mozilla
/
aom
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Move msa implementations of 2D-DCT to vpx_dsp 

Refactor and clean up the msa transform related code layout.

Change-Id: Ic5048bd3d62a6046589817da745370ea89448e44
This commit is contained in:
Jingning Han 2015-07-22 11:53:21 -07:00
Родитель d4657a7efb
Коммит 9aaf523ace
12 изменённых файлов: 525 добавлений и 492 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

3
test/dct16x16_test.cc
Просмотреть файл

@ -922,8 +922,7 @@ INSTANTIATE_TEST_CASE_P(
&idct16x16_256_add_12_sse2, 3167, VPX_BITS_12)));
#endif // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
// TODO(jingning) Re-enable the mips/msa unit test.
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE && 0
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
MSA, Trans16x16DCT,
::testing::Values(

3
test/fdct4x4_test.cc
Просмотреть файл

@ -538,8 +538,7 @@ INSTANTIATE_TEST_CASE_P(
make_tuple(&vp9_fht4x4_sse2, &vp9_iht4x4_16_add_c, 3, VPX_BITS_8)));
#endif // HAVE_SSE2 && CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
// TODO(jingning) Re-enable the mips/msa unit test.
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE && 0
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
MSA, Trans4x4DCT,
::testing::Values(

3
test/fdct8x8_test.cc
Просмотреть файл

@ -773,8 +773,7 @@ INSTANTIATE_TEST_CASE_P(
VPX_BITS_8)));
#endif
// TODO(jingning) Re-enable the mips/msa unit test.
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE && 0
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
MSA, FwdTrans8x8DCT,
::testing::Values(

173
vp9/encoder/mips/msa/vp9_fdct16x16_msa.c
Просмотреть файл

@ -12,178 +12,7 @@
#include "./vp9_rtcd.h"
#include "vp9/encoder/mips/msa/vp9_fdct_msa.h"
static void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
int32_t src_stride) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
-cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
v8i16 coeff2 = { -cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64,
0, 0, 0, 0 };
LD_SH16(input, src_stride,
in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in8, in9, in10, in11, 2);
SLLI_4V(in12, in13, in14, in15, 2);
ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
tmp_ptr += 16;
/* stp 1 */
ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
cnst4 = __msa_splati_h(coeff, 0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
cnst5 = __msa_splati_h(coeff, 1);
cnst5 = __msa_ilvev_h(cnst5, cnst4);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
stp24 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
stp23 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
/* stp2 */
BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
stp26 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
cnst0 = __msa_splati_h(coeff, 4);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp21 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
ILVRL_H2_SH(in15, in8, vec1, vec0);
SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr);
cnst0 = __msa_splati_h(coeff2, 0);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 224);
ILVRL_H2_SH(in14, in9, vec1, vec0);
SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 128);
cnst1 = __msa_splati_h(coeff2, 2);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 96);
SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
cnst1 = __msa_splati_h(coeff, 3);
cnst1 = __msa_ilvev_h(cnst0, cnst1);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
/* stp4 */
ADD2(stp34, stp25, stp33, stp22, in13, in10);
ILVRL_H2_SH(in13, in10, vec1, vec0);
SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 64);
cnst0 = __msa_splati_h(coeff2, 1);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 160);
SUB2(stp34, stp25, stp33, stp22, in12, in11);
ILVRL_H2_SH(in12, in11, vec1, vec0);
SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 192);
cnst1 = __msa_splati_h(coeff2, 3);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 32);
}
static void fdct16x8_1d_row(int16_t *input, int16_t *output) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
SRA_4V(in0, in1, in2, in3, 2);
SRA_4V(in4, in5, in6, in7, 2);
SRA_4V(in8, in9, in10, in11, 2);
SRA_4V(in12, in13, in14, in15, 2);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5,
tmp6, tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
}
void vp9_fdct16x16_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
int32_t i;
DECLARE_ALIGNED(32, int16_t, tmp_buf[16 * 16]);
/* column transform */
for (i = 0; i < 2; ++i) {
fdct8x16_1d_column((input + 8 * i), (&tmp_buf[0] + 8 * i), src_stride);
}
/* row transform */
for (i = 0; i < 2; ++i) {
fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i)));
}
}
#include "vpx_dsp/mips/fwd_txfm_msa.h"
void vp9_fdct16x16_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
out[1] = 0;

