generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

mips msa vp9 idct 8x8 optimization

Browse Source 
average improvement ~4x-6x

Change-Id: I5edf713721b9e24c7e0ce2e69d8fc3ecab625d91
This commit is contained in:
Parag Salasakar 2015-05-08 12:23:27 +05:30
parent a8a9c2bb45
commit 7c5f00f868
6 changed files with 676 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
test/fdct8x8_test.cc
View File

@ -777,4 +777,18 @@ INSTANTIATE_TEST_CASE_P(
make_tuple(&vp9_fdct8x8_ssse3, &vp9_idct8x8_64_add_ssse3, 0,
VPX_BITS_8)));
#endif
#if HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
INSTANTIATE_TEST_CASE_P(
MSA, FwdTrans8x8DCT,
::testing::Values(
make_tuple(&vp9_fdct8x8_c, &vp9_idct8x8_64_add_msa, 0, VPX_BITS_8)));
INSTANTIATE_TEST_CASE_P(
MSA, FwdTrans8x8HT,
::testing::Values(
make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_msa, 0, VPX_BITS_8),
make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_msa, 1, VPX_BITS_8),
make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_msa, 2, VPX_BITS_8),
make_tuple(&vp9_fht8x8_c, &vp9_iht8x8_64_add_msa, 3, VPX_BITS_8)));
#endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
} // namespace

10
test/partial_idct_test.cc
View File

@ -324,7 +324,15 @@ INSTANTIATE_TEST_CASE_P(
make_tuple(&vp9_fdct16x16_c,
&vp9_idct16x16_256_add_c,
&vp9_idct16x16_1_add_msa,
TX_16X16, 1)));
TX_16X16, 1),
make_tuple(&vp9_fdct8x8_c,
&vp9_idct8x8_64_add_c,
&vp9_idct8x8_12_add_msa,
TX_8X8, 10),
make_tuple(&vp9_fdct8x8_c,
&vp9_idct8x8_64_add_c,
&vp9_idct8x8_1_add_msa,
TX_8X8, 1)));
#endif // HAVE_MSA && !CONFIG_VP9_HIGHBITDEPTH && !CONFIG_EMULATE_HARDWARE
} // namespace

585
vp9/common/mips/msa/vp9_idct8x8_msa.c Normal file
View File

@ -0,0 +1,585 @@
/*
* Copyright (c) 2015 The WebM project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "vp9/common/vp9_idct.h"
#include "vp9/common/mips/msa/vp9_macros_msa.h"
#define SET_COSPI_PAIR(c0_h, c1_h) ({ \
v8i16 out0, r0_m, r1_m; \
\
r0_m = __msa_fill_h(c0_h); \
r1_m = __msa_fill_h(c1_h); \
out0 = __msa_ilvev_h(r1_m, r0_m); \
\
out0; \
})
#define VP9_SET_CONST_PAIR(mask_h, idx1_h, idx2_h) ({ \
v8i16 c0_m, c1_m; \
\
c0_m = __msa_splati_h((mask_h), (idx1_h)); \
c1_m = __msa_splati_h((mask_h), (idx2_h)); \
c0_m = __msa_ilvev_h(c1_m, c0_m); \
\
c0_m; \
})
#define VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, \
in0, in1, in2, in3) { \
uint64_t out0_m, out1_m, out2_m, out3_m; \
v8i16 res0_m, res1_m, res2_m, res3_m; \
v16u8 dest0_m, dest1_m, dest2_m, dest3_m; \
v16i8 tmp0_m, tmp1_m; \
v16i8 zero_m = { 0 }; \
uint8_t *dst_m = (uint8_t *)(dest); \
\
LOAD_4VECS_UB(dst_m, (dest_stride), \
dest0_m, dest1_m, dest2_m, dest3_m); \
\
res0_m = (v8i16)__msa_ilvr_b(zero_m, (v16i8)dest0_m); \
res1_m = (v8i16)__msa_ilvr_b(zero_m, (v16i8)dest1_m); \
res2_m = (v8i16)__msa_ilvr_b(zero_m, (v16i8)dest2_m); \
