Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

avcodec/mips: MSA (MIPS-SIMD-Arch) optimizations for VP8 functions

Browse Source 
Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
pull/143/merge
Shivraj Patil 10 years ago committed by Ronald S. Bultje
parent 63c442e3b1
commit dec16372df
8 changed files with 3474 additions and 0 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 4
      libavcodec/mips/Makefile
  2. 160
      libavcodec/mips/vp8_idct_msa.c
  3. 690
      libavcodec/mips/vp8_lpf_msa.c
  4. 2332
      libavcodec/mips/vp8_mc_msa.c
  5. 113
      libavcodec/mips/vp8dsp_init_mips.c
  6. 172
      libavcodec/mips/vp8dsp_mips.h
  7. 2
      libavcodec/vp8dsp.c
  8. 1
      libavcodec/vp8dsp.h

4
libavcodec/mips/Makefile
Unescape View File

@ -21,6 +21,7 @@ MIPSFPU-OBJS-$(CONFIG_AAC_ENCODER) += mips/iirfilter_mips.o
OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_init_mips.o \ OBJS-$(CONFIG_HEVC_DECODER) += mips/hevcdsp_init_mips.o \
mips/hevcpred_init_mips.o mips/hevcpred_init_mips.o
OBJS-$(CONFIG_VP9_DECODER) += mips/vp9dsp_init_mips.o OBJS-$(CONFIG_VP9_DECODER) += mips/vp9dsp_init_mips.o
OBJS-$(CONFIG_VP8_DECODER) += mips/vp8dsp_init_mips.o
OBJS-$(CONFIG_H264DSP) += mips/h264dsp_init_mips.o OBJS-$(CONFIG_H264DSP) += mips/h264dsp_init_mips.o
OBJS-$(CONFIG_H264QPEL) += mips/h264qpel_init_mips.o OBJS-$(CONFIG_H264QPEL) += mips/h264qpel_init_mips.o
OBJS-$(CONFIG_H264CHROMA) += mips/h264chroma_init_mips.o OBJS-$(CONFIG_H264CHROMA) += mips/h264chroma_init_mips.o
@ -47,6 +48,9 @@ MSA-OBJS-$(CONFIG_VP9_DECODER) += mips/vp9_mc_msa.o \
mips/vp9_lpf_msa.o \ mips/vp9_lpf_msa.o \
mips/vp9_idct_msa.o \ mips/vp9_idct_msa.o \
mips/vp9_intra_msa.o mips/vp9_intra_msa.o
MSA-OBJS-$(CONFIG_VP8_DECODER) += mips/vp8_mc_msa.o \
mips/vp8_idct_msa.o \
mips/vp8_lpf_msa.o
MSA-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_msa.o \ MSA-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_msa.o \
mips/h264idct_msa.o mips/h264idct_msa.o
MSA-OBJS-$(CONFIG_H264QPEL) += mips/h264qpel_msa.o MSA-OBJS-$(CONFIG_H264QPEL) += mips/h264qpel_msa.o

