Merge "Fix integer overflow issue in rtc coding flow intra mode search"
diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index b37d8e3..66ca4bb 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -933,14 +933,15 @@
INSTANTIATE_TEST_CASE_P(
MSA, Trans16x16DCT,
::testing::Values(
- make_tuple(&vp9_fdct16x16_c,
+ make_tuple(&vp9_fdct16x16_msa,
&vp9_idct16x16_256_add_msa, 0, VPX_BITS_8)));
INSTANTIATE_TEST_CASE_P(
MSA, Trans16x16HT,
::testing::Values(
- make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
- make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
- make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
- make_tuple(&vp9_fht16x16_c, &vp9_iht16x16_256_add_msa, 3, VPX_BITS_8)));
+ make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 0, VPX_BITS_8),
+ make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 1, VPX_BITS_8),
+ make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 2, VPX_BITS_8),
+ make_tuple(&vp9_fht16x16_msa, &vp9_iht16x16_256_add_msa, 3,
+ VPX_BITS_8)));
#endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
} // namespace
diff --git a/vp9/common/mips/msa/vp9_macros_msa.h b/vp9/common/mips/msa/vp9_macros_msa.h
index 2043e13..2f2390b 100644
--- a/vp9/common/mips/msa/vp9_macros_msa.h
+++ b/vp9/common/mips/msa/vp9_macros_msa.h
@@ -380,6 +380,17 @@
out3 = (v8i16)__msa_ilvl_d((v2i64)out2, (v2i64)out2); \
}
+/* Description : Load 2 vectors of signed word elements with stride
+ Arguments : Inputs - psrc (source pointer to load from)
+ - stride
+ Outputs - out0, out1
+ Return Type - signed word
+*/
+#define LD_SW2(psrc, stride, out0, out1) { \
+ out0 = LD_SW((psrc)); \
+ out1 = LD_SW((psrc) + stride); \
+}
+
/* Description : Store vectors of 16 byte elements with stride
Arguments : Inputs - in0, in1, stride
Outputs - pdst (destination pointer to store to)
@@ -777,6 +788,24 @@
CLIP_SH2_0_255(in2, in3); \
}
+/* Description : Addition of 4 signed word elements
+ 4 signed word elements of input vector are added together and
+ the resulting integer sum is returned
+ Arguments : Inputs - in (signed word vector)
+ Outputs - sum_m (i32 sum)
+ Return Type - signed word
+*/
+#define HADD_SW_S32(in) ({ \
+ v2i64 res0_m, res1_m; \
+ int32_t sum_m; \
+ \
+ res0_m = __msa_hadd_s_d((v4i32)in, (v4i32)in); \
+ res1_m = __msa_splati_d(res0_m, 1); \
+ res0_m = res0_m + res1_m; \
+ sum_m = __msa_copy_s_w((v4i32)res0_m, 0); \
+ sum_m; \
+})
+
/* Description : Horizontal addition of unsigned byte vector elements
Arguments : Inputs - in0, in1
Outputs - out0, out1
@@ -1073,8 +1102,8 @@
Outputs - in0, in1, in2, in3 (in place)
Return Type - unsigned halfword
Details : Each unsigned halfword element from 'in0' is saturated to the
- value generated with (sat_val+1) bit range
- Results are in placed to original vectors
+ value generated with (sat_val+1) bit range.
+ The results are in placed to original vectors
*/
#define SAT_UH2(RTYPE, in0, in1, sat_val) { \
in0 = (RTYPE)__msa_sat_u_h((v8u16)in0, sat_val); \
@@ -1096,7 +1125,7 @@
Return Type - unsigned halfword
Details : Each unsigned halfword element from 'in0' is saturated to the
value generated with (sat_val+1) bit range
- Results are in placed to original vectors
+ The results are in placed to original vectors
*/
#define SAT_SH2(RTYPE, in0, in1, sat_val) { \
in0 = (RTYPE)__msa_sat_s_h((v8i16)in0, sat_val); \
@@ -1216,10 +1245,10 @@
Outputs - in0, in1 (in-place)
Return Type - as per RTYPE
Details : Each unsigned byte element from input vector 'in0' is
- logically xor'ed with 128 and result is in-place stored in
+ logically xor'ed with 128 and the result is in-place stored in
'in0' vector
Each unsigned byte element from input vector 'in1' is
- logically xor'ed with 128 and result is in-place stored in
+ logically xor'ed with 128 and the result is in-place stored in
'in1' vector
Similar for other pairs
*/
@@ -1271,13 +1300,28 @@
}
#define ADDS_SH4_SH(...) ADDS_SH4(v8i16, __VA_ARGS__)
+/* Description : Shift left all elements of vector (generic for all data types)
+ Arguments : Inputs - in0, in1, in2, in3, shift
+ Outputs - in0, in1, in2, in3 (in place)
+ Return Type - as per input vector RTYPE
+ Details : Each element of vector 'in0' is left shifted by 'shift' and
+ the result is in place written to 'in0'
+ Similar for other pairs
+*/
+#define SLLI_4V(in0, in1, in2, in3, shift) { \
+ in0 = in0 << shift; \
+ in1 = in1 << shift; \
+ in2 = in2 << shift; \
+ in3 = in3 << shift; \
+}
+
/* Description : Arithmetic shift right all elements of vector
(generic for all data types)
Arguments : Inputs - in0, in1, in2, in3, shift
Outputs - in0, in1, in2, in3 (in place)
Return Type - as per input vector RTYPE
Details : Each element of vector 'in0' is right shifted by 'shift' and
- result is in place written to 'in0'
+ the result is in place written to 'in0'
Here, 'shift' is GP variable passed in
Similar for other pairs
*/
@@ -1427,6 +1471,34 @@
out7 = in0 - in7; \
}
+/* Description : Butterfly of 16 input vectors
+ Arguments : Inputs - in0 ... in15
+ Outputs - out0 .. out15
+ Details : Butterfly operation
+*/
+#define BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, \
+ in8, in9, in10, in11, in12, in13, in14, in15, \
+ out0, out1, out2, out3, out4, out5, out6, out7, \
+ out8, out9, out10, out11, out12, out13, out14, out15) { \
+ out0 = in0 + in15; \
+ out1 = in1 + in14; \
+ out2 = in2 + in13; \
+ out3 = in3 + in12; \
+ out4 = in4 + in11; \
+ out5 = in5 + in10; \
+ out6 = in6 + in9; \
+ out7 = in7 + in8; \
+ \
+ out8 = in7 - in8; \
+ out9 = in6 - in9; \
+ out10 = in5 - in10; \
+ out11 = in4 - in11; \
+ out12 = in3 - in12; \
+ out13 = in2 - in13; \
+ out14 = in1 - in14; \
+ out15 = in0 - in15; \
+}
+
/* Description : Transposes input 8x8 byte block
Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
(input 8x8 byte block)
@@ -1640,7 +1712,7 @@
Outputs - out_m
Return Type - unsigned byte
Details : Signed byte even elements from 'in0' and 'in1' are packed
- together in one vector and the resulted vector is xor'ed with
+ together in one vector and the resulting vector is xor'ed with
128 to shift the range from signed to unsigned byte
*/
#define PCKEV_XORI128_UB(in0, in1) ({ \
diff --git a/vp9/common/vp9_blockd.h b/vp9/common/vp9_blockd.h
index 53ae921..64d379c 100644
--- a/vp9/common/vp9_blockd.h
+++ b/vp9/common/vp9_blockd.h
@@ -161,6 +161,8 @@
int up_available;
int left_available;
+ const vp9_prob (*partition_probs)[PARTITION_TYPES - 1];
+
/* Distance of MB away from frame edges */
int mb_to_left_edge;
int mb_to_right_edge;
diff --git a/vp9/common/vp9_onyxc_int.h b/vp9/common/vp9_onyxc_int.h
index 3af2a41..1811d76 100644
--- a/vp9/common/vp9_onyxc_int.h
+++ b/vp9/common/vp9_onyxc_int.h
@@ -162,7 +162,8 @@
int show_existing_frame;
// Flag signaling that the frame is encoded using only INTRA modes.