res3_m = (v8i16)__msa_ilvr_b(zero_m, (v16i8)dest3_m); \
\
res0_m += (v8i16)(in0); \
res1_m += (v8i16)(in1); \
res2_m += (v8i16)(in2); \
res3_m += (v8i16)(in3); \
\
res0_m = CLIP_UNSIGNED_CHAR_H(res0_m); \
res1_m = CLIP_UNSIGNED_CHAR_H(res1_m); \
res2_m = CLIP_UNSIGNED_CHAR_H(res2_m); \
res3_m = CLIP_UNSIGNED_CHAR_H(res3_m); \
\
tmp0_m = __msa_pckev_b((v16i8)res1_m, (v16i8)res0_m); \
tmp1_m = __msa_pckev_b((v16i8)res3_m, (v16i8)res2_m); \
\
out0_m = __msa_copy_u_d((v2i64)tmp0_m, 0); \
out1_m = __msa_copy_u_d((v2i64)tmp0_m, 1); \
out2_m = __msa_copy_u_d((v2i64)tmp1_m, 0); \
out3_m = __msa_copy_u_d((v2i64)tmp1_m, 1); \
\
STORE_DWORD(dst_m, out0_m); \
dst_m += (dest_stride); \
STORE_DWORD(dst_m, out1_m); \
dst_m += (dest_stride); \
STORE_DWORD(dst_m, out2_m); \
dst_m += (dest_stride); \
STORE_DWORD(dst_m, out3_m); \
}
/* multiply and add macro */
#define VP9_MADD(inp0, inp1, inp2, inp3, \
cst0, cst1, cst2, cst3, \
out0, out1, out2, out3) { \
v8i16 madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m; \
v4i32 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
\
ILV_H_LRLR_SH(inp0, inp1, inp2, inp3, \
madd_s0_m, madd_s1_m, madd_s2_m, madd_s3_m); \
\
DOTP_S_W_4VECS_SW(madd_s1_m, cst0, madd_s0_m, cst0, \
madd_s1_m, cst1, madd_s0_m, cst1, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
\
SRARI_W_4VECS_SW(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
DCT_CONST_BITS); \
\
PCKEV_H_2VECS_SH(tmp1_m, tmp0_m, tmp3_m, tmp2_m, \
out0, out1); \
\
DOTP_S_W_4VECS_SW(madd_s3_m, cst2, madd_s2_m, cst2, \
madd_s3_m, cst3, madd_s2_m, cst3, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
\
SRARI_W_4VECS_SW(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
DCT_CONST_BITS); \
\
PCKEV_H_2VECS_SH(tmp1_m, tmp0_m, tmp3_m, tmp2_m, \
out2, out3); \
}
/* idct 8x8 macro */
#define VP9_IDCT8x8_1D_ODD(in1, in3, in5, in7, \
k0, k1, k2, k3, mask, \
out0, out1, out2, out3) { \
v8i16 res0_m, res1_m, res2_m, res3_m; \
v4i32 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
\
VP9_MADD(in1, in7, in3, in5, k0, k1, k2, k3, \
in1, in7, in3, in5); \
\
res0_m = in1 - in3; \
res1_m = in7 - in5; \
\
k0 = VP9_SET_CONST_PAIR(mask, 4, 7); \
k1 = __msa_splati_h(mask, 4); \
\
res2_m = __msa_ilvr_h(res0_m, res1_m); \
res3_m = __msa_ilvl_h(res0_m, res1_m); \
\
DOTP_S_W_4VECS_SW(res2_m, k0, res3_m, k0, \
res2_m, k1, res3_m, k1, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
\
SRARI_W_4VECS_SW(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
DCT_CONST_BITS); \
out0 = in1 + in3; \
PCKEV_H_2VECS_SH(tmp1_m, tmp0_m, tmp3_m, tmp2_m, \
out1, out2); \
out3 = in7 + in5; \
}
#define VP9_IDCT8x8_1D_EVEN(in0, in2, in4, in6, \
k0, k1, k2, k3, \
out0, out1, out2, out3) { \
k2 = SET_COSPI_PAIR(cospi_24_64, -cospi_8_64); \
k3 = SET_COSPI_PAIR(cospi_8_64, cospi_24_64); \
\