160
libavcodec/mips/vp8_idct_msa.c
Unescape View File

@ -0,0 +1,160 @@
/*
* Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <string.h>
#include "libavcodec/vp8dsp.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp8dsp_mips.h"
static const int cospi8sqrt2minus1 = 20091;
static const int sinpi8sqrt2 = 35468;
#define VP8_IDCT_1D_W(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
v4i32 a1_m, b1_m, c1_m, d1_m; \
v4i32 c_tmp1_m, c_tmp2_m, d_tmp1_m, d_tmp2_m; \
v4i32 const_cospi8sqrt2minus1_m, sinpi8_sqrt2_m; \
\
const_cospi8sqrt2minus1_m = __msa_fill_w(cospi8sqrt2minus1); \
sinpi8_sqrt2_m = __msa_fill_w(sinpi8sqrt2); \
a1_m = in0 + in2; \
b1_m = in0 - in2; \
c_tmp1_m = ((in1) * sinpi8_sqrt2_m) >> 16; \
c_tmp2_m = in3 + (((in3) * const_cospi8sqrt2minus1_m) >> 16); \
c1_m = c_tmp1_m - c_tmp2_m; \
d_tmp1_m = (in1) + (((in1) * const_cospi8sqrt2minus1_m) >> 16); \
d_tmp2_m = ((in3) * sinpi8_sqrt2_m) >> 16; \
d1_m = d_tmp1_m + d_tmp2_m; \
BUTTERFLY_4(a1_m, b1_m, c1_m, d1_m, out0, out1, out2, out3); \
}
void ff_vp8_idct_add_msa(uint8_t *dst, int16_t input[16], ptrdiff_t stride)
{
v8i16 input0, input1;
v4i32 in0, in1, in2, in3, hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3;
v4i32 res0, res1, res2, res3;
v16i8 zero = { 0 };
v16i8 pred0, pred1, pred2, pred3, dest0, dest1;
v16i8 mask = { 0, 4, 8, 12, 16, 20, 24, 28, 0, 0, 0, 0, 0, 0, 0, 0 };
/* load short vector elements of 4x4 block */
LD_SH2(input, 8, input0, input1);
UNPCK_SH_SW(input0, in0, in1);
UNPCK_SH_SW(input1, in2, in3);
VP8_IDCT_1D_W(in0, in1, in2, in3, hz0, hz1, hz2, hz3);
/* transpose the block */
TRANSPOSE4x4_SW_SW(hz0, hz1, hz2, hz3, hz0, hz1, hz2, hz3);
VP8_IDCT_1D_W(hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3);
SRARI_W4_SW(vt0, vt1, vt2, vt3, 3);
/* transpose the block */
TRANSPOSE4x4_SW_SW(vt0, vt1, vt2, vt3, vt0, vt1, vt2, vt3);
LD_SB4(dst, stride, pred0, pred1, pred2, pred3);
ILVR_B4_SW(zero, pred0, zero, pred1, zero, pred2, zero, pred3,
res0, res1, res2, res3);
ILVR_H4_SW(zero, res0, zero, res1, zero, res2, zero, res3,
res0, res1, res2, res3);
ADD4(res0, vt0, res1, vt1, res2, vt2, res3, vt3, res0, res1, res2, res3);
res0 = CLIP_SW_0_255(res0);
res1 = CLIP_SW_0_255(res1);
res2 = CLIP_SW_0_255(res2);
res3 = CLIP_SW_0_255(res3);
VSHF_B2_SB(res0, res1, res2, res3, mask, mask, dest0, dest1);
ST4x4_UB(dest0, dest1, 0, 1, 0, 1, dst, stride);
memset(input, 0, 4 * 4 * sizeof(*input));
}
void ff_vp8_idct_dc_add_msa(uint8_t *dst, int16_t in_dc[16], ptrdiff_t stride)
{
v8i16 vec;
v8i16 res0, res1, res2, res3;
v16i8 zero = { 0 };
v16i8 pred0, pred1, pred2, pred3, dest0, dest1;
v16i8 mask = { 0, 2, 4, 6, 16, 18, 20, 22, 0, 0, 0, 0, 0, 0, 0, 0 };
vec = __msa_fill_h(in_dc[0]);
vec = __msa_srari_h(vec, 3);
LD_SB4(dst, stride, pred0, pred1, pred2, pred3);
ILVR_B4_SH(zero, pred0, zero, pred1, zero, pred2, zero, pred3,
res0, res1, res2, res3);
ADD4(res0, vec, res1, vec, res2, vec, res3, vec, res0, res1, res2, res3);
CLIP_SH4_0_255(res0, res1, res2, res3);
VSHF_B2_SB(res0, res1, res2, res3, mask, mask, dest0, dest1);
ST4x4_UB(dest0, dest1, 0, 1, 0, 1, dst, stride);
in_dc[0] = 0;
}
void ff_vp8_luma_dc_wht_msa(int16_t block[4][4][16], int16_t input[16])
{
int16_t *mb_dq_coeff = &block[0][0][0];
v8i16 input0, input1;
v4i32 in0, in1, in2, in3, a1, b1, c1, d1;
v4i32 hz0, hz1, hz2, hz3, vt0, vt1, vt2, vt3;
/* load short vector elements of 4x4 block */
LD_SH2(input, 8, input0, input1);
UNPCK_SH_SW(input0, in0, in1);
UNPCK_SH_SW(input1, in2, in3);
BUTTERFLY_4(in0, in1, in2, in3, a1, b1, c1, d1);
BUTTERFLY_4(a1, d1, c1, b1, hz0, hz1, hz3, hz2);
/* transpose the block */
TRANSPOSE4x4_SW_SW(hz0, hz1, hz2, hz3, hz0, hz1, hz2, hz3);
BUTTERFLY_4(hz0, hz1, hz2, hz3, a1, b1, c1, d1);
BUTTERFLY_4(a1, d1, c1, b1, vt0, vt1, vt3, vt2);
ADD4(vt0, 3, vt1, 3, vt2, 3, vt3, 3, vt0, vt1, vt2, vt3);
SRA_4V(vt0, vt1, vt2, vt3, 3);
mb_dq_coeff[0] = __msa_copy_s_h((v8i16) vt0, 0);
mb_dq_coeff[16] = __msa_copy_s_h((v8i16) vt1, 0);
mb_dq_coeff[32] = __msa_copy_s_h((v8i16) vt2, 0);
mb_dq_coeff[48] = __msa_copy_s_h((v8i16) vt3, 0);
mb_dq_coeff[64] = __msa_copy_s_h((v8i16) vt0, 2);
mb_dq_coeff[80] = __msa_copy_s_h((v8i16) vt1, 2);
mb_dq_coeff[96] = __msa_copy_s_h((v8i16) vt2, 2);
mb_dq_coeff[112] = __msa_copy_s_h((v8i16) vt3, 2);
mb_dq_coeff[128] = __msa_copy_s_h((v8i16) vt0, 4);
mb_dq_coeff[144] = __msa_copy_s_h((v8i16) vt1, 4);
mb_dq_coeff[160] = __msa_copy_s_h((v8i16) vt2, 4);
mb_dq_coeff[176] = __msa_copy_s_h((v8i16) vt3, 4);
mb_dq_coeff[192] = __msa_copy_s_h((v8i16) vt0, 6);
mb_dq_coeff[208] = __msa_copy_s_h((v8i16) vt1, 6);
mb_dq_coeff[224] = __msa_copy_s_h((v8i16) vt2, 6);
mb_dq_coeff[240] = __msa_copy_s_h((v8i16) vt3, 6);
memset(input, 0, 4 * 4 * sizeof(int16_t));
}
void ff_vp8_idct_dc_add4y_msa(uint8_t *dst, int16_t block[4][16],
ptrdiff_t stride)
{
ff_vp8_idct_dc_add_msa(dst, &block[0][0], stride);
ff_vp8_idct_dc_add_msa(dst + 4, &block[1][0], stride);
ff_vp8_idct_dc_add_msa(dst + 8, &block[2][0], stride);
ff_vp8_idct_dc_add_msa(dst + 12, &block[3][0], stride);
}
void ff_vp8_idct_dc_add4uv_msa(uint8_t *dst, int16_t block[4][16],
ptrdiff_t stride)
{
ff_vp8_idct_dc_add_msa(dst, &block[0][0], stride);
ff_vp8_idct_dc_add_msa(dst + 4, &block[1][0], stride);
ff_vp8_idct_dc_add_msa(dst + stride * 4, &block[2][0], stride);
ff_vp8_idct_dc_add_msa(dst + stride * 4 + 4, &block[3][0], stride);
}

