Move the 32x32 2D-DCT implementations from vp9/ to vpx_dsp/.
Change-Id: Id3980696f8b69906ff7a59ff9fb2b9013d60047d
#include <stdlib.h>
#include <string.h>
-#include "third_party/googletest/src/include/gtest/gtest.h"
#include "test/acm_random.h"
#include "test/clear_system_state.h"
#include "test/register_state_check.h"
#include "test/util.h"
+#include "third_party/googletest/src/include/gtest/gtest.h"
-#include "./vpx_config.h"
#include "./vp9_rtcd.h"
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
#include "vp9/common/vp9_entropy.h"
#include "vpx/vpx_codec.h"
#include "vpx/vpx_integer.h"
add_proto qw/void vp9_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct32x32_1 sse2/;
-
- add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct32x32 sse2/;
-
- add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct32x32_rd sse2/;
} else {
add_proto qw/void vp9_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
specialize qw/vp9_fht4x4 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 msa/;
-
- add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct32x32 sse2 avx2 msa/;
-
- add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_fdct32x32_rd sse2 avx2 msa/;
}
#
add_proto qw/void vp9_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_highbd_fdct32x32_1/;
- add_proto qw/void vp9_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_highbd_fdct32x32 sse2/;
-
- add_proto qw/void vp9_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
- specialize qw/vp9_highbd_fdct32x32_rd sse2/;
-
add_proto qw/void vp9_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
specialize qw/vp9_highbd_temporal_filter_apply/;
* be found in the AUTHORS file in the root of the source tree.
*/
-#include "./vp9_rtcd.h"
#include "vp9/encoder/mips/msa/vp9_fdct_msa.h"
-static void fdct8x32_1d_column_load_butterfly(const int16_t *input,
- int32_t src_stride,
- int16_t *temp_buff) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 step0, step1, step2, step3;
- v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
- v8i16 step0_1, step1_1, step2_1, step3_1;
-
- /* 1st and 2nd set */
- LD_SH4(input, src_stride, in0, in1, in2, in3);
- LD_SH4(input + (28 * src_stride), src_stride, in4, in5, in6, in7);
- LD_SH4(input + (4 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
- LD_SH4(input + (24 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
- SLLI_4V(in0, in1, in2, in3, 2);
- SLLI_4V(in4, in5, in6, in7, 2);
- SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
- SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
- BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
- step0, step1, step2, step3, in4, in5, in6, in7);
- BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
- ST_SH4(step0, step1, step2, step3, temp_buff, 8);
- ST_SH4(in4, in5, in6, in7, temp_buff + (28 * 8), 8);
- ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (4 * 8), 8);
- ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (24 * 8), 8);
-
- /* 3rd and 4th set */
- LD_SH4(input + (8 * src_stride), src_stride, in0, in1, in2, in3);
- LD_SH4(input + (20 * src_stride), src_stride, in4, in5, in6, in7);
- LD_SH4(input + (12 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
- LD_SH4(input + (16 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
- SLLI_4V(in0, in1, in2, in3, 2);
- SLLI_4V(in4, in5, in6, in7, 2);
- SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
- SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
- BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
- step0, step1, step2, step3, in4, in5, in6, in7);
- BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
- ST_SH4(step0, step1, step2, step3, temp_buff + (8 * 8), 8);
- ST_SH4(in4, in5, in6, in7, temp_buff + (20 * 8), 8);
- ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (12 * 8), 8);
- ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (15 * 8) + 8, 8);
-}
-
-static void fdct8x32_1d_column_even_store(int16_t *input, int16_t *temp) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v8i16 temp0, temp1;
-
- /* fdct even */
- LD_SH4(input, 8, in0, in1, in2, in3);
- LD_SH4(input + 96, 8, in12, in13, in14, in15);
- BUTTERFLY_8(in0, in1, in2, in3, in12, in13, in14, in15,
- vec0, vec1, vec2, vec3, in12, in13, in14, in15);
- LD_SH4(input + 32, 8, in4, in5, in6, in7);
- LD_SH4(input + 64, 8, in8, in9, in10, in11);
- BUTTERFLY_8(in4, in5, in6, in7, in8, in9, in10, in11,
- vec4, vec5, vec6, vec7, in8, in9, in10, in11);
-
- /* Stage 3 */
- ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
- BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
- DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp);
- ST_SH(temp1, temp + 512);
-
- DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 256);
- ST_SH(temp1, temp + 768);
-
- SUB4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, vec7, vec6, vec5, vec4);
- DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
- ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
- DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 128);
- ST_SH(temp1, temp + 896);
-
- SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
- DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 640);
- ST_SH(temp1, temp + 384);
-
- DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
- DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
- ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
- DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
- ADD2(in0, in1, in2, in3, vec0, vec7);
- DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 64);
- ST_SH(temp1, temp + 960);
-
- SUB2(in0, in1, in2, in3, in0, in2);
- DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 576);
- ST_SH(temp1, temp + 448);
-
- SUB2(in9, vec2, in14, vec5, vec2, vec5);
- DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
- SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
- DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 320);
- ST_SH(temp1, temp + 704);
-
- ADD2(in3, in2, in0, in1, vec3, vec4);
- DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
- FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
- ST_SH(temp0, temp + 192);
- ST_SH(temp1, temp + 832);
-}
-
-static void fdct8x32_1d_column_odd_store(int16_t *input, int16_t *temp_ptr) {
- v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
- v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
-
- in20 = LD_SH(input + 32);
- in21 = LD_SH(input + 40);
- in26 = LD_SH(input + 80);
- in27 = LD_SH(input + 88);
-
- DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
- DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
-
- in18 = LD_SH(input + 16);
- in19 = LD_SH(input + 24);
- in28 = LD_SH(input + 96);
- in29 = LD_SH(input + 104);
-
- vec4 = in19 - in20;
- ST_SH(vec4, input + 32);
- vec4 = in18 - in21;
- ST_SH(vec4, input + 40);
- vec4 = in29 - in26;
- ST_SH(vec4, input + 80);
- vec4 = in28 - in27;
- ST_SH(vec4, input + 88);
-
- in21 = in18 + in21;
- in20 = in19 + in20;
- in27 = in28 + in27;
- in26 = in29 + in26;
-
- LD_SH4(input + 48, 8, in22, in23, in24, in25);
- DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
- DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
-
- in16 = LD_SH(input);
- in17 = LD_SH(input + 8);
- in30 = LD_SH(input + 112);
- in31 = LD_SH(input + 120);
-
- vec4 = in17 - in22;
- ST_SH(vec4, input + 16);
- vec4 = in16 - in23;
- ST_SH(vec4, input + 24);
- vec4 = in31 - in24;
- ST_SH(vec4, input + 96);
- vec4 = in30 - in25;
- ST_SH(vec4, input + 104);
-
- ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
- DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
- DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
- ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
- DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
- ADD2(in27, in26, in25, in24, in23, in20);
- DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr);
- ST_SH(vec4, temp_ptr + 960);
-
- SUB2(in27, in26, in25, in24, in22, in21);
- DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr + 448);
- ST_SH(vec4, temp_ptr + 512);
-
- SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
- DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
- SUB2(in26, in27, in24, in25, in23, in20);
- DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec4, temp_ptr + 704);
- ST_SH(vec5, temp_ptr + 256);
-
- ADD2(in26, in27, in24, in25, in22, in21);
- DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec4, temp_ptr + 192);
- ST_SH(vec5, temp_ptr + 768);
-
- LD_SH4(input + 16, 8, in22, in23, in20, in21);
- LD_SH4(input + 80, 8, in26, in27, in24, in25);
- in16 = in20;
- in17 = in21;
- DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
- DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
- SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
- DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
- ADD2(in28, in29, in31, in30, in16, in19);
- DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr + 832);
- ST_SH(vec4, temp_ptr + 128);
-
- SUB2(in28, in29, in31, in30, in17, in18);
- DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr + 320);
- ST_SH(vec4, temp_ptr + 640);
- ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
- DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
- SUB2(in29, in28, in30, in31, in16, in19);
- DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr + 576);
- ST_SH(vec4, temp_ptr + 384);
-
- ADD2(in29, in28, in30, in31, in17, in18);
- DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
- FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
- ST_SH(vec5, temp_ptr + 64);
- ST_SH(vec4, temp_ptr + 896);
-}
-
-static void fdct8x32_1d_column(const int16_t *input, int32_t src_stride,
- int16_t *tmp_buf, int16_t *tmp_buf_big) {
- fdct8x32_1d_column_load_butterfly(input, src_stride, tmp_buf);
- fdct8x32_1d_column_even_store(tmp_buf, tmp_buf_big);
- fdct8x32_1d_column_odd_store(tmp_buf + 128, (tmp_buf_big + 32));
-}
-
-static void fdct8x32_1d_row_load_butterfly(int16_t *temp_buff,
- int16_t *output) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
- v8i16 step0, step1, step2, step3, step4, step5, step6, step7;
-
- LD_SH8(temp_buff, 32, in0, in1, in2, in3, in4, in5, in6, in7);
- LD_SH8(temp_buff + 24, 32, 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);
- BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
- in8, in9, in10, in11, in12, in13, in14, in15,
- step0, step1, step2, step3, step4, step5, step6, step7,
- in8, in9, in10, in11, in12, in13, in14, in15);
- ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7, output, 8);
- ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 24 * 8), 8);
-
- /* 2nd set */
- LD_SH8(temp_buff + 8, 32, in0, in1, in2, in3, in4, in5, in6, in7);
- LD_SH8(temp_buff + 16, 32, 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);
- BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
- in8, in9, in10, in11, in12, in13, in14, in15,
- step0, step1, step2, step3, step4, step5, step6, step7,
- in8, in9, in10, in11, in12, in13, in14, in15);
- ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7,
- (output + 8 * 8), 8);
- ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 16 * 8), 8);
-}
-
-static void fdct8x32_1d_row_even_4x(int16_t *input, int16_t *interm_ptr,
- int16_t *out) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
- v4i32 vec0_l, vec1_l, vec2_l, vec3_l, vec4_l, vec5_l, vec6_l, vec7_l;
- v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec4_r, vec5_r, vec6_r, vec7_r;
- v4i32 tmp0_w, tmp1_w, tmp2_w, tmp3_w;
-
- /* fdct32 even */
- /* stage 2 */
- LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
- LD_SH8(input + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
-
- BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
- in8, in9, in10, in11, in12, in13, in14, in15,
- vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
- in8, in9, in10, in11, in12, in13, in14, in15);
- ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, interm_ptr, 8);
- ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, interm_ptr + 64, 8);
-
- /* Stage 3 */
- UNPCK_SH_SW(vec0, vec0_l, vec0_r);
- UNPCK_SH_SW(vec1, vec1_l, vec1_r);
- UNPCK_SH_SW(vec2, vec2_l, vec2_r);
- UNPCK_SH_SW(vec3, vec3_l, vec3_r);
- UNPCK_SH_SW(vec4, vec4_l, vec4_r);
- UNPCK_SH_SW(vec5, vec5_l, vec5_r);
- UNPCK_SH_SW(vec6, vec6_l, vec6_r);
- UNPCK_SH_SW(vec7, vec7_l, vec7_r);
- ADD4(vec0_r, vec7_r, vec1_r, vec6_r, vec2_r, vec5_r, vec3_r, vec4_r,
- tmp0_w, tmp1_w, tmp2_w, tmp3_w);
- BUTTERFLY_4(tmp0_w, tmp1_w, tmp2_w, tmp3_w, vec4_r, vec6_r, vec7_r, vec5_r);
- ADD4(vec0_l, vec7_l, vec1_l, vec6_l, vec2_l, vec5_l, vec3_l, vec4_l,
- vec0_r, vec1_r, vec2_r, vec3_r);
-
- tmp3_w = vec0_r + vec3_r;
- vec0_r = vec0_r - vec3_r;
- vec3_r = vec1_r + vec2_r;
- vec1_r = vec1_r - vec2_r;
-
- DOTP_CONST_PAIR_W(vec4_r, vec6_r, tmp3_w, vec3_r, cospi_16_64,
- cospi_16_64, vec4_r, tmp3_w, vec6_r, vec3_r);
- FDCT32_POSTPROC_NEG_W(vec4_r);
- FDCT32_POSTPROC_NEG_W(tmp3_w);
- FDCT32_POSTPROC_NEG_W(vec6_r);
- FDCT32_POSTPROC_NEG_W(vec3_r);
- PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
- ST_SH2(vec5, vec4, out, 8);
-
- DOTP_CONST_PAIR_W(vec5_r, vec7_r, vec0_r, vec1_r, cospi_24_64,
- cospi_8_64, vec4_r, tmp3_w, vec6_r, vec3_r);
- FDCT32_POSTPROC_NEG_W(vec4_r);
- FDCT32_POSTPROC_NEG_W(tmp3_w);
- FDCT32_POSTPROC_NEG_W(vec6_r);
- FDCT32_POSTPROC_NEG_W(vec3_r);
- PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
- ST_SH2(vec5, vec4, out + 16, 8);
-
- LD_SH8(interm_ptr, 8, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
- SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
- DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
- ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
- DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, in5, in4);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 32);
- ST_SH(in5, out + 56);
-
- SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
- DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, in5, in4);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 40);
- ST_SH(in5, out + 48);
-
- LD_SH8(interm_ptr + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
- DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
- DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
- ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
- DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
- ADD2(in0, in1, in2, in3, vec0, vec7);
- DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, in5, in4);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 64);
- ST_SH(in5, out + 120);
-
- SUB2(in0, in1, in2, in3, in0, in2);
- DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, in5, in4);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 72);
- ST_SH(in5, out + 112);
-
- SUB2(in9, vec2, in14, vec5, vec2, vec5);
- DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
- SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
- DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, in5, in4);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 80);
- ST_SH(in5, out + 104);
-
- ADD2(in3, in2, in0, in1, vec3, vec4);
- DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, in4, in5);
- FDCT_POSTPROC_2V_NEG_H(in4, in5);
- ST_SH(in4, out + 96);
- ST_SH(in5, out + 88);
-}
-
-static void fdct8x32_1d_row_even(int16_t *temp, int16_t *out) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1;
-
- /* fdct32 even */
- /* stage 2 */
- LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
- LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
-
- BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
- in8, in9, in10, in11, in12, in13, in14, in15,
- vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
- in8, in9, in10, in11, in12, in13, in14, in15);
-
- /* Stage 3 */
- ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
- BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
- DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out);
- ST_SH(temp1, out + 8);
-
- DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 16);
- ST_SH(temp1, out + 24);
-
- SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
- DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
- ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
- DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 32);
- ST_SH(temp1, out + 56);
-
- SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
- DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 40);
- ST_SH(temp1, out + 48);
-
- DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
- DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
- ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
- DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
- ADD2(in0, in1, in2, in3, vec0, vec7);
- DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 64);
- ST_SH(temp1, out + 120);
-
- SUB2(in0, in1, in2, in3, in0, in2);
- DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 72);
- ST_SH(temp1, out + 112);
-
- SUB2(in9, vec2, in14, vec5, vec2, vec5);
- DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
- SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5)
- DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 80);
- ST_SH(temp1, out + 104);
-
- ADD2(in3, in2, in0, in1, vec3, vec4);
- DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
- FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
- ST_SH(temp0, out + 96);
- ST_SH(temp1, out + 88);
-}
-
-static void fdct8x32_1d_row_odd(int16_t *temp, int16_t *interm_ptr,
- int16_t *out) {
- v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
- v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
-
- in20 = LD_SH(temp + 32);
- in21 = LD_SH(temp + 40);
- in26 = LD_SH(temp + 80);
- in27 = LD_SH(temp + 88);
-
- DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
- DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
-
- in18 = LD_SH(temp + 16);
- in19 = LD_SH(temp + 24);
- in28 = LD_SH(temp + 96);
- in29 = LD_SH(temp + 104);
-
- vec4 = in19 - in20;
- ST_SH(vec4, interm_ptr + 32);
- vec4 = in18 - in21;
- ST_SH(vec4, interm_ptr + 88);
- vec4 = in28 - in27;
- ST_SH(vec4, interm_ptr + 56);
- vec4 = in29 - in26;
- ST_SH(vec4, interm_ptr + 64);
-
- ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
-
- in22 = LD_SH(temp + 48);
- in23 = LD_SH(temp + 56);
- in24 = LD_SH(temp + 64);
- in25 = LD_SH(temp + 72);
-
- DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
- DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
-
- in16 = LD_SH(temp);
- in17 = LD_SH(temp + 8);
- in30 = LD_SH(temp + 112);
- in31 = LD_SH(temp + 120);
-
- vec4 = in17 - in22;
- ST_SH(vec4, interm_ptr + 40);
- vec4 = in30 - in25;
- ST_SH(vec4, interm_ptr + 48);
- vec4 = in31 - in24;
- ST_SH(vec4, interm_ptr + 72);
- vec4 = in16 - in23;
- ST_SH(vec4, interm_ptr + 80);
-
- ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
- DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
- DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
-
- ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
- DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
- ADD2(in27, in26, in25, in24, in23, in20);
-
- DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec5, out);
- ST_SH(vec4, out + 120);
-
- SUB2(in27, in26, in25, in24, in22, in21);
-
- DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec5, out + 112);
- ST_SH(vec4, out + 8);
-
- SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
- DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
- SUB2(in26, in27, in24, in25, in23, in20);
-
- DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec4, out + 16);
- ST_SH(vec5, out + 104);
-
- ADD2(in26, in27, in24, in25, in22, in21);
- DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec4, out + 24);
- ST_SH(vec5, out + 96);
-
- in20 = LD_SH(interm_ptr + 32);
- in21 = LD_SH(interm_ptr + 88);
- in27 = LD_SH(interm_ptr + 56);
- in26 = LD_SH(interm_ptr + 64);
-
- in16 = in20;
- in17 = in21;
- DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
- DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
-
- in22 = LD_SH(interm_ptr + 40);
- in25 = LD_SH(interm_ptr + 48);
- in24 = LD_SH(interm_ptr + 72);
- in23 = LD_SH(interm_ptr + 80);
-
- SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
- DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
- ADD2(in28, in29, in31, in30, in16, in19);
- DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec5, out + 32);
- ST_SH(vec4, out + 88);
-
- SUB2(in28, in29, in31, in30, in17, in18);
- DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec5, out + 40);
- ST_SH(vec4, out + 80);
-
- ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
- DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
- SUB2(in29, in28, in30, in31, in16, in19);
-
- DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec5, out + 72);
- ST_SH(vec4, out + 48);
-
- ADD2(in29, in28, in30, in31, in17, in18);
-
- DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
- FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
- ST_SH(vec4, out + 56);
- ST_SH(vec5, out + 64);
-}
-
-static void fdct8x32_1d_row_transpose_store(int16_t *temp, int16_t *output) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
-
- /* 1st set */
- in0 = LD_SH(temp);
- in4 = LD_SH(temp + 32);
- in2 = LD_SH(temp + 64);
- in6 = LD_SH(temp + 96);
- in1 = LD_SH(temp + 128);
- in7 = LD_SH(temp + 152);
- in3 = LD_SH(temp + 192);
- in5 = LD_SH(temp + 216);
-
- TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
- in0, in1, in2, in3, in4, in5, in6, in7);
-
- /* 2nd set */
- in0_1 = LD_SH(temp + 16);
- in1_1 = LD_SH(temp + 232);
- in2_1 = LD_SH(temp + 80);
- in3_1 = LD_SH(temp + 168);
- in4_1 = LD_SH(temp + 48);
- in5_1 = LD_SH(temp + 176);
- in6_1 = LD_SH(temp + 112);
- in7_1 = LD_SH(temp + 240);
-
- ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 32);
- TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
-
- /* 3rd set */
- in0 = LD_SH(temp + 8);
- in1 = LD_SH(temp + 136);
- in2 = LD_SH(temp + 72);
- in3 = LD_SH(temp + 200);
- in4 = LD_SH(temp + 40);
- in5 = LD_SH(temp + 208);
- in6 = LD_SH(temp + 104);
- in7 = LD_SH(temp + 144);
-
- ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- output + 8, 32);
- TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
- in0, in1, in2, in3, in4, in5, in6, in7);
- ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output + 16, 32);
-
- /* 4th set */
- in0_1 = LD_SH(temp + 24);
- in1_1 = LD_SH(temp + 224);
- in2_1 = LD_SH(temp + 88);
- in3_1 = LD_SH(temp + 160);
- in4_1 = LD_SH(temp + 56);
- in5_1 = LD_SH(temp + 184);
- in6_1 = LD_SH(temp + 120);
- in7_1 = LD_SH(temp + 248);
-
- TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
- ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
- output + 24, 32);
-}
-
-static void fdct32x8_1d_row(int16_t *temp, int16_t *temp_buf,
- int16_t *output) {
- fdct8x32_1d_row_load_butterfly(temp, temp_buf);
- fdct8x32_1d_row_even(temp_buf, temp_buf);
- fdct8x32_1d_row_odd(temp_buf + 128, temp, temp_buf + 128);
- fdct8x32_1d_row_transpose_store(temp_buf, output);
-}
-
-static void fdct32x8_1d_row_4x(int16_t *tmp_buf_big, int16_t *tmp_buf,
- int16_t *output) {
- fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf);
- fdct8x32_1d_row_even_4x(tmp_buf, tmp_buf_big, tmp_buf);
- fdct8x32_1d_row_odd(tmp_buf + 128, tmp_buf_big, tmp_buf + 128);
- fdct8x32_1d_row_transpose_store(tmp_buf, output);
-}
-
-void vp9_fdct32x32_msa(const int16_t *input, int16_t *output,
- int32_t src_stride) {
- int32_t i;
- DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]);
- DECLARE_ALIGNED(32, int16_t, tmp_buf[256]);
-
- /* column transform */
- for (i = 0; i < 4; ++i) {
- fdct8x32_1d_column(input + (8 * i), src_stride, tmp_buf,
- tmp_buf_big + (8 * i));
- }
-
- /* row transform */
- fdct32x8_1d_row_4x(tmp_buf_big, tmp_buf, output);
-
- /* row transform */
- for (i = 1; i < 4; ++i) {
- fdct32x8_1d_row(tmp_buf_big + (i * 256), tmp_buf, output + (i * 256));
- }
-}
-
void vp9_fdct32x32_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
out[1] = 0;
out[0] += LD_HADD(input + 32 * 24 + 24, stride);
out[0] >>= 3;
}
-
-static void fdct8x32_1d_row_even_rd(int16_t *temp, int16_t *out) {
- v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
- v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
- v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1;
-
- /* fdct32 even */
- /* stage 2 */
- LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
- LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
-
- BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
- in8, in9, in10, in11, in12, in13, in14, in15,
- vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
- in8, in9, in10, in11, in12, in13, in14, in15);
- FDCT_POSTPROC_2V_NEG_H(vec0, vec1);
- FDCT_POSTPROC_2V_NEG_H(vec2, vec3);
- FDCT_POSTPROC_2V_NEG_H(vec4, vec5);
- FDCT_POSTPROC_2V_NEG_H(vec6, vec7);
- FDCT_POSTPROC_2V_NEG_H(in8, in9);
- FDCT_POSTPROC_2V_NEG_H(in10, in11);
- FDCT_POSTPROC_2V_NEG_H(in12, in13);
- FDCT_POSTPROC_2V_NEG_H(in14, in15);
-
- /* Stage 3 */
- ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
-
- temp0 = in0 + in3;
- in0 = in0 - in3;
- in3 = in1 + in2;
- in1 = in1 - in2;
-
- DOTP_CONST_PAIR(temp0, in3, cospi_16_64, cospi_16_64, temp1, temp0);
- ST_SH(temp0, out);
- ST_SH(temp1, out + 8);
-
- DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
- ST_SH(temp0, out + 16);
- ST_SH(temp1, out + 24);
-
- SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
- DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
- ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
- DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
- ST_SH(temp0, out + 32);
- ST_SH(temp1, out + 56);
-
- SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
- DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
- ST_SH(temp0, out + 40);
- ST_SH(temp1, out + 48);
-
- DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
- DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
- ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
- DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
- ADD2(in0, in1, in2, in3, vec0, vec7);
- DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
- ST_SH(temp0, out + 64);
- ST_SH(temp1, out + 120);
-
- SUB2(in0, in1, in2, in3, in0, in2);
- DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
- ST_SH(temp0, out + 72);
- ST_SH(temp1, out + 112);
-
- SUB2(in9, vec2, in14, vec5, vec2, vec5);
- DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
- SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
- DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
- ST_SH(temp0, out + 80);
- ST_SH(temp1, out + 104);
-
- ADD2(in3, in2, in0, in1, vec3, vec4);
- DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
- ST_SH(temp0, out + 96);
- ST_SH(temp1, out + 88);
-}
-
-static void fdct8x32_1d_row_odd_rd(int16_t *temp, int16_t *interm_ptr,
- int16_t *out) {
- v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
- v8i16 in24, in25, in26, in27, in28, in29, in30, in31;
- v8i16 vec4, vec5;
-
- in20 = LD_SH(temp + 32);
- in21 = LD_SH(temp + 40);
- in26 = LD_SH(temp + 80);
- in27 = LD_SH(temp + 88);
-
- DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
- DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
-
- FDCT_POSTPROC_2V_NEG_H(in20, in21);
- FDCT_POSTPROC_2V_NEG_H(in26, in27);
-
- in18 = LD_SH(temp + 16);
- in19 = LD_SH(temp + 24);
- in28 = LD_SH(temp + 96);
- in29 = LD_SH(temp + 104);
-
- FDCT_POSTPROC_2V_NEG_H(in18, in19);
- FDCT_POSTPROC_2V_NEG_H(in28, in29);
-
- vec4 = in19 - in20;
- ST_SH(vec4, interm_ptr + 32);
- vec4 = in18 - in21;
- ST_SH(vec4, interm_ptr + 88);
- vec4 = in29 - in26;
- ST_SH(vec4, interm_ptr + 64);
- vec4 = in28 - in27;
- ST_SH(vec4, interm_ptr + 56);
-
- ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