VP9_MADD(in0, in4, in2, in6, k1, k0, k2, k3, \
in0, in4, in2, in6); \
\
out0 = in0 + in6; \
out1 = in4 + in2; \
out2 = in4 - in2; \
out3 = in0 - in6; \
}
#define VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 res0_m, res1_m, res2_m, res3_m, res4_m, res5_m, res6_m, res7_m; \
v8i16 k0_m, k1_m, k2_m, k3_m; \
v8i16 mask_m = { cospi_28_64, cospi_4_64, cospi_20_64, cospi_12_64, \
cospi_16_64, -cospi_4_64, -cospi_20_64, -cospi_16_64 \
}; \
\
k0_m = VP9_SET_CONST_PAIR(mask_m, 0, 5); \
k1_m = VP9_SET_CONST_PAIR(mask_m, 1, 0); \
k2_m = VP9_SET_CONST_PAIR(mask_m, 6, 3); \
k3_m = VP9_SET_CONST_PAIR(mask_m, 3, 2); \
\
VP9_IDCT8x8_1D_ODD(in1, in3, in5, in7, k0_m, k1_m, k2_m, k3_m, mask_m, \
res4_m, res5_m, res6_m, res7_m); \
\
VP9_IDCT8x8_1D_EVEN(in0, in2, in4, in6, k0_m, k1_m, k2_m, k3_m, \
res0_m, res1_m, res2_m, res3_m); \
\
BUTTERFLY_8(res0_m, res1_m, res2_m, res3_m, \
res4_m, res5_m, res6_m, res7_m, \
out0, out1, out2, out3, \
out4, out5, out6, out7); \
}
#define DOT_ADD_SUB_SRARI_PCK(in0, in1, in2, in3, in4, in5, in6, in7, \
dst0, dst1, dst2, dst3) { \
v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8, tmp9; \
\
tmp0 = __msa_dotp_s_w((in0), (in4)); \
tmp2 = __msa_dotp_s_w((in1), (in4)); \
tmp3 = __msa_dotp_s_w((in0), (in5)); \
tmp4 = __msa_dotp_s_w((in1), (in5)); \
tmp5 = __msa_dotp_s_w((in2), (in6)); \
tmp6 = __msa_dotp_s_w((in3), (in6)); \
tmp7 = __msa_dotp_s_w((in2), (in7)); \
tmp8 = __msa_dotp_s_w((in3), (in7)); \
\
BUTTERFLY_4(tmp0, tmp3, tmp7, tmp5, tmp1, tmp9, tmp7, tmp5); \
BUTTERFLY_4(tmp2, tmp4, tmp8, tmp6, tmp3, tmp0, tmp4, tmp2); \
\
SRARI_W_4VECS_SW(tmp1, tmp9, tmp7, tmp5, tmp1, tmp9, tmp7, tmp5, \
DCT_CONST_BITS); \
SRARI_W_4VECS_SW(tmp3, tmp0, tmp4, tmp2, tmp3, tmp0, tmp4, tmp2, \
DCT_CONST_BITS); \
\
PCKEV_H_4VECS_SH(tmp1, tmp3, tmp9, tmp0, tmp7, tmp4, tmp5, tmp2, \
dst0, dst1, dst2, dst3); \
}
#define DOT_SHIFT_RIGHT_PCK_H(in0, in1, in2) ({ \
v8i16 dst_m; \
v4i32 tp0_m, tp1_m; \
\
tp1_m = __msa_dotp_s_w((in0), (in2)); \
tp0_m = __msa_dotp_s_w((in1), (in2)); \
tp1_m = __msa_srari_w(tp1_m, DCT_CONST_BITS); \
tp0_m = __msa_srari_w(tp0_m, DCT_CONST_BITS); \
dst_m = __msa_pckev_h((v8i16)tp1_m, (v8i16)tp0_m); \
\
dst_m; \
})
#define VP9_ADST8_ROW(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 const0_m, const1_m, const2_m, const3_m, const4_m; \
v8i16 temp0_m, temp1_m, temp2_m, temp3_m, s0_m, s1_m; \
v8i16 coeff0_m = { cospi_2_64, cospi_6_64, cospi_10_64, \
cospi_14_64, cospi_18_64, cospi_22_64, cospi_26_64, cospi_30_64 \
}; \
v8i16 coeff1_m = { cospi_8_64, -cospi_8_64, cospi_16_64, \
-cospi_16_64, cospi_24_64, -cospi_24_64, 0, 0 \
}; \
\
const0_m = __msa_splati_h(coeff0_m, 0); \
const1_m = __msa_splati_h(coeff0_m, 7); \
const2_m = -const0_m; \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = __msa_ilvev_h(const2_m, const1_m); \
const2_m = __msa_splati_h(coeff0_m, 4); \
