Merge "Fix a bug in vpxenc reading raw input frame"
diff --git a/build/make/configure.sh b/build/make/configure.sh
index 0106737..8dcb9bb 100755
--- a/build/make/configure.sh
+++ b/build/make/configure.sh
@@ -1078,7 +1078,7 @@
msvs_arch_dir=x86-msvs
vc_version=${tgt_cc##vs}
case $vc_version in
- 7|8|9)
+ 7|8|9|10)
echo "${tgt_cc} does not support avx/avx2, disabling....."
RTCD_OPTIONS="${RTCD_OPTIONS}--disable-avx --disable-avx2 "
soft_disable avx
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 50cb98c..121947b 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -381,11 +381,6 @@
const int stride = 4 << b_width_log2(plane_bsize);
return base + raster_block_offset(plane_bsize, raster_block, stride);
}
-static uint8_t* raster_block_offset_uint8(BLOCK_SIZE plane_bsize,
- int raster_block, uint8_t *base,
- int stride) {
- return base + raster_block_offset(plane_bsize, raster_block, stride);
-}
static void txfrm_block_to_raster_xy(BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, int block,
diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index 2a33844..c65184f 100644
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -174,12 +174,10 @@
static INLINE void transpose_4x4(__m128i *res) {
const __m128i tr0_0 = _mm_unpacklo_epi16(res[0], res[1]);
- const __m128i tr0_1 = _mm_unpacklo_epi16(res[2], res[3]);
- res[0] = _mm_unpacklo_epi32(tr0_0, tr0_1);
- res[2] = _mm_unpackhi_epi32(tr0_0, tr0_1);
+ const __m128i tr0_1 = _mm_unpackhi_epi16(res[0], res[1]);
- res[1] = _mm_unpackhi_epi64(res[0], res[0]);
- res[3] = _mm_unpackhi_epi64(res[2], res[2]);
+ res[0] = _mm_unpacklo_epi16(tr0_0, tr0_1);
+ res[1] = _mm_unpackhi_epi16(tr0_0, tr0_1);
}
static void idct4_1d_sse2(__m128i *in) {
@@ -192,8 +190,8 @@
transpose_4x4(in);
// stage 1
- u[0] = _mm_unpacklo_epi16(in[0], in[2]);
- u[1] = _mm_unpacklo_epi16(in[1], in[3]);
+ u[0] = _mm_unpacklo_epi16(in[0], in[1]);
+ u[1] = _mm_unpackhi_epi16(in[0], in[1]);
v[0] = _mm_madd_epi16(u[0], k__cospi_p16_p16);
v[1] = _mm_madd_epi16(u[0], k__cospi_p16_m16);
v[2] = _mm_madd_epi16(u[1], k__cospi_p24_m08);
@@ -209,16 +207,13 @@
v[2] = _mm_srai_epi32(u[2], DCT_CONST_BITS);
v[3] = _mm_srai_epi32(u[3], DCT_CONST_BITS);
- u[0] = _mm_packs_epi32(v[0], v[2]);
- u[1] = _mm_packs_epi32(v[1], v[3]);
- u[2] = _mm_unpackhi_epi64(u[0], u[0]);
- u[3] = _mm_unpackhi_epi64(u[1], u[1]);
+ u[0] = _mm_packs_epi32(v[0], v[1]);
+ u[1] = _mm_packs_epi32(v[3], v[2]);
// stage 2
- in[0] = _mm_add_epi16(u[0], u[3]);
- in[1] = _mm_add_epi16(u[1], u[2]);
- in[2] = _mm_sub_epi16(u[1], u[2]);
- in[3] = _mm_sub_epi16(u[0], u[3]);
+ in[0] = _mm_add_epi16(u[0], u[1]);
+ in[1] = _mm_sub_epi16(u[0], u[1]);
+ in[1] = _mm_shuffle_epi32(in[1], 0x4E);
}
static void iadst4_1d_sse2(__m128i *in) {
@@ -232,13 +227,14 @@
__m128i u[8], v[8], in7;
transpose_4x4(in);
- in7 = _mm_add_epi16(in[0], in[3]);
