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

VP8: armv6 optimizations.

Browse Source 
From 52.503s (~40fps) to 27.973sec (~80fps) decoding of 480p sintel
trailer, i.e. a ~2x speedup overall, on a Nexus S.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Ronald S. Bultje 2011-08-24 13:58:37 -07:00 committed by Michael Niedermayer
parent c3a774969a
commit a5dfeb612e
5 changed files with 2577 additions and 119 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
libavcodec/arm/Makefile
View File

@ -11,7 +11,8 @@ ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o
OBJS-$(CONFIG_VP5_DECODER) += arm/vp56dsp_init_arm.o
OBJS-$(CONFIG_VP6_DECODER) += arm/vp56dsp_init_arm.o
OBJS-$(CONFIG_VP8_DECODER) += arm/vp8dsp_init_arm.o
ARMV6-OBJS-$(CONFIG_VP8_DECODER) += arm/vp8_armv6.o
ARMV6-OBJS-$(CONFIG_VP8_DECODER) += arm/vp8_armv6.o \
arm/vp8dsp_armv6.o
OBJS-$(CONFIG_H264DSP) += arm/h264dsp_init_arm.o
OBJS-$(CONFIG_H264PRED) += arm/h264pred_init_arm.o

12
libavcodec/arm/asm.S
View File

@ -97,6 +97,12 @@ T add \rn, \rn, \rm
T ldr \rt, [\rn]
.endm
.macro ldr_dpren rt, rn, rm:vararg
A ldr \rt, [\rn, -\rm]
T sub \rt, \rn, \rm
T ldr \rt, [\rt]
.endm
.macro ldr_post rt, rn, rm:vararg
A ldr \rt, [\rn], \rm
T ldr \rt, [\rn]
@ -133,6 +139,12 @@ T ldrh \rt, [\rn]
T add \rn, \rn, \rm
.endm
.macro ldrb_post rt, rn, rm
A ldrb \rt, [\rn], \rm
T ldrb \rt, [\rn]
T add \rn, \rn, \rm
.endm
.macro str_post rt, rn, rm:vararg
A str \rt, [\rn], \rm
T str \rt, [\rn]