30
vp9/encoder/mips/msa/vp9_fdct4x4_msa.c
Просмотреть файл

@ -45,36 +45,6 @@ void vp9_fwht4x4_msa(const int16_t *input, int16_t *output,
ST4x2_UB(in2, output + 12, 4);
}
void vp9_fdct4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, src_stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 vec, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
vec = __msa_ceqi_h(in0, 0);
vec = vec ^ 255;
vec = mask & vec;
in0 += vec;
}
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
void vp9_fht4x4_msa(const int16_t *input, int16_t *output, int32_t stride,
int32_t tx_type) {
v8i16 in0, in1, in2, in3;

19
vp9/encoder/mips/msa/vp9_fdct8x8_msa.c
Просмотреть файл

@ -13,25 +13,6 @@
#include "./vp9_rtcd.h"
#include "vp9/encoder/mips/msa/vp9_fdct_msa.h"
void vp9_fdct8x8_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}
void vp9_fdct8x8_1_msa(const int16_t *input, int16_t *out, int32_t stride) {
out[0] = LD_HADD(input, stride);
out[1] = 0;

263
vp9/encoder/mips/msa/vp9_fdct_msa.h
Просмотреть файл

@ -11,10 +11,10 @@
#ifndef VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
#define VP9_ENCODER_MIPS_MSA_VP9_FDCT_MSA_H_
#include "vpx_ports/mem.h"
#include "vp9/common/vp9_idct.h"
#include "vpx_dsp/mips/macros_msa.h"
#include "vpx_dsp/mips/fwd_txfm_msa.h"
#include "vpx_dsp/mips/txfm_macros_msa.h"
#include "vpx_ports/mem.h"
#define VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
@ -113,33 +113,6 @@
vec1 >>= 2; \
}
#define VP9_FDCT4(in0, in1, in2, in3, out0, out1, out2, out3) { \
v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m; \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m; \
v4i32 vec4_m, vec5_m, vec6_m, vec7_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, -cospi_8_64, 0, 0, 0 }; \
\
BUTTERFLY_4(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m); \
ILVR_H2_SH(vec1_m, vec0_m, vec3_m, vec2_m, vec0_m, vec2_m); \
SPLATI_H2_SH(coeff_m, 0, 1, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
vec5_m = __msa_dotp_s_w(vec0_m, cnst1_m); \
\
SPLATI_H2_SH(coeff_m, 4, 3, cnst2_m, cnst3_m); \
cnst2_m = __msa_ilvev_h(cnst3_m, cnst2_m); \
vec7_m = __msa_dotp_s_w(vec2_m, cnst2_m); \
\
vec4_m = __msa_dotp_s_w(vec0_m, cnst0_m); \
cnst2_m = __msa_splati_h(coeff_m, 2); \
cnst2_m = __msa_ilvev_h(cnst2_m, cnst3_m); \
vec6_m = __msa_dotp_s_w(vec2_m, cnst2_m); \
\
SRARI_W4_SW(vec4_m, vec5_m, vec6_m, vec7_m, DCT_CONST_BITS); \
PCKEV_H4_SH(vec4_m, vec4_m, vec5_m, vec5_m, vec6_m, vec6_m, \
vec7_m, vec7_m, out0, out2, out1, out3); \
}
#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; \
@ -174,238 +147,6 @@
s3_m, s3_m, out0, out1, out2, out3); \
}
#define VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m; \
v8i16 s7_m, x0_m, x1_m, x2_m, x3_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, cospi_4_64, cospi_28_64, \
cospi_12_64, cospi_20_64 }; \
\
/* FDCT stage1 */ \
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m); \
BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m); \
ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m); \
ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m); \
SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x1_m, x0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m); \
x2_m = -x2_m; \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
x2_m = __msa_splati_h(coeff_m, 2); \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
/* stage2 */ \
ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m); \
\
s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
/* stage3 */ \
BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m); \
\
/* stage4 */ \
ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m); \
ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m); \
\
SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x0_m, x1_m); \
out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m); \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
\
x1_m = __msa_splati_h(coeff_m, 5); \
x0_m = -x0_m; \
x0_m = __msa_ilvev_h(x1_m, x0_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m); \
\
x2_m = __msa_splati_h(coeff_m, 6); \
x3_m = -x3_m; \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
}
#define SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7) { \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
\
SRLI_H4_SH(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m, 15); \
SRLI_H4_SH(in4, in5, in6, in7, vec4_m, vec5_m, vec6_m, vec7_m, 15); \
AVE_SH4_SH(vec0_m, in0, vec1_m, in1, vec2_m, in2, vec3_m, in3, \
in0, in1, in2, in3); \
AVE_SH4_SH(vec4_m, in4, vec5_m, in5, vec6_m, in6, vec7_m, in7, \
in4, in5, in6, in7); \
}
#define FDCT8x16_EVEN(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m; \
v8i16 x0_m, x1_m, x2_m, x3_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64, \