- int intra_only;
+ uint8_t intra_only;
+ uint8_t last_intra_only;
int allow_high_precision_mv;
@@ -335,6 +336,18 @@
return ALIGN_POWER_OF_TWO(n_mis, MI_BLOCK_SIZE_LOG2);
}
+static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
+ return cm->frame_type == KEY_FRAME || cm->intra_only;
+}
+
+static INLINE void set_partition_probs(const VP9_COMMON *const cm,
+ MACROBLOCKD *const xd) {
+ xd->partition_probs =
+ frame_is_intra_only(cm) ?
+ &vp9_kf_partition_probs[0] :
+ (const vp9_prob (*)[PARTITION_TYPES - 1])cm->fc->partition_prob;
+}
+
static INLINE void init_macroblockd(VP9_COMMON *cm, MACROBLOCKD *xd) {
int i;
@@ -355,16 +368,13 @@
xd->above_seg_context = cm->above_seg_context;
xd->mi_stride = cm->mi_stride;
xd->error_info = &cm->error;
+
+ set_partition_probs(cm, xd);
}
-static INLINE int frame_is_intra_only(const VP9_COMMON *const cm) {
- return cm->frame_type == KEY_FRAME || cm->intra_only;
-}
-
-static INLINE const vp9_prob* get_partition_probs(const VP9_COMMON *cm,
+static INLINE const vp9_prob* get_partition_probs(const MACROBLOCKD *xd,
int ctx) {
- return frame_is_intra_only(cm) ? vp9_kf_partition_probs[ctx]
- : cm->fc->partition_prob[ctx];
+ return xd->partition_probs[ctx];
}
static INLINE void set_skip_context(MACROBLOCKD *xd, int mi_row, int mi_col) {
diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index 5326087..4a9970c 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -1029,7 +1029,7 @@
specialize qw/vp9_fht8x8 sse2/;
add_proto qw/void vp9_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
- specialize qw/vp9_fht16x16 sse2/;
+ specialize qw/vp9_fht16x16 sse2 msa/;
add_proto qw/void vp9_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fwht4x4/, "$mmx_x86inc";
@@ -1047,10 +1047,10 @@
specialize qw/vp9_fdct8x8 sse2 neon/, "$ssse3_x86_64";
add_proto qw/void vp9_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct16x16_1 sse2/;
+ specialize qw/vp9_fdct16x16_1 sse2 msa/;
add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct16x16 sse2/;
+ specialize qw/vp9_fdct16x16 sse2 msa/;
add_proto qw/void vp9_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct32x32_1 sse2/;
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 6954721..9311d8d 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -771,14 +771,11 @@
xd->corrupted |= vp9_reader_has_error(r);
}
-static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd,
- int hbs,
- int mi_row, int mi_col, BLOCK_SIZE bsize,
- vp9_reader *r) {
+static PARTITION_TYPE read_partition(MACROBLOCKD *xd, int mi_row, int mi_col,
+ BLOCK_SIZE bsize, vp9_reader *r,
+ int has_rows, int has_cols) {
const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
- const vp9_prob *const probs = get_partition_probs(cm, ctx);
- const int has_rows = (mi_row + hbs) < cm->mi_rows;
- const int has_cols = (mi_col + hbs) < cm->mi_cols;
+ const vp9_prob *const probs = get_partition_probs(xd, ctx);
FRAME_COUNTS *counts = xd->counts;
PARTITION_TYPE p;
@@ -805,11 +802,13 @@
const int hbs = num_8x8_blocks_wide_lookup[bsize] / 2;
PARTITION_TYPE partition;
BLOCK_SIZE subsize;
+ const int has_rows = (mi_row + hbs) < cm->mi_rows;
+ const int has_cols = (mi_col + hbs) < cm->mi_cols;
if (mi_row >= cm->mi_rows || mi_col >= cm->mi_cols)
return;
- partition = read_partition(cm, xd, hbs, mi_row, mi_col, bsize, r);
+ partition = read_partition(xd, mi_row, mi_col, bsize, r, has_rows, has_cols);
subsize = get_subsize(bsize, partition);
if (bsize == BLOCK_8X8) {
decode_block(pbi, xd, tile, mi_row, mi_col, r, subsize);
@@ -820,12 +819,12 @@
break;
case PARTITION_HORZ:
decode_block(pbi, xd, tile, mi_row, mi_col, r, subsize);
- if (mi_row + hbs < cm->mi_rows)
+ if (has_rows)
decode_block(pbi, xd, tile, mi_row + hbs, mi_col, r, subsize);
break;
case PARTITION_VERT:
decode_block(pbi, xd, tile, mi_row, mi_col, r, subsize);
- if (mi_col + hbs < cm->mi_cols)
+ if (has_cols)
decode_block(pbi, xd, tile, mi_row, mi_col + hbs, r, subsize);
break;
case PARTITION_SPLIT:
@@ -1648,6 +1647,7 @@
size_t sz;
cm->last_frame_type = cm->frame_type;
+ cm->last_intra_only = cm->intra_only;
if (vp9_rb_read_literal(rb, 2) != VP9_FRAME_MARKER)
vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
@@ -1996,8 +1996,9 @@
cm->use_prev_frame_mvs = !cm->error_resilient_mode &&
cm->width == cm->last_width &&
cm->height == cm->last_height &&
- !cm->intra_only &&
- cm->last_show_frame;
+ !cm->last_intra_only &&
+ cm->last_show_frame &&
+ (cm->last_frame_type != KEY_FRAME);
vp9_setup_block_planes(xd, cm->subsampling_x, cm->subsampling_y);
diff --git a/vp9/decoder/vp9_decodemv.c b/vp9/decoder/vp9_decodemv.c
index d34926d..8a8d8dd 100644
--- a/vp9/decoder/vp9_decodemv.c
+++ b/vp9/decoder/vp9_decodemv.c
@@ -599,19 +599,20 @@
MV_REF* frame_mvs = cm->cur_frame->mvs + mi_row * cm->mi_cols + mi_col;
int w, h;
- if (frame_is_intra_only(cm))
+ if (frame_is_intra_only(cm)) {
read_intra_frame_mode_info(cm, xd, mi_row, mi_col, r);
- else
+ } else {
read_inter_frame_mode_info(pbi, xd, tile, mi_row, mi_col, r);
- for (h = 0; h < y_mis; ++h) {
- MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
- for (w = 0; w < x_mis; ++w) {
- MV_REF *const mv = frame_mv + w;
- mv->ref_frame[0] = mi->mbmi.ref_frame[0];
- mv->ref_frame[1] = mi->mbmi.ref_frame[1];
- mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
- mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
+ for (h = 0; h < y_mis; ++h) {
+ MV_REF *const frame_mv = frame_mvs + h * cm->mi_cols;
+ for (w = 0; w < x_mis; ++w) {
+ MV_REF *const mv = frame_mv + w;
+ mv->ref_frame[0] = mi->mbmi.ref_frame[0];
+ mv->ref_frame[1] = mi->mbmi.ref_frame[1];
+ mv->mv[0].as_int = mi->mbmi.mv[0].as_int;
+ mv->mv[1].as_int = mi->mbmi.mv[1].as_int;
+ }
}
}
}
diff --git a/vp9/encoder/mips/msa/vp9_fdct16x16_msa.c b/vp9/encoder/mips/msa/vp9_fdct16x16_msa.c
new file mode 100644
index 0000000..a3ebfab
--- /dev/null
+++ b/vp9/encoder/mips/msa/vp9_fdct16x16_msa.c