2328
libavcodec/arm/vp8dsp_armv6.S Normal file
View File

File diff suppressed because it is too large Load Diff

324
libavcodec/arm/vp8dsp_init_arm.c
View File

@ -19,13 +19,17 @@
#include <stdint.h>
#include "libavcodec/vp8dsp.h"
void ff_vp8_luma_dc_wht_neon(DCTELEM block[4][4][16], DCTELEM dc[16]);
void ff_vp8_luma_dc_wht_dc_neon(DCTELEM block[4][4][16], DCTELEM dc[16]);
void ff_vp8_luma_dc_wht_dc_armv6(DCTELEM block[4][4][16], DCTELEM dc[16]);
void ff_vp8_idct_add_neon(uint8_t *dst, DCTELEM block[16], int stride);
void ff_vp8_idct_dc_add_neon(uint8_t *dst, DCTELEM block[16], int stride);
void ff_vp8_idct_dc_add4y_neon(uint8_t *dst, DCTELEM block[4][16], int stride);
void ff_vp8_idct_dc_add4uv_neon(uint8_t *dst, DCTELEM block[4][16], int stride);
#define idct_funcs(opt) \
void ff_vp8_luma_dc_wht_ ## opt(DCTELEM block[4][4][16], DCTELEM dc[16]); \
void ff_vp8_idct_add_ ## opt(uint8_t *dst, DCTELEM block[16], int stride); \
void ff_vp8_idct_dc_add_ ## opt(uint8_t *dst, DCTELEM block[16], int stride); \
void ff_vp8_idct_dc_add4y_ ## opt(uint8_t *dst, DCTELEM block[4][16], int stride); \
void ff_vp8_idct_dc_add4uv_ ## opt(uint8_t *dst, DCTELEM block[4][16], int stride)
idct_funcs(neon);
idct_funcs(armv6);
void ff_vp8_v_loop_filter16_neon(uint8_t *dst, int stride,
int flim_E, int flim_I, int hev_thresh);
@ -47,29 +51,106 @@ void ff_vp8_h_loop_filter8uv_inner_neon(uint8_t *dstU, uint8_t *dstV,
int stride, int flim_E, int flim_I,
int hev_thresh);
void ff_vp8_v_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim);
void ff_vp8_h_loop_filter16_simple_neon(uint8_t *dst, int stride, int flim);
void ff_vp8_v_loop_filter_inner_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I,
int hev_thresh, int count);
void ff_vp8_h_loop_filter_inner_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I,
int hev_thresh, int count);
void ff_vp8_v_loop_filter_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I,
int hev_thresh, int count);
void ff_vp8_h_loop_filter_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I,
int hev_thresh, int count);
static void ff_vp8_v_loop_filter16_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_v_loop_filter_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4);
}
#define VP8_MC(n) \
void ff_put_vp8_##n##_neon(uint8_t *dst, int dststride, \
uint8_t *src, int srcstride, \
int h, int x, int y)
static void ff_vp8_h_loop_filter16_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_h_loop_filter_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4);
}
static void ff_vp8_v_loop_filter8uv_armv6(uint8_t *dstU, uint8_t *dstV, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_v_loop_filter_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2);
ff_vp8_v_loop_filter_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2);
}
static void ff_vp8_h_loop_filter8uv_armv6(uint8_t *dstU, uint8_t *dstV, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_h_loop_filter_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2);
ff_vp8_h_loop_filter_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2);
}
static void ff_vp8_v_loop_filter16_inner_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_v_loop_filter_inner_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4);
}
static void ff_vp8_h_loop_filter16_inner_armv6(uint8_t *dst, int stride,
int flim_E, int flim_I, int hev_thresh)
{
ff_vp8_h_loop_filter_inner_armv6(dst, stride, flim_E, flim_I, hev_thresh, 4);
}
static void ff_vp8_v_loop_filter8uv_inner_armv6(uint8_t *dstU, uint8_t *dstV,
int stride, int flim_E, int flim_I,
int hev_thresh)
{
ff_vp8_v_loop_filter_inner_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2);
ff_vp8_v_loop_filter_inner_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2);
}
static void ff_vp8_h_loop_filter8uv_inner_armv6(uint8_t *dstU, uint8_t *dstV,
int stride, int flim_E, int flim_I,
int hev_thresh)
{
ff_vp8_h_loop_filter_inner_armv6(dstU, stride, flim_E, flim_I, hev_thresh, 2);
ff_vp8_h_loop_filter_inner_armv6(dstV, stride, flim_E, flim_I, hev_thresh, 2);
}
#define simple_lf_funcs(opt) \
void ff_vp8_v_loop_filter16_simple_ ## opt(uint8_t *dst, int stride, int flim); \
void ff_vp8_h_loop_filter16_simple_ ## opt(uint8_t *dst, int stride, int flim)
simple_lf_funcs(neon);
simple_lf_funcs(armv6);
#define VP8_MC_OPT(n, opt) \
void ff_put_vp8_##n##_##opt(uint8_t *dst, int dststride, \
uint8_t *src, int srcstride, \
int h, int x, int y)
#define VP8_MC(n) \
VP8_MC_OPT(n, neon)
#define VP8_EPEL(w) \
VP8_MC(pixels ## w); \
VP8_MC(epel ## w ## _h4); \
VP8_MC(epel ## w ## _h6); \
VP8_MC(epel ## w ## _v4); \
VP8_MC(epel ## w ## _h4v4); \
VP8_MC(epel ## w ## _h6v4); \
VP8_MC(epel ## w ## _v4); \
VP8_MC(epel ## w ## _v6); \
VP8_MC(epel ## w ## _h4v6); \
VP8_MC(epel ## w ## _h6v6)
VP8_EPEL(16);
VP8_MC(pixels16);
VP8_MC_OPT(pixels16, armv6);
VP8_EPEL(8);
VP8_MC(pixels8);
VP8_MC_OPT(pixels8, armv6);
VP8_EPEL(4);
VP8_MC_OPT(pixels4, armv6);
VP8_MC(bilin16_h);