cospi_4_64, cospi_28_64, cospi_12_64, cospi_20_64 }; \
\
/* FDCT stage1 */ \
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m); \
BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m); \
ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m); \
ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m); \
SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x1_m, x0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m); \
x2_m = -x2_m; \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
x2_m = __msa_splati_h(coeff_m, 2); \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
/* stage2 */ \
ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m); \
\
s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
/* stage3 */ \
BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m); \
\
/* stage4 */ \
ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m); \
ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m); \
\
SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x0_m, x1_m); \
out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m); \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
\
x1_m = __msa_splati_h(coeff_m, 5); \
x0_m = -x0_m; \
x0_m = __msa_ilvev_h(x1_m, x0_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m); \
\
x2_m = __msa_splati_h(coeff_m, 6); \
x3_m = -x3_m; \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
}
#define FDCT8x16_ODD(input0, input1, input2, input3, \
input4, input5, input6, input7, \
out1, out3, out5, out7, \
out9, out11, out13, out15) { \
v8i16 stp21_m, stp22_m, stp23_m, stp24_m, stp25_m, stp26_m; \
v8i16 stp30_m, stp31_m, stp32_m, stp33_m, stp34_m, stp35_m; \
v8i16 stp36_m, stp37_m, vec0_m, vec1_m; \
v8i16 vec2_m, vec3_m, vec4_m, vec5_m, vec6_m; \
v8i16 cnst0_m, cnst1_m, cnst4_m, cnst5_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, -cospi_8_64, -cospi_24_64, \
cospi_12_64, cospi_20_64 }; \
v8i16 coeff1_m = { cospi_2_64, cospi_30_64, cospi_14_64, \
cospi_18_64, cospi_10_64, cospi_22_64, \
cospi_6_64, cospi_26_64 }; \
v8i16 coeff2_m = { -cospi_2_64, -cospi_10_64, -cospi_18_64, \
-cospi_26_64, 0, 0, 0, 0 }; \
\
/* stp 1 */ \
ILVL_H2_SH(input2, input5, input3, input4, vec2_m, vec4_m); \
ILVR_H2_SH(input2, input5, input3, input4, vec3_m, vec5_m); \
\
cnst4_m = __msa_splati_h(coeff_m, 0); \
stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst4_m); \
\
cnst5_m = __msa_splati_h(coeff_m, 1); \
cnst5_m = __msa_ilvev_h(cnst5_m, cnst4_m); \
stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst5_m); \
stp24_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst4_m); \
stp23_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst5_m); \
\
/* stp2 */ \
BUTTERFLY_4(input0, input1, stp22_m, stp23_m, \
stp30_m, stp31_m, stp32_m, stp33_m); \
BUTTERFLY_4(input7, input6, stp25_m, stp24_m, \
stp37_m, stp36_m, stp35_m, stp34_m); \
\
ILVL_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec2_m, vec4_m); \
ILVR_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec3_m, vec5_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
stp26_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff_m, 4); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
stp21_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
\
SPLATI_H2_SH(coeff_m, 5, 2, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
\
cnst0_m = __msa_splati_h(coeff_m, 3); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
\
/* stp4 */ \
BUTTERFLY_4(stp30_m, stp37_m, stp26_m, stp21_m, \
vec6_m, vec2_m, vec4_m, vec5_m); \
BUTTERFLY_4(stp33_m, stp34_m, stp25_m, stp22_m, \
stp21_m, stp23_m, stp24_m, stp31_m); \
\
ILVRL_H2_SH(vec2_m, vec6_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 0, 1, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
\
out1 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff2_m, 0); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
out15 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(vec4_m, vec5_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
out9 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
\
cnst1_m = __msa_splati_h(coeff2_m, 2); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(stp23_m, stp21_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 4, 5, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff2_m, 1); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
out11 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(stp24_m, stp31_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 6, 7, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
out13 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
\
cnst1_m = __msa_splati_h(coeff2_m, 3); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
}
#define FDCT32_POSTPROC_NEG_W(vec) { \
v4i32 temp_m; \
v4i32 one_m = __msa_ldi_w(1); \