690
libavcodec/mips/vp8_lpf_msa.c
Unescape View File

@ -0,0 +1,690 @@
/*
* Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavcodec/vp8dsp.h"
#include "libavutil/mips/generic_macros_msa.h"
#include "vp8dsp_mips.h"
#define VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask) \
{ \
v16u8 p1_a_sub_q1, p0_a_sub_q0; \
\
p0_a_sub_q0 = __msa_asub_u_b(p0, q0); \
p1_a_sub_q1 = __msa_asub_u_b(p1, q1); \
p1_a_sub_q1 = (v16u8) __msa_srli_b((v16i8) p1_a_sub_q1, 1); \
p0_a_sub_q0 = __msa_adds_u_b(p0_a_sub_q0, p0_a_sub_q0); \
mask = __msa_adds_u_b(p0_a_sub_q0, p1_a_sub_q1); \
mask = ((v16u8) mask <= b_limit); \
}
#define VP8_LPF_FILTER4_4W(p1_in_out, p0_in_out, q0_in_out, q1_in_out, \
mask_in, hev_in) \
{ \
v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, filt_sign; \
v16i8 filt, filt1, filt2, cnst4b, cnst3b; \
v8i16 q0_sub_p0_r, q0_sub_p0_l, filt_l, filt_r, cnst3h; \
\
p1_m = (v16i8) __msa_xori_b(p1_in_out, 0x80); \
p0_m = (v16i8) __msa_xori_b(p0_in_out, 0x80); \
q0_m = (v16i8) __msa_xori_b(q0_in_out, 0x80); \
q1_m = (v16i8) __msa_xori_b(q1_in_out, 0x80); \
\
filt = __msa_subs_s_b(p1_m, q1_m); \
\
filt = filt & (v16i8) hev_in; \
\
q0_sub_p0 = q0_m - p0_m; \
filt_sign = __msa_clti_s_b(filt, 0); \
\
cnst3h = __msa_ldi_h(3); \
q0_sub_p0_r = (v8i16) __msa_ilvr_b(q0_sub_p0, q0_sub_p0); \
q0_sub_p0_r = __msa_dotp_s_h((v16i8) q0_sub_p0_r, (v16i8) cnst3h); \
filt_r = (v8i16) __msa_ilvr_b(filt_sign, filt); \
filt_r += q0_sub_p0_r; \
filt_r = __msa_sat_s_h(filt_r, 7); \
\
q0_sub_p0_l = (v8i16) __msa_ilvl_b(q0_sub_p0, q0_sub_p0); \
q0_sub_p0_l = __msa_dotp_s_h((v16i8) q0_sub_p0_l, (v16i8) cnst3h); \
filt_l = (v8i16) __msa_ilvl_b(filt_sign, filt); \
filt_l += q0_sub_p0_l; \
filt_l = __msa_sat_s_h(filt_l, 7); \
\
filt = __msa_pckev_b((v16i8) filt_l, (v16i8) filt_r); \
filt = filt & (v16i8) mask_in; \
\
cnst4b = __msa_ldi_b(4); \
filt1 = __msa_adds_s_b(filt, cnst4b); \
filt1 >>= 3; \
\
cnst3b = __msa_ldi_b(3); \
filt2 = __msa_adds_s_b(filt, cnst3b); \
filt2 >>= 3; \
\
q0_m = __msa_subs_s_b(q0_m, filt1); \
q0_in_out = __msa_xori_b((v16u8) q0_m, 0x80); \
p0_m = __msa_adds_s_b(p0_m, filt2); \
p0_in_out = __msa_xori_b((v16u8) p0_m, 0x80); \
\
filt = __msa_srari_b(filt1, 1); \
hev_in = __msa_xori_b((v16u8) hev_in, 0xff); \
filt = filt & (v16i8) hev_in; \
\
q1_m = __msa_subs_s_b(q1_m, filt); \
q1_in_out = __msa_xori_b((v16u8) q1_m, 0x80); \
p1_m = __msa_adds_s_b(p1_m, filt); \
p1_in_out = __msa_xori_b((v16u8) p1_m, 0x80); \
}
#define VP8_SIMPLE_FILT(p1_in, p0_in, q0_in, q1_in, mask) \
{ \
v16i8 p1_m, p0_m, q0_m, q1_m, q0_sub_p0, q0_sub_p0_sign; \
v16i8 filt, filt1, filt2, cnst4b, cnst3b, filt_sign; \
v8i16 q0_sub_p0_r, q0_sub_p0_l, filt_l, filt_r, cnst3h; \
\
p1_m = (v16i8) __msa_xori_b(p1_in, 0x80); \
p0_m = (v16i8) __msa_xori_b(p0_in, 0x80); \
q0_m = (v16i8) __msa_xori_b(q0_in, 0x80); \
q1_m = (v16i8) __msa_xori_b(q1_in, 0x80); \
\
filt = __msa_subs_s_b(p1_m, q1_m); \
\
q0_sub_p0 = q0_m - p0_m; \
filt_sign = __msa_clti_s_b(filt, 0); \
\
cnst3h = __msa_ldi_h(3); \
q0_sub_p0_sign = __msa_clti_s_b(q0_sub_p0, 0); \
q0_sub_p0_r = (v8i16) __msa_ilvr_b(q0_sub_p0_sign, q0_sub_p0); \
q0_sub_p0_r *= cnst3h; \
filt_r = (v8i16) __msa_ilvr_b(filt_sign, filt); \
filt_r += q0_sub_p0_r; \
filt_r = __msa_sat_s_h(filt_r, 7); \
\
q0_sub_p0_l = (v8i16) __msa_ilvl_b(q0_sub_p0_sign, q0_sub_p0); \
q0_sub_p0_l *= cnst3h; \
filt_l = (v8i16) __msa_ilvl_b(filt_sign, filt); \
filt_l += q0_sub_p0_l; \
filt_l = __msa_sat_s_h(filt_l, 7); \
\
filt = __msa_pckev_b((v16i8) filt_l, (v16i8) filt_r); \
filt = filt & (v16i8) (mask); \
\
cnst4b = __msa_ldi_b(4); \
filt1 = __msa_adds_s_b(filt, cnst4b); \
filt1 >>= 3; \
\
cnst3b = __msa_ldi_b(3); \
filt2 = __msa_adds_s_b(filt, cnst3b); \
filt2 >>= 3; \
\
q0_m = __msa_subs_s_b(q0_m, filt1); \
p0_m = __msa_adds_s_b(p0_m, filt2); \
q0_in = __msa_xori_b((v16u8) q0_m, 0x80); \
p0_in = __msa_xori_b((v16u8) p0_m, 0x80); \
}
#define VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev) \
{ \
v16i8 p2_m, p1_m, p0_m, q2_m, q1_m, q0_m; \
v16i8 filt, q0_sub_p0, cnst4b, cnst3b; \
v16i8 u, filt1, filt2, filt_sign, q0_sub_p0_sign; \
v8i16 q0_sub_p0_r, q0_sub_p0_l, filt_r, u_r, u_l, filt_l; \
v8i16 cnst3h, cnst27h, cnst18h, cnst63h; \
\
cnst3h = __msa_ldi_h(3); \
\
p2_m = (v16i8) __msa_xori_b(p2, 0x80); \
p1_m = (v16i8) __msa_xori_b(p1, 0x80); \
p0_m = (v16i8) __msa_xori_b(p0, 0x80); \
q0_m = (v16i8) __msa_xori_b(q0, 0x80); \
q1_m = (v16i8) __msa_xori_b(q1, 0x80); \
q2_m = (v16i8) __msa_xori_b(q2, 0x80); \
\
filt = __msa_subs_s_b(p1_m, q1_m); \
q0_sub_p0 = q0_m - p0_m; \
q0_sub_p0_sign = __msa_clti_s_b(q0_sub_p0, 0); \
filt_sign = __msa_clti_s_b(filt, 0); \
\