- in7 = _mm_sub_epi16(in7, in[2]);
+ in7 = _mm_srli_si128(in[1], 8);
+ in7 = _mm_add_epi16(in7, in[0]);
+ in7 = _mm_sub_epi16(in7, in[1]);
- u[0] = _mm_unpacklo_epi16(in[0], in[2]);
- u[1] = _mm_unpacklo_epi16(in[1], in[3]);
+ u[0] = _mm_unpacklo_epi16(in[0], in[1]);
+ u[1] = _mm_unpackhi_epi16(in[0], in[1]);
u[2] = _mm_unpacklo_epi16(in7, kZero);
- u[3] = _mm_unpacklo_epi16(in[1], kZero);
+ u[3] = _mm_unpackhi_epi16(in[0], kZero);
v[0] = _mm_madd_epi16(u[0], k__sinpi_p01_p04); // s0 + s3
v[1] = _mm_madd_epi16(u[1], k__sinpi_p03_p02); // s2 + s5
@@ -265,22 +261,18 @@
u[2] = _mm_srai_epi32(v[2], DCT_CONST_BITS);
u[3] = _mm_srai_epi32(v[3], DCT_CONST_BITS);
- in[0] = _mm_packs_epi32(u[0], u[2]);
- in[1] = _mm_packs_epi32(u[1], u[3]);
- in[2] = _mm_unpackhi_epi64(in[0], in[0]);
- in[3] = _mm_unpackhi_epi64(in[1], in[1]);
+ in[0] = _mm_packs_epi32(u[0], u[1]);
+ in[1] = _mm_packs_epi32(u[2], u[3]);
}
void vp9_iht4x4_16_add_sse2(const int16_t *input, uint8_t *dest, int stride,
int tx_type) {
- __m128i in[4];
+ __m128i in[2];
const __m128i zero = _mm_setzero_si128();
const __m128i eight = _mm_set1_epi16(8);
- in[0] = _mm_loadl_epi64((const __m128i *)input);
- in[1] = _mm_loadl_epi64((const __m128i *)(input + 4));
- in[2] = _mm_loadl_epi64((const __m128i *)(input + 8));
- in[3] = _mm_loadl_epi64((const __m128i *)(input + 12));
+ in[0]= _mm_loadu_si128((const __m128i *)(input));
+ in[1]= _mm_loadu_si128((const __m128i *)(input + 8));
switch (tx_type) {
case 0: // DCT_DCT
@@ -307,18 +299,35 @@
// Final round and shift
in[0] = _mm_add_epi16(in[0], eight);
in[1] = _mm_add_epi16(in[1], eight);
- in[2] = _mm_add_epi16(in[2], eight);
- in[3] = _mm_add_epi16(in[3], eight);
in[0] = _mm_srai_epi16(in[0], 4);
in[1] = _mm_srai_epi16(in[1], 4);
- in[2] = _mm_srai_epi16(in[2], 4);
- in[3] = _mm_srai_epi16(in[3], 4);
- RECON_AND_STORE4X4(dest, in[0]);
- RECON_AND_STORE4X4(dest, in[1]);
- RECON_AND_STORE4X4(dest, in[2]);
- RECON_AND_STORE4X4(dest, in[3]);
+ // Reconstruction and Store
+ {
+ __m128i d0 = _mm_cvtsi32_si128(*(const int *)(dest));
+ __m128i d2 = _mm_cvtsi32_si128(*(const int *)(dest + stride * 2));
+ d0 = _mm_unpacklo_epi32(d0,
+ _mm_cvtsi32_si128(*(const int *) (dest + stride)));
+ d2 = _mm_unpacklo_epi32(d2, _mm_cvtsi32_si128(
+ *(const int *) (dest + stride * 3)));
+ d0 = _mm_unpacklo_epi8(d0, zero);
+ d2 = _mm_unpacklo_epi8(d2, zero);
+ d0 = _mm_add_epi16(d0, in[0]);
+ d2 = _mm_add_epi16(d2, in[1]);
+ d0 = _mm_packus_epi16(d0, d2);
+ // store result[0]
+ *(int *)dest = _mm_cvtsi128_si32(d0);
+ // store result[1]
+ d0 = _mm_srli_si128(d0, 4);
+ *(int *)(dest + stride) = _mm_cvtsi128_si32(d0);
+ // store result[2]
+ d0 = _mm_srli_si128(d0, 4);
+ *(int *)(dest + stride * 2) = _mm_cvtsi128_si32(d0);
+ // store result[3]
+ d0 = _mm_srli_si128(d0, 4);
+ *(int *)(dest + stride * 3) = _mm_cvtsi128_si32(d0);
+ }