const3_m = __msa_splati_h(coeff0_m, 3); \
const4_m = -const2_m; \
const2_m = __msa_ilvev_h(const3_m, const2_m); \
const3_m = __msa_ilvev_h(const4_m, const3_m); \
\
ILV_H_LRLR_SH(in7, in0, in3, in4, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const1_m, const2_m, const3_m, \
in7, in0, in4, in3); \
\
const0_m = __msa_splati_h(coeff0_m, 2); \
const1_m = __msa_splati_h(coeff0_m, 5); \
const2_m = -const0_m; \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = __msa_ilvev_h(const2_m, const1_m); \
const2_m = __msa_splati_h(coeff0_m, 6); \
const3_m = __msa_splati_h(coeff0_m, 1); \
const4_m = -const2_m; \
const2_m = __msa_ilvev_h(const3_m, const2_m); \
const3_m = __msa_ilvev_h(const4_m, const3_m); \
\
ILV_H_LRLR_SH(in5, in2, in1, in6, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const1_m, const2_m, const3_m, \
in5, in2, in6, in1); \
\
BUTTERFLY_4(in7, in0, in2, in5, s1_m, s0_m, in2, in5); \
out7 = -s0_m; \
out0 = s1_m; \
\
SPLATI_H_4VECS_SH(coeff1_m, 0, 4, 1, 5, \
const0_m, const1_m, const2_m, const3_m); \
\
const3_m = __msa_ilvev_h(const0_m, const3_m); \
const2_m = __msa_ilvev_h(const2_m, const1_m); \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = const0_m; \
\
ILV_H_LRLR_SH(in3, in4, in1, in6, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const2_m, const3_m, const1_m, \
out1, out6, s0_m, s1_m); \
\
const0_m = __msa_splati_h(coeff1_m, 2); \
const1_m = __msa_splati_h(coeff1_m, 3); \
const1_m = __msa_ilvev_h(const1_m, const0_m); \
\
ILV_H_LRLR_SH(in5, in2, s1_m, s0_m, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
out3 = DOT_SHIFT_RIGHT_PCK_H(temp0_m, temp1_m, const0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(temp0_m, temp1_m, const1_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(temp2_m, temp3_m, const0_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(temp2_m, temp3_m, const1_m); \
\
out1 = -out1; \
out3 = -out3; \
out5 = -out5; \
}
#define VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7, \
out0, out1, out2, out3, out4, out5, out6, out7) { \
v8i16 const0_m, const1_m, const2_m, const3_m, const4_m; \
v8i16 temp0_m, temp1_m, temp2_m, temp3_m, s0_m, s1_m; \
\
const0_m = __msa_fill_h(cospi_2_64); \
const1_m = __msa_fill_h(cospi_30_64); \
const2_m = -const0_m; \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = __msa_ilvev_h(const2_m, const1_m); \
const2_m = __msa_fill_h(cospi_18_64); \
const3_m = __msa_fill_h(cospi_14_64); \
const4_m = -const2_m; \
const2_m = __msa_ilvev_h(const3_m, const2_m); \
const3_m = __msa_ilvev_h(const4_m, const3_m); \
\
ILV_H_LRLR_SH(in7, in0, in3, in4, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const1_m, const2_m, const3_m, \
in7, in0, in4, in3); \
\
const0_m = __msa_fill_h(cospi_10_64); \
const1_m = __msa_fill_h(cospi_22_64); \