VP8_MC(bilin16_v);
@ -81,83 +162,148 @@ VP8_MC(bilin4_h);
VP8_MC(bilin4_v);
VP8_MC(bilin4_hv);
#define VP8_V6_MC(n) \
void ff_put_vp8_##n##_armv6(uint8_t *dst, int dststride, uint8_t *src, \
int srcstride, int w, int h, int mxy)
VP8_V6_MC(epel_v6);
VP8_V6_MC(epel_h6);
VP8_V6_MC(epel_v4);
VP8_V6_MC(epel_h4);
VP8_V6_MC(bilin_v);
VP8_V6_MC(bilin_h);
#define VP8_EPEL_HV(SIZE, TAPNUMX, TAPNUMY, NAME, HNAME, VNAME, MAXHEIGHT) \
static void ff_put_vp8_##NAME##SIZE##_##HNAME##VNAME##_armv6( \
uint8_t *dst, int dststride, uint8_t *src, \
int srcstride, int h, int mx, int my) \
{ \
DECLARE_ALIGNED(4, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \
uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \
src -= srcstride * (TAPNUMY / 2 - 1); \
ff_put_vp8_ ## NAME ## _ ## HNAME ## _armv6(tmp, SIZE, src, srcstride, \
SIZE, h + TAPNUMY - 1, mx); \
ff_put_vp8_ ## NAME ## _ ## VNAME ## _armv6(dst, dststride, tmpptr, SIZE, \
SIZE, h, my); \
}
VP8_EPEL_HV(16, 6, 6, epel, h6, v6, 16);
VP8_EPEL_HV(16, 2, 2, bilin, h, v, 16);
VP8_EPEL_HV(8, 6, 6, epel, h6, v6, 16);
VP8_EPEL_HV(8, 4, 6, epel, h4, v6, 16);
VP8_EPEL_HV(8, 6, 4, epel, h6, v4, 16);
VP8_EPEL_HV(8, 4, 4, epel, h4, v4, 16);
VP8_EPEL_HV(8, 2, 2, bilin, h, v, 16);
VP8_EPEL_HV(4, 6, 6, epel, h6, v6, 8);
VP8_EPEL_HV(4, 4, 6, epel, h4, v6, 8);
VP8_EPEL_HV(4, 6, 4, epel, h6, v4, 8);
VP8_EPEL_HV(4, 4, 4, epel, h4, v4, 8);
VP8_EPEL_HV(4, 2, 2, bilin, h, v, 8);
extern void put_vp8_epel4_v6_c(uint8_t *dst, int d, uint8_t *src, int s, int h, int mx, int my);
#undef printf
#define VP8_EPEL_H_OR_V(SIZE, NAME, HV) \
static void ff_put_vp8_##NAME##SIZE##_##HV##_armv6( \
uint8_t *dst, int dststride, uint8_t *src, \
int srcstride, int h, int mx, int my) \
{ \
ff_put_vp8_## NAME ## _ ## HV ## _armv6(dst, dststride, src, srcstride, \
SIZE, h, mx | my); \
}
VP8_EPEL_H_OR_V(4, epel, h6);
VP8_EPEL_H_OR_V(4, epel, h4);
VP8_EPEL_H_OR_V(4, epel, v6);
VP8_EPEL_H_OR_V(4, epel, v4);
VP8_EPEL_H_OR_V(4, bilin, v);
VP8_EPEL_H_OR_V(4, bilin, h);
VP8_EPEL_H_OR_V(8, epel, h6);
VP8_EPEL_H_OR_V(8, epel, h4);
VP8_EPEL_H_OR_V(8, epel, v6);
VP8_EPEL_H_OR_V(8, epel, v4);
VP8_EPEL_H_OR_V(8, bilin, v);
VP8_EPEL_H_OR_V(8, bilin, h);
VP8_EPEL_H_OR_V(16, epel, h6);
VP8_EPEL_H_OR_V(16, epel, v6);
VP8_EPEL_H_OR_V(16, bilin, v);
VP8_EPEL_H_OR_V(16, bilin, h);
av_cold void ff_vp8dsp_init_arm(VP8DSPContext *dsp)
{
#define set_func_ptrs(opt) \
dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_##opt; \
dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_armv6; \
\
dsp->vp8_idct_add = ff_vp8_idct_add_##opt; \
dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_##opt; \
dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_##opt; \
dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_##opt; \
\
dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_##opt; \
dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_##opt; \
dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_##opt; \
dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_##opt; \
\
dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_##opt; \
dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_##opt; \
dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_##opt; \
dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_##opt; \
\
dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_##opt; \
dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_##opt; \
\
dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_##opt; \
dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_##opt; \
dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_##opt; \
dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_##opt; \
\
dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_##opt; \
dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_##opt; \
dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_##opt; \
dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_##opt; \
dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_##opt; \
dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_##opt; \
dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_##opt; \
dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_##opt; \
dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_##opt; \
\
dsp->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_armv6; \
dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_##opt; \
dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_##opt; \
dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_##opt; \
dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_##opt; \
dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_##opt; \
dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_##opt; \
dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_##opt; \
dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_##opt; \
\
dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_##opt; \
dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_##opt; \
dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_##opt; \
dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_##opt; \
\
dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_##opt; \
\
dsp->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_armv6; \
dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_##opt; \
dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_##opt
if (HAVE_NEON) {
dsp->vp8_luma_dc_wht = ff_vp8_luma_dc_wht_neon;
dsp->vp8_luma_dc_wht_dc = ff_vp8_luma_dc_wht_dc_neon;
dsp->vp8_idct_add = ff_vp8_idct_add_neon;
dsp->vp8_idct_dc_add = ff_vp8_idct_dc_add_neon;
dsp->vp8_idct_dc_add4y = ff_vp8_idct_dc_add4y_neon;
dsp->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_neon;
dsp->vp8_v_loop_filter16y = ff_vp8_v_loop_filter16_neon;
dsp->vp8_h_loop_filter16y = ff_vp8_h_loop_filter16_neon;
dsp->vp8_v_loop_filter8uv = ff_vp8_v_loop_filter8uv_neon;
dsp->vp8_h_loop_filter8uv = ff_vp8_h_loop_filter8uv_neon;
dsp->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16_inner_neon;
dsp->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16_inner_neon;
dsp->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_neon;
dsp->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_neon;
dsp->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter16_simple_neon;
dsp->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter16_simple_neon;
dsp->put_vp8_epel_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
dsp->put_vp8_epel_pixels_tab[0][0][2] = ff_put_vp8_epel16_h6_neon;
dsp->put_vp8_epel_pixels_tab[0][2][0] = ff_put_vp8_epel16_v6_neon;
dsp->put_vp8_epel_pixels_tab[0][2][2] = ff_put_vp8_epel16_h6v6_neon;
dsp->put_vp8_epel_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
dsp->put_vp8_epel_pixels_tab[1][0][1] = ff_put_vp8_epel8_h4_neon;
dsp->put_vp8_epel_pixels_tab[1][0][2] = ff_put_vp8_epel8_h6_neon;
dsp->put_vp8_epel_pixels_tab[1][1][0] = ff_put_vp8_epel8_v4_neon;
dsp->put_vp8_epel_pixels_tab[1][1][1] = ff_put_vp8_epel8_h4v4_neon;
dsp->put_vp8_epel_pixels_tab[1][1][2] = ff_put_vp8_epel8_h6v4_neon;
dsp->put_vp8_epel_pixels_tab[1][2][0] = ff_put_vp8_epel8_v6_neon;
dsp->put_vp8_epel_pixels_tab[1][2][1] = ff_put_vp8_epel8_h4v6_neon;
dsp->put_vp8_epel_pixels_tab[1][2][2] = ff_put_vp8_epel8_h6v6_neon;
dsp->put_vp8_epel_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon;
dsp->put_vp8_epel_pixels_tab[2][0][1] = ff_put_vp8_epel4_h4_neon;
dsp->put_vp8_epel_pixels_tab[2][0][2] = ff_put_vp8_epel4_h6_neon;
dsp->put_vp8_epel_pixels_tab[2][1][0] = ff_put_vp8_epel4_v4_neon;
dsp->put_vp8_epel_pixels_tab[2][1][1] = ff_put_vp8_epel4_h4v4_neon;
dsp->put_vp8_epel_pixels_tab[2][1][2] = ff_put_vp8_epel4_h6v4_neon;
dsp->put_vp8_epel_pixels_tab[2][2][0] = ff_put_vp8_epel4_v6_neon;
dsp->put_vp8_epel_pixels_tab[2][2][1] = ff_put_vp8_epel4_h4v6_neon;
dsp->put_vp8_epel_pixels_tab[2][2][2] = ff_put_vp8_epel4_h6v6_neon;
dsp->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_neon;
dsp->put_vp8_bilinear_pixels_tab[0][0][1] = ff_put_vp8_bilin16_h_neon;
dsp->put_vp8_bilinear_pixels_tab[0][0][2] = ff_put_vp8_bilin16_h_neon;
dsp->put_vp8_bilinear_pixels_tab[0][1][0] = ff_put_vp8_bilin16_v_neon;
dsp->put_vp8_bilinear_pixels_tab[0][1][1] = ff_put_vp8_bilin16_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[0][1][2] = ff_put_vp8_bilin16_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[0][2][0] = ff_put_vp8_bilin16_v_neon;
dsp->put_vp8_bilinear_pixels_tab[0][2][1] = ff_put_vp8_bilin16_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[0][2][2] = ff_put_vp8_bilin16_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_neon;
dsp->put_vp8_bilinear_pixels_tab[1][0][1] = ff_put_vp8_bilin8_h_neon;
dsp->put_vp8_bilinear_pixels_tab[1][0][2] = ff_put_vp8_bilin8_h_neon;
dsp->put_vp8_bilinear_pixels_tab[1][1][0] = ff_put_vp8_bilin8_v_neon;
dsp->put_vp8_bilinear_pixels_tab[1][1][1] = ff_put_vp8_bilin8_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[1][1][2] = ff_put_vp8_bilin8_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[1][2][0] = ff_put_vp8_bilin8_v_neon;
dsp->put_vp8_bilinear_pixels_tab[1][2][1] = ff_put_vp8_bilin8_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[1][2][2] = ff_put_vp8_bilin8_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[2][0][0] = ff_put_vp8_pixels4_neon;
dsp->put_vp8_bilinear_pixels_tab[2][0][1] = ff_put_vp8_bilin4_h_neon;
dsp->put_vp8_bilinear_pixels_tab[2][0][2] = ff_put_vp8_bilin4_h_neon;
dsp->put_vp8_bilinear_pixels_tab[2][1][0] = ff_put_vp8_bilin4_v_neon;
dsp->put_vp8_bilinear_pixels_tab[2][1][1] = ff_put_vp8_bilin4_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[2][1][2] = ff_put_vp8_bilin4_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[2][2][0] = ff_put_vp8_bilin4_v_neon;
dsp->put_vp8_bilinear_pixels_tab[2][2][1] = ff_put_vp8_bilin4_hv_neon;
dsp->put_vp8_bilinear_pixels_tab[2][2][2] = ff_put_vp8_bilin4_hv_neon;
set_func_ptrs(neon);
} else if (HAVE_ARMV6) {
set_func_ptrs(armv6);
}
}

