generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

vp9: coef context reading optimization.

Browse Source
This commit is contained in:
Ronald S. Bultje
2014-01-21 20:08:18 -05:00
committed by Clément Bœsch
parent 9aeca1c572
commit cdfcd06126
1 changed files with 17 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
libavcodec/vp9.c
View File

@@ -2057,14 +2057,20 @@ static void decode_coeffs(AVCodecContext *ctx)
const int16_t *y_band_counts = band_counts[b->tx]; const int16_t *y_band_counts = band_counts[b->tx];
const int16_t *uv_band_counts = band_counts[b->uvtx]; const int16_t *uv_band_counts = band_counts[b->uvtx];
#define MERGE(la, end, step, rd) \
for (n = 0; n < end; n += step) \
la[n] = !!rd(&la[n])
#define MERGE_CTX(step, rd) \
do { \
MERGE(l, end_y, step, rd); \
MERGE(a, end_x, step, rd); \
} while (0)
/* y tokens */ /* y tokens */
if (b->tx > TX_4X4) { // FIXME slow switch (b->tx) {
for (y = 0; y < end_y; y += step1d) case TX_8X8: MERGE_CTX(2, AV_RN16A); break;
for (x = 1; x < step1d; x++) case TX_16X16: MERGE_CTX(4, AV_RN32A); break;
l[y] |= l[y + x]; case TX_32X32: MERGE_CTX(8, AV_RN64A); break;
for (x = 0; x < end_x; x += step1d)
for (y = 1; y < step1d; y++)
a[x] |= a[x + y];
} }
for (n = 0, y = 0; y < end_y; y += step1d) { for (n = 0, y = 0; y < end_y; y += step1d) {
for (x = 0; x < end_x; x += step1d, n += step) { for (x = 0; x < end_x; x += step1d, n += step) {
@@ -2100,13 +2106,10 @@ static void decode_coeffs(AVCodecContext *ctx)
for (pl = 0; pl < 2; pl++) { for (pl = 0; pl < 2; pl++) {
a = &s->above_uv_nnz_ctx[pl][col]; a = &s->above_uv_nnz_ctx[pl][col];
l = &s->left_uv_nnz_ctx[pl][row & 7]; l = &s->left_uv_nnz_ctx[pl][row & 7];
if (b->uvtx > TX_4X4) { // FIXME slow switch (b->uvtx) {
for (y = 0; y < end_y; y += uvstep1d) case TX_8X8: MERGE_CTX(2, AV_RN16A); break;
for (x = 1; x < uvstep1d; x++) case TX_16X16: MERGE_CTX(4, AV_RN32A); break;
l[y] |= l[y + x]; case TX_32X32: MERGE_CTX(8, AV_RN64A); break;
for (x = 0; x < end_x; x += uvstep1d)
for (y = 1; y < uvstep1d; y++)
a[x] |= a[x + y];
} }
for (n = 0, y = 0; y < end_y; y += uvstep1d) { for (n = 0, y = 0; y < end_y; y += uvstep1d) {
for (x = 0; x < end_x; x += uvstep1d, n += uvstep) { for (x = 0; x < end_x; x += uvstep1d, n += uvstep) {