232
vpx_dsp/mips/fwd_txfm_msa.c Normal file
Просмотреть файл

@ -0,0 +1,232 @@
/*
* 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"
void fdct8x16_1d_column(const int16_t *input, int16_t *tmp_ptr,
int32_t src_stride) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
v8i16 stp21, stp22, stp23, stp24, stp25, stp26, stp30;
v8i16 stp31, stp32, stp33, stp34, stp35, stp36, stp37;
v8i16 vec0, vec1, vec2, vec3, vec4, vec5, cnst0, cnst1, cnst4, cnst5;
v8i16 coeff = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64,
-cospi_8_64, -cospi_24_64, cospi_12_64, cospi_20_64 };
v8i16 coeff1 = { cospi_2_64, cospi_30_64, cospi_14_64, cospi_18_64,
cospi_10_64, cospi_22_64, cospi_6_64, cospi_26_64 };
v8i16 coeff2 = { -cospi_2_64, -cospi_10_64, -cospi_18_64, -cospi_26_64,
0, 0, 0, 0 };
LD_SH16(input, src_stride,
in0, in1, in2, in3, in4, in5, in6, in7,
in8, in9, in10, in11, in12, in13, in14, in15);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
SLLI_4V(in8, in9, in10, in11, 2);
SLLI_4V(in12, in13, in14, in15, 2);
ADD4(in0, in15, in1, in14, in2, in13, in3, in12, tmp0, tmp1, tmp2, tmp3);
ADD4(in4, in11, in5, in10, in6, in9, in7, in8, tmp4, tmp5, tmp6, tmp7);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
ST_SH8(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp_ptr, 32);
SUB4(in0, in15, in1, in14, in2, in13, in3, in12, in15, in14, in13, in12);
SUB4(in4, in11, in5, in10, in6, in9, in7, in8, in11, in10, in9, in8);
tmp_ptr += 16;
/* stp 1 */
ILVL_H2_SH(in10, in13, in11, in12, vec2, vec4);
ILVR_H2_SH(in10, in13, in11, in12, vec3, vec5);
cnst4 = __msa_splati_h(coeff, 0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst4);
cnst5 = __msa_splati_h(coeff, 1);
cnst5 = __msa_ilvev_h(cnst5, cnst4);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst5);
stp24 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst4);
stp23 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst5);
/* stp2 */
BUTTERFLY_4(in8, in9, stp22, stp23, stp30, stp31, stp32, stp33);
BUTTERFLY_4(in15, in14, stp25, stp24, stp37, stp36, stp35, stp34);
ILVL_H2_SH(stp36, stp31, stp35, stp32, vec2, vec4);
ILVR_H2_SH(stp36, stp31, stp35, stp32, vec3, vec5);
SPLATI_H2_SH(coeff, 2, 3, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
stp26 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst0);
cnst0 = __msa_splati_h(coeff, 4);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp21 = DOT_SHIFT_RIGHT_PCK_H(vec2, vec3, cnst1);
BUTTERFLY_4(stp30, stp37, stp26, stp21, in8, in15, in14, in9);
ILVRL_H2_SH(in15, in8, vec1, vec0);
SPLATI_H2_SH(coeff1, 0, 1, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr);
cnst0 = __msa_splati_h(coeff2, 0);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 224);
ILVRL_H2_SH(in14, in9, vec1, vec0);
SPLATI_H2_SH(coeff1, 2, 3, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 128);
cnst1 = __msa_splati_h(coeff2, 2);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 96);
SPLATI_H2_SH(coeff, 2, 5, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
stp25 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
cnst1 = __msa_splati_h(coeff, 3);
cnst1 = __msa_ilvev_h(cnst0, cnst1);
stp22 = DOT_SHIFT_RIGHT_PCK_H(vec4, vec5, cnst1);
/* stp4 */
ADD2(stp34, stp25, stp33, stp22, in13, in10);
ILVRL_H2_SH(in13, in10, vec1, vec0);
SPLATI_H2_SH(coeff1, 4, 5, cnst0, cnst1);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 64);
cnst0 = __msa_splati_h(coeff2, 1);