const2_m = -const0_m; \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = __msa_ilvev_h(const2_m, const1_m); \
const2_m = __msa_fill_h(cospi_26_64); \
const3_m = __msa_fill_h(cospi_6_64); \
const4_m = -const2_m; \
const2_m = __msa_ilvev_h(const3_m, const2_m); \
const3_m = __msa_ilvev_h(const4_m, const3_m); \
\
ILV_H_LRLR_SH(in5, in2, in1, in6, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const1_m, const2_m, const3_m, \
in5, in2, in6, in1); \
\
BUTTERFLY_4(in7, in0, in2, in5, s1_m, s0_m, in2, in5); \
out7 = -s0_m; \
out0 = s1_m; \
\
const1_m = __msa_fill_h(cospi_24_64); \
const0_m = __msa_fill_h(cospi_8_64); \
const3_m = -const1_m; \
const2_m = -const0_m; \
\
const3_m = __msa_ilvev_h(const0_m, const3_m); \
const2_m = __msa_ilvev_h(const2_m, const1_m); \
const0_m = __msa_ilvev_h(const1_m, const0_m); \
const1_m = const0_m; \
\
ILV_H_LRLR_SH(in3, in4, in1, in6, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
DOT_ADD_SUB_SRARI_PCK(temp0_m, temp1_m, temp2_m, temp3_m, \
const0_m, const2_m, const3_m, const1_m, \
out1, out6, s0_m, s1_m); \
\
const0_m = __msa_fill_h(cospi_16_64); \
const1_m = -const0_m; \
const1_m = __msa_ilvev_h(const1_m, const0_m); \
\
ILV_H_LRLR_SH(in5, in2, s1_m, s0_m, \
temp0_m, temp1_m, temp2_m, temp3_m); \
\
out3 = DOT_SHIFT_RIGHT_PCK_H(temp0_m, temp1_m, const0_m); \
out4 = DOT_SHIFT_RIGHT_PCK_H(temp0_m, temp1_m, const1_m); \
out2 = DOT_SHIFT_RIGHT_PCK_H(temp2_m, temp3_m, const0_m); \
out5 = DOT_SHIFT_RIGHT_PCK_H(temp2_m, temp3_m, const1_m); \
\
out1 = -out1; \
out3 = -out3; \
out5 = -out5; \
}
void vp9_idct8x8_64_add_msa(const int16_t *input, uint8_t *dest,
int32_t dest_stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* load vector elements of 8x8 block */
LOAD_8VECS_SH(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
/* rows transform */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* 1D idct8x8 */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* columns transform */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* 1D idct8x8 */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H_4VECS_SH(in0, in1, in2, in3, in0, in1, in2, in3, 5);
SRARI_H_4VECS_SH(in4, in5, in6, in7, in4, in5, in6, in7, 5);
/* add block and store 8x8 */
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in0, in1, in2, in3);
dest += (4 * dest_stride);
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in4, in5, in6, in7);
}
void vp9_idct8x8_12_add_msa(const int16_t *input, uint8_t *dest,
int32_t dest_stride) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
v8i16 s0, s1, s2, s3, s4, s5, s6, s7;
v8i16 k0, k1, k2, k3, m0, m1, m2, m3;
v4i32 tmp0, tmp1, tmp2, tmp3;
v8i16 zero = { 0 };
/* load vector elements of 8x8 block */
LOAD_8VECS_SH(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8X4_H(in0, in1, in2, in3, in0, in1, in2, in3);
/* stage1 */
s0 = __msa_ilvl_h(in3, in0);