@@ -0,0 +1,688 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <assert.h>
+
+#include "./vp9_rtcd.h"
+#include "vp9/encoder/mips/msa/vp9_fdct_msa.h"
+
+static void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
+ int32_t src_stride) {
+ v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
+ v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
+ v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
+ -cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
+ v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
+ cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
+ v8i16 coeff2 = { -cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64,
+ 0, 0, 0, 0 };
+
+ LD_SH16(input, src_stride,
+ in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ SLLI_4V(in0, in1, in2, in3, 2);
+ SLLI_4V(in4, in5, in6, in7, 2);
+ SLLI_4V(in8, in9, in10, in11, 2);
+ SLLI_4V(in12, in13, in14, in15, 2);
+ ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
+ ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
+ VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
+ ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
+ SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
+ SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
+
+ tmp_ptr += 16;
+
+ /* stp 1 */
+ ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
+ ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
+
+ cnst4 = __msa_splati_h(coeff, 0);
+ stp25 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
+
+ cnst5 = __msa_splati_h(coeff, 1);
+ cnst5 = __msa_ilvev_h(cnst5, cnst4);
+ stp22 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
+ stp24 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
+ stp23 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
+
+ /* stp2 */
+ BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
+ BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
+ ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
+ ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
+ SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
+ cnst0 = __msa_ilvev_h(cnst0, cnst1);
+ stp26 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
+
+ cnst0 = __msa_splati_h(coeff, 4);
+ cnst1 = __msa_ilvev_h(cnst1, cnst0);
+ stp21 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
+
+ BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
+ ILVRL_H2_SH(in15, in8, vec1, vec0);
+ SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
+ cnst0 = __msa_ilvev_h(cnst0, cnst1);
+
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr);
+
+ cnst0 = __msa_splati_h(coeff2, 0);
+ cnst0 = __msa_ilvev_h(cnst1, cnst0);
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr + 224);
+
+ ILVRL_H2_SH(in14, in9, vec1, vec0);
+ SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
+ cnst1 = __msa_ilvev_h(cnst1, cnst0);
+
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
+ ST_SH(in8, tmp_ptr + 128);
+
+ cnst1 = __msa_splati_h(coeff2, 2);
+ cnst0 = __msa_ilvev_h(cnst0, cnst1);
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr + 96);
+
+ SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
+ cnst1 = __msa_ilvev_h(cnst1, cnst0);
+
+ stp25 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
+
+ cnst1 = __msa_splati_h(coeff, 3);
+ cnst1 = __msa_ilvev_h(cnst0, cnst1);
+ stp22 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
+
+ /* stp4 */
+ ADD2(stp34, stp25, stp33, stp22, in13, in10);
+
+ ILVRL_H2_SH(in13, in10, vec1, vec0);
+ SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
+ cnst0 = __msa_ilvev_h(cnst0, cnst1);
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr + 64);
+
+ cnst0 = __msa_splati_h(coeff2, 1);
+ cnst0 = __msa_ilvev_h(cnst1, cnst0);
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr + 160);
+
+ SUB2(stp34, stp25, stp33, stp22, in12, in11);
+ ILVRL_H2_SH(in12, in11, vec1, vec0);
+ SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
+ cnst1 = __msa_ilvev_h(cnst1, cnst0);
+
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
+ ST_SH(in8, tmp_ptr + 192);
+
+ cnst1 = __msa_splati_h(coeff2, 3);
+ cnst0 = __msa_ilvev_h(cnst0, cnst1);
+ in8 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
+ ST_SH(in8, tmp_ptr + 32);
+}
+
+static void fdct16x8_1d_row(int16_t *input, int16_t *output) {
+ v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+
+ LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+ TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
+ ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
+ ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
+ ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
+ SRA_4V(in0, in1, in2, in3, 2);
+ SRA_4V(in4, in5, in6, in7, 2);
+ SRA_4V(in8, in9, in10, in11, 2);
+ SRA_4V(in12, in13, in14, in15, 2);
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
+ in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5,
+ tmp6, tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
+ ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
+ VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
+ LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
+ VP9_FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+ TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
+ tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
+ ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
+ TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
+ tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
+ ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
+}
+
+void vp9_fdct16x16_msa(const int16_t *input, int16_t *output,
+ int32_t src_stride) {
+ int32_t i;
+ DECLARE_ALIGNED(32, int16_t, tmp_buf[16 * 16]);
+
+ /* column transform */
+ for (i = 0; i < 2; ++i) {
+ fdct8x16_1d_column((input + 8 * i), (&tmp_buf[0] + 8 * i), src_stride);
+ }
+
+ /* row transform */
+ for (i = 0; i < 2; ++i) {
+ fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i)));
+ }
+}
+
+void vp9_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
+ out[1] = 0;
+
+ out[0] = VP9_LD_HADD(input, stride);
+ out[0] += VP9_LD_HADD(input + 8, stride);
+ out[0] += VP9_LD_HADD(input + 16 * 8, stride);
+ out[0] += VP9_LD_HADD(input + 16 * 8 + 8, stride);
+ out[0] >>= 1;
+}
+
+static void fadst16_cols_step1_msa(const int16_t *input, int32_t stride,
+ const int32_t *const0, int16_t *int_buf) {
+ v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
+ v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
+ v4i32 k0, k1, k2, k3;
+
+ /* load input data */
+ r0 = LD_SH(input);
+ r15 = LD_SH(input + 15 * stride);
+ r7 = LD_SH(input + 7 * stride);
+ r8 = LD_SH(input + 8 * stride);
+ SLLI_4V(r0, r15, r7, r8, 2);
+
+ /* stage 1 */
+ LD_SW2(const0, 4, k0, k1);
+ LD_SW2(const0 + 8, 4, k2, k3);
+ VP9_MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
+
+ r3 = LD_SH(input + 3 * stride);
+ r4 = LD_SH(input + 4 * stride);
+ r11 = LD_SH(input + 11 * stride);
+ r12 = LD_SH(input + 12 * stride);
+ SLLI_4V(r3, r4, r11, r12, 2);
+
+ LD_SW2(const0 + 4 * 4, 4, k0, k1);
+ LD_SW2(const0 + 4 * 6, 4, k2, k3);
+ VP9_MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
+
+ /* stage 2 */
+ BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
+ ST_SH2(tp0, tp2, int_buf, 8);
+ ST_SH2(tp1, tp3, int_buf + 4 * 8, 8);
+
+ LD_SW2(const0 + 4 * 8, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 10);
+ VP9_MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
+
+ ST_SH2(h0, h1, int_buf + 8 * 8, 8);
+ ST_SH2(h3, h2, int_buf + 12 * 8, 8);
+
+ r9 = LD_SH(input + 9 * stride);
+ r6 = LD_SH(input + 6 * stride);
+ r1 = LD_SH(input + stride);
+ r14 = LD_SH(input + 14 * stride);
+ SLLI_4V(r9, r6, r1, r14, 2);
+
+ LD_SW2(const0 + 4 * 11, 4, k0, k1);
+ LD_SW2(const0 + 4 * 13, 4, k2, k3);
+ VP9_MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
+
+ ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
+
+ r13 = LD_SH(input + 13 * stride);
+ r2 = LD_SH(input + 2 * stride);
+ r5 = LD_SH(input + 5 * stride);
+ r10 = LD_SH(input + 10 * stride);
+ SLLI_4V(r13, r2, r5, r10, 2);
+
+ LD_SW2(const0 + 4 * 15, 4, k0, k1);
+ LD_SW2(const0 + 4 * 17, 4, k2, k3);
+ VP9_MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
+
+ ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
+
+ BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
+ ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
+}
+
+static void fadst16_cols_step2_msa(int16_t *int_buf, const int32_t *const0,
+ int16_t *out) {
+ int16_t *out_ptr = out + 128;
+ v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
+ v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
+ v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
+ v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
+ v4i32 k0, k1, k2, k3;
+
+ LD_SH2(int_buf + 3 * 8, 4 * 8, g13, g15);
+ LD_SH2(int_buf + 11 * 8, 4 * 8, g5, g7);
+ LD_SW2(const0 + 4 * 19, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 21);
+ VP9_MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
+
+ tp0 = LD_SH(int_buf + 4 * 8);
+ tp1 = LD_SH(int_buf + 5 * 8);
+ tp3 = LD_SH(int_buf + 10 * 8);
+ tp2 = LD_SH(int_buf + 14 * 8);
+ LD_SW2(const0 + 4 * 22, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 24);
+ VP9_MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
+ out4 = -out4;
+ ST_SH(out4, (out + 3 * 16));
+ ST_SH(out5, (out_ptr + 4 * 16));
+
+ h1 = LD_SH(int_buf + 9 * 8);
+ h3 = LD_SH(int_buf + 12 * 8);
+ VP9_MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
+ out13 = -out13;
+ ST_SH(out12, (out + 2 * 16));
+ ST_SH(out13, (out_ptr + 5 * 16));
+
+ tp0 = LD_SH(int_buf);
+ tp1 = LD_SH(int_buf + 8);
+ tp2 = LD_SH(int_buf + 2 * 8);
+ tp3 = LD_SH(int_buf + 6 * 8);
+
+ BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
+ out1 = -out1;
+ ST_SH(out0, (out));
+ ST_SH(out1, (out_ptr + 7 * 16));
+
+ h0 = LD_SH(int_buf + 8 * 8);
+ h2 = LD_SH(int_buf + 13 * 8);
+
+ BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
+ out8 = -out8;
+ ST_SH(out8, (out + 16));
+ ST_SH(out9, (out_ptr + 6 * 16));
+
+ /* stage 4 */
+ LD_SW2(const0 + 4 * 25, 4, k0, k1);
+ LD_SW2(const0 + 4 * 27, 4, k2, k3);
+ VP9_MADD_SHORT(h10, h11, k1, k2, out2, out3);
+ ST_SH(out2, (out + 7 * 16));
+ ST_SH(out3, (out_ptr));
+
+ VP9_MADD_SHORT(out6, out7, k0, k3, out6, out7);
+ ST_SH(out6, (out + 4 * 16));
+ ST_SH(out7, (out_ptr + 3 * 16));
+
+ VP9_MADD_SHORT(out10, out11, k0, k3, out10, out11);
+ ST_SH(out10, (out + 6 * 16));
+ ST_SH(out11, (out_ptr + 16));
+
+ VP9_MADD_SHORT(out14, out15, k1, k2, out14, out15);
+ ST_SH(out14, (out + 5 * 16));
+ ST_SH(out15, (out_ptr + 2 * 16));
+}
+
+static void fadst16_transpose_postproc_msa(int16_t *input, int16_t *out) {
+ v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
+ v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
+
+ /* load input data */
+ LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
+ TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
+ r0, r1, r2, r3, r4, r5, r6, r7);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r0, r1);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r2, r3);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r4, r5);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r6, r7);
+ ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
+ out += 64;
+
+ LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
+ TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
+ r8, r9, r10, r11, r12, r13, r14, r15);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r8, r9);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r10, r11);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r12, r13);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r14, r15);
+ ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
+ out += 64;
+
+ /* load input data */
+ input += 128;
