ffmpeg/libswscale/slice.c
Pedro Arthur 737aa902f0 swscale: process horizontal lines in batches
Process more lines in a single pass to improve performance

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2015-08-18 01:33:32 +02:00

296 lines
8.3 KiB
C

#include "swscale_internal.h"
static void free_lines(SwsSlice *s)
{
int i;
for (i = 0; i < 2; ++i) {
int n = s->plane[i].available_lines;
int j;
for (j = 0; j < n; ++j) {
av_freep(&s->plane[i].line[j]);
if (s->is_ring)
s->plane[i].line[j+n] = NULL;
}
}
for (i = 0; i < 4; ++i)
memset(s->plane[i].line, 0, sizeof(uint8_t*) * s->plane[i].available_lines * (s->is_ring ? 3 : 1));
s->should_free_lines = 0;
}
/*
slice lines contains extra bytes for vetorial code thus @size
is the allocated memory size and @width is the number of pixels
*/
static int alloc_lines(SwsSlice *s, int size, int width)
{
int i;
int idx[2] = {3, 2};
s->should_free_lines = 1;
s->width = width;
for (i = 0; i < 2; ++i) {
int n = s->plane[i].available_lines;
int j;
int ii = idx[i];
av_assert0(n == s->plane[ii].available_lines);
for (j = 0; j < n; ++j) {
// chroma plane line U and V are expected to be contiguous in memory
// by mmx vertical scaler code
s->plane[i].line[j] = av_malloc(size * 2 + 32);
if (!s->plane[i].line[j]) {
free_lines(s);
return AVERROR(ENOMEM);
}
s->plane[ii].line[j] = s->plane[i].line[j] + size + 16;
if (s->is_ring) {
s->plane[i].line[j+n] = s->plane[i].line[j];
s->plane[ii].line[j+n] = s->plane[ii].line[j];
}
}
}
return 0;
}
static int alloc_slice(SwsSlice *s, enum AVPixelFormat fmt, int lumLines, int chrLines, int h_sub_sample, int v_sub_sample, int ring)
{
int i;
int size[4] = { lumLines,
chrLines,
chrLines,
lumLines };
s->h_chr_sub_sample = h_sub_sample;
s->v_chr_sub_sample = v_sub_sample;
s->fmt = fmt;
s->is_ring = ring;
s->should_free_lines = 0;
for (i = 0; i < 4; ++i) {
int n = size[i] * ( ring == 0 ? 1 : 3);
s->plane[i].line = av_mallocz_array(sizeof(uint8_t*), n);
if (!s->plane[i].line)
return AVERROR(ENOMEM);
s->plane[i].tmp = ring ? s->plane[i].line + size[i] * 2 : NULL;
s->plane[i].available_lines = size[i];
s->plane[i].sliceY = 0;
s->plane[i].sliceH = 0;
}
return 0;
}
static void free_slice(SwsSlice *s)
{
int i;
if (s) {
if (s->should_free_lines)
free_lines(s);
for (i = 0; i < 4; ++i) {
av_freep(&s->plane[i].line);
s->plane[i].tmp = NULL;
}
}
}
int ff_rotate_slice(SwsSlice *s, int lum, int chr)
{
int i;
if (lum) {
for (i = 0; i < 4; i+=3) {
int n = s->plane[i].available_lines;
int l = lum - s->plane[i].sliceY;
if (l >= n * 2) {
s->plane[i].sliceY += n;
s->plane[i].sliceH -= n;
}
}
}
if (chr) {
for (i = 1; i < 3; ++i) {
int n = s->plane[i].available_lines;
int l = chr - s->plane[i].sliceY;
if (l >= n * 2) {
s->plane[i].sliceY += n;
s->plane[i].sliceH -= n;
}
}
}
return 0;
}
int ff_init_slice_from_src(SwsSlice * s, uint8_t *src[4], int stride[4], int srcW, int lumY, int lumH, int chrY, int chrH)
{
int i = 0;
const int start[4] = {lumY,
chrY,
chrY,
lumY};
const int end[4] = {lumY +lumH,
chrY + chrH,
chrY + chrH,
lumY + lumH};
s->width = srcW;
for (i = 0; i < 4; ++i) {
int j;
int lines = end[i];
lines = s->plane[i].available_lines < lines ? s->plane[i].available_lines : lines;
if (end[i] > s->plane[i].sliceY+s->plane[i].sliceH) {
if (start[i] <= s->plane[i].sliceY+1)
s->plane[i].sliceY = FFMIN(start[i], s->plane[i].sliceY);
else
s->plane[i].sliceY = start[i];
s->plane[i].sliceH = end[i] - s->plane[i].sliceY;
} else {
if (end[i] >= s->plane[i].sliceY)
