= 0);
assert(!wrk->y_expand);
if (yscale) {
for (x_out = 0; x_out < x_out_max; ++x_out) {
const int frac = (int)MULT_FIX(frow[x_out], yscale);
const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
irow[x_out] = frac; // new fractional start
}
wrk->y_accum += wrk->y_add;
wrk->dst += wrk->dst_stride;
} else {
for (x_out = 0; x_out < x_out_max; ++x_out) {
const int v = (int)MULT_FIX(irow[x_out], wrk->fxy_scale);
dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
irow[x_out] = 0;
}
}
}
@ -96,10 +143,39 @@ void WebPRescalerExportRowC(WebPRescaler* const wrk, int x_out) {
#undef ROUNDER
//------------------------------------------------------------------------------
// Main entry calls
void (*WebPRescalerImportRow)(struct WebPRescaler* const wrk,
const uint8_t* const src, int channel);
void (*WebPRescalerExportRow)(struct WebPRescaler* const wrk, int x_out);
void WebPRescalerImportRow(WebPRescaler* const wrk,
const uint8_t* const src, int channel) {
assert(!WebPRescalerInputDone(wrk));
if (!wrk->x_expand) {
WebPRescalerImportRowShrink(wrk, src, channel);
} else {
WebPRescalerImportRowExpand(wrk, src, channel);
}
}
void WebPRescalerExportRow(WebPRescaler* const wrk) {
if (wrk->y_accum <= 0) {
assert(!WebPRescalerOutputDone(wrk));
if (wrk->y_expand) {
WebPRescalerExportRowExpand(wrk);
} else {
WebPRescalerExportRowShrink(wrk);
}
wrk->y_accum += wrk->y_add;
wrk->dst += wrk->dst_stride;
++wrk->dst_y;
}
}
//------------------------------------------------------------------------------
WebPRescalerImportRowFunc WebPRescalerImportRowExpand;
WebPRescalerImportRowFunc WebPRescalerImportRowShrink;
WebPRescalerExportRowFunc WebPRescalerExportRowExpand;
WebPRescalerExportRowFunc WebPRescalerExportRowShrink;
extern void WebPRescalerDspInitMIPS32(void);
extern void WebPRescalerDspInitMIPSdspR2(void);
@ -110,8 +186,11 @@ static volatile VP8CPUInfo rescaler_last_cpuinfo_used =
WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
if (rescaler_last_cpuinfo_used == VP8GetCPUInfo) return;
WebPRescalerImportRow = RescalerImportRowC;
WebPRescalerExportRow = WebPRescalerExportRowC;
WebPRescalerImportRowExpand = WebPRescalerImportRowExpandC;
WebPRescalerImportRowShrink = WebPRescalerImportRowShrinkC;
WebPRescalerExportRowExpand = WebPRescalerExportRowExpandC;
WebPRescalerExportRowShrink = WebPRescalerExportRowShrinkC;
if (VP8GetCPUInfo != NULL) {
#if defined(WEBP_USE_MIPS32)
if (VP8GetCPUInfo(kMIPS32)) {