+ LD_SH8(input, 16, l0, l1, l2, l3, l4, l5, l6, l7);
+ TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
+ r0, r1, r2, r3, r4, r5, r6, r7);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r0, r1);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r2, r3);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r4, r5);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r6, r7);
+ ST_SH8(r0, r1, r2, r3, r4, r5, r6, r7, out, 8);
+ out += 64;
+
+ LD_SH8(input + 8, 16, l8, l9, l10, l11, l12, l13, l14, l15);
+ TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
+ r8, r9, r10, r11, r12, r13, r14, r15);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r8, r9);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r10, r11);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r12, r13);
+ VP9_FDCT_POSTPROC_2V_NEG_H(r14, r15);
+ ST_SH8(r8, r9, r10, r11, r12, r13, r14, r15, out, 8);
+}
+
+static void fadst16_rows_step1_msa(int16_t *input, const int32_t *const0,
+ int16_t *int_buf) {
+ v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
+ v8i16 tp0, tp1, tp2, tp3, g0, g1, g2, g3, g8, g9, g10, g11, h0, h1, h2, h3;
+ v4i32 k0, k1, k2, k3;
+
+ /* load input data */
+ r0 = LD_SH(input);
+ r7 = LD_SH(input + 7 * 8);
+ r8 = LD_SH(input + 8 * 8);
+ r15 = LD_SH(input + 15 * 8);
+
+ /* stage 1 */
+ LD_SW2(const0, 4, k0, k1);
+ LD_SW2(const0 + 4 * 2, 4, k2, k3);
+ VP9_MADD_BF(r15, r0, r7, r8, k0, k1, k2, k3, g0, g1, g2, g3);
+
+ r3 = LD_SH(input + 3 * 8);
+ r4 = LD_SH(input + 4 * 8);
+ r11 = LD_SH(input + 11 * 8);
+ r12 = LD_SH(input + 12 * 8);
+
+ LD_SW2(const0 + 4 * 4, 4, k0, k1);
+ LD_SW2(const0 + 4 * 6, 4, k2, k3);
+ VP9_MADD_BF(r11, r4, r3, r12, k0, k1, k2, k3, g8, g9, g10, g11);
+
+ /* stage 2 */
+ BUTTERFLY_4(g0, g2, g10, g8, tp0, tp2, tp3, tp1);
+ ST_SH2(tp0, tp1, int_buf, 4 * 8);
+ ST_SH2(tp2, tp3, int_buf + 8, 4 * 8);
+
+ LD_SW2(const0 + 4 * 8, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 10);
+ VP9_MADD_BF(g1, g3, g9, g11, k0, k1, k2, k0, h0, h1, h2, h3);
+ ST_SH2(h0, h3, int_buf + 8 * 8, 4 * 8);
+ ST_SH2(h1, h2, int_buf + 9 * 8, 4 * 8);
+
+ r1 = LD_SH(input + 8);
+ r6 = LD_SH(input + 6 * 8);
+ r9 = LD_SH(input + 9 * 8);
+ r14 = LD_SH(input + 14 * 8);
+
+ LD_SW2(const0 + 4 * 11, 4, k0, k1);
+ LD_SW2(const0 + 4 * 13, 4, k2, k3);
+ VP9_MADD_BF(r9, r6, r1, r14, k0, k1, k2, k3, g0, g1, g2, g3);
+ ST_SH2(g1, g3, int_buf + 3 * 8, 4 * 8);
+
+ r2 = LD_SH(input + 2 * 8);
+ r5 = LD_SH(input + 5 * 8);
+ r10 = LD_SH(input + 10 * 8);
+ r13 = LD_SH(input + 13 * 8);
+
+ LD_SW2(const0 + 4 * 15, 4, k0, k1);
+ LD_SW2(const0 + 4 * 17, 4, k2, k3);
+ VP9_MADD_BF(r13, r2, r5, r10, k0, k1, k2, k3, h0, h1, h2, h3);
+ ST_SH2(h1, h3, int_buf + 11 * 8, 4 * 8);
+ BUTTERFLY_4(h0, h2, g2, g0, tp0, tp1, tp2, tp3);
+ ST_SH4(tp0, tp1, tp2, tp3, int_buf + 2 * 8, 4 * 8);
+}
+
+static void fadst16_rows_step2_msa(int16_t *int_buf, const int32_t *const0,
+ int16_t *out) {
+ int16_t *out_ptr = out + 8;
+ v8i16 tp0, tp1, tp2, tp3, g5, g7, g13, g15;
+ v8i16 h0, h1, h2, h3, h4, h5, h6, h7, h10, h11;
+ v8i16 out0, out1, out2, out3, out4, out5, out6, out7;
+ v8i16 out8, out9, out10, out11, out12, out13, out14, out15;
+ v4i32 k0, k1, k2, k3;
+
+ g13 = LD_SH(int_buf + 3 * 8);
+ g15 = LD_SH(int_buf + 7 * 8);
+ g5 = LD_SH(int_buf + 11 * 8);
+ g7 = LD_SH(int_buf + 15 * 8);
+
+ LD_SW2(const0 + 4 * 19, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 21);
+ VP9_MADD_BF(g7, g5, g15, g13, k0, k1, k2, k0, h4, h5, h6, h7);
+
+ tp0 = LD_SH(int_buf + 4 * 8);
+ tp1 = LD_SH(int_buf + 5 * 8);
+ tp3 = LD_SH(int_buf + 10 * 8);
+ tp2 = LD_SH(int_buf + 14 * 8);
+
+ LD_SW2(const0 + 4 * 22, 4, k0, k1);
+ k2 = LD_SW(const0 + 4 * 24);
+ VP9_MADD_BF(tp0, tp1, tp2, tp3, k0, k1, k2, k0, out4, out6, out5, out7);
+ out4 = -out4;
+ ST_SH(out4, (out + 3 * 16));
+ ST_SH(out5, (out_ptr + 4 * 16));
+
+ h1 = LD_SH(int_buf + 9 * 8);
+ h3 = LD_SH(int_buf + 12 * 8);
+ VP9_MADD_BF(h1, h3, h5, h7, k0, k1, k2, k0, out12, out14, out13, out15);
+ out13 = -out13;
+ ST_SH(out12, (out + 2 * 16));
+ ST_SH(out13, (out_ptr + 5 * 16));
+
+ tp0 = LD_SH(int_buf);
+ tp1 = LD_SH(int_buf + 8);
+ tp2 = LD_SH(int_buf + 2 * 8);
+ tp3 = LD_SH(int_buf + 6 * 8);
+
+ BUTTERFLY_4(tp0, tp1, tp3, tp2, out0, out1, h11, h10);
+ out1 = -out1;
+ ST_SH(out0, (out));
+ ST_SH(out1, (out_ptr + 7 * 16));
+
+ h0 = LD_SH(int_buf + 8 * 8);
+ h2 = LD_SH(int_buf + 13 * 8);
+ BUTTERFLY_4(h0, h2, h6, h4, out8, out9, out11, out10);
+ out8 = -out8;
+ ST_SH(out8, (out + 16));
+ ST_SH(out9, (out_ptr + 6 * 16));
+
+ /* stage 4 */
+ LD_SW2(const0 + 4 * 25, 4, k0, k1);
+ LD_SW2(const0 + 4 * 27, 4, k2, k3);
+ VP9_MADD_SHORT(h10, h11, k1, k2, out2, out3);
+ ST_SH(out2, (out + 7 * 16));
+ ST_SH(out3, (out_ptr));
+
+ VP9_MADD_SHORT(out6, out7, k0, k3, out6, out7);
+ ST_SH(out6, (out + 4 * 16));
+ ST_SH(out7, (out_ptr + 3 * 16));
+
+ VP9_MADD_SHORT(out10, out11, k0, k3, out10, out11);
+ ST_SH(out10, (out + 6 * 16));
+ ST_SH(out11, (out_ptr + 16));
+
+ VP9_MADD_SHORT(out14, out15, k1, k2, out14, out15);
+ ST_SH(out14, (out + 5 * 16));
+ ST_SH(out15, (out_ptr + 2 * 16));
+}
+
+static void fadst16_transpose_msa(int16_t *input, int16_t *out) {
+ v8i16 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, r13, r14, r15;
+ v8i16 l0, l1, l2, l3, l4, l5, l6, l7, l8, l9, l10, l11, l12, l13, l14, l15;
+
+ /* load input data */
+ LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
+ l4, l12, l5, l13, l6, l14, l7, l15);
+ TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
+ r0, r1, r2, r3, r4, r5, r6, r7);
+ TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
+ r8, r9, r10, r11, r12, r13, r14, r15);
+ ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
+ ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
+ out += 16 * 8;
+
+ /* load input data */
+ input += 128;
+ LD_SH16(input, 8, l0, l8, l1, l9, l2, l10, l3, l11,
+ l4, l12, l5, l13, l6, l14, l7, l15);