s->plane[i].sliceH = s->plane[i].sliceY + s->plane[i].sliceH - start[i];
else
s->plane[i].sliceH = end[i] - start[i];
s->plane[i].sliceY = start[i];
}
for (j = start[i]; j < lines; j+= 1)
s->plane[i].line[j] = src[i] + (start[i] + j) * stride[i];
}
return 0;
}
static void fill_ones(SwsSlice *s, int n, int is16bit)
{
int i;
for (i = 0; i < 4; ++i) {
int j;
int size = s->plane[i].available_lines;
for (j = 0; j < size; ++j) {
int k;
int end = is16bit ? n>>1: n;
// fill also one extra element
end += 1;
if (is16bit)
for (k = 0; k < end; ++k)
((int32_t*)(s->plane[i].line[j]))[k] = 1<<18;
else
for (k = 0; k < end; ++k)
((int16_t*)(s->plane[i].line[j]))[k] = 1<<14;
}
}
}
int ff_init_filters(SwsContext * c)
{
int i;
int index;
int num_ydesc;
int num_cdesc;
int need_lum_conv = c->lumToYV12 || c->readLumPlanar || c->alpToYV12 || c->readAlpPlanar;
int need_chr_conv = c->chrToYV12 || c->readChrPlanar;
int srcIdx, dstIdx;
int dst_stride = FFALIGN(c->dstW * sizeof(int16_t) + 66, 16);
uint32_t * pal = usePal(c->srcFormat) ? c->pal_yuv : (uint32_t*)c->input_rgb2yuv_table;
int res = 0;
if (c->dstBpc == 16)
dst_stride <<= 1;
num_ydesc = need_lum_conv ? 2 : 1;
num_cdesc = need_chr_conv ? 2 : 1;
c->numSlice = FFMAX(num_ydesc, num_cdesc) + 1;
c->numDesc = num_ydesc + num_cdesc;
c->descIndex[0] = num_ydesc;
c->descIndex[1] = num_ydesc + num_cdesc;
c->desc = av_mallocz_array(sizeof(SwsFilterDescriptor), c->numDesc);
if (!c->desc)
return AVERROR(ENOMEM);
c->slice = av_mallocz_array(sizeof(SwsSlice), c->numSlice);
res = alloc_slice(&c->slice[0], c->srcFormat, c->srcH, c->chrSrcH, c->chrSrcHSubSample, c->chrSrcVSubSample, 0);
if (res < 0) goto cleanup;
for (i = 1; i < c->numSlice-1; ++i) {
res = alloc_slice(&c->slice[i], c->srcFormat, c->vLumFilterSize + MAX_LINES_AHEAD, c->vChrFilterSize + MAX_LINES_AHEAD, c->chrSrcHSubSample, c->chrSrcVSubSample, 0);
if (res < 0) goto cleanup;
res = alloc_lines(&c->slice[i], FFALIGN(c->srcW*2+78, 16), c->srcW);
if (res < 0) goto cleanup;
}
res = alloc_slice(&c->slice[i], c->srcFormat, c->vLumFilterSize + MAX_LINES_AHEAD, c->vChrFilterSize + MAX_LINES_AHEAD, c->chrDstHSubSample, c->chrDstVSubSample, 1);
if (res < 0) goto cleanup;
res = alloc_lines(&c->slice[i], dst_stride, c->dstW);
if (res < 0) goto cleanup;
fill_ones(&c->slice[i], dst_stride>>1, c->dstBpc == 16);
index = 0;
srcIdx = 0;
dstIdx = 1;
if (need_lum_conv) {
ff_init_desc_fmt_convert(&c->desc[index], &c->slice[srcIdx], &c->slice[dstIdx], pal);
c->desc[index].alpha = c->alpPixBuf != 0;
++index;
srcIdx = dstIdx;
}
dstIdx = FFMAX(num_ydesc, num_cdesc);
ff_init_desc_hscale(&c->desc[index], &c->slice[index], &c->slice[dstIdx], c->hLumFilter, c->hLumFilterPos, c->hLumFilterSize, c->lumXInc);
c->desc[index].alpha = c->alpPixBuf != 0;
++index;
{
srcIdx = 0;
dstIdx = 1;
if (need_chr_conv) {
ff_init_desc_cfmt_convert(&c->desc[index], &c->slice[srcIdx], &c->slice[dstIdx], pal);
++index;
srcIdx = dstIdx;
}
dstIdx = FFMAX(num_ydesc, num_cdesc);
if (c->needs_hcscale)
ff_init_desc_chscale(&c->desc[index], &c->slice[srcIdx], &c->slice[dstIdx], c->hChrFilter, c->hChrFilterPos, c->hChrFilterSize, c->chrXInc);
else
ff_init_desc_no_chr(&c->desc[index], &c->slice[srcIdx], &c->slice[dstIdx]);
}
return 0;
cleanup:
ff_free_filters(c);
return res;
}
int ff_free_filters(SwsContext *c)
{
int i;
if (c->desc) {
for (i = 0; i < c->numDesc; ++i)
av_freep(&c->desc[i].instance);
av_freep(&c->desc);
}
if (c->slice) {
for (i = 0; i < c->numSlice; ++i)
free_slice(&c->slice[i]);
av_freep(&c->slice);
}
return 0;
}