dilim/FFmpeg
1
0 Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2026-11-06 05:17:42 +00:00
Code Issues Packages Projects Releases Wiki Activity

swscale/ops_memcpy: merge bulk memset/memcpy paths into per-line loops

Browse Source 
Instead of having separate paths for bulk and per-line memset/memcpy
operations, use the same per-line loops for the bulk operations, by
modifying the loop count and size parameters to have the loop run only
once.
This commit is contained in:
Ramiro Polla committed 2026-11-05 19:43:24 +00:00
1 parent f5fd844e9b
commit 2b507718fc
1 file changed
+13 -12
Download Patch File Download Diff File Expand all files Collapse all files
libswscale/ops_memcpy.c
+13 -12
View File
@@ -42,7 +42,6 @@ static void process_memcpy(const SwsOpExec *exec, const void *priv,
int x_start, int y_start, int x_end, int y_end)
{
const MemcpyPriv *p = priv;
const int lines = y_end - y_start;
av_assert1(x_start == 0 && x_end == exec->width);
for (int i = 0; i < p->num_planes; i++) {
@@ -57,21 +56,23 @@ static void process_memcpy(const SwsOpExec *exec, const void *priv,
continue; /* plane was already ref'd */
}
const int bytes = x_end * exec->block_size_out[i];
const int use_loop = out_stride < 0 ||
out_stride > bytes + SWS_MAX_PADDING;
const size_t span = (size_t) (lines - 1) * out_stride + bytes;
if (!in && !use_loop) {
memset(out, p->clear_value[i], span);
} else if (!in) {
for (int y = y_start; y < y_end; y++) {
size_t bytes = x_end * exec->block_size_out[i];
int lines = y_end - y_start;
/* Replace the per-line loop with a bulk memset/memcpy when possible. */
if ((!in || out_stride == in_stride) &&
out_stride >= 0 && (size_t) out_stride <= bytes + SWS_MAX_PADDING) {
bytes += (size_t) (lines - 1) * out_stride;
lines = 1;
}
if (!in) {
for (int y = 0; y < lines; y++) {
memset(out, p->clear_value[i], bytes);
out += out_stride;
}
} else if (out_stride == in_stride && !use_loop) {
memcpy(out, in, span);
} else {
for (int y = y_start; y < y_end; y++) {
for (int y = 0; y < lines; y++) {
memcpy(out, in, bytes);
out += out_stride;
in += in_stride;