s1 = __msa_ilvl_h(in2, in1);
k0 = SET_COSPI_PAIR(cospi_28_64, -cospi_4_64);
k1 = SET_COSPI_PAIR(cospi_4_64, cospi_28_64);
k2 = SET_COSPI_PAIR(-cospi_20_64, cospi_12_64);
k3 = SET_COSPI_PAIR(cospi_12_64, cospi_20_64);
DOTP_S_W_4VECS_SW(s0, k0, s0, k1, s1, k2, s1, k3, tmp0, tmp1, tmp2, tmp3);
SRARI_W_4VECS_SW(tmp0, tmp1, tmp2, tmp3,
tmp0, tmp1, tmp2, tmp3, DCT_CONST_BITS);
PCKEV_H_2VECS_SH(zero, tmp0, zero, tmp1, s0, s1);
PCKEV_H_2VECS_SH(zero, tmp2, zero, tmp3, s2, s3);
BUTTERFLY_4(s0, s1, s3, s2, s4, s7, s6, s5);
/* stage2 */
s0 = __msa_ilvr_h(in2, in0);
s1 = __msa_ilvr_h(in3, in1);
k0 = SET_COSPI_PAIR(cospi_16_64, cospi_16_64);
k1 = SET_COSPI_PAIR(cospi_16_64, -cospi_16_64);
k2 = SET_COSPI_PAIR(cospi_24_64, -cospi_8_64);
k3 = SET_COSPI_PAIR(cospi_8_64, cospi_24_64);
DOTP_S_W_4VECS_SW(s0, k0, s0, k1, s1, k2, s1, k3, tmp0, tmp1, tmp2, tmp3);
SRARI_W_4VECS_SW(tmp0, tmp1, tmp2, tmp3,
tmp0, tmp1, tmp2, tmp3, DCT_CONST_BITS);
PCKEV_H_2VECS_SH(zero, tmp0, zero, tmp1, s0, s1);
PCKEV_H_2VECS_SH(zero, tmp2, zero, tmp3, s2, s3);
BUTTERFLY_4(s0, s1, s2, s3, m0, m1, m2, m3);
/* stage3 */
s0 = __msa_ilvr_h(s6, s5);
k1 = SET_COSPI_PAIR(-cospi_16_64, cospi_16_64);
tmp0 = __msa_dotp_s_w(s0, k1);
tmp1 = __msa_dotp_s_w(s0, k0);
tmp0 = __msa_srari_w(tmp0, DCT_CONST_BITS);
tmp1 = __msa_srari_w(tmp1, DCT_CONST_BITS);
PCKEV_H_2VECS_SH(zero, tmp0, zero, tmp1, s2, s3);
/* stage4 */
BUTTERFLY_8(m0, m1, m2, m3, s4, s2, s3, s7,
in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE4X8_H(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H_4VECS_SH(in0, in1, in2, in3, in0, in1, in2, in3, 5);
SRARI_H_4VECS_SH(in4, in5, in6, in7, in4, in5, in6, in7, 5);
/* add block and store 8x8 */
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in0, in1, in2, in3);
dest += (4 * dest_stride);
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in4, in5, in6, in7);
}
void vp9_idct8x8_1_add_msa(const int16_t *input, uint8_t *dest,
int32_t dest_stride) {
int16_t out;
int32_t const1;
v8i16 const2;
out = dct_const_round_shift(input[0] * cospi_16_64);
out = dct_const_round_shift(out * cospi_16_64);
const1 = ROUND_POWER_OF_TWO(out, 5);
const2 = __msa_fill_h(const1);
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride,
const2, const2, const2, const2);
dest += (4 * dest_stride);
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride,
const2, const2, const2, const2);
}
void vp9_iht8x8_64_add_msa(const int16_t *input, uint8_t *dest,
int32_t dest_stride, int32_t tx_type) {
v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
/* load vector elements of 8x8 block */
LOAD_8VECS_SH(input, 8, in0, in1, in2, in3, in4, in5, in6, in7);
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
switch (tx_type) {
case DCT_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_DCT:
/* DCT in horizontal */