+ TRANSPOSE8x8_SH_SH(l0, l1, l2, l3, l4, l5, l6, l7,
+ r0, r1, r2, r3, r4, r5, r6, r7);
+ TRANSPOSE8x8_SH_SH(l8, l9, l10, l11, l12, l13, l14, l15,
+ r8, r9, r10, r11, r12, r13, r14, r15);
+ ST_SH8(r0, r8, r1, r9, r2, r10, r3, r11, out, 8);
+ ST_SH8(r4, r12, r5, r13, r6, r14, r7, r15, (out + 64), 8);
+}
+
+static void postproc_fdct16x8_1d_row(int16_t *intermediate, int16_t *output) {
+ int16_t *temp = intermediate;
+ int16_t *out = output;
+ v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11;
+ v8i16 in12, in13, in14, in15;
+
+ LD_SH8(temp, 16, in0, in1, in2, in3, in4, in5, in6, in7);
+ temp = intermediate + 8;
+ LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+ TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in0, in1);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in2, in3);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in6, in7);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in8, in9);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in10, in11);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in12, in13);
+ VP9_FDCT_POSTPROC_2V_NEG_H(in14, in15);
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ temp = intermediate;
+ ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, temp, 16);
+ VP9_FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
+ tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
+ temp = intermediate;
+ LD_SH8(temp, 16, in8, in9, in10, in11, in12, in13, in14, in15);
+ VP9_FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+ TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
+ tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
+ ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, out, 16);
+ TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
+ tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
+ out = output + 8;
+ ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, out, 16);
+}
+
+void vp9_fht16x16_msa(const int16_t *input, int16_t *output,
+ int32_t stride, int32_t tx_type) {
+ DECLARE_ALIGNED(32, int16_t, tmp[256]);
+ DECLARE_ALIGNED(32, int16_t, trans_buf[256]);
+ DECLARE_ALIGNED(32, int16_t, tmp_buf[128]);
+ int32_t i;
+ int16_t *ptmpbuf = &tmp_buf[0];
+ int16_t *trans = &trans_buf[0];
+ const int32_t const_arr[29 * 4] = {
+ 52707308, 52707308, 52707308, 52707308,
+ -1072430300, -1072430300, -1072430300, -1072430300,
+ 795618043, 795618043, 795618043, 795618043,
+ -721080468, -721080468, -721080468, -721080468,
+ 459094491, 459094491, 459094491, 459094491,
+ -970646691, -970646691, -970646691, -970646691,
+ 1010963856, 1010963856, 1010963856, 1010963856,
+ -361743294, -361743294, -361743294, -361743294,
+ 209469125, 209469125, 209469125, 209469125,
+ -1053094788, -1053094788, -1053094788, -1053094788,
+ 1053160324, 1053160324, 1053160324, 1053160324,
+ 639644520, 639644520, 639644520, 639644520,
+ -862444000, -862444000, -862444000, -862444000,
+ 1062144356, 1062144356, 1062144356, 1062144356,
+ -157532337, -157532337, -157532337, -157532337,
+ 260914709, 260914709, 260914709, 260914709,
+ -1041559667, -1041559667, -1041559667, -1041559667,
+ 920985831, 920985831, 920985831, 920985831,
+ -551995675, -551995675, -551995675, -551995675,
+ 596522295, 596522295, 596522295, 596522295,
+ 892853362, 892853362, 892853362, 892853362,
+ -892787826, -892787826, -892787826, -892787826,
+ 410925857, 410925857, 410925857, 410925857,
+ -992012162, -992012162, -992012162, -992012162,
+ 992077698, 992077698, 992077698, 992077698,
+ 759246145, 759246145, 759246145, 759246145,
+ -759180609, -759180609, -759180609, -759180609,
+ -759222975, -759222975, -759222975, -759222975,
+ 759288511, 759288511, 759288511, 759288511 };
+
+ switch (tx_type) {
+ case DCT_DCT:
+ /* column transform */
+ for (i = 0; i < 2; ++i) {
+ fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
+ }
+
+ /* row transform */
+ for (i = 0; i < 2; ++i) {
+ fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
+ }
+ break;
+ case ADST_DCT:
+ /* column transform */
+ for (i = 0; i < 2; ++i) {
+ fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
+ fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
+ }
+
+ /* row transform */
+ for (i = 0; i < 2; ++i) {
+ postproc_fdct16x8_1d_row(tmp + (128 * i), output + (128 * i));
+ }
+ break;
+ case DCT_ADST:
+ /* column transform */
+ for (i = 0; i < 2; ++i) {
+ fdct8x16_1d_column(input + 8 * i, tmp + 8 * i, stride);
+ }
+
+ fadst16_transpose_postproc_msa(tmp, trans);
+
+ /* row transform */
+ for (i = 0; i < 2; ++i) {
+ fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
+ fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
+ }
+
+ fadst16_transpose_msa(tmp, output);
+ break;
+ case ADST_ADST:
+ /* column transform */
+ for (i = 0; i < 2; ++i) {
+ fadst16_cols_step1_msa(input + (i << 3), stride, const_arr, ptmpbuf);
+ fadst16_cols_step2_msa(ptmpbuf, const_arr, tmp + (i << 3));
+ }
+
+ fadst16_transpose_postproc_msa(tmp, trans);
+
+ /* row transform */
+ for (i = 0; i < 2; ++i) {
+ fadst16_rows_step1_msa(trans + (i << 7), const_arr, ptmpbuf);
+ fadst16_rows_step2_msa(ptmpbuf, const_arr, tmp + (i << 7));
+ }
+
+ fadst16_transpose_msa(tmp, output);
+ break;
+ default:
+ assert(0);
+ break;
+ }
+}
diff --git a/vp9/encoder/mips/msa/vp9_fdct_msa.h b/vp9/encoder/mips/msa/vp9_fdct_msa.h
new file mode 100644
index 0000000..f461650
--- /dev/null
+++ b/vp9/encoder/mips/msa/vp9_fdct_msa.h
@@ -0,0 +1,253 @@
+/*
+ * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
+#define VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
+
+#include "vpx_ports/mem.h"
+#include "vp9/common/vp9_idct.h"
+#include "vp9/common/mips/msa/vp9_macros_msa.h"
+
+#define VP9_DOT_SHIFT_RIGHT_PCK_H(in0, in1, in2) ({ \
+ v8i16 dst_m; \
+ v4i32 tp0_m, tp1_m; \
+ \
+ DOTP_SH2_SW(in0, in1, in2, in2, tp1_m, tp0_m); \
+ SRARI_W2_SW(tp1_m, tp0_m, DCT_CONST_BITS); \
+ dst_m = __msa_pckev_h((v8i16)tp1_m, (v8i16)tp0_m); \
+ \