cnst0 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 160);
SUB2(stp34, stp25, stp33, stp22, in12, in11);
ILVRL_H2_SH(in12, in11, vec1, vec0);
SPLATI_H2_SH(coeff1, 6, 7, cnst0, cnst1);
cnst1 = __msa_ilvev_h(cnst1, cnst0);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst1);
ST_SH(in8, tmp_ptr + 192);
cnst1 = __msa_splati_h(coeff2, 3);
cnst0 = __msa_ilvev_h(cnst0, cnst1);
in8 = DOT_SHIFT_RIGHT_PCK_H(vec0, vec1, cnst0);
ST_SH(in8, tmp_ptr + 32);
}
void fdct16x8_1d_row(int16_t *input, int16_t *output) {
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
LD_SH8(input, 16, in0, in1, in2, in3, in4, in5, in6, in7);
LD_SH8((input + 8), 16, in8, in9, in10, in11, in12, in13, in14, in15);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
in8, in9, in10, in11, in12, in13, in14, in15);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
ADD4(in4, 1, in5, 1, in6, 1, in7, 1, in4, in5, in6, in7);
ADD4(in8, 1, in9, 1, in10, 1, in11, 1, in8, in9, in10, in11);
ADD4(in12, 1, in13, 1, in14, 1, in15, 1, in12, in13, in14, in15);
SRA_4V(in0, in1, in2, in3, 2);
SRA_4V(in4, in5, in6, in7, 2);
SRA_4V(in8, in9, in10, in11, 2);
SRA_4V(in12, in13, in14, in15, 2);
BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, in8, in9, in10, in11,
in12, in13, in14, in15, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5,
tmp6, tmp7, in8, in9, in10, in11, in12, in13, in14, in15);
ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, input, 16);
FDCT8x16_EVEN(tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7);
LD_SH8(input, 16, in8, in9, in10, in11, in12, in13, in14, in15);
FDCT8x16_ODD(in8, in9, in10, in11, in12, in13, in14, in15,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3,
tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3);
ST_SH8(tmp0, in0, tmp1, in1, tmp2, in2, tmp3, in3, output, 16);
TRANSPOSE8x8_SH_SH(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7,
tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7);
ST_SH8(tmp4, in4, tmp5, in5, tmp6, in6, tmp7, in7, output + 8, 16);
}
void vp9_fdct4x4_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3;
LD_SH4(input, src_stride, in0, in1, in2, in3);
/* fdct4 pre-process */
{
v8i16 vec, mask;
v16i8 zero = { 0 };
v16i8 one = __msa_ldi_b(1);
mask = (v8i16)__msa_sldi_b(zero, one, 15);
SLLI_4V(in0, in1, in2, in3, 4);
vec = __msa_ceqi_h(in0, 0);
vec = vec ^ 255;
vec = mask & vec;
in0 += vec;
}
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
VP9_FDCT4(in0, in1, in2, in3, in0, in1, in2, in3);
TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, in0, in1, in2, in3);
ADD4(in0, 1, in1, 1, in2, 1, in3, 1, in0, in1, in2, in3);
SRA_4V(in0, in1, in2, in3, 2);
PCKEV_D2_SH(in1, in0, in3, in2, in0, in2);
ST_SH2(in0, in2, output, 8);
}
void vp9_fdct8x8_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
LD_SH8(input, src_stride, in0, in1, in2, in3, in4, in5, in6, in7);
SLLI_4V(in0, in1, in2, in3, 2);
SLLI_4V(in4, in5, in6, in7, 2);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7);
ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 8);
}
void vp9_fdct16x16_msa(const int16_t *input, int16_t *output,
int32_t src_stride) {
int32_t i;
DECLARE_ALIGNED(32, int16_t, tmp_buf[16 * 16]);
/* column transform */
for (i = 0; i < 2; ++i) {
fdct8x16_1d_column((input + 8 * i), (&tmp_buf[0] + 8 * i), src_stride);
}
/* row transform */
for (i = 0; i < 2; ++i) {
fdct16x8_1d_row((&tmp_buf[0] + (128 * i)), (output + (128 * i)));
}
}