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case DCT_ADST:
/* ADST in horizontal */
VP9_ADST8_ROW(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* DCT in vertical */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_IDCT8x8_1D(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
case ADST_ADST:
/* ADST in horizontal */
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
/* ADST in vertical */
TRANSPOSE8x8_H_SH(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
VP9_ADST8(in0, in1, in2, in3, in4, in5, in6, in7,
in0, in1, in2, in3, in4, in5, in6, in7);
break;
default:
assert(0);
break;
}
/* final rounding (add 2^4, divide by 2^5) and shift */
SRARI_H_4VECS_SH(in0, in1, in2, in3, in0, in1, in2, in3, 5);
SRARI_H_4VECS_SH(in4, in5, in6, in7, in4, in5, in6, in7, 5);
/* add block and store 8x8 */
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in0, in1, in2, in3);
dest += (4 * dest_stride);
VP9_ADDBLK_CLIP_AND_STORE_8_BYTES_4(dest, dest_stride, in4, in5, in6, in7);
}

63
vp9/common/mips/msa/vp9_macros_msa.h
View File

@ -388,6 +388,44 @@
out_m; \
})
#define TRANSPOSE4X8_H(in0, in1, in2, in3, \
in4, in5, in6, in7, \
out0, out1, out2, out3, \
out4, out5, out6, out7) { \
v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
v8i16 tmp0_n, tmp1_n, tmp2_n, tmp3_n; \
v8i16 zero_m = { 0 }; \
\
tmp0_n = __msa_ilvr_h((v8i16)(in1), (v8i16)(in0)); \
tmp1_n = __msa_ilvr_h((v8i16)(in3), (v8i16)(in2)); \
tmp2_n = __msa_ilvr_h((v8i16)(in5), (v8i16)(in4)); \
tmp3_n = __msa_ilvr_h((v8i16)(in7), (v8i16)(in6)); \
\
ILV_W_LRLR_SH((tmp0_n), (tmp1_n), (tmp2_n), (tmp3_n), \
tmp2_m, tmp0_m, tmp3_m, tmp1_m); \
\
out1 = (v8i16)__msa_ilvl_d((v2i64)tmp1_m, (v2i64)tmp0_m); \
out0 = (v8i16)__msa_ilvr_d((v2i64)tmp1_m, (v2i64)tmp0_m); \
out3 = (v8i16)__msa_ilvl_d((v2i64)tmp3_m, (v2i64)tmp2_m); \
out2 = (v8i16)__msa_ilvr_d((v2i64)tmp3_m, (v2i64)tmp2_m); \
\
out4 = zero_m; \
out5 = zero_m; \
out6 = zero_m; \
out7 = zero_m; \
}
#define TRANSPOSE8X4_H(in0, in1, in2, in3, \
out0, out1, out2, out3) { \
v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
\
ILV_H_LRLR_SH((in0), (in1), (in2), (in3), \
tmp2_m, tmp0_m, tmp3_m, tmp1_m); \
\
ILV_W_LRLR_SH(tmp0_m, tmp1_m, tmp2_m, tmp3_m, \
out1, out0, out3, out2); \
}
/* halfword 8x8 transpose macro */
#define TRANSPOSE8x8_H_SH(in0, in1, in2, in3, \
in4, in5, in6, in7, \
@ -445,6 +483,14 @@
out3 = __msa_ilvr_h((v8i16)(in3), (v8i16)(in2)); \
}
#define ILV_W_LRLR_SH(in0, in1, in2, in3, \
out0, out1, out2, out3) { \
out0 = (v8i16)__msa_ilvl_w((v4i32)(in1), (v4i32)(in0)); \
out1 = (v8i16)__msa_ilvr_w((v4i32)(in1), (v4i32)(in0)); \
out2 = (v8i16)__msa_ilvl_w((v4i32)(in3), (v4i32)(in2)); \
out3 = (v8i16)__msa_ilvr_w((v4i32)(in3), (v4i32)(in2)); \
}
#define ILV_H_LR_SH(in0, in1, out0, out1) { \
out0 = __msa_ilvl_h((v8i16)(in1), (v8i16)(in0)); \