29
libavcodec/arm/vp8dsp_neon.S
View File

@ -76,18 +76,6 @@ function ff_vp8_luma_dc_wht_neon, export=1
bx lr
endfunc
function ff_vp8_luma_dc_wht_dc_neon, export=1
ldrsh r2, [r1]
mov r3, #0
add r2, r2, #3
strh r3, [r1]
asr r2, r2, #3
.rept 16
strh r2, [r0], #32
.endr
bx lr
endfunc
function ff_vp8_idct_add_neon, export=1
vld1.16 {q0-q1}, [r1,:128]
movw r3, #20091
@ -741,23 +729,6 @@ function ff_put_vp8_pixels8_neon, export=1
bx lr
endfunc
function ff_put_vp8_pixels4_neon, export=1
ldr r12, [sp, #0] @ h
push {r4-r6,lr}
1:
subs r12, r12, #4
ldr_post r4, r2, r3
ldr_post r5, r2, r3
ldr_post r6, r2, r3
ldr_post lr, r2, r3
str_post r4, r0, r1
str_post r5, r0, r1
str_post r6, r0, r1
str_post lr, r0, r1
bgt 1b
pop {r4-r6,pc}
endfunc
/* 4/6-tap 8th-pel MC */
.macro vp8_epel8_h6 d, a, b