dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-12-27 06:08:21 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

Move ARGB->YUV functions from dec/vp8l.c to dsp/yuv.c

Browse Source 
also switch to using ExtractAlpha() instead of hard-coding the loop.

The ARGBToY/UV functions are rather easy to port to SSE2 / NEON.

Change-Id: I8f1346a9ca427a36ce2d6c848369ca7964d8b3c7
This commit is contained in:
Pascal Massimino 2015-10-27 02:37:45 -07:00
parent 469ba2cdfd
commit fa8927efe4
3 changed files with 100 additions and 49 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
src/dec/vp8l.c
View File

@ -19,6 +19,7 @@
#include "../dsp/dsp.h"
#include "../dsp/lossless.h"
#include "../dsp/yuv.h"
#include "../utils/endian_inl.h"
#include "../utils/huffman.h"
#include "../utils/utils.h"
@ -504,67 +505,29 @@ static int EmitRows(WEBP_CSP_MODE colorspace,
//------------------------------------------------------------------------------
// Export to YUVA
// TODO(skal): should be in yuv.c
static void ConvertToYUVA(const uint32_t* const src, int width, int y_pos,
const WebPDecBuffer* const output) {
const WebPYUVABuffer* const buf = &output->u.YUVA;
// first, the luma plane
{
int i;
uint8_t* const y = buf->y + y_pos * buf->y_stride;
for (i = 0; i < width; ++i) {
const uint32_t p = src[i];
y[i] = VP8RGBToY((p >> 16) & 0xff, (p >> 8) & 0xff, (p >> 0) & 0xff,
YUV_HALF);
}
}
WebPConvertARGBToY(src, buf->y + y_pos * buf->y_stride, width);
// then U/V planes
{
uint8_t* const u = buf->u + (y_pos >> 1) * buf->u_stride;
uint8_t* const v = buf->v + (y_pos >> 1) * buf->v_stride;
const int uv_width = width >> 1;
int i;
for (i = 0; i < uv_width; ++i) {
const uint32_t v0 = src[2 * i + 0];
const uint32_t v1 = src[2 * i + 1];
// VP8RGBToU/V expects four accumulated pixels. Hence we need to
// scale r/g/b value by a factor 2. We just shift v0/v1 one bit less.
const int r = ((v0 >> 15) & 0x1fe) + ((v1 >> 15) & 0x1fe);
const int g = ((v0 >> 7) & 0x1fe) + ((v1 >> 7) & 0x1fe);
const int b = ((v0 << 1) & 0x1fe) + ((v1 << 1) & 0x1fe);
if (!(y_pos & 1)) { // even lines: store values
u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
} else { // odd lines: average with previous values
const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
// Approximated average-of-four. But it's an acceptable diff.
u[i] = (u[i] + tmp_u + 1) >> 1;
v[i] = (v[i] + tmp_v + 1) >> 1;
}
}
if (width & 1) { // last pixel
const uint32_t v0 = src[2 * i + 0];
const int r = (v0 >> 14) & 0x3fc;
const int g = (v0 >> 6) & 0x3fc;
const int b = (v0 << 2) & 0x3fc;
if (!(y_pos & 1)) { // even lines
u[i] = VP8RGBToU(r, g, b, YUV_HALF << 2);
v[i] = VP8RGBToV(r, g, b, YUV_HALF << 2);
} else { // odd lines (note: we could just skip this)
const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
u[i] = (u[i] + tmp_u + 1) >> 1;
v[i] = (v[i] + tmp_v + 1) >> 1;
}
}
// even lines: store values
// odd lines: average with previous values
WebPConvertARGBToUV(src, u, v, width, !(y_pos & 1));
}
// Lastly, store alpha if needed.
if (buf->a != NULL) {
int i;
uint8_t* const a = buf->a + y_pos * buf->a_stride;
for (i = 0; i < width; ++i) a[i] = (src[i] >> 24);
#if defined(WORDS_BIGENDIAN)
WebPExtractAlpha((uint8_t*)src + 0, 0, width, 1, a, 0);
#else
WebPExtractAlpha((uint8_t*)src + 3, 0, width, 1, a, 0);
#endif
}
}
@ -713,7 +676,7 @@ static void ProcessRows(VP8LDecoder* const dec, int row) {
// Nothing to output (this time).
} else {
const WebPDecBuffer* const output = dec->output_;
if (output->colorspace < MODE_YUV) { // convert to RGBA
if (WebPIsRGBMode(output->colorspace)) { // convert to RGBA
const WebPRGBABuffer* const buf = &output->u.RGBA;
uint8_t* const rgba = buf->rgba + dec->last_out_row_ * buf->stride;
const int num_rows_out = io->use_scaling ?
@ -1552,6 +1515,10 @@ int VP8LDecodeImage(VP8LDecoder* const dec) {
// need the alpha-multiply functions for premultiplied output or rescaling
WebPInitAlphaProcessing();
}
if (!WebPIsRGBMode(dec->output_->colorspace)) {
WebPInitConvertARGBToYUV();
if (dec->output_->u.YUVA.a != NULL) WebPInitAlphaProcessing();
}
if (dec->incremental_) {
if (dec->hdr_.color_cache_size_ > 0 &&
dec->hdr_.saved_color_cache_.colors_ == NULL) {