279
vpx_dsp/mips/fwd_txfm_msa.h Normal file
Просмотреть файл

@ -0,0 +1,279 @@
/*
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef VPX_DSP_MIPS_FWD_TXFM_MSA_H_
#define VPX_DSP_MIPS_FWD_TXFM_MSA_H_
#include "vpx_dsp/mips/txfm_macros_msa.h"
#include "vp9/common/vp9_idct.h"
#define VP9_FDCT4(in0, in1, in2, in3, out0, out1, out2, out3) { \
v8i16 cnst0_m, cnst1_m, cnst2_m, cnst3_m; \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m; \
v4i32 vec4_m, vec5_m, vec6_m, vec7_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, -cospi_8_64, 0, 0, 0 }; \
\
BUTTERFLY_4(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m); \
ILVR_H2_SH(vec1_m, vec0_m, vec3_m, vec2_m, vec0_m, vec2_m); \
SPLATI_H2_SH(coeff_m, 0, 1, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
vec5_m = __msa_dotp_s_w(vec0_m, cnst1_m); \
\
SPLATI_H2_SH(coeff_m, 4, 3, cnst2_m, cnst3_m); \
cnst2_m = __msa_ilvev_h(cnst3_m, cnst2_m); \
vec7_m = __msa_dotp_s_w(vec2_m, cnst2_m); \
\
vec4_m = __msa_dotp_s_w(vec0_m, cnst0_m); \
cnst2_m = __msa_splati_h(coeff_m, 2); \
cnst2_m = __msa_ilvev_h(cnst2_m, cnst3_m); \
vec6_m = __msa_dotp_s_w(vec2_m, cnst2_m); \
\
SRARI_W4_SW(vec4_m, vec5_m, vec6_m, vec7_m, DCT_CONST_BITS); \
PCKEV_H4_SH(vec4_m, vec4_m, vec5_m, vec5_m, vec6_m, vec6_m, \
vec7_m, vec7_m, out0, out2, out1, out3); \
}
#define SRLI_AVE_S_4V_H(in0, in1, in2, in3, in4, in5, in6, in7) { \
v8i16 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \
\
SRLI_H4_SH(in0, in1, in2, in3, vec0_m, vec1_m, vec2_m, vec3_m, 15); \
SRLI_H4_SH(in4, in5, in6, in7, vec4_m, vec5_m, vec6_m, vec7_m, 15); \
AVE_SH4_SH(vec0_m, in0, vec1_m, in1, vec2_m, in2, vec3_m, in3, \
in0, in1, in2, in3); \
AVE_SH4_SH(vec4_m, in4, vec5_m, in5, vec6_m, in6, vec7_m, in7, \
in4, in5, in6, in7); \
}
#define VP9_FDCT8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m; \
v8i16 s7_m, x0_m, x1_m, x2_m, x3_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, cospi_4_64, cospi_28_64, \
cospi_12_64, cospi_20_64 }; \
\
/* FDCT stage1 */ \
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m); \
BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m); \
ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m); \
ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m); \
SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x1_m, x0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m); \
x2_m = -x2_m; \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
x2_m = __msa_splati_h(coeff_m, 2); \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
/* stage2 */ \
ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m); \
\
s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
/* stage3 */ \
BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m); \
\
/* stage4 */ \
ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m); \
ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m); \
\
SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x0_m, x1_m); \
out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m); \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
\
x1_m = __msa_splati_h(coeff_m, 5); \
x0_m = -x0_m; \
x0_m = __msa_ilvev_h(x1_m, x0_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m); \
\
x2_m = __msa_splati_h(coeff_m, 6); \
x3_m = -x3_m; \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
}
#define FDCT8x16_EVEN(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m; \
v8i16 x0_m, x1_m, x2_m, x3_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, cospi_24_64, \
cospi_4_64, cospi_28_64, cospi_12_64, cospi_20_64 }; \
\
/* FDCT stage1 */ \
BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
s0_m, s1_m, s2_m, s3_m, s4_m, s5_m, s6_m, s7_m); \
BUTTERFLY_4(s0_m, s1_m, s2_m, s3_m, x0_m, x1_m, x2_m, x3_m); \
ILVL_H2_SH(x1_m, x0_m, x3_m, x2_m, s0_m, s2_m); \
ILVR_H2_SH(x1_m, x0_m, x3_m, x2_m, s1_m, s3_m); \
SPLATI_H2_SH(coeff_m, 0, 1, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x1_m, x0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, x2_m, x3_m); \
x2_m = -x2_m; \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out6 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
out0 = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
x2_m = __msa_splati_h(coeff_m, 2); \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(s2_m, s3_m, x2_m); \
\
/* stage2 */ \
ILVRL_H2_SH(s5_m, s6_m, s1_m, s0_m); \
\
s6_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x0_m); \
s5_m = DOT_SHIFT_RIGHT_PCK_H(s0_m, s1_m, x1_m); \
\
/* stage3 */ \
BUTTERFLY_4(s4_m, s7_m, s6_m, s5_m, x0_m, x3_m, x2_m, x1_m); \