-
- in22 = LD_SH(temp + 48);
- in23 = LD_SH(temp + 56);
- in24 = LD_SH(temp + 64);
- in25 = LD_SH(temp + 72);
-
- DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
- DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
- FDCT_POSTPROC_2V_NEG_H(in22, in23);
- FDCT_POSTPROC_2V_NEG_H(in24, in25);
-
- in16 = LD_SH(temp);
- in17 = LD_SH(temp + 8);
- in30 = LD_SH(temp + 112);
- in31 = LD_SH(temp + 120);
-
- FDCT_POSTPROC_2V_NEG_H(in16, in17);
- FDCT_POSTPROC_2V_NEG_H(in30, in31);
-
- vec4 = in17 - in22;
- ST_SH(vec4, interm_ptr + 40);
- vec4 = in30 - in25;
- ST_SH(vec4, interm_ptr + 48);
- vec4 = in31 - in24;
- ST_SH(vec4, interm_ptr + 72);
- vec4 = in16 - in23;
- ST_SH(vec4, interm_ptr + 80);
-
- ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
- DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
- DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
- ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
- DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
- ADD2(in27, in26, in25, in24, in23, in20);
- DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
- ST_SH(vec5, out);
- ST_SH(vec4, out + 120);
-
- SUB2(in27, in26, in25, in24, in22, in21);
- DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
- ST_SH(vec5, out + 112);
- ST_SH(vec4, out + 8);
-
- SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
- DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
- SUB2(in26, in27, in24, in25, in23, in20);
- DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
- ST_SH(vec4, out + 16);
- ST_SH(vec5, out + 104);
-
- ADD2(in26, in27, in24, in25, in22, in21);
- DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
- ST_SH(vec4, out + 24);
- ST_SH(vec5, out + 96);
-
- in20 = LD_SH(interm_ptr + 32);
- in21 = LD_SH(interm_ptr + 88);
- in27 = LD_SH(interm_ptr + 56);
- in26 = LD_SH(interm_ptr + 64);
-
- in16 = in20;
- in17 = in21;
- DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
- DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
-
- in22 = LD_SH(interm_ptr + 40);
- in25 = LD_SH(interm_ptr + 48);
- in24 = LD_SH(interm_ptr + 72);
- in23 = LD_SH(interm_ptr + 80);
-
- SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
- DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
- in16 = in28 + in29;
- in19 = in31 + in30;
- DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
- ST_SH(vec5, out + 32);
- ST_SH(vec4, out + 88);
-
- SUB2(in28, in29, in31, in30, in17, in18);
- DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
- ST_SH(vec5, out + 40);
- ST_SH(vec4, out + 80);
-
- ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
- DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
- SUB2(in29, in28, in30, in31, in16, in19);
- DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
- ST_SH(vec5, out + 72);
- ST_SH(vec4, out + 48);
-
- ADD2(in29, in28, in30, in31, in17, in18);
- DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
- ST_SH(vec4, out + 56);
- ST_SH(vec5, out + 64);
-}
-
-static void fdct32x8_1d_row_rd(int16_t *tmp_buf_big, int16_t *tmp_buf,
- int16_t *output) {
- fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf);
- fdct8x32_1d_row_even_rd(tmp_buf, tmp_buf);
- fdct8x32_1d_row_odd_rd((tmp_buf + 128), tmp_buf_big, (tmp_buf + 128));
- fdct8x32_1d_row_transpose_store(tmp_buf, output);
-}
-
-void vp9_fdct32x32_rd_msa(const int16_t *input, int16_t *out,
- int32_t src_stride) {
- int32_t i;
- DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]);
- DECLARE_ALIGNED(32, int16_t, tmp_buf[256]);
-
- /* column transform */
- for (i = 0; i < 4; ++i) {
- fdct8x32_1d_column(input + (8 * i), src_stride, &tmp_buf[0],
- &tmp_buf_big[0] + (8 * i));
- }
-
- /* row transform */
- for (i = 0; i < 4; ++i) {
- fdct32x8_1d_row_rd(&tmp_buf_big[0] + (8 * i * 32), &tmp_buf[0],
- out + (8 * i * 32));
- }
-}
HADD_SW_S32(vec_w_m); \
})
-#define 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_FADST4(in0, in1, in2, in3, out0, out1, out2, out3) { \
v4i32 s0_m, s1_m, s2_m, s3_m, constant_m; \
v4i32 in0_r_m, in1_r_m, in2_r_m, in3_r_m; \
PCKEV_H4_SH(in0_r_m, in0_r_m, in1_r_m, in1_r_m, s2_m, s2_m, \
s3_m, s3_m, out0, out1, out2, out3); \
}
-
-#define FDCT32_POSTPROC_NEG_W(vec) { \
- v4i32 temp_m; \
- v4i32 one_m = __msa_ldi_w(1); \
- \
- temp_m = __msa_clti_s_w(vec, 0); \
- vec += 1; \
- temp_m = one_m & temp_m; \
- vec += temp_m; \
- vec >>= 2; \
-}
-
-#define FDCT32_POSTPROC_2V_POS_H(vec0, vec1) { \
- v8i16 tp0_m, tp1_m; \
- v8i16 one = __msa_ldi_h(1); \
- \
- tp0_m = __msa_clei_s_h(vec0, 0); \
- tp1_m = __msa_clei_s_h(vec1, 0); \
- tp0_m = (v8i16)__msa_xori_b((v16u8)tp0_m, 255); \
- tp1_m = (v8i16)__msa_xori_b((v16u8)tp1_m, 255); \
- vec0 += 1; \
- vec1 += 1; \
- tp0_m = one & tp0_m; \
- tp1_m = one & tp1_m; \
- vec0 += tp0_m; \
- vec1 += tp1_m; \
- vec0 >>= 2; \
- vec1 >>= 2; \
-}
-
-#define DOTP_CONST_PAIR_W(reg0_left, reg1_left, reg0_right, \
- reg1_right, const0, const1, \
- out0, out1, out2, out3) { \
- v4i32 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m; \
- v2i64 tp0_m, tp1_m, tp2_m, tp3_m; \
- v4i32 k0_m = __msa_fill_w((int32_t) const0); \
- \
- s0_m = __msa_fill_w((int32_t) const1); \
- k0_m = __msa_ilvev_w(s0_m, k0_m); \
- \
- ILVRL_W2_SW(-reg1_left, reg0_left, s1_m, s0_m); \
- ILVRL_W2_SW(reg0_left, reg1_left, s3_m, s2_m); \
- ILVRL_W2_SW(-reg1_right, reg0_right, s5_m, s4_m); \
- ILVRL_W2_SW(reg0_right, reg1_right, s7_m, s6_m); \
- \
- DOTP_SW2_SD(s0_m, s1_m, k0_m, k0_m, tp0_m, tp1_m); \
- DOTP_SW2_SD(s4_m, s5_m, k0_m, k0_m, tp2_m, tp3_m); \
- tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS); \
- tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS); \
- tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS); \
- tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS); \
- out0 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m); \
- out1 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m); \
- \
- DOTP_SW2_SD(s2_m, s3_m, k0_m, k0_m, tp0_m, tp1_m); \
- DOTP_SW2_SD(s6_m, s7_m, k0_m, k0_m, tp2_m, tp3_m); \
- tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS); \
- tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS); \
- tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS); \
- tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS); \
- out2 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m); \
- out3 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m); \
-}
#endif /* VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_ */
}
}
-static INLINE tran_high_t dct_32_round(tran_high_t input) {
- tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
- // TODO(debargha, peter.derivaz): Find new bounds for this assert,
- // and make the bounds consts.
- // assert(-131072 <= rv && rv <= 131071);
- return rv;
-}
-
-static INLINE tran_high_t half_round_shift(tran_high_t input) {
- tran_high_t rv = (input + 1 + (input < 0)) >> 2;
- return rv;
-}
-
-void vp9_fdct32(const tran_high_t *input, tran_high_t *output, int round) {
- tran_high_t step[32];
- // Stage 1
- step[0] = input[0] + input[(32 - 1)];
- step[1] = input[1] + input[(32 - 2)];
- step[2] = input[2] + input[(32 - 3)];
- step[3] = input[3] + input[(32 - 4)];
- step[4] = input[4] + input[(32 - 5)];
- step[5] = input[5] + input[(32 - 6)];
- step[6] = input[6] + input[(32 - 7)];
- step[7] = input[7] + input[(32 - 8)];
- step[8] = input[8] + input[(32 - 9)];
- step[9] = input[9] + input[(32 - 10)];
- step[10] = input[10] + input[(32 - 11)];
- step[11] = input[11] + input[(32 - 12)];
- step[12] = input[12] + input[(32 - 13)];
- step[13] = input[13] + input[(32 - 14)];
- step[14] = input[14] + input[(32 - 15)];
- step[15] = input[15] + input[(32 - 16)];
- step[16] = -input[16] + input[(32 - 17)];
- step[17] = -input[17] + input[(32 - 18)];
- step[18] = -input[18] + input[(32 - 19)];
- step[19] = -input[19] + input[(32 - 20)];
- step[20] = -input[20] + input[(32 - 21)];
- step[21] = -input[21] + input[(32 - 22)];
- step[22] = -input[22] + input[(32 - 23)];
- step[23] = -input[23] + input[(32 - 24)];
- step[24] = -input[24] + input[(32 - 25)];
- step[25] = -input[25] + input[(32 - 26)];
- step[26] = -input[26] + input[(32 - 27)];
- step[27] = -input[27] + input[(32 - 28)];
- step[28] = -input[28] + input[(32 - 29)];
- step[29] = -input[29] + input[(32 - 30)];
- step[30] = -input[30] + input[(32 - 31)];
- step[31] = -input[31] + input[(32 - 32)];
-
- // Stage 2
- output[0] = step[0] + step[16 - 1];
- output[1] = step[1] + step[16 - 2];
- output[2] = step[2] + step[16 - 3];
- output[3] = step[3] + step[16 - 4];
- output[4] = step[4] + step[16 - 5];
- output[5] = step[5] + step[16 - 6];
- output[6] = step[6] + step[16 - 7];
- output[7] = step[7] + step[16 - 8];
- output[8] = -step[8] + step[16 - 9];
- output[9] = -step[9] + step[16 - 10];
- output[10] = -step[10] + step[16 - 11];
- output[11] = -step[11] + step[16 - 12];
- output[12] = -step[12] + step[16 - 13];
- output[13] = -step[13] + step[16 - 14];
- output[14] = -step[14] + step[16 - 15];
- output[15] = -step[15] + step[16 - 16];
-
- output[16] = step[16];
- output[17] = step[17];
- output[18] = step[18];
- output[19] = step[19];
-
- output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
- output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
- output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
- output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
-
- output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
- output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
- output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
- output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
-
- output[28] = step[28];
- output[29] = step[29];
- output[30] = step[30];
- output[31] = step[31];
-
- // dump the magnitude by 4, hence the intermediate values are within
- // the range of 16 bits.
- if (round) {
- output[0] = half_round_shift(output[0]);
- output[1] = half_round_shift(output[1]);
- output[2] = half_round_shift(output[2]);
- output[3] = half_round_shift(output[3]);
- output[4] = half_round_shift(output[4]);
- output[5] = half_round_shift(output[5]);
- output[6] = half_round_shift(output[6]);
- output[7] = half_round_shift(output[7]);
- output[8] = half_round_shift(output[8]);
- output[9] = half_round_shift(output[9]);
- output[10] = half_round_shift(output[10]);
- output[11] = half_round_shift(output[11]);
- output[12] = half_round_shift(output[12]);
- output[13] = half_round_shift(output[13]);
- output[14] = half_round_shift(output[14]);
- output[15] = half_round_shift(output[15]);
-
- output[16] = half_round_shift(output[16]);
- output[17] = half_round_shift(output[17]);
- output[18] = half_round_shift(output[18]);
- output[19] = half_round_shift(output[19]);
- output[20] = half_round_shift(output[20]);
- output[21] = half_round_shift(output[21]);
- output[22] = half_round_shift(output[22]);
- output[23] = half_round_shift(output[23]);
- output[24] = half_round_shift(output[24]);
- output[25] = half_round_shift(output[25]);
- output[26] = half_round_shift(output[26]);
- output[27] = half_round_shift(output[27]);
- output[28] = half_round_shift(output[28]);
- output[29] = half_round_shift(output[29]);
- output[30] = half_round_shift(output[30]);
- output[31] = half_round_shift(output[31]);
- }
-
- // Stage 3
- step[0] = output[0] + output[(8 - 1)];
- step[1] = output[1] + output[(8 - 2)];
- step[2] = output[2] + output[(8 - 3)];
- step[3] = output[3] + output[(8 - 4)];
- step[4] = -output[4] + output[(8 - 5)];
- step[5] = -output[5] + output[(8 - 6)];
- step[6] = -output[6] + output[(8 - 7)];
- step[7] = -output[7] + output[(8 - 8)];
- step[8] = output[8];
- step[9] = output[9];
- step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
- step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
- step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
- step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
- step[14] = output[14];
- step[15] = output[15];
-
- step[16] = output[16] + output[23];
- step[17] = output[17] + output[22];
- step[18] = output[18] + output[21];
- step[19] = output[19] + output[20];
- step[20] = -output[20] + output[19];
- step[21] = -output[21] + output[18];
- step[22] = -output[22] + output[17];
- step[23] = -output[23] + output[16];
- step[24] = -output[24] + output[31];
- step[25] = -output[25] + output[30];
- step[26] = -output[26] + output[29];
- step[27] = -output[27] + output[28];
- step[28] = output[28] + output[27];
- step[29] = output[29] + output[26];
- step[30] = output[30] + output[25];
- step[31] = output[31] + output[24];
-
- // Stage 4
- output[0] = step[0] + step[3];
- output[1] = step[1] + step[2];
- output[2] = -step[2] + step[1];
- output[3] = -step[3] + step[0];
- output[4] = step[4];
- output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
- output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
- output[7] = step[7];
- output[8] = step[8] + step[11];
- output[9] = step[9] + step[10];
- output[10] = -step[10] + step[9];
- output[11] = -step[11] + step[8];
- output[12] = -step[12] + step[15];
- output[13] = -step[13] + step[14];
- output[14] = step[14] + step[13];
- output[15] = step[15] + step[12];
-
- output[16] = step[16];
- output[17] = step[17];
- output[18] = dct_32_round(step[18] * -cospi_8_64 + step[29] * cospi_24_64);
- output[19] = dct_32_round(step[19] * -cospi_8_64 + step[28] * cospi_24_64);
- output[20] = dct_32_round(step[20] * -cospi_24_64 + step[27] * -cospi_8_64);
- output[21] = dct_32_round(step[21] * -cospi_24_64 + step[26] * -cospi_8_64);
- output[22] = step[22];
- output[23] = step[23];
- output[24] = step[24];
- output[25] = step[25];
- output[26] = dct_32_round(step[26] * cospi_24_64 + step[21] * -cospi_8_64);
- output[27] = dct_32_round(step[27] * cospi_24_64 + step[20] * -cospi_8_64);
- output[28] = dct_32_round(step[28] * cospi_8_64 + step[19] * cospi_24_64);