13
src/dsp/dsp.h
View File

@ -324,6 +324,19 @@ void WebPInitSamplers(void);
// Must be called before using WebPYUV444Converters[]
void WebPInitYUV444Converters(void);
//------------------------------------------------------------------------------
// ARGB -> YUV converters (for lossless decoding)
// Convert ARGB samples to luma Y.
extern void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
// Convert ARGB samples to U/V with downsampling. do_store should be '1' for
// even lines and '0' for odd ones. 'src_width' is the original width, not
// the U/V one.
extern void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
int src_width, int do_store);
// Must be called before using WebPConvertARGBToXXX
void WebPInitConvertARGBToYUV(void);
//------------------------------------------------------------------------------
// Rescaler

71
src/dsp/yuv.c
View File

@ -164,3 +164,74 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplers(void) {
}
//-----------------------------------------------------------------------------
// ARGB -> YUV converters (for lossless decoding)
static void ConvertARGBToY(const uint32_t* argb, uint8_t* y, int width) {
int i;
for (i = 0; i < width; ++i) {
const uint32_t p = argb[i];
y[i] = VP8RGBToY((p >> 16) & 0xff, (p >> 8) & 0xff, (p >> 0) & 0xff,
YUV_HALF);
}
}
static void ConvertARGBToUV(const uint32_t* argb, uint8_t* u, uint8_t* v,
int src_width, int do_store) {
// No rounding. Last pixel is dealt with separately.
const int uv_width = src_width >> 1;
int i;
for (i = 0; i < uv_width; ++i) {
const uint32_t v0 = argb[2 * i + 0];
const uint32_t v1 = argb[2 * i + 1];
// VP8RGBToU/V expects four accumulated pixels. Hence we need to
// scale r/g/b value by a factor 2. We just shift v0/v1 one bit less.
const int r = ((v0 >> 15) & 0x1fe) + ((v1 >> 15) & 0x1fe);
const int g = ((v0 >> 7) & 0x1fe) + ((v1 >> 7) & 0x1fe);
const int b = ((v0 << 1) & 0x1fe) + ((v1 << 1) & 0x1fe);
const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
if (do_store) {
u[i] = tmp_u;
v[i] = tmp_v;
} else {
// Approximated average-of-four. But it's an acceptable diff.
u[i] = (u[i] + tmp_u + 1) >> 1;
v[i] = (v[i] + tmp_v + 1) >> 1;
}
}
if (src_width & 1) { // last pixel
const uint32_t v0 = argb[2 * i + 0];
const int r = (v0 >> 14) & 0x3fc;
const int g = (v0 >> 6) & 0x3fc;
const int b = (v0 << 2) & 0x3fc;
const int tmp_u = VP8RGBToU(r, g, b, YUV_HALF << 2);
const int tmp_v = VP8RGBToV(r, g, b, YUV_HALF << 2);
if (do_store) {
u[i] = tmp_u;
v[i] = tmp_v;
} else {
u[i] = (u[i] + tmp_u + 1) >> 1;
v[i] = (v[i] + tmp_v + 1) >> 1;
}
}
}
void (*WebPConvertARGBToY)(const uint32_t* argb, uint8_t* y, int width);
void (*WebPConvertARGBToUV)(const uint32_t* argb, uint8_t* u, uint8_t* v,
int src_width, int do_store);
static volatile VP8CPUInfo rgba_to_yuv_last_cpuinfo_used =
(VP8CPUInfo)&rgba_to_yuv_last_cpuinfo_used;
WEBP_TSAN_IGNORE_FUNCTION void WebPInitConvertARGBToYUV(void) {
if (rgba_to_yuv_last_cpuinfo_used == VP8GetCPUInfo) return;
WebPConvertARGBToY = ConvertARGBToY;
WebPConvertARGBToUV = ConvertARGBToUV;
if (VP8GetCPUInfo != NULL) {
#if defined(WEBP_USE_SSE2)
if (VP8GetCPUInfo(kSSE2)) {
}
#endif // WEBP_USE_SSE2
}
rgba_to_yuv_last_cpuinfo_used = VP8GetCPUInfo;
}