\
/* stage4 */ \
ILVL_H2_SH(x3_m, x0_m, x2_m, x1_m, s4_m, s6_m); \
ILVR_H2_SH(x3_m, x0_m, x2_m, x1_m, s5_m, s7_m); \
\
SPLATI_H2_SH(coeff_m, 4, 5, x0_m, x1_m); \
x1_m = __msa_ilvev_h(x0_m, x1_m); \
out1 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x1_m); \
\
SPLATI_H2_SH(coeff_m, 6, 7, x2_m, x3_m); \
x2_m = __msa_ilvev_h(x3_m, x2_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
\
x1_m = __msa_splati_h(coeff_m, 5); \
x0_m = -x0_m; \
x0_m = __msa_ilvev_h(x1_m, x0_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(s4_m, s5_m, x0_m); \
\
x2_m = __msa_splati_h(coeff_m, 6); \
x3_m = -x3_m; \
x2_m = __msa_ilvev_h(x2_m, x3_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(s6_m, s7_m, x2_m); \
}
#define FDCT8x16_ODD(input0, input1, input2, input3, \
input4, input5, input6, input7, \
out1, out3, out5, out7, \
out9, out11, out13, out15) { \
v8i16 stp21_m, stp22_m, stp23_m, stp24_m, stp25_m, stp26_m; \
v8i16 stp30_m, stp31_m, stp32_m, stp33_m, stp34_m, stp35_m; \
v8i16 stp36_m, stp37_m, vec0_m, vec1_m; \
v8i16 vec2_m, vec3_m, vec4_m, vec5_m, vec6_m; \
v8i16 cnst0_m, cnst1_m, cnst4_m, cnst5_m; \
v8i16 coeff_m = { cospi_16_64, -cospi_16_64, cospi_8_64, \
cospi_24_64, -cospi_8_64, -cospi_24_64, \
cospi_12_64, cospi_20_64 }; \
v8i16 coeff1_m = { cospi_2_64, cospi_30_64, cospi_14_64, \
cospi_18_64, cospi_10_64, cospi_22_64, \
cospi_6_64, cospi_26_64 }; \
v8i16 coeff2_m = { -cospi_2_64, -cospi_10_64, -cospi_18_64, \
-cospi_26_64, 0, 0, 0, 0 }; \
\
/* stp 1 */ \
ILVL_H2_SH(input2, input5, input3, input4, vec2_m, vec4_m); \
ILVR_H2_SH(input2, input5, input3, input4, vec3_m, vec5_m); \
\
cnst4_m = __msa_splati_h(coeff_m, 0); \
stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst4_m); \
\
cnst5_m = __msa_splati_h(coeff_m, 1); \
cnst5_m = __msa_ilvev_h(cnst5_m, cnst4_m); \
stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst5_m); \
stp24_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst4_m); \
stp23_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst5_m); \
\
/* stp2 */ \
BUTTERFLY_4(input0, input1, stp22_m, stp23_m, \
stp30_m, stp31_m, stp32_m, stp33_m); \
BUTTERFLY_4(input7, input6, stp25_m, stp24_m, \
stp37_m, stp36_m, stp35_m, stp34_m); \
\
ILVL_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec2_m, vec4_m); \
ILVR_H2_SH(stp36_m, stp31_m, stp35_m, stp32_m, vec3_m, vec5_m); \
\
SPLATI_H2_SH(coeff_m, 2, 3, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
stp26_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff_m, 4); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
stp21_m = DOT_SHIFT_RIGHT_PCK_H(vec2_m, vec3_m, cnst1_m); \
\
SPLATI_H2_SH(coeff_m, 5, 2, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
stp25_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
\
cnst0_m = __msa_splati_h(coeff_m, 3); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
stp22_m = DOT_SHIFT_RIGHT_PCK_H(vec4_m, vec5_m, cnst1_m); \
\
/* stp4 */ \
BUTTERFLY_4(stp30_m, stp37_m, stp26_m, stp21_m, \
vec6_m, vec2_m, vec4_m, vec5_m); \
BUTTERFLY_4(stp33_m, stp34_m, stp25_m, stp22_m, \
stp21_m, stp23_m, stp24_m, stp31_m); \
\
ILVRL_H2_SH(vec2_m, vec6_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 0, 1, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
\
out1 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff2_m, 0); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
out15 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(vec4_m, vec5_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 2, 3, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
out9 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
\
cnst1_m = __msa_splati_h(coeff2_m, 2); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out7 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(stp23_m, stp21_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 4, 5, cnst0_m, cnst1_m); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
cnst0_m = __msa_splati_h(coeff2_m, 1); \
cnst0_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
out11 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_m); \
\
ILVRL_H2_SH(stp24_m, stp31_m, vec1_m, vec0_m); \
SPLATI_H2_SH(coeff1_m, 6, 7, cnst0_m, cnst1_m); \
cnst1_m = __msa_ilvev_h(cnst1_m, cnst0_m); \
\
out13 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst1_m); \
\
cnst1_m = __msa_splati_h(coeff2_m, 3); \
cnst0_m = __msa_ilvev_h(cnst0_m, cnst1_m); \
out3 = DOT_SHIFT_RIGHT_PCK_H(vec0_m, vec1_m, cnst0_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);
#endif // VPX_DSP_MIPS_FWD_TXFM_MSA_H_