- output[29] = dct_32_round(step[29] * cospi_8_64 + step[18] * cospi_24_64);
- output[30] = step[30];
- output[31] = step[31];
-
- // Stage 5
- step[0] = dct_32_round((output[0] + output[1]) * cospi_16_64);
- step[1] = dct_32_round((-output[1] + output[0]) * cospi_16_64);
- step[2] = dct_32_round(output[2] * cospi_24_64 + output[3] * cospi_8_64);
- step[3] = dct_32_round(output[3] * cospi_24_64 - output[2] * cospi_8_64);
- step[4] = output[4] + output[5];
- step[5] = -output[5] + output[4];
- step[6] = -output[6] + output[7];
- step[7] = output[7] + output[6];
- step[8] = output[8];
- step[9] = dct_32_round(output[9] * -cospi_8_64 + output[14] * cospi_24_64);
- step[10] = dct_32_round(output[10] * -cospi_24_64 + output[13] * -cospi_8_64);
- step[11] = output[11];
- step[12] = output[12];
- step[13] = dct_32_round(output[13] * cospi_24_64 + output[10] * -cospi_8_64);
- step[14] = dct_32_round(output[14] * cospi_8_64 + output[9] * cospi_24_64);
- step[15] = output[15];
-
- step[16] = output[16] + output[19];
- step[17] = output[17] + output[18];
- step[18] = -output[18] + output[17];
- step[19] = -output[19] + output[16];
- step[20] = -output[20] + output[23];
- step[21] = -output[21] + output[22];
- step[22] = output[22] + output[21];
- step[23] = output[23] + output[20];
- step[24] = output[24] + output[27];
- step[25] = output[25] + output[26];
- step[26] = -output[26] + output[25];
- step[27] = -output[27] + output[24];
- step[28] = -output[28] + output[31];
- step[29] = -output[29] + output[30];
- step[30] = output[30] + output[29];
- step[31] = output[31] + output[28];
-
- // Stage 6
- output[0] = step[0];
- output[1] = step[1];
- output[2] = step[2];
- output[3] = step[3];
- output[4] = dct_32_round(step[4] * cospi_28_64 + step[7] * cospi_4_64);
- output[5] = dct_32_round(step[5] * cospi_12_64 + step[6] * cospi_20_64);
- output[6] = dct_32_round(step[6] * cospi_12_64 + step[5] * -cospi_20_64);
- output[7] = dct_32_round(step[7] * cospi_28_64 + step[4] * -cospi_4_64);
- output[8] = step[8] + step[9];
- output[9] = -step[9] + step[8];
- output[10] = -step[10] + step[11];
- output[11] = step[11] + step[10];
- output[12] = step[12] + step[13];
- output[13] = -step[13] + step[12];
- output[14] = -step[14] + step[15];
- output[15] = step[15] + step[14];
-
- output[16] = step[16];
- output[17] = dct_32_round(step[17] * -cospi_4_64 + step[30] * cospi_28_64);
- output[18] = dct_32_round(step[18] * -cospi_28_64 + step[29] * -cospi_4_64);
- output[19] = step[19];
- output[20] = step[20];
- output[21] = dct_32_round(step[21] * -cospi_20_64 + step[26] * cospi_12_64);
- output[22] = dct_32_round(step[22] * -cospi_12_64 + step[25] * -cospi_20_64);
- output[23] = step[23];
- output[24] = step[24];
- output[25] = dct_32_round(step[25] * cospi_12_64 + step[22] * -cospi_20_64);
- output[26] = dct_32_round(step[26] * cospi_20_64 + step[21] * cospi_12_64);
- output[27] = step[27];
- output[28] = step[28];
- output[29] = dct_32_round(step[29] * cospi_28_64 + step[18] * -cospi_4_64);
- output[30] = dct_32_round(step[30] * cospi_4_64 + step[17] * cospi_28_64);
- output[31] = step[31];
-
- // Stage 7
- step[0] = output[0];
- step[1] = output[1];
- step[2] = output[2];
- step[3] = output[3];
- step[4] = output[4];
- step[5] = output[5];
- step[6] = output[6];
- step[7] = output[7];
- step[8] = dct_32_round(output[8] * cospi_30_64 + output[15] * cospi_2_64);
- step[9] = dct_32_round(output[9] * cospi_14_64 + output[14] * cospi_18_64);
- step[10] = dct_32_round(output[10] * cospi_22_64 + output[13] * cospi_10_64);
- step[11] = dct_32_round(output[11] * cospi_6_64 + output[12] * cospi_26_64);
- step[12] = dct_32_round(output[12] * cospi_6_64 + output[11] * -cospi_26_64);
- step[13] = dct_32_round(output[13] * cospi_22_64 + output[10] * -cospi_10_64);
- step[14] = dct_32_round(output[14] * cospi_14_64 + output[9] * -cospi_18_64);
- step[15] = dct_32_round(output[15] * cospi_30_64 + output[8] * -cospi_2_64);
-
- step[16] = output[16] + output[17];
- step[17] = -output[17] + output[16];
- step[18] = -output[18] + output[19];
- step[19] = output[19] + output[18];
- step[20] = output[20] + output[21];
- step[21] = -output[21] + output[20];
- step[22] = -output[22] + output[23];
- step[23] = output[23] + output[22];
- step[24] = output[24] + output[25];
- step[25] = -output[25] + output[24];
- step[26] = -output[26] + output[27];
- step[27] = output[27] + output[26];
- step[28] = output[28] + output[29];
- step[29] = -output[29] + output[28];
- step[30] = -output[30] + output[31];
- step[31] = output[31] + output[30];
-
- // Final stage --- outputs indices are bit-reversed.
- output[0] = step[0];
- output[16] = step[1];
- output[8] = step[2];
- output[24] = step[3];
- output[4] = step[4];
- output[20] = step[5];
- output[12] = step[6];
- output[28] = step[7];
- output[2] = step[8];
- output[18] = step[9];
- output[10] = step[10];
- output[26] = step[11];
- output[6] = step[12];
- output[22] = step[13];
- output[14] = step[14];
- output[30] = step[15];
-
- output[1] = dct_32_round(step[16] * cospi_31_64 + step[31] * cospi_1_64);
- output[17] = dct_32_round(step[17] * cospi_15_64 + step[30] * cospi_17_64);
- output[9] = dct_32_round(step[18] * cospi_23_64 + step[29] * cospi_9_64);
- output[25] = dct_32_round(step[19] * cospi_7_64 + step[28] * cospi_25_64);
- output[5] = dct_32_round(step[20] * cospi_27_64 + step[27] * cospi_5_64);
- output[21] = dct_32_round(step[21] * cospi_11_64 + step[26] * cospi_21_64);
- output[13] = dct_32_round(step[22] * cospi_19_64 + step[25] * cospi_13_64);
- output[29] = dct_32_round(step[23] * cospi_3_64 + step[24] * cospi_29_64);
- output[3] = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
- output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
- output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
- output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
- output[7] = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
- output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
- output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
- output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
-}
-
-void vp9_fdct32x32_1_c(const int16_t *input, tran_low_t *output, int stride) {
- int r, c;
- tran_low_t sum = 0;
- for (r = 0; r < 32; ++r)
- for (c = 0; c < 32; ++c)
- sum += input[r * stride + c];
-
- output[0] = sum >> 3;
- output[1] = 0;
-}
-
-void vp9_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
- int i, j;
- tran_high_t output[32 * 32];
-
- // Columns
- for (i = 0; i < 32; ++i) {
- tran_high_t temp_in[32], temp_out[32];
- for (j = 0; j < 32; ++j)
- temp_in[j] = input[j * stride + i] * 4;
- vp9_fdct32(temp_in, temp_out, 0);
- for (j = 0; j < 32; ++j)
- output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
- }
-
- // Rows
- for (i = 0; i < 32; ++i) {
- tran_high_t temp_in[32], temp_out[32];
- for (j = 0; j < 32; ++j)
- temp_in[j] = output[j + i * 32];
- vp9_fdct32(temp_in, temp_out, 0);
- for (j = 0; j < 32; ++j)
- out[j + i * 32] =
- (tran_low_t)((temp_out[j] + 1 + (temp_out[j] < 0)) >> 2);
- }
-}
-
-// Note that although we use dct_32_round in dct32 computation flow,
-// this 2d fdct32x32 for rate-distortion optimization loop is operating
-// within 16 bits precision.
-void vp9_fdct32x32_rd_c(const int16_t *input, tran_low_t *out, int stride) {
- int i, j;
- tran_high_t output[32 * 32];
-
- // Columns
- for (i = 0; i < 32; ++i) {
- tran_high_t temp_in[32], temp_out[32];
- for (j = 0; j < 32; ++j)
- temp_in[j] = input[j * stride + i] * 4;
- vp9_fdct32(temp_in, temp_out, 0);
- for (j = 0; j < 32; ++j)
- // TODO(cd): see quality impact of only doing
- // output[j * 32 + i] = (temp_out[j] + 1) >> 2;
- // PS: also change code in vp9/encoder/x86/vp9_dct_sse2.c
- output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
- }
-
- // Rows
- for (i = 0; i < 32; ++i) {
- tran_high_t temp_in[32], temp_out[32];
- for (j = 0; j < 32; ++j)
- temp_in[j] = output[j + i * 32];
- vp9_fdct32(temp_in, temp_out, 1);
- for (j = 0; j < 32; ++j)
- out[j + i * 32] = (tran_low_t)temp_out[j];
- }
-}
-
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_fht4x4_c(const int16_t *input, tran_low_t *output,
int stride, int tx_type) {
int stride) {
vp9_fdct32x32_1_c(input, out, stride);
}
-
-void vp9_highbd_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
- vp9_fdct32x32_c(input, out, stride);
-}
-
-void vp9_highbd_fdct32x32_rd_c(const int16_t *input, tran_low_t *out,
- int stride) {
- vp9_fdct32x32_rd_c(input, out, stride);
-}
#endif // CONFIG_VP9_HIGHBITDEPTH
in1 = _mm_srai_epi32(in1, 3);
store_output(&in1, output);
}
-
-/*
- * The DCTnxn functions are defined using the macros below. The main code for
- * them is in separate files (vp9/encoder/x86/vp9_dct_sse2_impl.h &
- * vp9/encoder/x86/vp9_dct32x32_sse2_impl.h) which are used by both the 8 bit code
- * and the high bit depth code.
- */
-
-#define DCT_HIGH_BIT_DEPTH 0
-
-#define FDCT32x32_2D vp9_fdct32x32_rd_sse2
-#define FDCT32x32_HIGH_PRECISION 0
-#include "vp9/encoder/x86/vp9_dct32x32_sse2_impl.h"
-#undef FDCT32x32_2D
-#undef FDCT32x32_HIGH_PRECISION
-
-#define FDCT32x32_2D vp9_fdct32x32_sse2
-#define FDCT32x32_HIGH_PRECISION 1
-#include "vp9/encoder/x86/vp9_dct32x32_sse2_impl.h" // NOLINT
-#undef FDCT32x32_2D
-#undef FDCT32x32_HIGH_PRECISION
-
-#undef DCT_HIGH_BIT_DEPTH
-
-
-#if CONFIG_VP9_HIGHBITDEPTH
-
-#define DCT_HIGH_BIT_DEPTH 1
-
-#define FDCT32x32_2D vp9_highbd_fdct32x32_rd_sse2
-#define FDCT32x32_HIGH_PRECISION 0
-#include "vp9/encoder/x86/vp9_dct32x32_sse2_impl.h" // NOLINT
-#undef FDCT32x32_2D
-#undef FDCT32x32_HIGH_PRECISION
-
-#define FDCT32x32_2D vp9_highbd_fdct32x32_sse2
-#define FDCT32x32_HIGH_PRECISION 1
-#include "vp9/encoder/x86/vp9_dct32x32_sse2_impl.h" // NOLINT
-#undef FDCT32x32_2D
-#undef FDCT32x32_HIGH_PRECISION
-
-#undef DCT_HIGH_BIT_DEPTH
-
-#endif // CONFIG_VP9_HIGHBITDEPTH
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct_sse2.c
VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_dct_ssse3.c
-VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_dct32x32_sse2_impl.h
ifeq ($(CONFIG_VP9_TEMPORAL_DENOISING),yes)
VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_denoiser_sse2.c
endif
-VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_dct32x32_avx2_impl.h
-VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_dct_avx2.c
VP9_CX_SRCS-$(HAVE_AVX2) += encoder/x86/vp9_error_intrin_avx2.c
ifneq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
}
}
+static INLINE tran_high_t dct_32_round(tran_high_t input) {
+ tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
+ // TODO(debargha, peter.derivaz): Find new bounds for this assert,
+ // and make the bounds consts.
+ // assert(-131072 <= rv && rv <= 131071);
+ return rv;
+}
+
+static INLINE tran_high_t half_round_shift(tran_high_t input) {
+ tran_high_t rv = (input + 1 + (input < 0)) >> 2;
+ return rv;
+}
+
+void vp9_fdct32(const tran_high_t *input, tran_high_t *output, int round) {
+ tran_high_t step[32];
+ // Stage 1
+ step[0] = input[0] + input[(32 - 1)];
+ step[1] = input[1] + input[(32 - 2)];
+ step[2] = input[2] + input[(32 - 3)];
+ step[3] = input[3] + input[(32 - 4)];
+ step[4] = input[4] + input[(32 - 5)];
+ step[5] = input[5] + input[(32 - 6)];
+ step[6] = input[6] + input[(32 - 7)];
+ step[7] = input[7] + input[(32 - 8)];
+ step[8] = input[8] + input[(32 - 9)];
+ step[9] = input[9] + input[(32 - 10)];
+ step[10] = input[10] + input[(32 - 11)];
+ step[11] = input[11] + input[(32 - 12)];
+ step[12] = input[12] + input[(32 - 13)];
+ step[13] = input[13] + input[(32 - 14)];
+ step[14] = input[14] + input[(32 - 15)];
+ step[15] = input[15] + input[(32 - 16)];
+ step[16] = -input[16] + input[(32 - 17)];
+ step[17] = -input[17] + input[(32 - 18)];
+ step[18] = -input[18] + input[(32 - 19)];
+ step[19] = -input[19] + input[(32 - 20)];
+ step[20] = -input[20] + input[(32 - 21)];
+ step[21] = -input[21] + input[(32 - 22)];
+ step[22] = -input[22] + input[(32 - 23)];
+ step[23] = -input[23] + input[(32 - 24)];
+ step[24] = -input[24] + input[(32 - 25)];
+ step[25] = -input[25] + input[(32 - 26)];
+ step[26] = -input[26] + input[(32 - 27)];
+ step[27] = -input[27] + input[(32 - 28)];
+ step[28] = -input[28] + input[(32 - 29)];
+ step[29] = -input[29] + input[(32 - 30)];
+ step[30] = -input[30] + input[(32 - 31)];
+ step[31] = -input[31] + input[(32 - 32)];
+
+ // Stage 2
+ output[0] = step[0] + step[16 - 1];
+ output[1] = step[1] + step[16 - 2];
+ output[2] = step[2] + step[16 - 3];
+ output[3] = step[3] + step[16 - 4];
+ output[4] = step[4] + step[16 - 5];
+ output[5] = step[5] + step[16 - 6];
+ output[6] = step[6] + step[16 - 7];
+ output[7] = step[7] + step[16 - 8];
+ output[8] = -step[8] + step[16 - 9];
+ output[9] = -step[9] + step[16 - 10];
+ output[10] = -step[10] + step[16 - 11];
+ output[11] = -step[11] + step[16 - 12];
+ output[12] = -step[12] + step[16 - 13];
+ output[13] = -step[13] + step[16 - 14];
+ output[14] = -step[14] + step[16 - 15];
+ output[15] = -step[15] + step[16 - 16];
+
+ output[16] = step[16];
+ output[17] = step[17];
+ output[18] = step[18];
+ output[19] = step[19];
+
+ output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
+ output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
+ output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
+ output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
+
+ output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
+ output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
+ output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
+ output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
+
+ output[28] = step[28];
+ output[29] = step[29];
+ output[30] = step[30];
+ output[31] = step[31];
+
+ // dump the magnitude by 4, hence the intermediate values are within
+ // the range of 16 bits.