out1 = __msa_ilvr_h((v8i16)(in1), (v8i16)(in0)); \
@ -572,12 +618,29 @@
out3 = __msa_dotp_s_w((v8i16)(m3), (v8i16)(c3)); \
}
#define SPLATI_H_4VECS_SH(coeff, val0, val1, val2, val3, \
out0, out1, out2, out3) { \
out0 = __msa_splati_h((v8i16)(coeff), (val0)); \
out1 = __msa_splati_h((v8i16)(coeff), (val1)); \
out2 = __msa_splati_h((v8i16)(coeff), (val2)); \
out3 = __msa_splati_h((v8i16)(coeff), (val3)); \
}
#define PCKEV_H_2VECS_SH(in0_l, in0_r, in1_l, in1_r, \
out0, out1) { \
out0 = __msa_pckev_h((v8i16)(in0_l), (v8i16)(in0_r)); \
out1 = __msa_pckev_h((v8i16)(in1_l), (v8i16)(in1_r)); \
}
#define PCKEV_H_4VECS_SH(in0_l, in0_r, in1_l, in1_r, \
in2_l, in2_r, in3_l, in3_r, \
out0, out1, out2, out3) { \
out0 = __msa_pckev_h((v8i16)(in0_l), (v8i16)(in0_r)); \
out1 = __msa_pckev_h((v8i16)(in1_l), (v8i16)(in1_r)); \
out2 = __msa_pckev_h((v8i16)(in2_l), (v8i16)(in2_r)); \
out3 = __msa_pckev_h((v8i16)(in3_l), (v8i16)(in3_r)); \
}
#define XORI_B_2VECS_UB(val0, val1, \
out0, out1, xor_val) { \
out0 = __msa_xori_b((v16u8)(val0), (xor_val)); \

8
vp9/common/vp9_rtcd_defs.pl
View File

@ -425,13 +425,13 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
specialize qw/vp9_idct4x4_16_add sse2 neon dspr2/;
add_proto qw/void vp9_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vp9_idct8x8_1_add sse2 neon dspr2/;
specialize qw/vp9_idct8x8_1_add sse2 neon dspr2 msa/;
add_proto qw/void vp9_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vp9_idct8x8_64_add sse2 neon dspr2/, "$ssse3_x86_64";
specialize qw/vp9_idct8x8_64_add sse2 neon dspr2 msa/, "$ssse3_x86_64";
add_proto qw/void vp9_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vp9_idct8x8_12_add sse2 neon dspr2/, "$ssse3_x86_64";
specialize qw/vp9_idct8x8_12_add sse2 neon dspr2 msa/, "$ssse3_x86_64";
add_proto qw/void vp9_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
specialize qw/vp9_idct16x16_1_add sse2 neon dspr2 msa/;
@ -457,7 +457,7 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
specialize qw/vp9_iht4x4_16_add sse2 neon dspr2/;
add_proto qw/void vp9_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
specialize qw/vp9_iht8x8_64_add sse2 neon dspr2/;
specialize qw/vp9_iht8x8_64_add sse2 neon dspr2 msa/;
add_proto qw/void vp9_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
specialize qw/vp9_iht16x16_256_add sse2 dspr2 msa/;

1
vp9/vp9_common.mk
View File

@ -137,6 +137,7 @@ VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve8_vert_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_avg_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_copy_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_convolve_msa.h
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct8x8_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct16x16_msa.c
VP9_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/vp9_idct32x32_msa.c