diff options
author | Dmitry Kovalev <dkovalev@google.com> | 2013-10-22 13:05:24 -0700 |
---|---|---|
committer | Gerrit Code Review <gerrit@gerrit.golo.chromium.org> | 2013-10-22 13:05:24 -0700 |
commit | 9f09618bd4d1b90b0897fa8ddcb0d3c73b3d9b25 (patch) | |
tree | 6e0865490136e90e5c8de2aae33e7393f471b68b /vp9/encoder | |
parent | c9af3de1557a6c99fada8b9b608ef14ba8f24a85 (diff) | |
parent | 190c2b4591039a2acef4964581e0d24d82de0d61 (diff) | |
download | libvpx-9f09618bd4d1b90b0897fa8ddcb0d3c73b3d9b25.tar libvpx-9f09618bd4d1b90b0897fa8ddcb0d3c73b3d9b25.tar.gz libvpx-9f09618bd4d1b90b0897fa8ddcb0d3c73b3d9b25.tar.bz2 libvpx-9f09618bd4d1b90b0897fa8ddcb0d3c73b3d9b25.zip |
Merge "Using stride (# of elements) instead of pitch (bytes) in fdct4x4."
Diffstat (limited to 'vp9/encoder')
-rw-r--r-- | vp9/encoder/vp9_dct.c | 14 | ||||
-rw-r--r-- | vp9/encoder/vp9_encodemb.c | 4 | ||||
-rw-r--r-- | vp9/encoder/vp9_onyx_if.c | 4 | ||||
-rw-r--r-- | vp9/encoder/vp9_rdopt.c | 4 | ||||
-rw-r--r-- | vp9/encoder/x86/vp9_dct_sse2.c | 3 |
5 files changed, 13 insertions, 16 deletions
diff --git a/vp9/encoder/vp9_dct.c b/vp9/encoder/vp9_dct.c index 461df635e..550cdee60 100644 --- a/vp9/encoder/vp9_dct.c +++ b/vp9/encoder/vp9_dct.c @@ -36,14 +36,13 @@ static void fdct4(const int16_t *input, int16_t *output) { output[3] = dct_const_round_shift(temp2); } -void vp9_short_fdct4x4_c(int16_t *input, int16_t *output, int pitch) { +void vp9_short_fdct4x4_c(int16_t *input, int16_t *output, int stride) { // The 2D transform is done with two passes which are actually pretty // similar. In the first one, we transform the columns and transpose // the results. In the second one, we transform the rows. To achieve that, // as the first pass results are transposed, we tranpose the columns (that // is the transposed rows) and transpose the results (so that it goes back // in normal/row positions). - const int stride = pitch >> 1; int pass; // We need an intermediate buffer between passes. int16_t intermediate[4 * 4]; @@ -586,18 +585,17 @@ void vp9_short_fht8x8_c(int16_t *input, int16_t *output, /* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per pixel. */ -void vp9_short_walsh4x4_c(int16_t *input, int16_t *output, int pitch) { +void vp9_short_walsh4x4_c(int16_t *input, int16_t *output, int stride) { int i; int a1, b1, c1, d1, e1; int16_t *ip = input; int16_t *op = output; - int pitch_short = pitch >> 1; for (i = 0; i < 4; i++) { - a1 = ip[0 * pitch_short]; - b1 = ip[1 * pitch_short]; - c1 = ip[2 * pitch_short]; - d1 = ip[3 * pitch_short]; + a1 = ip[0 * stride]; + b1 = ip[1 * stride]; + c1 = ip[2 * stride]; + d1 = ip[3 * stride]; a1 += b1; d1 = d1 - c1; diff --git a/vp9/encoder/vp9_encodemb.c b/vp9/encoder/vp9_encodemb.c index c3a231702..3358fbbe9 100644 --- a/vp9/encoder/vp9_encodemb.c +++ b/vp9/encoder/vp9_encodemb.c @@ -402,7 +402,7 @@ void vp9_xform_quant(int plane, int block, BLOCK_SIZE plane_bsize, xoff = 4 * (block & twmask); yoff = 4 * (block >> twl); src_diff = p->src_diff + 4 * bw * yoff + xoff; - x->fwd_txm4x4(src_diff, coeff, bw * 8); + x->fwd_txm4x4(src_diff, coeff, bw * 4); vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant, p->quant_shift, qcoeff, dqcoeff, pd->dequant, p->zbin_extra, eob, scan, iscan); @@ -612,7 +612,7 @@ void vp9_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize, if (tx_type != DCT_DCT) vp9_short_fht4x4(src_diff, coeff, bw * 4, tx_type); else - x->fwd_txm4x4(src_diff, coeff, bw * 8); + x->fwd_txm4x4(src_diff, coeff, bw * 4); vp9_quantize_b(coeff, 16, x->skip_block, p->zbin, p->round, p->quant, p->quant_shift, qcoeff, dqcoeff, pd->dequant, p->zbin_extra, eob, scan, iscan); diff --git a/vp9/encoder/vp9_onyx_if.c b/vp9/encoder/vp9_onyx_if.c index f6b2a2876..a2556f4e8 100644 --- a/vp9/encoder/vp9_onyx_if.c +++ b/vp9/encoder/vp9_onyx_if.c @@ -959,9 +959,9 @@ void vp9_set_speed_features(VP9_COMP *cpi) { sf->optimize_coefficients = 0; } - cpi->mb.fwd_txm4x4 = vp9_short_fdct4x4; + cpi->mb.fwd_txm4x4 = vp9_short_fdct4x4; if (cpi->oxcf.lossless || cpi->mb.e_mbd.lossless) { - cpi->mb.fwd_txm4x4 = vp9_short_walsh4x4; + cpi->mb.fwd_txm4x4 = vp9_short_walsh4x4; } cpi->mb.quantize_b_4x4 = vp9_regular_quantize_b_4x4; diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c index 2fd0fbf94..b6f37e9e1 100644 --- a/vp9/encoder/vp9_rdopt.c +++ b/vp9/encoder/vp9_rdopt.c @@ -1089,7 +1089,7 @@ static int64_t rd_pick_intra4x4block(VP9_COMP *cpi, MACROBLOCK *x, int ib, vp9_short_fht4x4(src_diff, coeff, 8, tx_type); x->quantize_b_4x4(x, block, tx_type, 16); } else { - x->fwd_txm4x4(src_diff, coeff, 16); + x->fwd_txm4x4(src_diff, coeff, 8); x->quantize_b_4x4(x, block, tx_type, 16); } @@ -1563,7 +1563,7 @@ static int64_t encode_inter_mb_segment(VP9_COMP *cpi, k += (idy * 2 + idx); coeff = BLOCK_OFFSET(p->coeff, k); x->fwd_txm4x4(raster_block_offset_int16(BLOCK_8X8, k, p->src_diff), - coeff, 16); + coeff, 8); x->quantize_b_4x4(x, k, DCT_DCT, 16); thisdistortion += vp9_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff, k), 16, &ssz); diff --git a/vp9/encoder/x86/vp9_dct_sse2.c b/vp9/encoder/x86/vp9_dct_sse2.c index 88c133bf5..fa60e80eb 100644 --- a/vp9/encoder/x86/vp9_dct_sse2.c +++ b/vp9/encoder/x86/vp9_dct_sse2.c @@ -12,14 +12,13 @@ #include "vp9/common/vp9_idct.h" // for cospi constants #include "vpx_ports/mem.h" -void vp9_short_fdct4x4_sse2(int16_t *input, int16_t *output, int pitch) { +void vp9_short_fdct4x4_sse2(int16_t *input, int16_t *output, int stride) { // The 2D transform is done with two passes which are actually pretty // similar. In the first one, we transform the columns and transpose // the results. In the second one, we transform the rows. To achieve that, // as the first pass results are transposed, we tranpose the columns (that // is the transposed rows) and transpose the results (so that it goes back // in normal/row positions). - const int stride = pitch >> 1; int pass; // Constants // When we use them, in one case, they are all the same. In all others |