/* right part */ \
q0_sub_p0_r = (v8i16) __msa_ilvr_b(q0_sub_p0_sign, q0_sub_p0); \
q0_sub_p0_r *= cnst3h; \
filt_r = (v8i16) __msa_ilvr_b(filt_sign, filt); \
filt_r = filt_r + q0_sub_p0_r; \
filt_r = __msa_sat_s_h(filt_r, 7); \
\
/* left part */ \
q0_sub_p0_l = (v8i16) __msa_ilvl_b(q0_sub_p0_sign, q0_sub_p0); \
q0_sub_p0_l *= cnst3h; \
filt_l = (v8i16) __msa_ilvl_b(filt_sign, filt); \
filt_l = filt_l + q0_sub_p0_l; \
filt_l = __msa_sat_s_h(filt_l, 7); \
\
/* combine left and right part */ \
filt = __msa_pckev_b((v16i8) filt_l, (v16i8) filt_r); \
filt = filt & (v16i8) mask; \
filt2 = filt & (v16i8) hev; \
\
/* filt_val &= ~hev */ \
hev = __msa_xori_b(hev, 0xff); \
filt = filt & (v16i8) hev; \
cnst4b = __msa_ldi_b(4); \
filt1 = __msa_adds_s_b(filt2, cnst4b); \
filt1 >>= 3; \
cnst3b = __msa_ldi_b(3); \
filt2 = __msa_adds_s_b(filt2, cnst3b); \
filt2 >>= 3; \
q0_m = __msa_subs_s_b(q0_m, filt1); \
p0_m = __msa_adds_s_b(p0_m, filt2); \
\
filt_sign = __msa_clti_s_b(filt, 0); \
ILVRL_B2_SH(filt_sign, filt, filt_r, filt_l); \
\
cnst27h = __msa_ldi_h(27); \
cnst63h = __msa_ldi_h(63); \
\
/* right part */ \
u_r = filt_r * cnst27h; \
u_r += cnst63h; \
u_r >>= 7; \
u_r = __msa_sat_s_h(u_r, 7); \
/* left part */ \
u_l = filt_l * cnst27h; \
u_l += cnst63h; \
u_l >>= 7; \
u_l = __msa_sat_s_h(u_l, 7); \
/* combine left and right part */ \
u = __msa_pckev_b((v16i8) u_l, (v16i8) u_r); \
q0_m = __msa_subs_s_b(q0_m, u); \
q0 = __msa_xori_b((v16u8) q0_m, 0x80); \
p0_m = __msa_adds_s_b(p0_m, u); \
p0 = __msa_xori_b((v16u8) p0_m, 0x80); \
cnst18h = __msa_ldi_h(18); \
u_r = filt_r * cnst18h; \
u_r += cnst63h; \
u_r >>= 7; \
u_r = __msa_sat_s_h(u_r, 7); \
\
/* left part */ \
u_l = filt_l * cnst18h; \
u_l += cnst63h; \
u_l >>= 7; \
u_l = __msa_sat_s_h(u_l, 7); \
/* combine left and right part */ \
u = __msa_pckev_b((v16i8) u_l, (v16i8) u_r); \
q1_m = __msa_subs_s_b(q1_m, u); \
q1 = __msa_xori_b((v16u8) q1_m, 0x80); \
p1_m = __msa_adds_s_b(p1_m, u); \
p1 = __msa_xori_b((v16u8) p1_m, 0x80); \
u_r = filt_r << 3; \
u_r += filt_r + cnst63h; \
u_r >>= 7; \
u_r = __msa_sat_s_h(u_r, 7); \
\
/* left part */ \
u_l = filt_l << 3; \
u_l += filt_l + cnst63h; \
u_l >>= 7; \
u_l = __msa_sat_s_h(u_l, 7); \
/* combine left and right part */ \
u = __msa_pckev_b((v16i8) u_l, (v16i8) u_r); \
q2_m = __msa_subs_s_b(q2_m, u); \
q2 = __msa_xori_b((v16u8) q2_m, 0x80); \
p2_m = __msa_adds_s_b(p2_m, u); \
p2 = __msa_xori_b((v16u8) p2_m, 0x80); \
}
#define LPF_MASK_HEV(p3_in, p2_in, p1_in, p0_in, \
q0_in, q1_in, q2_in, q3_in, \
limit_in, b_limit_in, thresh_in, \
hev_out, mask_out, flat_out) \
{ \
v16u8 p3_asub_p2_m, p2_asub_p1_m, p1_asub_p0_m, q1_asub_q0_m; \
v16u8 p1_asub_q1_m, p0_asub_q0_m, q3_asub_q2_m, q2_asub_q1_m; \
\
/* absolute subtraction of pixel values */ \
p3_asub_p2_m = __msa_asub_u_b((p3_in), (p2_in)); \
p2_asub_p1_m = __msa_asub_u_b((p2_in), (p1_in)); \
p1_asub_p0_m = __msa_asub_u_b((p1_in), (p0_in)); \
q1_asub_q0_m = __msa_asub_u_b((q1_in), (q0_in)); \
q2_asub_q1_m = __msa_asub_u_b((q2_in), (q1_in)); \
q3_asub_q2_m = __msa_asub_u_b((q3_in), (q2_in)); \
p0_asub_q0_m = __msa_asub_u_b((p0_in), (q0_in)); \
p1_asub_q1_m = __msa_asub_u_b((p1_in), (q1_in)); \
/* calculation of hev */ \
flat_out = __msa_max_u_b(p1_asub_p0_m, q1_asub_q0_m); \
hev_out = (thresh_in) < (v16u8) flat_out; \
/* calculation of mask */ \
p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p0_asub_q0_m); \
p1_asub_q1_m >>= 1; \
p0_asub_q0_m = __msa_adds_u_b(p0_asub_q0_m, p1_asub_q1_m); \
mask_out = (b_limit_in) < p0_asub_q0_m; \
mask_out = __msa_max_u_b(flat_out, mask_out); \
p3_asub_p2_m = __msa_max_u_b(p3_asub_p2_m, p2_asub_p1_m); \
mask_out = __msa_max_u_b(p3_asub_p2_m, mask_out); \
q2_asub_q1_m = __msa_max_u_b(q2_asub_q1_m, q3_asub_q2_m); \
mask_out = __msa_max_u_b(q2_asub_q1_m, mask_out); \
mask_out = (limit_in) < (v16u8) mask_out; \
mask_out = __msa_xori_b(mask_out, 0xff); \
}
#define VP8_ST6x1_UB(in0, in0_idx, in1, in1_idx, pdst, stride) \
{ \
uint16_t tmp0_h; \
uint32_t tmp0_w; \
\
tmp0_w = __msa_copy_u_w((v4i32) in0, in0_idx); \
tmp0_h = __msa_copy_u_h((v8i16) in1, in1_idx); \
SW(tmp0_w, pdst); \
SH(tmp0_h, pdst + stride); \
}
void ff_vp8_v_loop_filter16_msa(uint8_t *src, ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
/* load vector elements */
temp_src = src - (pitch << 2);
LD_UB8(temp_src, pitch, p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
/* store vector elements */
temp_src = src - 3 * pitch;
ST_UB4(p2, p1, p0, q0, temp_src, pitch);
temp_src += (4 * pitch);
ST_UB2(q1, q2, temp_src, pitch);
}
void ff_vp8_v_loop_filter8uv_msa(uint8_t *src_u, uint8_t *src_v,
ptrdiff_t pitch, int b_limit_in, int limit_in,
