diff options
| author | Ronald S. Bultje <rsbultje@gmail.com> | 2014-01-21 20:08:18 -0500 |
|---|---|---|
| committer | Clément Bœsch <u@pkh.me> | 2014-02-09 18:07:14 +0100 |
| commit | cdfcd061260176c39ac353fb0a5091d750f6fc11 (patch) | |
| tree | 7af4ce88096100f4b78cdb95dd5bb70f079c7ac1 /libavcodec/vp9.c | |
| parent | 9aeca1c572dcd446bba340eb6c5fa4f65e18d1e8 (diff) | |
vp9: coef context reading optimization.
Diffstat (limited to 'libavcodec/vp9.c')
| -rw-r--r-- | libavcodec/vp9.c | 31 |
1 files changed, 17 insertions, 14 deletions
diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c index 8099c7609b..8419c446d6 100644 --- a/libavcodec/vp9.c +++ b/libavcodec/vp9.c @@ -2057,14 +2057,20 @@ static void decode_coeffs(AVCodecContext *ctx) const int16_t *y_band_counts = band_counts[b->tx]; 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 */ - if (b->tx > TX_4X4) { // FIXME slow - for (y = 0; y < end_y; y += step1d) - for (x = 1; x < step1d; x++) - l[y] |= l[y + x]; - for (x = 0; x < end_x; x += step1d) - for (y = 1; y < step1d; y++) - a[x] |= a[x + y]; + switch (b->tx) { + case TX_8X8: MERGE_CTX(2, AV_RN16A); break; + case TX_16X16: MERGE_CTX(4, AV_RN32A); break; + case TX_32X32: MERGE_CTX(8, AV_RN64A); break; } for (n = 0, y = 0; y < end_y; y += step1d) { 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++) { a = &s->above_uv_nnz_ctx[pl][col]; l = &s->left_uv_nnz_ctx[pl][row & 7]; - if (b->uvtx > TX_4X4) { // FIXME slow - for (y = 0; y < end_y; y += uvstep1d) - for (x = 1; x < uvstep1d; x++) - l[y] |= l[y + x]; - for (x = 0; x < end_x; x += uvstep1d) - for (y = 1; y < uvstep1d; y++) - a[x] |= a[x + y]; + switch (b->uvtx) { + case TX_8X8: MERGE_CTX(2, AV_RN16A); break; + case TX_16X16: MERGE_CTX(4, AV_RN32A); break; + case TX_32X32: MERGE_CTX(8, AV_RN64A); break; } for (n = 0, y = 0; y < end_y; y += uvstep1d) { for (x = 0; x < end_x; x += uvstep1d, n += uvstep) { |