2
vpx_dsp/mips/txfm_macros_msa.h
Просмотреть файл

@ -11,6 +11,8 @@
#ifndef VPX_DSP_MIPS_TXFM_MACROS_MIPS_MSA_H_
#define VPX_DSP_MIPS_TXFM_MACROS_MIPS_MSA_H_
#include "vpx_dsp/mips/macros_msa.h"
#define DOTP_CONST_PAIR(reg0, reg1, cnst0, cnst1, out0, out1) { \
v8i16 k0_m = __msa_fill_h(cnst0); \
v4i32 s0_m, s1_m, s2_m, s3_m; \

4
vpx_dsp/vpx_dsp.mk
Просмотреть файл

@ -61,6 +61,7 @@ ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
DSP_SRCS-$(HAVE_SSE2) += x86/highbd_loopfilter_sse2.c
endif # CONFIG_VP9_HIGHBITDEPTH
DSP_SRCS-$(HAVE_MSA) += mips/txfm_macros_msa.h
# forward transform
ifeq ($(CONFIG_VP9_ENCODER),yes)
DSP_SRCS-yes += fwd_txfm.c
@ -73,8 +74,9 @@ DSP_SRCS-$(HAVE_SSSE3) += x86/fwd_txfm_ssse3_x86_64.asm
endif
endif
DSP_SRCS-$(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
endif # CONFIG_VP9_ENCODER
DSP_SRCS-$(HAVE_MSA) += mips/txfm_macros_msa.h
# quantization
ifeq ($(CONFIG_VP9_ENCODER),yes)

6
vpx_dsp/vpx_dsp_rtcd_defs.pl
Просмотреть файл

@ -148,13 +148,13 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
specialize qw/vp9_highbd_fdct16x16 sse2/;
} else {
add_proto qw/void vp9_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct4x4 sse2/;
specialize qw/vp9_fdct4x4 sse2 msa/;
add_proto qw/void vp9_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct8x8 sse2 neon/, "$ssse3_x86_64_x86inc";
specialize qw/vp9_fdct8x8 sse2 neon msa/, "$ssse3_x86_64_x86inc";
add_proto qw/void vp9_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
specialize qw/vp9_fdct16x16 sse2/;
specialize qw/vp9_fdct16x16 sse2 msa/;
} # CONFIG_VP9_HIGHBITDEPTH
} # CONFIG_VP9_ENCODER