int thresh_in)
{
uint8_t *temp_src;
uint64_t p2_d, p1_d, p0_d, q0_d, q1_d, q2_d;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
temp_src = src_u - (pitch << 2);
LD_UB8(temp_src, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
temp_src = src_v - (pitch << 2);
LD_UB8(temp_src, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
/* rht 8 element of p3 are u pixel and left 8 element of p3 are v pixel */
ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
p2_d = __msa_copy_u_d((v2i64) p2, 0);
p1_d = __msa_copy_u_d((v2i64) p1, 0);
p0_d = __msa_copy_u_d((v2i64) p0, 0);
q0_d = __msa_copy_u_d((v2i64) q0, 0);
q1_d = __msa_copy_u_d((v2i64) q1, 0);
q2_d = __msa_copy_u_d((v2i64) q2, 0);
src_u -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_u, pitch);
src_u += 4 * pitch;
SD(q1_d, src_u);
src_u += pitch;
SD(q2_d, src_u);
p2_d = __msa_copy_u_d((v2i64) p2, 1);
p1_d = __msa_copy_u_d((v2i64) p1, 1);
p0_d = __msa_copy_u_d((v2i64) p0, 1);
q0_d = __msa_copy_u_d((v2i64) q0, 1);
q1_d = __msa_copy_u_d((v2i64) q1, 1);
q2_d = __msa_copy_u_d((v2i64) q2, 1);
src_v -= (pitch * 3);
SD4(p2_d, p1_d, p0_d, q0_d, src_v, pitch);
src_v += 4 * pitch;
SD(q1_d, src_v);
src_v += pitch;
SD(q2_d, src_v);
}
void ff_vp8_h_loop_filter16_msa(uint8_t *src, ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint8_t *temp_src;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
temp_src = src - 4;
LD_UB8(temp_src, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
temp_src += (8 * pitch);
LD_UB8(temp_src, pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
ILVRL_B2_SH(q2, q1, tmp2, tmp5);
temp_src = src - 3;
VP8_ST6x1_UB(tmp3, 0, tmp2, 0, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 1, tmp2, 1, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 2, tmp2, 2, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp3, 3, tmp2, 3, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 0, tmp2, 4, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 1, tmp2, 5, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 2, tmp2, 6, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp4, 3, tmp2, 7, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 0, tmp5, 0, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 1, tmp5, 1, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 2, tmp5, 2, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp6, 3, tmp5, 3, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 0, tmp5, 4, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 1, tmp5, 5, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 2, tmp5, 6, temp_src, 4);
temp_src += pitch;
VP8_ST6x1_UB(tmp7, 3, tmp5, 7, temp_src, 4);
}
void ff_vp8_h_loop_filter8uv_msa(uint8_t *src_u, uint8_t *src_v,
ptrdiff_t pitch, int b_limit_in, int limit_in,
int thresh_in)
{
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
b_limit = (v16u8) __msa_fill_b(b_limit_in);
limit = (v16u8) __msa_fill_b(limit_in);
thresh = (v16u8) __msa_fill_b(thresh_in);
LD_UB8(src_u - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src_v - 4, pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev);
ILVR_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp3, tmp4);
ILVL_B2_SH(p1, p2, q0, p0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp6, tmp7);
ILVRL_B2_SH(q2, q1, tmp2, tmp5);
src_u -= 3;
VP8_ST6x1_UB(tmp3, 0, tmp2, 0, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 1, tmp2, 1, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 2, tmp2, 2, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp3, 3, tmp2, 3, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 0, tmp2, 4, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 1, tmp2, 5, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 2, tmp2, 6, src_u, 4);
src_u += pitch;
VP8_ST6x1_UB(tmp4, 3, tmp2, 7, src_u, 4);
src_v -= 3;
VP8_ST6x1_UB(tmp6, 0, tmp5, 0, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 1, tmp5, 1, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 2, tmp5, 2, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp6, 3, tmp5, 3, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 0, tmp5, 4, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 1, tmp5, 5, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 2, tmp5, 6, src_v, 4);
src_v += pitch;
VP8_ST6x1_UB(tmp7, 3, tmp5, 7, src_v, 4);
}
void ff_vp8_v_loop_filter_simple_msa(uint8_t *src, ptrdiff_t pitch,
int b_limit_ptr)
{
v16u8 p1, p0, q1, q0;
v16u8 mask, b_limit;
b_limit = (v16u8) __msa_fill_b(b_limit_ptr);
/* load vector elements */
LD_UB4(src - (pitch << 1), pitch, p1, p0, q0, q1);
VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
ST_UB2(p0, q0, (src - pitch), pitch);