+ if (round) {
+ output[0] = half_round_shift(output[0]);
+ output[1] = half_round_shift(output[1]);
+ output[2] = half_round_shift(output[2]);
+ output[3] = half_round_shift(output[3]);
+ output[4] = half_round_shift(output[4]);
+ output[5] = half_round_shift(output[5]);
+ output[6] = half_round_shift(output[6]);
+ output[7] = half_round_shift(output[7]);
+ output[8] = half_round_shift(output[8]);
+ output[9] = half_round_shift(output[9]);
+ output[10] = half_round_shift(output[10]);
+ output[11] = half_round_shift(output[11]);
+ output[12] = half_round_shift(output[12]);
+ output[13] = half_round_shift(output[13]);
+ output[14] = half_round_shift(output[14]);
+ output[15] = half_round_shift(output[15]);
+
+ output[16] = half_round_shift(output[16]);
+ output[17] = half_round_shift(output[17]);
+ output[18] = half_round_shift(output[18]);
+ output[19] = half_round_shift(output[19]);
+ output[20] = half_round_shift(output[20]);
+ output[21] = half_round_shift(output[21]);
+ output[22] = half_round_shift(output[22]);
+ output[23] = half_round_shift(output[23]);
+ output[24] = half_round_shift(output[24]);
+ output[25] = half_round_shift(output[25]);
+ output[26] = half_round_shift(output[26]);
+ output[27] = half_round_shift(output[27]);
+ output[28] = half_round_shift(output[28]);
+ output[29] = half_round_shift(output[29]);
+ output[30] = half_round_shift(output[30]);
+ output[31] = half_round_shift(output[31]);
+ }
+
+ // Stage 3
+ step[0] = output[0] + output[(8 - 1)];
+ step[1] = output[1] + output[(8 - 2)];
+ step[2] = output[2] + output[(8 - 3)];
+ step[3] = output[3] + output[(8 - 4)];
+ step[4] = -output[4] + output[(8 - 5)];
+ step[5] = -output[5] + output[(8 - 6)];
+ step[6] = -output[6] + output[(8 - 7)];
+ step[7] = -output[7] + output[(8 - 8)];
+ step[8] = output[8];
+ step[9] = output[9];
+ step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
+ step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
+ step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
+ step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
+ step[14] = output[14];
+ step[15] = output[15];
+
+ step[16] = output[16] + output[23];
+ step[17] = output[17] + output[22];
+ step[18] = output[18] + output[21];
+ step[19] = output[19] + output[20];
+ step[20] = -output[20] + output[19];
+ step[21] = -output[21] + output[18];
+ step[22] = -output[22] + output[17];
+ step[23] = -output[23] + output[16];
+ step[24] = -output[24] + output[31];
+ step[25] = -output[25] + output[30];
+ step[26] = -output[26] + output[29];
+ step[27] = -output[27] + output[28];
+ step[28] = output[28] + output[27];
+ step[29] = output[29] + output[26];
+ step[30] = output[30] + output[25];
+ step[31] = output[31] + output[24];
+
+ // Stage 4
+ output[0] = step[0] + step[3];
+ output[1] = step[1] + step[2];
+ output[2] = -step[2] + step[1];
+ output[3] = -step[3] + step[0];
+ output[4] = step[4];
+ output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
+ output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
+ output[7] = step[7];
+ output[8] = step[8] + step[11];
+ output[9] = step[9] + step[10];
+ output[10] = -step[10] + step[9];
+ output[11] = -step[11] + step[8];
+ output[12] = -step[12] + step[15];
+ output[13] = -step[13] + step[14];
+ output[14] = step[14] + step[13];
+ output[15] = step[15] + step[12];
+
+ output[16] = step[16];
+ output[17] = step[17];
+ output[18] = dct_32_round(step[18] * -cospi_8_64 + step[29] * cospi_24_64);
+ output[19] = dct_32_round(step[19] * -cospi_8_64 + step[28] * cospi_24_64);
+ output[20] = dct_32_round(step[20] * -cospi_24_64 + step[27] * -cospi_8_64);
+ output[21] = dct_32_round(step[21] * -cospi_24_64 + step[26] * -cospi_8_64);
+ output[22] = step[22];
+ output[23] = step[23];
+ output[24] = step[24];
+ output[25] = step[25];
+ output[26] = dct_32_round(step[26] * cospi_24_64 + step[21] * -cospi_8_64);
+ output[27] = dct_32_round(step[27] * cospi_24_64 + step[20] * -cospi_8_64);
+ output[28] = dct_32_round(step[28] * cospi_8_64 + step[19] * cospi_24_64);
+ output[29] = dct_32_round(step[29] * cospi_8_64 + step[18] * cospi_24_64);
+ output[30] = step[30];
+ output[31] = step[31];
+
+ // Stage 5
+ step[0] = dct_32_round((output[0] + output[1]) * cospi_16_64);
+ step[1] = dct_32_round((-output[1] + output[0]) * cospi_16_64);
+ step[2] = dct_32_round(output[2] * cospi_24_64 + output[3] * cospi_8_64);
+ step[3] = dct_32_round(output[3] * cospi_24_64 - output[2] * cospi_8_64);
+ step[4] = output[4] + output[5];
+ step[5] = -output[5] + output[4];
+ step[6] = -output[6] + output[7];
+ step[7] = output[7] + output[6];
+ step[8] = output[8];
+ step[9] = dct_32_round(output[9] * -cospi_8_64 + output[14] * cospi_24_64);
+ step[10] = dct_32_round(output[10] * -cospi_24_64 + output[13] * -cospi_8_64);
+ step[11] = output[11];
+ step[12] = output[12];
+ step[13] = dct_32_round(output[13] * cospi_24_64 + output[10] * -cospi_8_64);
+ step[14] = dct_32_round(output[14] * cospi_8_64 + output[9] * cospi_24_64);
+ step[15] = output[15];
+
+ step[16] = output[16] + output[19];
+ step[17] = output[17] + output[18];
+ step[18] = -output[18] + output[17];
+ step[19] = -output[19] + output[16];
+ step[20] = -output[20] + output[23];
+ step[21] = -output[21] + output[22];
+ step[22] = output[22] + output[21];
+ step[23] = output[23] + output[20];
+ step[24] = output[24] + output[27];
+ step[25] = output[25] + output[26];
+ step[26] = -output[26] + output[25];
+ step[27] = -output[27] + output[24];
+ step[28] = -output[28] + output[31];
+ step[29] = -output[29] + output[30];
+ step[30] = output[30] + output[29];
+ step[31] = output[31] + output[28];
+
+ // Stage 6
+ output[0] = step[0];
+ output[1] = step[1];
+ output[2] = step[2];
+ output[3] = step[3];
+ output[4] = dct_32_round(step[4] * cospi_28_64 + step[7] * cospi_4_64);
+ output[5] = dct_32_round(step[5] * cospi_12_64 + step[6] * cospi_20_64);
+ output[6] = dct_32_round(step[6] * cospi_12_64 + step[5] * -cospi_20_64);
+ output[7] = dct_32_round(step[7] * cospi_28_64 + step[4] * -cospi_4_64);
+ output[8] = step[8] + step[9];
+ output[9] = -step[9] + step[8];
+ output[10] = -step[10] + step[11];
+ output[11] = step[11] + step[10];
+ output[12] = step[12] + step[13];
+ output[13] = -step[13] + step[12];
+ output[14] = -step[14] + step[15];
+ output[15] = step[15] + step[14];
+
+ output[16] = step[16];
+ output[17] = dct_32_round(step[17] * -cospi_4_64 + step[30] * cospi_28_64);
+ output[18] = dct_32_round(step[18] * -cospi_28_64 + step[29] * -cospi_4_64);
+ output[19] = step[19];
+ output[20] = step[20];
+ output[21] = dct_32_round(step[21] * -cospi_20_64 + step[26] * cospi_12_64);
+ output[22] = dct_32_round(step[22] * -cospi_12_64 + step[25] * -cospi_20_64);
+ output[23] = step[23];
+ output[24] = step[24];
+ output[25] = dct_32_round(step[25] * cospi_12_64 + step[22] * -cospi_20_64);
+ output[26] = dct_32_round(step[26] * cospi_20_64 + step[21] * cospi_12_64);
+ output[27] = step[27];
+ output[28] = step[28];
+ output[29] = dct_32_round(step[29] * cospi_28_64 + step[18] * -cospi_4_64);
+ output[30] = dct_32_round(step[30] * cospi_4_64 + step[17] * cospi_28_64);
+ output[31] = step[31];
+
+ // Stage 7
+ step[0] = output[0];
+ step[1] = output[1];
+ step[2] = output[2];
+ step[3] = output[3];
+ step[4] = output[4];
+ step[5] = output[5];
+ step[6] = output[6];
+ step[7] = output[7];
+ step[8] = dct_32_round(output[8] * cospi_30_64 + output[15] * cospi_2_64);
+ step[9] = dct_32_round(output[9] * cospi_14_64 + output[14] * cospi_18_64);
+ step[10] = dct_32_round(output[10] * cospi_22_64 + output[13] * cospi_10_64);
+ step[11] = dct_32_round(output[11] * cospi_6_64 + output[12] * cospi_26_64);
+ step[12] = dct_32_round(output[12] * cospi_6_64 + output[11] * -cospi_26_64);
+ step[13] = dct_32_round(output[13] * cospi_22_64 + output[10] * -cospi_10_64);
+ step[14] = dct_32_round(output[14] * cospi_14_64 + output[9] * -cospi_18_64);
+ step[15] = dct_32_round(output[15] * cospi_30_64 + output[8] * -cospi_2_64);
+
+ step[16] = output[16] + output[17];
+ step[17] = -output[17] + output[16];
+ step[18] = -output[18] + output[19];
+ step[19] = output[19] + output[18];
+ step[20] = output[20] + output[21];
+ step[21] = -output[21] + output[20];
+ step[22] = -output[22] + output[23];
+ step[23] = output[23] + output[22];
+ step[24] = output[24] + output[25];
+ step[25] = -output[25] + output[24];
+ step[26] = -output[26] + output[27];
+ step[27] = output[27] + output[26];
+ step[28] = output[28] + output[29];
+ step[29] = -output[29] + output[28];
+ step[30] = -output[30] + output[31];
+ step[31] = output[31] + output[30];
+
+ // Final stage --- outputs indices are bit-reversed.
+ output[0] = step[0];
+ output[16] = step[1];
+ output[8] = step[2];
+ output[24] = step[3];
+ output[4] = step[4];
+ output[20] = step[5];
+ output[12] = step[6];
+ output[28] = step[7];
+ output[2] = step[8];
+ output[18] = step[9];
+ output[10] = step[10];
+ output[26] = step[11];
+ output[6] = step[12];
+ output[22] = step[13];
+ output[14] = step[14];
+ output[30] = step[15];
+
+ output[1] = dct_32_round(step[16] * cospi_31_64 + step[31] * cospi_1_64);
+ output[17] = dct_32_round(step[17] * cospi_15_64 + step[30] * cospi_17_64);
+ output[9] = dct_32_round(step[18] * cospi_23_64 + step[29] * cospi_9_64);
+ output[25] = dct_32_round(step[19] * cospi_7_64 + step[28] * cospi_25_64);
+ output[5] = dct_32_round(step[20] * cospi_27_64 + step[27] * cospi_5_64);
+ output[21] = dct_32_round(step[21] * cospi_11_64 + step[26] * cospi_21_64);
+ output[13] = dct_32_round(step[22] * cospi_19_64 + step[25] * cospi_13_64);
+ output[29] = dct_32_round(step[23] * cospi_3_64 + step[24] * cospi_29_64);
+ output[3] = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
+ output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
+ output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
+ output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
+ output[7] = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
+ output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
+ output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
+ output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
+}
+
+void vp9_fdct32x32_1_c(const int16_t *input, tran_low_t *output, int stride) {
+ int r, c;
+ tran_low_t sum = 0;
+ for (r = 0; r < 32; ++r)
+ for (c = 0; c < 32; ++c)
+ sum += input[r * stride + c];
+
+ output[0] = sum >> 3;
+ output[1] = 0;
+}
+
+void vp9_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
+ int i, j;
+ tran_high_t output[32 * 32];
+
+ // Columns
+ for (i = 0; i < 32; ++i) {
+ tran_high_t temp_in[32], temp_out[32];
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = input[j * stride + i] * 4;
+ vp9_fdct32(temp_in, temp_out, 0);
+ for (j = 0; j < 32; ++j)
+ output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
+ }
+
+ // Rows
+ for (i = 0; i < 32; ++i) {
+ tran_high_t temp_in[32], temp_out[32];
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = output[j + i * 32];
+ vp9_fdct32(temp_in, temp_out, 0);
+ for (j = 0; j < 32; ++j)
+ out[j + i * 32] =
+ (tran_low_t)((temp_out[j] + 1 + (temp_out[j] < 0)) >> 2);
+ }
+}
+
+// Note that although we use dct_32_round in dct32 computation flow,
+// this 2d fdct32x32 for rate-distortion optimization loop is operating
+// within 16 bits precision.