+ dst_m; \
+})
+
+#define VP9_MADD_SHORT(m0, m1, c0, c1, res0, res1) { \
+ v4i32 madd0_m, madd1_m, madd2_m, madd3_m; \
+ v8i16 madd_s0_m, madd_s1_m; \
+ \
+ ILVRL_H2_SH(m1, m0, madd_s0_m, madd_s1_m); \
+ DOTP_SH4_SW(madd_s0_m, madd_s1_m, madd_s0_m, madd_s1_m, \
+ c0, c0, c1, c1, madd0_m, madd1_m, madd2_m, madd3_m); \
+ SRARI_W4_SW(madd0_m, madd1_m, madd2_m, madd3_m, DCT_CONST_BITS); \
+ PCKEV_H2_SH(madd1_m, madd0_m, madd3_m, madd2_m, res0, res1); \
+}
+
+#define VP9_MADD_BF(inp0, inp1, inp2, inp3, cst0, cst1, cst2, cst3, \
+ out0, out1, out2, out3) { \
+ v8i16 madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m; \
+ v4i32 tmp0_m, tmp1_m, tmp2_m, tmp3_m, m4_m, m5_m; \
+ \
+ ILVRL_H2_SH(inp1, inp0, madd_s0_m, madd_s1_m); \
+ ILVRL_H2_SH(inp3, inp2, madd_s2_m, madd_s3_m); \
+ DOTP_SH4_SW(madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m, \
+ cst0, cst0, cst2, cst2, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
+ BUTTERFLY_4(tmp0_m, tmp1_m, tmp3_m, tmp2_m, \
+ m4_m, m5_m, tmp3_m, tmp2_m); \
+ SRARI_W4_SW(m4_m, m5_m, tmp2_m, tmp3_m, DCT_CONST_BITS); \
+ PCKEV_H2_SH(m5_m, m4_m, tmp3_m, tmp2_m, out0, out1); \
+ DOTP_SH4_SW(madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m, \
+ cst1, cst1, cst3, cst3, tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
+ BUTTERFLY_4(tmp0_m, tmp1_m, tmp3_m, tmp2_m, \
+ m4_m, m5_m, tmp3_m, tmp2_m); \
+ SRARI_W4_SW(m4_m, m5_m, tmp2_m, tmp3_m, DCT_CONST_BITS); \
+ PCKEV_H2_SH(m5_m, m4_m, tmp3_m, tmp2_m, out2, out3); \
+}
+
+#define VP9_LD_HADD(psrc, stride) ({ \
+ v8i16 in0_m, in1_m, in2_m, in3_m, in4_m, in5_m, in6_m, in7_m; \
+ v4i32 vec_w_m; \
+ \
+ LD_SH4((psrc), stride, in0_m, in1_m, in2_m, in3_m); \
+ ADD2(in0_m, in1_m, in2_m, in3_m, in0_m, in2_m); \
+ LD_SH4(((psrc) + 4 * stride), stride, in4_m, in5_m, in6_m, in7_m); \
+ ADD4(in4_m, in5_m, in6_m, in7_m, in0_m, in2_m, in4_m, in6_m, \
+ in4_m, in6_m, in0_m, in4_m); \
+ in0_m += in4_m; \
+ \
+ vec_w_m = __msa_hadd_s_w(in0_m, in0_m); \
+ HADD_SW_S32(vec_w_m); \
+})
+
+#define VP9_FDCT_POSTPROC_2V_NEG_H(vec0, vec1) { \
+ v8i16 tp0_m, tp1_m; \
+ v8i16 one_m = __msa_ldi_h(1); \
+ \
+ tp0_m = __msa_clti_s_h(vec0, 0); \
+ tp1_m = __msa_clti_s_h(vec1, 0); \
+ vec0 += 1; \
+ vec1 += 1; \
+ tp0_m = one_m & tp0_m; \
+ tp1_m = one_m & tp1_m; \
+ vec0 += tp0_m; \
+ vec1 += tp1_m; \
+ vec0 >>= 2; \
+ vec1 >>= 2; \
+}
+
+#define VP9_FDCT8x16_EVEN(in0, in1, in2, in3, in4, in5, in6, in7, \
+ out0, out1, out2, out3, out4, out5, out6, out7) { \
+ v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m; \
+ v8i16 x0_m, x1_m, x2_m, x3_m; \
+ v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64, \
+ cospi_4_64, cospi_28_64, cospi_12_64, cospi_20_64 }; \
+ \
+ /* FDCT stage1 */ \
+ BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
+ s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m); \
+ BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m); \
+ ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m); \
+ ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m); \
+ SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m); \
+ x1_m = __msa_ilvev_h(x1_m, x0_m); \
+ out4 = VP9_DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
+ \
+ SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m); \
+ x2_m = -x2_m; \
+ x2_m = __msa_ilvev_h(x3_m, x2_m); \
+ out6 = VP9_DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
+ \
+ out0 = VP9_DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
+ x2_m = __msa_splati_h(coeff_m, 2); \
+ x2_m = __msa_ilvev_h(x2_m, x3_m); \
+ out2 = VP9_DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
+ \
+ /* stage2 */ \
+ ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m); \
+ \
+ s6_m = VP9_DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
+ s5_m = VP9_DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
+ \
+ /* stage3 */ \
+ BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m); \
+ \
+ /* stage4 */ \
+ ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m); \
+ ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m); \
+ \
+ SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m); \
+ x1_m = __msa_ilvev_h(x0_m, x1_m); \
+ out1 = VP9_DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m); \
+ \
+ SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m); \
+ x2_m = __msa_ilvev_h(x3_m, x2_m); \
+ out5 = VP9_DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
+ \
+ x1_m = __msa_splati_h(coeff_m, 5); \
+ x0_m = -x0_m; \
+ x0_m = __msa_ilvev_h(x1_m, x0_m); \
+ out7 = VP9_DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m); \
+ \
+ x2_m = __msa_splati_h(coeff_m, 6); \
+ x3_m = -x3_m; \
+ x2_m = __msa_ilvev_h(x2_m, x3_m); \
+ out3 = VP9_DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
+}
+
+/* FDCT16x16 specific */
+#define VP9_FDCT8x16_ODD(input0, input1, input2, input3, \
+ input4, input5, input6, input7, \
+ out1, out3, out5, out7, \
+ out9, out11, out13, out15) { \
+ v8i16 stp21_m, stp22_m, stp23_m, stp24_m, stp25_m, stp26_m; \
+ v8i16 stp30_m, stp31_m, stp32_m, stp33_m, stp34_m, stp35_m; \
+ v8i16 stp36_m, stp37_m, vec0_m, vec1_m; \
+ v8i16 vec2_m, vec3_m, vec4_m, vec5_m, vec6_m; \
+ v8i16 cnst0_m, cnst1_m, cnst4_m, cnst5_m; \
+ v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
+ cospi_24_64, -cospi_8_64, -cospi_24_64, \
+ cospi_12_64, cospi_20_64 }; \
+ v8i16 coeff1_m = { cospi_2_64, cospi_30_64, cospi_14_64, \
+ cospi_18_64, cospi_10_64, cospi_22_64, \
+ cospi_6_64, cospi_26_64 }; \
+ v8i16 coeff2_m = { -cospi_2_64, -cospi_10_64, -cospi_18_64, \
+ -cospi_26_64, 0, 0, 0, 0 }; \
+ \
+ /* stp 1 */ \
+ ILVL_H2_SH(input2, input5, input3, input4, vec2_m, vec4_m); \
+ ILVR_H2_SH(input2, input5, input3, input4, vec3_m, vec5_m); \
+ \
+ cnst4_m = __msa_splati_h(coeff_m, 0); \
+ stp25_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst4_m); \
+ \
+ cnst5_m = __msa_splati_h(coeff_m, 1); \