}
void ff_vp8_h_loop_filter_simple_msa(uint8_t *src, ptrdiff_t pitch,
int b_limit_ptr)
{
uint8_t *temp_src;
v16u8 p1, p0, q1, q0;
v16u8 mask, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1;
b_limit = (v16u8) __msa_fill_b(b_limit_ptr);
temp_src = src - 2;
LD_UB8(temp_src, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
temp_src += (8 * pitch);
LD_UB8(temp_src, pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x4_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p1, p0, q0, q1);
VP8_SIMPLE_MASK(p1, p0, q0, q1, b_limit, mask);
VP8_SIMPLE_FILT(p1, p0, q0, q1, mask);
ILVRL_B2_SH(q0, p0, tmp1, tmp0);
src -= 1;
ST2x4_UB(tmp1, 0, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp1, 4, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp0, 0, src, pitch);
src += 4 * pitch;
ST2x4_UB(tmp0, 4, src, pitch);
src += 4 * pitch;
}
void ff_vp8_v_loop_filter8uv_inner_msa(uint8_t *src_u, uint8_t *src_v,
ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint64_t p1_d, p0_d, q0_d, q1_d;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u;
v16u8 p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v;
thresh = (v16u8) __msa_fill_b(thresh_in);
limit = (v16u8) __msa_fill_b(limit_in);
b_limit = (v16u8) __msa_fill_b(b_limit_in);
src_u = src_u - (pitch << 2);
LD_UB8(src_u, pitch, p3_u, p2_u, p1_u, p0_u, q0_u, q1_u, q2_u, q3_u);
src_u += (5 * pitch);
src_v = src_v - (pitch << 2);
LD_UB8(src_v, pitch, p3_v, p2_v, p1_v, p0_v, q0_v, q1_v, q2_v, q3_v);
src_v += (5 * pitch);
/* right 8 element of p3 are u pixel and
left 8 element of p3 are v pixel */
ILVR_D4_UB(p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, p2, p1, p0);
ILVR_D4_UB(q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
p1_d = __msa_copy_u_d((v2i64) p1, 0);
p0_d = __msa_copy_u_d((v2i64) p0, 0);
q0_d = __msa_copy_u_d((v2i64) q0, 0);
q1_d = __msa_copy_u_d((v2i64) q1, 0);
SD4(q1_d, q0_d, p0_d, p1_d, src_u, (- pitch));
p1_d = __msa_copy_u_d((v2i64) p1, 1);
p0_d = __msa_copy_u_d((v2i64) p0, 1);
q0_d = __msa_copy_u_d((v2i64) q0, 1);
q1_d = __msa_copy_u_d((v2i64) q1, 1);
SD4(q1_d, q0_d, p0_d, p1_d, src_v, (- pitch));
}
void ff_vp8_h_loop_filter8uv_inner_msa(uint8_t *src_u, uint8_t *src_v,
ptrdiff_t pitch, int b_limit_in,
int limit_in, int thresh_in)
{
uint8_t *temp_src_u, *temp_src_v;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 mask, hev, flat, thresh, limit, b_limit;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7, row8;
v16u8 row9, row10, row11, row12, row13, row14, row15;
v4i32 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
thresh = (v16u8) __msa_fill_b(thresh_in);
limit = (v16u8) __msa_fill_b(limit_in);
b_limit = (v16u8) __msa_fill_b(b_limit_in);
LD_UB8(src_u - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src_v - 4, pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ILVR_B2_SW(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SW(tmp1, tmp0, tmp2, tmp3);
tmp0 = (v4i32) __msa_ilvl_b((v16i8) p0, (v16i8) p1);
tmp1 = (v4i32) __msa_ilvl_b((v16i8) q1, (v16i8) q0);
ILVRL_H2_SW(tmp1, tmp0, tmp4, tmp5);
temp_src_u = src_u - 2;
ST4x4_UB(tmp2, tmp2, 0, 1, 2, 3, temp_src_u, pitch);
temp_src_u += 4 * pitch;
ST4x4_UB(tmp3, tmp3, 0, 1, 2, 3, temp_src_u, pitch);
temp_src_v = src_v - 2;
ST4x4_UB(tmp4, tmp4, 0, 1, 2, 3, temp_src_v, pitch);
temp_src_v += 4 * pitch;
ST4x4_UB(tmp5, tmp5, 0, 1, 2, 3, temp_src_v, pitch);
}
void ff_vp8_v_loop_filter16_inner_msa(uint8_t *src, ptrdiff_t pitch,
int32_t e, int32_t i, int32_t h)
{
v16u8 mask, hev, flat;
v16u8 thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
/* load vector elements */
LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8) __msa_fill_b(h);
b_limit = (v16u8) __msa_fill_b(e);
limit = (v16u8) __msa_fill_b(i);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
}
void ff_vp8_h_loop_filter16_inner_msa(uint8_t *src, ptrdiff_t pitch,
int32_t e, int32_t i, int32_t h)
{
v16u8 mask, hev, flat;
v16u8 thresh, b_limit, limit;
v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
LD_UB8(src - 4 + (8 * pitch), pitch,
row8, row9, row10, row11, row12, row13, row14, row15);
TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7,
row8, row9, row10, row11, row12, row13, row14, row15,
p3, p2, p1, p0, q0, q1, q2, q3);
thresh = (v16u8) __msa_fill_b(h);
b_limit = (v16u8) __msa_fill_b(e);
limit = (v16u8) __msa_fill_b(i);
LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh,
hev, mask, flat);
VP8_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev);
ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
src -= 2;
ST4x8_UB(tmp2, tmp3, src, pitch);
src += (8 * pitch);
ST4x8_UB(tmp4, tmp5, src, pitch);
}