+void vp9_fdct32x32_rd_c(const int16_t *input, tran_low_t *out, int stride) {
+ int i, j;
+ tran_high_t output[32 * 32];
+
+ // Columns
+ for (i = 0; i < 32; ++i) {
+ tran_high_t temp_in[32], temp_out[32];
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = input[j * stride + i] * 4;
+ vp9_fdct32(temp_in, temp_out, 0);
+ for (j = 0; j < 32; ++j)
+ // TODO(cd): see quality impact of only doing
+ // output[j * 32 + i] = (temp_out[j] + 1) >> 2;
+ // PS: also change code in vp9/encoder/x86/vp9_dct_sse2.c
+ output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
+ }
+
+ // Rows
+ for (i = 0; i < 32; ++i) {
+ tran_high_t temp_in[32], temp_out[32];
+ for (j = 0; j < 32; ++j)
+ temp_in[j] = output[j + i * 32];
+ vp9_fdct32(temp_in, temp_out, 1);
+ for (j = 0; j < 32; ++j)
+ out[j + i * 32] = (tran_low_t)temp_out[j];
+ }
+}
+
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_highbd_fdct4x4_c(const int16_t *input, tran_low_t *output,
int stride) {
int stride) {
vp9_fdct16x16_c(input, output, stride);
}
+
+void vp9_highbd_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
+ vp9_fdct32x32_c(input, out, stride);
+}
+
+void vp9_highbd_fdct32x32_rd_c(const int16_t *input, tran_low_t *out,
+ int stride) {
+ vp9_fdct32x32_rd_c(input, out, stride);
+}
#endif // CONFIG_VP9_HIGHBITDEPTH
* be found in the AUTHORS file in the root of the source tree.
*/
+#ifndef VPX_DSP_FWD_TXFM_H_
+#define VPX_DSP_FWD_TXFM_H_
+
#include "vpx_dsp/txfm_common.h"
static INLINE tran_high_t fdct_round_shift(tran_high_t input) {
// assert(INT16_MIN <= rv && rv <= INT16_MAX);
return rv;
}
+
+void vp9_fdct32(const tran_high_t *input, tran_high_t *output, int round);
+#endif // VPX_DSP_FWD_TXFM_H_
--- /dev/null
+/*
+ * 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 "vpx_dsp/mips/fwd_txfm_msa.h"
+
+static void fdct8x32_1d_column_load_butterfly(const int16_t *input,
+ int32_t src_stride,
+ int16_t *temp_buff) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 step0, step1, step2, step3;
+ v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
+ v8i16 step0_1, step1_1, step2_1, step3_1;
+
+ /* 1st and 2nd set */
+ LD_SH4(input, src_stride, in0, in1, in2, in3);
+ LD_SH4(input + (28 * src_stride), src_stride, in4, in5, in6, in7);
+ LD_SH4(input + (4 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
+ LD_SH4(input + (24 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
+ SLLI_4V(in0, in1, in2, in3, 2);
+ SLLI_4V(in4, in5, in6, in7, 2);
+ SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
+ SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
+ BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
+ step0, step1, step2, step3, in4, in5, in6, in7);
+ BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
+ ST_SH4(step0, step1, step2, step3, temp_buff, 8);
+ ST_SH4(in4, in5, in6, in7, temp_buff + (28 * 8), 8);
+ ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (4 * 8), 8);
+ ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (24 * 8), 8);
+
+ /* 3rd and 4th set */
+ LD_SH4(input + (8 * src_stride), src_stride, in0, in1, in2, in3);
+ LD_SH4(input + (20 * src_stride), src_stride, in4, in5, in6, in7);
+ LD_SH4(input + (12 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1);
+ LD_SH4(input + (16 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1);
+ SLLI_4V(in0, in1, in2, in3, 2);
+ SLLI_4V(in4, in5, in6, in7, 2);
+ SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2);
+ SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2);
+ BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7,
+ step0, step1, step2, step3, in4, in5, in6, in7);
+ BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1);
+ ST_SH4(step0, step1, step2, step3, temp_buff + (8 * 8), 8);
+ ST_SH4(in4, in5, in6, in7, temp_buff + (20 * 8), 8);
+ ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (12 * 8), 8);
+ ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (15 * 8) + 8, 8);
+}
+
+static void fdct8x32_1d_column_even_store(int16_t *input, int16_t *temp) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v8i16 temp0, temp1;
+
+ /* fdct even */
+ LD_SH4(input, 8, in0, in1, in2, in3);
+ LD_SH4(input + 96, 8, in12, in13, in14, in15);
+ BUTTERFLY_8(in0, in1, in2, in3, in12, in13, in14, in15,
+ vec0, vec1, vec2, vec3, in12, in13, in14, in15);
+ LD_SH4(input + 32, 8, in4, in5, in6, in7);
+ LD_SH4(input + 64, 8, in8, in9, in10, in11);
+ BUTTERFLY_8(in4, in5, in6, in7, in8, in9, in10, in11,
+ vec4, vec5, vec6, vec7, in8, in9, in10, in11);
+
+ /* Stage 3 */
+ ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
+ BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
+ DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp);
+ ST_SH(temp1, temp + 512);
+
+ DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 256);
+ ST_SH(temp1, temp + 768);
+
+ SUB4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, vec7, vec6, vec5, vec4);
+ DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
+ ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
+ DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 128);
+ ST_SH(temp1, temp + 896);
+
+ SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
+ DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 640);
+ ST_SH(temp1, temp + 384);
+
+ DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
+ DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
+ ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
+ DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
+ ADD2(in0, in1, in2, in3, vec0, vec7);
+ DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 64);
+ ST_SH(temp1, temp + 960);
+
+ SUB2(in0, in1, in2, in3, in0, in2);
+ DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 576);
+ ST_SH(temp1, temp + 448);
+
+ SUB2(in9, vec2, in14, vec5, vec2, vec5);
+ DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
+ SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
+ DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 320);
+ ST_SH(temp1, temp + 704);
+
+ ADD2(in3, in2, in0, in1, vec3, vec4);
+ DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
+ FDCT32_POSTPROC_2V_POS_H(temp0, temp1);
+ ST_SH(temp0, temp + 192);
+ ST_SH(temp1, temp + 832);
+}
+
+static void fdct8x32_1d_column_odd_store(int16_t *input, int16_t *temp_ptr) {
+ v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
+ v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
+
+ in20 = LD_SH(input + 32);
+ in21 = LD_SH(input + 40);
+ in26 = LD_SH(input + 80);
+ in27 = LD_SH(input + 88);
+
+ DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
+ DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
+
+ in18 = LD_SH(input + 16);
+ in19 = LD_SH(input + 24);
+ in28 = LD_SH(input + 96);
+ in29 = LD_SH(input + 104);
+
+ vec4 = in19 - in20;
+ ST_SH(vec4, input + 32);
+ vec4 = in18 - in21;
+ ST_SH(vec4, input + 40);
+ vec4 = in29 - in26;
+ ST_SH(vec4, input + 80);
+ vec4 = in28 - in27;
+ ST_SH(vec4, input + 88);
+
+ in21 = in18 + in21;
+ in20 = in19 + in20;
+ in27 = in28 + in27;
+ in26 = in29 + in26;
+
+ LD_SH4(input + 48, 8, in22, in23, in24, in25);
+ DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
+ DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
+
+ in16 = LD_SH(input);
+ in17 = LD_SH(input + 8);
+ in30 = LD_SH(input + 112);
+ in31 = LD_SH(input + 120);
+
+ vec4 = in17 - in22;
+ ST_SH(vec4, input + 16);
+ vec4 = in16 - in23;
+ ST_SH(vec4, input + 24);
+ vec4 = in31 - in24;
+ ST_SH(vec4, input + 96);
+ vec4 = in30 - in25;
+ ST_SH(vec4, input + 104);
+
+ ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
+ DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
+ DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
+ ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
+ DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
+ ADD2(in27, in26, in25, in24, in23, in20);
+ DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr);
+ ST_SH(vec4, temp_ptr + 960);
+
+ SUB2(in27, in26, in25, in24, in22, in21);
+ DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr + 448);
+ ST_SH(vec4, temp_ptr + 512);
+
+ SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
+ DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
+ SUB2(in26, in27, in24, in25, in23, in20);
+ DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec4, temp_ptr + 704);
+ ST_SH(vec5, temp_ptr + 256);
+
+ ADD2(in26, in27, in24, in25, in22, in21);
+ DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec4, temp_ptr + 192);
+ ST_SH(vec5, temp_ptr + 768);
+
+ LD_SH4(input + 16, 8, in22, in23, in20, in21);
+ LD_SH4(input + 80, 8, in26, in27, in24, in25);
+ in16 = in20;
+ in17 = in21;
+ DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
+ DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
+ SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
+ DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
+ ADD2(in28, in29, in31, in30, in16, in19);
+ DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr + 832);
+ ST_SH(vec4, temp_ptr + 128);
+
+ SUB2(in28, in29, in31, in30, in17, in18);
+ DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr + 320);
+ ST_SH(vec4, temp_ptr + 640);
+ ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
+ DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
+ SUB2(in29, in28, in30, in31, in16, in19);
+ DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr + 576);
+ ST_SH(vec4, temp_ptr + 384);
+
+ ADD2(in29, in28, in30, in31, in17, in18);
+ DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
+ FDCT32_POSTPROC_2V_POS_H(vec5, vec4);
+ ST_SH(vec5, temp_ptr + 64);
+ ST_SH(vec4, temp_ptr + 896);
+}
+
+static void fdct8x32_1d_column(const int16_t *input, int32_t src_stride,
+ int16_t *tmp_buf, int16_t *tmp_buf_big) {
+ fdct8x32_1d_column_load_butterfly(input, src_stride, tmp_buf);
+ fdct8x32_1d_column_even_store(tmp_buf, tmp_buf_big);
+ fdct8x32_1d_column_odd_store(tmp_buf + 128, (tmp_buf_big + 32));
+}
+
+static void fdct8x32_1d_row_load_butterfly(int16_t *temp_buff,
+ int16_t *output) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 step0, step1, step2, step3, step4, step5, step6, step7;
+
+ LD_SH8(temp_buff, 32, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8(temp_buff + 24, 32, 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);
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ step0, step1, step2, step3, step4, step5, step6, step7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7, output, 8);
+ ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 24 * 8), 8);
+
+ /* 2nd set */
+ LD_SH8(temp_buff + 8, 32, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8(temp_buff + 16, 32, 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);
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ step0, step1, step2, step3, step4, step5, step6, step7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7,
+ (output + 8 * 8), 8);
+ ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 16 * 8), 8);
+}
+
+static void fdct8x32_1d_row_even_4x(int16_t *input, int16_t *interm_ptr,
+ int16_t *out) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
+ v4i32 vec0_l, vec1_l, vec2_l, vec3_l, vec4_l, vec5_l, vec6_l, vec7_l;
+ v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec4_r, vec5_r, vec6_r, vec7_r;
+ v4i32 tmp0_w, tmp1_w, tmp2_w, tmp3_w;
+
+ /* fdct32 even */
+ /* stage 2 */
+ LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8(input + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
+
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, interm_ptr, 8);
+ ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, interm_ptr + 64, 8);
+
+ /* Stage 3 */
+ UNPCK_SH_SW(vec0, vec0_l, vec0_r);
+ UNPCK_SH_SW(vec1, vec1_l, vec1_r);
+ UNPCK_SH_SW(vec2, vec2_l, vec2_r);
+ UNPCK_SH_SW(vec3, vec3_l, vec3_r);
+ UNPCK_SH_SW(vec4, vec4_l, vec4_r);
+ UNPCK_SH_SW(vec5, vec5_l, vec5_r);
+ UNPCK_SH_SW(vec6, vec6_l, vec6_r);
+ UNPCK_SH_SW(vec7, vec7_l, vec7_r);
+ ADD4(vec0_r, vec7_r, vec1_r, vec6_r, vec2_r, vec5_r, vec3_r, vec4_r,
+ tmp0_w, tmp1_w, tmp2_w, tmp3_w);
+ BUTTERFLY_4(tmp0_w, tmp1_w, tmp2_w, tmp3_w, vec4_r, vec6_r, vec7_r, vec5_r);
+ ADD4(vec0_l, vec7_l, vec1_l, vec6_l, vec2_l, vec5_l, vec3_l, vec4_l,
+ vec0_r, vec1_r, vec2_r, vec3_r);
+
+ tmp3_w = vec0_r + vec3_r;
+ vec0_r = vec0_r - vec3_r;
+ vec3_r = vec1_r + vec2_r;
+ vec1_r = vec1_r - vec2_r;
+
+ DOTP_CONST_PAIR_W(vec4_r, vec6_r, tmp3_w, vec3_r, cospi_16_64,
+ cospi_16_64, vec4_r, tmp3_w, vec6_r, vec3_r);
+ FDCT32_POSTPROC_NEG_W(vec4_r);
+ FDCT32_POSTPROC_NEG_W(tmp3_w);
+ FDCT32_POSTPROC_NEG_W(vec6_r);
+ FDCT32_POSTPROC_NEG_W(vec3_r);
+ PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
+ ST_SH2(vec5, vec4, out, 8);
+
+ DOTP_CONST_PAIR_W(vec5_r, vec7_r, vec0_r, vec1_r, cospi_24_64,
+ cospi_8_64, vec4_r, tmp3_w, vec6_r, vec3_r);
+ FDCT32_POSTPROC_NEG_W(vec4_r);
+ FDCT32_POSTPROC_NEG_W(tmp3_w);
+ FDCT32_POSTPROC_NEG_W(vec6_r);
+ FDCT32_POSTPROC_NEG_W(vec3_r);
+ PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5);
+ ST_SH2(vec5, vec4, out + 16, 8);
+
+ LD_SH8(interm_ptr, 8, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
+ SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
+ DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
+ ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
+ DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, in5, in4);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 32);
+ ST_SH(in5, out + 56);
+
+ SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
+ DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, in5, in4);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 40);
+ ST_SH(in5, out + 48);
+
+ LD_SH8(interm_ptr + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
+ DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
+ DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
+ ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
+ DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
+ ADD2(in0, in1, in2, in3, vec0, vec7);
+ DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, in5, in4);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 64);
+ ST_SH(in5, out + 120);
+
+ SUB2(in0, in1, in2, in3, in0, in2);
+ DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, in5, in4);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 72);
+ ST_SH(in5, out + 112);
+
+ SUB2(in9, vec2, in14, vec5, vec2, vec5);
+ DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
+ SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
+ DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, in5, in4);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 80);
+ ST_SH(in5, out + 104);
+
+ ADD2(in3, in2, in0, in1, vec3, vec4);
+ DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, in4, in5);
+ FDCT_POSTPROC_2V_NEG_H(in4, in5);
+ ST_SH(in4, out + 96);
+ ST_SH(in5, out + 88);
+}
+
+static void fdct8x32_1d_row_even(int16_t *temp, int16_t *out) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1;
+
+ /* fdct32 even */
+ /* stage 2 */
+ LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
+
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+
+ /* Stage 3 */
+ ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
+ BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0);
+ DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out);
+ ST_SH(temp1, out + 8);
+
+ DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 16);
+ ST_SH(temp1, out + 24);
+
+ SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
+ DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
+ ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
+ DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 32);
+ ST_SH(temp1, out + 56);
+
+ SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
+ DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 40);
+ ST_SH(temp1, out + 48);
+
+ DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
+ DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
+ ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
+ DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
+ ADD2(in0, in1, in2, in3, vec0, vec7);
+ DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 64);
+ ST_SH(temp1, out + 120);
+
+ SUB2(in0, in1, in2, in3, in0, in2);
+ DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 72);
+ ST_SH(temp1, out + 112);
+
+ SUB2(in9, vec2, in14, vec5, vec2, vec5);
+ DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
+ SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5)
+ DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 80);
+ ST_SH(temp1, out + 104);
+
+ ADD2(in3, in2, in0, in1, vec3, vec4);
+ DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
+ FDCT_POSTPROC_2V_NEG_H(temp0, temp1);
+ ST_SH(temp0, out + 96);
+ ST_SH(temp1, out + 88);
+}
+
+static void fdct8x32_1d_row_odd(int16_t *temp, int16_t *interm_ptr,
+ int16_t *out) {
+ v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
+ v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5;
+
+ in20 = LD_SH(temp + 32);
+ in21 = LD_SH(temp + 40);
+ in26 = LD_SH(temp + 80);
+ in27 = LD_SH(temp + 88);
+
+ DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
+ DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
+
+ in18 = LD_SH(temp + 16);
+ in19 = LD_SH(temp + 24);
+ in28 = LD_SH(temp + 96);
+ in29 = LD_SH(temp + 104);
+
+ vec4 = in19 - in20;
+ ST_SH(vec4, interm_ptr + 32);
+ vec4 = in18 - in21;
+ ST_SH(vec4, interm_ptr + 88);
+ vec4 = in28 - in27;
+ ST_SH(vec4, interm_ptr + 56);
+ vec4 = in29 - in26;
+ ST_SH(vec4, interm_ptr + 64);
+
+ ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
+
+ in22 = LD_SH(temp + 48);
+ in23 = LD_SH(temp + 56);
+ in24 = LD_SH(temp + 64);
+ in25 = LD_SH(temp + 72);
+
+ DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
+ DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
+
+ in16 = LD_SH(temp);
+ in17 = LD_SH(temp + 8);
+ in30 = LD_SH(temp + 112);
+ in31 = LD_SH(temp + 120);
+
+ vec4 = in17 - in22;
+ ST_SH(vec4, interm_ptr + 40);
+ vec4 = in30 - in25;
+ ST_SH(vec4, interm_ptr + 48);
+ vec4 = in31 - in24;
+ ST_SH(vec4, interm_ptr + 72);
+ vec4 = in16 - in23;
+ ST_SH(vec4, interm_ptr + 80);
+
+ ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
+ DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
+ DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
+
+ ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
+ DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
+ ADD2(in27, in26, in25, in24, in23, in20);
+
+ DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec5, out);
+ ST_SH(vec4, out + 120);
+
+ SUB2(in27, in26, in25, in24, in22, in21);
+
+ DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec5, out + 112);
+ ST_SH(vec4, out + 8);
+
+ SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
+ DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
+ SUB2(in26, in27, in24, in25, in23, in20);
+
+ DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec4, out + 16);
+ ST_SH(vec5, out + 104);
+
+ ADD2(in26, in27, in24, in25, in22, in21);
+ DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec4, out + 24);
+ ST_SH(vec5, out + 96);
+
+ in20 = LD_SH(interm_ptr + 32);
+ in21 = LD_SH(interm_ptr + 88);
+ in27 = LD_SH(interm_ptr + 56);
+ in26 = LD_SH(interm_ptr + 64);
+
+ in16 = in20;
+ in17 = in21;
+ DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
+ DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
+
+ in22 = LD_SH(interm_ptr + 40);
+ in25 = LD_SH(interm_ptr + 48);
+ in24 = LD_SH(interm_ptr + 72);
+ in23 = LD_SH(interm_ptr + 80);
+
+ SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
+ DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
+ ADD2(in28, in29, in31, in30, in16, in19);
+ DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec5, out + 32);
+ ST_SH(vec4, out + 88);
+
+ SUB2(in28, in29, in31, in30, in17, in18);
+ DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec5, out + 40);
+ ST_SH(vec4, out + 80);
+
+ ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
+ DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
+ SUB2(in29, in28, in30, in31, in16, in19);
+
+ DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec5, out + 72);
+ ST_SH(vec4, out + 48);
+
+ ADD2(in29, in28, in30, in31, in17, in18);
+
+ DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
+ FDCT_POSTPROC_2V_NEG_H(vec5, vec4);
+ ST_SH(vec4, out + 56);
+ ST_SH(vec5, out + 64);
+}
+
+static void fdct8x32_1d_row_transpose_store(int16_t *temp, int16_t *output) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1;
+
+ /* 1st set */
+ in0 = LD_SH(temp);
+ in4 = LD_SH(temp + 32);
+ in2 = LD_SH(temp + 64);
+ in6 = LD_SH(temp + 96);
+ in1 = LD_SH(temp + 128);
+ in7 = LD_SH(temp + 152);
+ in3 = LD_SH(temp + 192);
+ in5 = LD_SH(temp + 216);
+
+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+
+ /* 2nd set */
+ in0_1 = LD_SH(temp + 16);
+ in1_1 = LD_SH(temp + 232);
+ in2_1 = LD_SH(temp + 80);
+ in3_1 = LD_SH(temp + 168);
+ in4_1 = LD_SH(temp + 48);
+ in5_1 = LD_SH(temp + 176);
+ in6_1 = LD_SH(temp + 112);
+ in7_1 = LD_SH(temp + 240);
+
+ ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 32);
+ TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
+
+ /* 3rd set */
+ in0 = LD_SH(temp + 8);
+ in1 = LD_SH(temp + 136);
+ in2 = LD_SH(temp + 72);
+ in3 = LD_SH(temp + 200);
+ in4 = LD_SH(temp + 40);
+ in5 = LD_SH(temp + 208);
+ in6 = LD_SH(temp + 104);
+ in7 = LD_SH(temp + 144);
+
+ ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ output + 8, 32);
+ TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
+ in0, in1, in2, in3, in4, in5, in6, in7);
+ ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output + 16, 32);
+
+ /* 4th set */
+ in0_1 = LD_SH(temp + 24);
+ in1_1 = LD_SH(temp + 224);
+ in2_1 = LD_SH(temp + 88);
+ in3_1 = LD_SH(temp + 160);
+ in4_1 = LD_SH(temp + 56);
+ in5_1 = LD_SH(temp + 184);
+ in6_1 = LD_SH(temp + 120);
+ in7_1 = LD_SH(temp + 248);
+
+ TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1);
+ ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1,
+ output + 24, 32);
+}
+
+static void fdct32x8_1d_row(int16_t *temp, int16_t *temp_buf,
+ int16_t *output) {
+ fdct8x32_1d_row_load_butterfly(temp, temp_buf);
+ fdct8x32_1d_row_even(temp_buf, temp_buf);
+ fdct8x32_1d_row_odd(temp_buf + 128, temp, temp_buf + 128);
+ fdct8x32_1d_row_transpose_store(temp_buf, output);
+}
+
+static void fdct32x8_1d_row_4x(int16_t *tmp_buf_big, int16_t *tmp_buf,
+ int16_t *output) {
+ fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf);
+ fdct8x32_1d_row_even_4x(tmp_buf, tmp_buf_big, tmp_buf);
+ fdct8x32_1d_row_odd(tmp_buf + 128, tmp_buf_big, tmp_buf + 128);
+ fdct8x32_1d_row_transpose_store(tmp_buf, output);
+}
+
+void vp9_fdct32x32_msa(const int16_t *input, int16_t *output,
+ int32_t src_stride) {
+ int32_t i;
+ DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]);
+ DECLARE_ALIGNED(32, int16_t, tmp_buf[256]);
+
+ /* column transform */
+ for (i = 0; i < 4; ++i) {
+ fdct8x32_1d_column(input + (8 * i), src_stride, tmp_buf,
+ tmp_buf_big + (8 * i));
+ }
+
+ /* row transform */
+ fdct32x8_1d_row_4x(tmp_buf_big, tmp_buf, output);
+
+ /* row transform */
+ for (i = 1; i < 4; ++i) {
+ fdct32x8_1d_row(tmp_buf_big + (i * 256), tmp_buf, output + (i * 256));
+ }
+}
+
+static void fdct8x32_1d_row_even_rd(int16_t *temp, int16_t *out) {
+ v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
+ v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
+ v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1;
+
+ /* fdct32 even */
+ /* stage 2 */
+ LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7);
+ LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15);
+
+ BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7,
+ in8, in9, in10, in11, in12, in13, in14, in15,
+ vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7,
+ in8, in9, in10, in11, in12, in13, in14, in15);
+ FDCT_POSTPROC_2V_NEG_H(vec0, vec1);
+ FDCT_POSTPROC_2V_NEG_H(vec2, vec3);
+ FDCT_POSTPROC_2V_NEG_H(vec4, vec5);
+ FDCT_POSTPROC_2V_NEG_H(vec6, vec7);
+ FDCT_POSTPROC_2V_NEG_H(in8, in9);
+ FDCT_POSTPROC_2V_NEG_H(in10, in11);
+ FDCT_POSTPROC_2V_NEG_H(in12, in13);
+ FDCT_POSTPROC_2V_NEG_H(in14, in15);
+
+ /* Stage 3 */
+ ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3);
+
+ temp0 = in0 + in3;
+ in0 = in0 - in3;
+ in3 = in1 + in2;
+ in1 = in1 - in2;
+
+ DOTP_CONST_PAIR(temp0, in3, cospi_16_64, cospi_16_64, temp1, temp0);
+ ST_SH(temp0, out);
+ ST_SH(temp1, out + 8);
+
+ DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0);
+ ST_SH(temp0, out + 16);
+ ST_SH(temp1, out + 24);
+
+ SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7);
+ DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6);
+ ADD2(vec4, vec5, vec7, vec6, vec0, vec1);
+ DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0);
+ ST_SH(temp0, out + 32);
+ ST_SH(temp1, out + 56);
+
+ SUB2(vec4, vec5, vec7, vec6, vec4, vec7);
+ DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0);
+ ST_SH(temp0, out + 40);
+ ST_SH(temp1, out + 48);
+
+ DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5);
+ DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4);
+ ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2);
+ DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3);
+ ADD2(in0, in1, in2, in3, vec0, vec7);
+ DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0);
+ ST_SH(temp0, out + 64);
+ ST_SH(temp1, out + 120);
+
+ SUB2(in0, in1, in2, in3, in0, in2);
+ DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0);
+ ST_SH(temp0, out + 72);
+ ST_SH(temp1, out + 112);
+
+ SUB2(in9, vec2, in14, vec5, vec2, vec5);
+ DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1);
+ SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5);
+ DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0);
+ ST_SH(temp0, out + 80);
+ ST_SH(temp1, out + 104);
+
+ ADD2(in3, in2, in0, in1, vec3, vec4);
+ DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1);
+ ST_SH(temp0, out + 96);
+ ST_SH(temp1, out + 88);
+}
+
+static void fdct8x32_1d_row_odd_rd(int16_t *temp, int16_t *interm_ptr,
+ int16_t *out) {
+ v8i16 in16, in17, in18, in19, in20, in21, in22, in23;
+ v8i16 in24, in25, in26, in27, in28, in29, in30, in31;
+ v8i16 vec4, vec5;
+
+ in20 = LD_SH(temp + 32);
+ in21 = LD_SH(temp + 40);
+ in26 = LD_SH(temp + 80);
+ in27 = LD_SH(temp + 88);
+
+ DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27);
+ DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26);
+
+ FDCT_POSTPROC_2V_NEG_H(in20, in21);
+ FDCT_POSTPROC_2V_NEG_H(in26, in27);
+
+ in18 = LD_SH(temp + 16);
+ in19 = LD_SH(temp + 24);
+ in28 = LD_SH(temp + 96);
+ in29 = LD_SH(temp + 104);
+
+ FDCT_POSTPROC_2V_NEG_H(in18, in19);
+ FDCT_POSTPROC_2V_NEG_H(in28, in29);
+
+ vec4 = in19 - in20;
+ ST_SH(vec4, interm_ptr + 32);
+ vec4 = in18 - in21;
+ ST_SH(vec4, interm_ptr + 88);
+ vec4 = in29 - in26;
+ ST_SH(vec4, interm_ptr + 64);
+ vec4 = in28 - in27;
+ ST_SH(vec4, interm_ptr + 56);
+
+ ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26);
+
+ in22 = LD_SH(temp + 48);
+ in23 = LD_SH(temp + 56);
+ in24 = LD_SH(temp + 64);
+ in25 = LD_SH(temp + 72);
+
+ DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25);
+ DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24);
+ FDCT_POSTPROC_2V_NEG_H(in22, in23);
+ FDCT_POSTPROC_2V_NEG_H(in24, in25);
+
+ in16 = LD_SH(temp);
+ in17 = LD_SH(temp + 8);
+ in30 = LD_SH(temp + 112);
+ in31 = LD_SH(temp + 120);
+
+ FDCT_POSTPROC_2V_NEG_H(in16, in17);
+ FDCT_POSTPROC_2V_NEG_H(in30, in31);
+
+ vec4 = in17 - in22;
+ ST_SH(vec4, interm_ptr + 40);
+ vec4 = in30 - in25;
+ ST_SH(vec4, interm_ptr + 48);
+ vec4 = in31 - in24;
+ ST_SH(vec4, interm_ptr + 72);
+ vec4 = in16 - in23;
+ ST_SH(vec4, interm_ptr + 80);
+
+ ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31);
+ DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29);
+ DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28);
+ ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25);
+ DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24);
+ ADD2(in27, in26, in25, in24, in23, in20);
+ DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5);
+ ST_SH(vec5, out);
+ ST_SH(vec4, out + 120);
+
+ SUB2(in27, in26, in25, in24, in22, in21);
+ DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4);
+ ST_SH(vec5, out + 112);
+ ST_SH(vec4, out + 8);
+
+ SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20);
+ DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25);
+ SUB2(in26, in27, in24, in25, in23, in20);
+ DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5);
+ ST_SH(vec4, out + 16);
+ ST_SH(vec5, out + 104);
+
+ ADD2(in26, in27, in24, in25, in22, in21);
+ DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5);
+ ST_SH(vec4, out + 24);
+ ST_SH(vec5, out + 96);
+
+ in20 = LD_SH(interm_ptr + 32);
+ in21 = LD_SH(interm_ptr + 88);
+ in27 = LD_SH(interm_ptr + 56);
+ in26 = LD_SH(interm_ptr + 64);
+
+ in16 = in20;
+ in17 = in21;
+ DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27);
+ DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26);
+
+ in22 = LD_SH(interm_ptr + 40);
+ in25 = LD_SH(interm_ptr + 48);
+ in24 = LD_SH(interm_ptr + 72);
+ in23 = LD_SH(interm_ptr + 80);
+
+ SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31);
+ DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30);
+ in16 = in28 + in29;
+ in19 = in31 + in30;
+ DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4);
+ ST_SH(vec5, out + 32);
+ ST_SH(vec4, out + 88);
+
+ SUB2(in28, in29, in31, in30, in17, in18);
+ DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4);
+ ST_SH(vec5, out + 40);
+ ST_SH(vec4, out + 80);
+
+ ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19);
+ DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31);
+ SUB2(in29, in28, in30, in31, in16, in19);
+ DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4);
+ ST_SH(vec5, out + 72);
+ ST_SH(vec4, out + 48);
+
+ ADD2(in29, in28, in30, in31, in17, in18);
+ DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4);
+ ST_SH(vec4, out + 56);
+ ST_SH(vec5, out + 64);
+}
+
+static void fdct32x8_1d_row_rd(int16_t *tmp_buf_big, int16_t *tmp_buf,
+ int16_t *output) {
+ fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf);
+ fdct8x32_1d_row_even_rd(tmp_buf, tmp_buf);
+ fdct8x32_1d_row_odd_rd((tmp_buf + 128), tmp_buf_big, (tmp_buf + 128));
+ fdct8x32_1d_row_transpose_store(tmp_buf, output);
+}
+
+void vp9_fdct32x32_rd_msa(const int16_t *input, int16_t *out,
+ int32_t src_stride) {
+ int32_t i;
+ DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]);
+ DECLARE_ALIGNED(32, int16_t, tmp_buf[256]);
+
+ /* column transform */
+ for (i = 0; i < 4; ++i) {
+ fdct8x32_1d_column(input + (8 * i), src_stride, &tmp_buf[0],
+ &tmp_buf_big[0] + (8 * i));
+ }
+
+ /* row transform */
+ for (i = 0; i < 4; ++i) {
+ fdct32x8_1d_row_rd(&tmp_buf_big[0] + (8 * i * 32), &tmp_buf[0],
+ out + (8 * i * 32));
+ }
+}
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
}
+#define 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 FDCT32_POSTPROC_NEG_W(vec) { \
+ v4i32 temp_m; \
+ v4i32 one_m = __msa_ldi_w(1); \
+ \
+ temp_m = __msa_clti_s_w(vec, 0); \
+ vec += 1; \
+ temp_m = one_m & temp_m; \
+ vec += temp_m; \
+ vec >>= 2; \
+}
+
+#define FDCT32_POSTPROC_2V_POS_H(vec0, vec1) { \
+ v8i16 tp0_m, tp1_m; \
+ v8i16 one = __msa_ldi_h(1); \
+ \
+ tp0_m = __msa_clei_s_h(vec0, 0); \
+ tp1_m = __msa_clei_s_h(vec1, 0); \
+ tp0_m = (v8i16)__msa_xori_b((v16u8)tp0_m, 255); \
+ tp1_m = (v8i16)__msa_xori_b((v16u8)tp1_m, 255); \
+ vec0 += 1; \
+ vec1 += 1; \
+ tp0_m = one & tp0_m; \
+ tp1_m = one & tp1_m; \
+ vec0 += tp0_m; \
+ vec1 += tp1_m; \
+ vec0 >>= 2; \
+ vec1 >>= 2; \
+}
+
+#define DOTP_CONST_PAIR_W(reg0_left, reg1_left, reg0_right, \
+ reg1_right, const0, const1, \
+ out0, out1, out2, out3) { \
+ v4i32 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m; \
+ v2i64 tp0_m, tp1_m, tp2_m, tp3_m; \
+ v4i32 k0_m = __msa_fill_w((int32_t) const0); \
+ \
+ s0_m = __msa_fill_w((int32_t) const1); \
+ k0_m = __msa_ilvev_w(s0_m, k0_m); \
+ \
+ ILVRL_W2_SW(-reg1_left, reg0_left, s1_m, s0_m); \
+ ILVRL_W2_SW(reg0_left, reg1_left, s3_m, s2_m); \
+ ILVRL_W2_SW(-reg1_right, reg0_right, s5_m, s4_m); \
+ ILVRL_W2_SW(reg0_right, reg1_right, s7_m, s6_m); \
+ \
+ DOTP_SW2_SD(s0_m, s1_m, k0_m, k0_m, tp0_m, tp1_m); \
+ DOTP_SW2_SD(s4_m, s5_m, k0_m, k0_m, tp2_m, tp3_m); \
+ tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS); \
+ tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS); \
+ tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS); \
+ tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS); \
+ out0 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m); \
+ out1 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m); \
+ \
+ DOTP_SW2_SD(s2_m, s3_m, k0_m, k0_m, tp0_m, tp1_m); \
+ DOTP_SW2_SD(s6_m, s7_m, k0_m, k0_m, tp2_m, tp3_m); \
+ tp0_m = __msa_srari_d(tp0_m, DCT_CONST_BITS); \
+ tp1_m = __msa_srari_d(tp1_m, DCT_CONST_BITS); \
+ tp2_m = __msa_srari_d(tp2_m, DCT_CONST_BITS); \
+ tp3_m = __msa_srari_d(tp3_m, DCT_CONST_BITS); \
+ out2 = __msa_pckev_w((v4i32)tp0_m, (v4i32)tp1_m); \
+ out3 = __msa_pckev_w((v4i32)tp2_m, (v4i32)tp3_m); \
+}
+
void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
int32_t src_stride);
void fdct16x8_1d_row(int16_t *input, int16_t *output);
DSP_SRCS-$(HAVE_SSE2) += x86/fwd_txfm_sse2.h
DSP_SRCS-$(HAVE_SSE2) += x86/fwd_txfm_sse2.c
DSP_SRCS-$(HAVE_SSE2) += x86/fwd_txfm_impl_sse2.h
+DSP_SRCS-$(HAVE_SSE2) += x86/fwd_dct32x32_impl_sse2.h
ifeq ($(ARCH_X86_64),yes)
ifeq ($(CONFIG_USE_X86INC),yes)
DSP_SRCS-$(HAVE_SSSE3) += x86/fwd_txfm_ssse3_x86_64.asm
endif
endif
+DSP_SRCS-$(HAVE_AVX2) += x86/fwd_txfm_avx2.c
+DSP_SRCS-$(HAVE_AVX2) += x86/fwd_dct32x32_impl_avx2.h
DSP_SRCS-$(HAVE_NEON) += arm/fwd_txfm_neon.c
DSP_SRCS-$(HAVE_MSA) += mips/fwd_txfm_msa.h
DSP_SRCS-$(HAVE_MSA) += mips/fwd_txfm_msa.c
+DSP_SRCS-$(HAVE_MSA) += mips/fwd_dct32x32_msa.c
endif # CONFIG_VP9_ENCODER
# quantization
add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct16x16 sse2/;
+ add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_fdct32x32 sse2/;
+
+ add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_fdct32x32_rd sse2/;
+
add_proto qw/void vp9_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_highbd_fdct4x4 sse2/;
add_proto qw/void vp9_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_highbd_fdct16x16 sse2/;
+
+ add_proto qw/void vp9_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_highbd_fdct32x32 sse2/;
+
+ add_proto qw/void vp9_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_highbd_fdct32x32_rd sse2/;
} else {
add_proto qw/void vp9_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct4x4 sse2 msa/;
add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct16x16 sse2 msa/;
+
+ add_proto qw/void vp9_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_fdct32x32 sse2 avx2 msa/;
+
+ add_proto qw/void vp9_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
+ specialize qw/vp9_fdct32x32_rd sse2 avx2 msa/;
} # CONFIG_VP9_HIGHBITDEPTH
} # CONFIG_VP9_ENCODER
#include <immintrin.h> // AVX2
-#include "./vp9_rtcd.h"
#include "vpx_dsp/txfm_common.h"
-#include "vpx_ports/mem.h"
#define pair256_set_epi16(a, b) \
_mm256_set_epi16((int16_t)(b), (int16_t)(a), (int16_t)(b), (int16_t)(a), \
#include <emmintrin.h> // SSE2
-#include "vp9/encoder/vp9_dct.h"
+#include "vpx_dsp/fwd_txfm.h"
#include "vpx_dsp/txfm_common.h"
#include "vpx_dsp/x86/txfm_common_sse2.h"
-#include "vpx_ports/mem.h"
+// TODO(jingning) The high bit-depth version needs re-work for performance.
+// The current SSE2 implementation also causes cross reference to the static
+// functions in the C implementation file.
#if DCT_HIGH_BIT_DEPTH
#define ADD_EPI16 _mm_adds_epi16
#define SUB_EPI16 _mm_subs_epi16
* be found in the AUTHORS file in the root of the source tree.
*/
-#include <immintrin.h> // AVX2
-#include "vp9/common/vp9_idct.h" // for cospi constants
-#include "vpx_ports/mem.h"
-
+#include "./vpx_config.h"
#define FDCT32x32_2D_AVX2 vp9_fdct32x32_rd_avx2
#define FDCT32x32_HIGH_PRECISION 0
-#include "vp9/encoder/x86/vp9_dct32x32_avx2_impl.h"
+#include "vpx_dsp/x86/fwd_dct32x32_impl_avx2.h"
#undef FDCT32x32_2D_AVX2
#undef FDCT32x32_HIGH_PRECISION
#define FDCT32x32_2D_AVX2 vp9_fdct32x32_avx2
#define FDCT32x32_HIGH_PRECISION 1
-#include "vp9/encoder/x86/vp9_dct32x32_avx2_impl.h" // NOLINT
+#include "vpx_dsp/x86/fwd_dct32x32_impl_avx2.h" // NOLINT
#undef FDCT32x32_2D_AVX2
#undef FDCT32x32_HIGH_PRECISION
#include "./vpx_config.h"
#define DCT_HIGH_BIT_DEPTH 0
-
#define FDCT4x4_2D vp9_fdct4x4_sse2
#define FDCT8x8_2D vp9_fdct8x8_sse2
#define FDCT16x16_2D vp9_fdct16x16_sse2
#undef FDCT4x4_2D
#undef FDCT8x8_2D
#undef FDCT16x16_2D
+
+#define FDCT32x32_2D vp9_fdct32x32_rd_sse2
+#define FDCT32x32_HIGH_PRECISION 0
+#include "vpx_dsp/x86/fwd_dct32x32_impl_sse2.h"
+#undef FDCT32x32_2D
+#undef FDCT32x32_HIGH_PRECISION
+
+#define FDCT32x32_2D vp9_fdct32x32_sse2
+#define FDCT32x32_HIGH_PRECISION 1
+#include "vpx_dsp/x86/fwd_dct32x32_impl_sse2.h" // NOLINT
+#undef FDCT32x32_2D
+#undef FDCT32x32_HIGH_PRECISION
#undef DCT_HIGH_BIT_DEPTH
#if CONFIG_VP9_HIGHBITDEPTH
#undef FDCT4x4_2D
#undef FDCT8x8_2D
#undef FDCT16x16_2D
+
+#define FDCT32x32_2D vp9_highbd_fdct32x32_rd_sse2
+#define FDCT32x32_HIGH_PRECISION 0
+#include "vpx_dsp/x86/fwd_dct32x32_impl_sse2.h" // NOLINT
+#undef FDCT32x32_2D
+#undef FDCT32x32_HIGH_PRECISION
+
+#define FDCT32x32_2D vp9_highbd_fdct32x32_sse2
+#define FDCT32x32_HIGH_PRECISION 1
+#include "vpx_dsp/x86/fwd_dct32x32_impl_sse2.h" // NOLINT
+#undef FDCT32x32_2D
+#undef FDCT32x32_HIGH_PRECISION
#undef DCT_HIGH_BIT_DEPTH
#endif // CONFIG_VP9_HIGHBITDEPTH
projects / platform / upstream / libvpx.git / commitdiff