}
#define TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, \
diff --git a/vp9/encoder/vp9_rdopt.c b/vp9/encoder/vp9_rdopt.c
index e92beb3..63be00e 100644
--- a/vp9/encoder/vp9_rdopt.c
+++ b/vp9/encoder/vp9_rdopt.c
@@ -1027,10 +1027,10 @@
struct macroblockd_plane *pd = &xd->plane[0];
const int src_stride = p->src.stride;
const int dst_stride = pd->dst.stride;
- uint8_t *src_init = raster_block_offset_uint8(BLOCK_8X8, ib,
- p->src.buf, src_stride);
- uint8_t *dst_init = raster_block_offset_uint8(BLOCK_8X8, ib,
- pd->dst.buf, dst_stride);
+ const uint8_t *src_init = &p->src.buf[raster_block_offset(BLOCK_8X8, ib,
+ src_stride)];
+ uint8_t *dst_init = &pd->dst.buf[raster_block_offset(BLOCK_8X8, ib,
+ dst_stride)];
int16_t *src_diff, *coeff;
ENTROPY_CONTEXT ta[2], tempa[2];
@@ -1072,7 +1072,7 @@
int64_t ssz;
const int16_t *scan;
const int16_t *nb;
- uint8_t *src = src_init + idx * 4 + idy * 4 * src_stride;
+ const uint8_t *src = src_init + idx * 4 + idy * 4 * src_stride;
uint8_t *dst = dst_init + idx * 4 + idy * 4 * dst_stride;
const int block = ib + idy * 2 + idx;
TX_TYPE tx_type;
@@ -1554,16 +1554,16 @@
const int height = plane_block_height(bsize, pd);
int idx, idy;
- uint8_t *const src = raster_block_offset_uint8(BLOCK_8X8, i,
- p->src.buf, p->src.stride);
- uint8_t *const dst = raster_block_offset_uint8(BLOCK_8X8, i,
- pd->dst.buf, pd->dst.stride);
+ const uint8_t *const src = &p->src.buf[raster_block_offset(BLOCK_8X8, i,
+ p->src.stride)];
+ uint8_t *const dst = &pd->dst.buf[raster_block_offset(BLOCK_8X8, i,
+ pd->dst.stride)];
int64_t thisdistortion = 0, thissse = 0;
int thisrate = 0, ref;
const int is_compound = has_second_ref(&mi->mbmi);
for (ref = 0; ref < 1 + is_compound; ++ref) {
- const uint8_t *pre = raster_block_offset_uint8(BLOCK_8X8, i,
- pd->pre[ref].buf, pd->pre[ref].stride);
+ const uint8_t *pre = &pd->pre[ref].buf[raster_block_offset(BLOCK_8X8, i,
+ pd->pre[ref].stride)];
vp9_build_inter_predictor(pre, pd->pre[ref].stride,
dst, pd->dst.stride,
&mi->bmi[i].as_mv[ref].as_mv,
@@ -1651,14 +1651,13 @@
struct macroblock_plane *const p = &x->plane[0];
struct macroblockd_plane *const pd = &x->e_mbd.plane[0];
- p->src.buf = raster_block_offset_uint8(BLOCK_8X8, i, p->src.buf,
- p->src.stride);
+ p->src.buf = &p->src.buf[raster_block_offset(BLOCK_8X8, i, p->src.stride)];
assert(((intptr_t)pd->pre[0].buf & 0x7) == 0);
- pd->pre[0].buf = raster_block_offset_uint8(BLOCK_8X8, i, pd->pre[0].buf,
- pd->pre[0].stride);
+ pd->pre[0].buf = &pd->pre[0].buf[raster_block_offset(BLOCK_8X8, i,
+ pd->pre[0].stride)];
if (has_second_ref(mbmi))
- pd->pre[1].buf = raster_block_offset_uint8(BLOCK_8X8, i, pd->pre[1].buf,
- pd->pre[1].stride);
+ pd->pre[1].buf = &pd->pre[1].buf[raster_block_offset(BLOCK_8X8, i,
+ pd->pre[1].stride)];
}
static INLINE void mi_buf_restore(MACROBLOCK *x, struct buf_2d orig_src,