2332
libavcodec/mips/vp8_mc_msa.c
View File

File diff suppressed because it is too large Load Diff

113
libavcodec/mips/vp8dsp_init_mips.c
Unescape View File

@ -0,0 +1,113 @@
/*
* Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
/**
* @file
* VP8 compatible video decoder
*/
#include "config.h"
#include "libavutil/attributes.h"
#include "libavcodec/vp8dsp.h"
#include "vp8dsp_mips.h"
#define VP8_MC_MIPS_FUNC(IDX, SIZE) \
dsp->put_vp8_epel_pixels_tab[IDX][0][1] = \
ff_put_vp8_epel##SIZE##_h4_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][0][2] = \
ff_put_vp8_epel##SIZE##_h6_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][1][0] = \
ff_put_vp8_epel##SIZE##_v4_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][1][1] = \
ff_put_vp8_epel##SIZE##_h4v4_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][1][2] = \
ff_put_vp8_epel##SIZE##_h6v4_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][2][0] = \
ff_put_vp8_epel##SIZE##_v6_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][2][1] = \
ff_put_vp8_epel##SIZE##_h4v6_msa; \
dsp->put_vp8_epel_pixels_tab[IDX][2][2] = \
ff_put_vp8_epel##SIZE##_h6v6_msa
#define VP8_BILINEAR_MC_MIPS_FUNC(IDX, SIZE) \
dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = \
ff_put_vp8_bilinear##SIZE##_h_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = \
ff_put_vp8_bilinear##SIZE##_h_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = \
ff_put_vp8_bilinear##SIZE##_v_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = \
ff_put_vp8_bilinear##SIZE##_hv_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = \
ff_put_vp8_bilinear##SIZE##_hv_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = \
ff_put_vp8_bilinear##SIZE##_v_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = \
ff_put_vp8_bilinear##SIZE##_hv_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = \
ff_put_vp8_bilinear##SIZE##_hv_msa
#define VP8_MC_MIPS_COPY(IDX, SIZE) \
dsp->put_vp8_epel_pixels_tab[IDX][0][0] = \
ff_put_vp8_pixels##SIZE##_msa; \
dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = \
ff_put_vp8_pixels##SIZE##_msa;
#if HAVE_MSA
static av_cold void vp8dsp_init_msa(VP8DSPContext *dsp)
{
dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_msa;
dsp->vp8_idct_add = ff_vp8_idct_add_msa;
dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_msa;
dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_msa;
dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_msa;
VP8_MC_MIPS_FUNC(0, 16);
VP8_MC_MIPS_FUNC(1, 8);
VP8_MC_MIPS_FUNC(2, 4);
VP8_BILINEAR_MC_MIPS_FUNC(0, 16);
VP8_BILINEAR_MC_MIPS_FUNC(1, 8);
VP8_BILINEAR_MC_MIPS_FUNC(2, 4);
VP8_MC_MIPS_COPY(0, 16);
VP8_MC_MIPS_COPY(1, 8);
dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_msa;
dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_msa;
dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_msa;
dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_msa;
dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_msa;
dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_msa;
dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_msa;
dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_msa;
dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_msa;
dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_msa;
}
#endif // #if HAVE_MSA
av_cold void ff_vp8dsp_init_mips(VP8DSPContext *dsp)
{
#if HAVE_MSA
vp8dsp_init_msa(dsp);
#endif // #if HAVE_MSA
}