+ cnst5_m = __msa_ilvev_h(cnst5_m, cnst4_m); \
+ stp22_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst5_m); \
+ stp24_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst4_m); \
+ stp23_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst5_m); \
+ \
+ /* stp2 */ \
+ BUTTERFLY_4(input0, input1, stp22_m, stp23_m, \
+ stp30_m, stp31_m, stp32_m, stp33_m); \
+ BUTTERFLY_4(input7, input6, stp25_m, stp24_m, \
+ stp37_m, stp36_m, stp35_m, stp34_m); \
+ \
+ ILVL_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec2_m, vec4_m); \
+ ILVR_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec3_m, vec5_m); \
+ \
+ SPLATI_H2_SH(coeff_m, 2, 3, cnst0_m, cnst1_m); \
+ cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ stp26_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
+ \
+ cnst0_m = __msa_splati_h(coeff_m, 4); \
+ cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ stp21_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
+ \
+ SPLATI_H2_SH(coeff_m, 5, 2, cnst0_m, cnst1_m); \
+ cnst1_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ stp25_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
+ \
+ cnst0_m = __msa_splati_h(coeff_m, 3); \
+ cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ stp22_m = VP9_DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
+ \
+ /* stp4 */ \
+ BUTTERFLY_4(stp30_m, stp37_m, stp26_m, stp21_m, \
+ vec6_m, vec2_m, vec4_m, vec5_m); \
+ BUTTERFLY_4(stp33_m, stp34_m, stp25_m, stp22_m, \
+ stp21_m, stp23_m, stp24_m, stp31_m); \
+ \
+ ILVRL_H2_SH(vec2_m, vec6_m, vec1_m, vec0_m); \
+ SPLATI_H2_SH(coeff1_m, 0, 1, cnst0_m, cnst1_m); \
+ cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ \
+ out1 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+ \
+ cnst0_m = __msa_splati_h(coeff2_m, 0); \
+ cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ out15 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+ \
+ ILVRL_H2_SH(vec4_m, vec5_m, vec1_m, vec0_m); \
+ SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
+ cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ \
+ out9 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
+ \
+ cnst1_m = __msa_splati_h(coeff2_m, 2); \
+ cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ out7 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+ \
+ ILVRL_H2_SH(stp23_m, stp21_m, vec1_m, vec0_m); \
+ SPLATI_H2_SH(coeff1_m, 4, 5, cnst0_m, cnst1_m); \
+ cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ out5 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+ \
+ cnst0_m = __msa_splati_h(coeff2_m, 1); \
+ cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ out11 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+ \
+ ILVRL_H2_SH(stp24_m, stp31_m, vec1_m, vec0_m); \
+ SPLATI_H2_SH(coeff1_m, 6, 7, cnst0_m, cnst1_m); \
+ cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
+ \
+ out13 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
+ \
+ cnst1_m = __msa_splati_h(coeff2_m, 3); \
+ cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
+ out3 = VP9_DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
+}
+#endif /* VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_ */
diff --git a/vp9/encoder/vp9_bitstream.c b/vp9/encoder/vp9_bitstream.c
index 092d265..1ebdd06 100644
--- a/vp9/encoder/vp9_bitstream.c
+++ b/vp9/encoder/vp9_bitstream.c
@@ -403,7 +403,7 @@
int hbs, int mi_row, int mi_col,
PARTITION_TYPE p, BLOCK_SIZE bsize, vp9_writer *w) {
const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
- const vp9_prob *const probs = get_partition_probs(cm, ctx);
+ const vp9_prob *const probs = xd->partition_probs[ctx];
const int has_rows = (mi_row + hbs) < cm->mi_rows;
const int has_cols = (mi_col + hbs) < cm->mi_cols;
@@ -481,9 +481,12 @@
static void write_modes(VP9_COMP *cpi,
const TileInfo *const tile, vp9_writer *w,
TOKENEXTRA **tok, const TOKENEXTRA *const tok_end) {
+ const VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
int mi_row, mi_col;
+ set_partition_probs(cm, xd);
+
for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
mi_row += MI_BLOCK_SIZE) {
vp9_zero(xd->left_seg_context);
diff --git a/vp9/encoder/vp9_encoder.c b/vp9/encoder/vp9_encoder.c
index ba38d64..ecd60d7 100644
--- a/vp9/encoder/vp9_encoder.c
+++ b/vp9/encoder/vp9_encoder.c
@@ -4165,7 +4165,7 @@
(is_two_pass_svc(cpi) &&
cpi->svc.encode_empty_frame_state != ENCODING))) {
vp9_rc_get_second_pass_params(cpi);
- } else {
+ } else if (oxcf->pass == 1) {
set_frame_size(cpi);
}
diff --git a/vp9/encoder/vp9_rd.c b/vp9/encoder/vp9_rd.c
index bbcbfe9..90ee1e4 100644
--- a/vp9/encoder/vp9_rd.c
+++ b/vp9/encoder/vp9_rd.c
@@ -265,6 +265,7 @@
void vp9_initialize_rd_consts(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
MACROBLOCK *const x = &cpi->td.mb;
+ MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
RD_OPT *const rd = &cpi->rd;
int i;
@@ -280,6 +281,7 @@
cm->frame_type != KEY_FRAME) ? 0 : 1;
set_block_thresholds(cm, rd);
+ set_partition_probs(cm, xd);
if (!cpi->sf.use_nonrd_pick_mode || cm->frame_type == KEY_FRAME)
fill_token_costs(x->token_costs, cm->fc->coef_probs);
@@ -287,7 +289,7 @@
if (cpi->sf.partition_search_type != VAR_BASED_PARTITION ||
cm->frame_type == KEY_FRAME) {
for (i = 0; i < PARTITION_CONTEXTS; ++i)
- vp9_cost_tokens(cpi->partition_cost[i], get_partition_probs(cm, i),
+ vp9_cost_tokens(cpi->partition_cost[i], get_partition_probs(xd, i),
vp9_partition_tree);
}
diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk
index 5415215..0926dff 100644
--- a/vp9/vp9cx.mk
+++ b/vp9/vp9cx.mk
@@ -152,4 +152,7 @@
VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_subtract_neon.c
VP9_CX_SRCS-$(HAVE_NEON) += encoder/arm/neon/vp9_variance_neon.c
+VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct16x16_msa.c
+VP9_CX_SRCS-$(HAVE_MSA) += encoder/mips/msa/vp9_fdct_msa.h
+
VP9_CX_SRCS-yes := $(filter-out $(VP9_CX_SRCS_REMOVE-yes),$(VP9_CX_SRCS-yes))