172
libavcodec/mips/vp8dsp_mips.h
Unescape View File

@ -0,0 +1,172 @@
/*
* Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef AVCODEC_MIPS_VP8DSP_MIPS_H
#define AVCODEC_MIPS_VP8DSP_MIPS_H
void ff_put_vp8_pixels4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int x, int y);
void ff_put_vp8_pixels8_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int x, int y);
void ff_put_vp8_pixels16_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int x, int y);
void ff_put_vp8_epel16_h4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_h6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_h4v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_h6v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_h4v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel16_h6v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h4v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h6v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h4v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel8_h6v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h4v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h6v4_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h4v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_epel4_h6v6_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear16_h_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear16_v_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear16_hv_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear8_h_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear8_v_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear8_hv_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear4_h_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear4_v_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
void ff_put_vp8_bilinear4_hv_msa(uint8_t *dst, ptrdiff_t dststride,
uint8_t *src, ptrdiff_t srcstride,
int h, int mx, int my);
/* loop filter */
void ff_vp8_h_loop_filter16_inner_msa(uint8_t *dst, ptrdiff_t stride,
int32_t e, int32_t i, int32_t h);
void ff_vp8_v_loop_filter16_inner_msa(uint8_t *dst, ptrdiff_t stride,
int32_t e, int32_t i, int32_t h);
void ff_vp8_h_loop_filter8uv_inner_msa(uint8_t *dst_u, uint8_t *dst_v,
ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_v_loop_filter8uv_inner_msa(uint8_t *dst_u, uint8_t *dst_v,
ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_h_loop_filter16_msa(uint8_t *dst, ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_v_loop_filter16_msa(uint8_t *dst, ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_h_loop_filter8uv_msa(uint8_t *dst_u, uint8_t *dst_v,
ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_v_loop_filter8uv_msa(uint8_t *dst_u, uint8_t *dst_v,
ptrdiff_t stride,
int flim_e, int flim_i, int hev_thresh);
void ff_vp8_h_loop_filter_simple_msa(uint8_t *dst, ptrdiff_t stride, int flim);
void ff_vp8_v_loop_filter_simple_msa(uint8_t *dst, ptrdiff_t stride, int flim);
/* Idct functions */
void ff_vp8_luma_dc_wht_msa(int16_t block[4][4][16], int16_t dc[16]);
void ff_vp8_idct_add_msa(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
void ff_vp8_idct_dc_add_msa(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
void ff_vp8_idct_dc_add4uv_msa(uint8_t *dst, int16_t block[4][16],
ptrdiff_t stride);
void ff_vp8_idct_dc_add4y_msa(uint8_t *dst, int16_t block[4][16],
ptrdiff_t stride);
#endif // #ifndef AVCODEC_MIPS_VP8DSP_MIPS_H

2
libavcodec/vp8dsp.c
Unescape View File

@ -735,5 +735,7 @@ av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
ff_vp8dsp_init_arm(dsp); ff_vp8dsp_init_arm(dsp);
if (ARCH_X86) if (ARCH_X86)
ff_vp8dsp_init_x86(dsp); ff_vp8dsp_init_x86(dsp);
if (ARCH_MIPS)
ff_vp8dsp_init_mips(dsp);
} }
#endif /* CONFIG_VP8_DECODER */ #endif /* CONFIG_VP8_DECODER */

1
libavcodec/vp8dsp.h
Unescape View File

@ -98,6 +98,7 @@ void ff_vp78dsp_init_x86(VP8DSPContext *c);
void ff_vp8dsp_init(VP8DSPContext *c); void ff_vp8dsp_init(VP8DSPContext *c);
void ff_vp8dsp_init_arm(VP8DSPContext *c); void ff_vp8dsp_init_arm(VP8DSPContext *c);
void ff_vp8dsp_init_x86(VP8DSPContext *c); void ff_vp8dsp_init_x86(VP8DSPContext *c);
void ff_vp8dsp_init_mips(VP8DSPContext *c);
#define IS_VP7 1 #define IS_VP7 1
#define IS_VP8 0 #define IS_VP8 0

Write Preview
Loading…
Cancel
Save