Commit 07ff7fa8 authored by Dragan Mrdjan's avatar Dragan Mrdjan Committed by Johann

VP8 optimizations for MIPS dspr2

Signed-off-by: default avatarRaghu Gandham <raghu@mips.com>

Change-Id: I3a8bca425cd3dab746a6328c8fc8843c8e87aea6
parent bb3d510a
......@@ -943,13 +943,16 @@ process_common_toolchain() {
esac
;;
mips*)
CROSS=${CROSS:-mipsel-linux-uclibc-}
link_with_cc=gcc
setup_gnu_toolchain
tune_cflags="-mtune="
if enabled dspr2; then
check_add_cflags -mips32r2 -mdspr2
disable fast_unaligned
fi
check_add_cflags -march=${tgt_isa}
check_add_asflags -march=${tgt_isa}
check_add_asflags -KPIC
check_add_asflags -march=${tgt_isa}
check_add_asflags -KPIC
;;
ppc*)
enable ppc
......
......@@ -278,6 +278,29 @@ EOF
}
mips() {
determine_indirection c $ALL_ARCHS
cat <<EOF
$(common_top)
#include "vpx_config.h"
void ${symbol:-rtcd}(void);
#ifdef RTCD_C
void ${symbol:-rtcd}(void)
{
$(set_function_pointers c)
#if HAVE_DSPR2
void dsputil_static_init();
dsputil_static_init();
#endif
}
#endif
$(common_bottom)
EOF
}
unoptimized() {
determine_indirection c
cat <<EOF
......@@ -309,6 +332,15 @@ case $arch in
require $(filter $REQUIRES)
x86
;;
mips32)
ALL_ARCHS=$(filter mips32)
dspr2=$([ -f "$config_file" ] && eval echo $(grep HAVE_DSPR2 "$config_file"))
HAVE_DSPR2="${dspr2#*=}"
if [ "$HAVE_DSPR2" = "yes" ]; then
ALL_ARCHS=$(filter mips32 dspr2)
fi
mips
;;
armv5te)
ALL_ARCHS=$(filter edsp)
arm
......
......@@ -209,6 +209,7 @@ ARCH_EXT_LIST="
neon
mips32
dspr2
mmx
sse
......
/*
* Copyright (c) 2012 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_config.h"
#include "vpx_rtcd.h"
#include "vpx_mem/vpx_mem.h"
#if HAVE_DSPR2
void vp8_dequant_idct_add_dspr2(short *input, short *dq,
unsigned char *dest, int stride)
{
int i;
for (i = 0; i < 16; i++)
{
input[i] = dq[i] * input[i];
}
vp8_short_idct4x4llm_dspr2(input, dest, stride, dest, stride);
vpx_memset(input, 0, 32);
}
#endif
This source diff could not be displayed because it is too large. You can view the blob instead.
/*
* Copyright (c) 2012 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_config.h"
#include "vpx_rtcd.h"
#if HAVE_DSPR2
void vp8_dequant_idct_add_y_block_dspr2
(short *q, short *dq,
unsigned char *dst, int stride, char *eobs)
{
int i, j;
for (i = 0; i < 4; i++)
{
for (j = 0; j < 4; j++)
{
if (*eobs++ > 1)
vp8_dequant_idct_add_dspr2(q, dq, dst, stride);
else
{
vp8_dc_only_idct_add_dspr2(q[0]*dq[0], dst, stride, dst, stride);
((int *)q)[0] = 0;
}
q += 16;
dst += 4;
}
dst += 4 * stride - 16;
}
}
void vp8_dequant_idct_add_uv_block_dspr2
(short *q, short *dq,
unsigned char *dstu, unsigned char *dstv, int stride, char *eobs)
{
int i, j;
for (i = 0; i < 2; i++)
{
for (j = 0; j < 2; j++)
{
if (*eobs++ > 1)
vp8_dequant_idct_add_dspr2(q, dq, dstu, stride);
else
{
vp8_dc_only_idct_add_dspr2(q[0]*dq[0], dstu, stride, dstu, stride);
((int *)q)[0] = 0;
}
q += 16;
dstu += 4;
}
dstu += 4 * stride - 8;
}
for (i = 0; i < 2; i++)
{
for (j = 0; j < 2; j++)
{
if (*eobs++ > 1)
vp8_dequant_idct_add_dspr2(q, dq, dstv, stride);
else
{
vp8_dc_only_idct_add_dspr2(q[0]*dq[0], dstv, stride, dstv, stride);
((int *)q)[0] = 0;
}
q += 16;
dstv += 4;
}
dstv += 4 * stride - 8;
}
}
#endif
This diff is collapsed.
This diff is collapsed.
/*
* Copyright (c) 2012 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_config.h"
#include "vpx_rtcd.h"
#include "vpx/vpx_integer.h"
#if HAVE_DSPR2
inline void prefetch_load_int(unsigned char *src)
{
__asm__ __volatile__ (
"pref 0, 0(%[src]) \n\t"
:
: [src] "r" (src)
);
}
__inline void vp8_copy_mem16x16_dspr2(
unsigned char *RESTRICT src,
int src_stride,
unsigned char *RESTRICT dst,
int dst_stride)
{
int r;
unsigned int a0, a1, a2, a3;
for (r = 16; r--;)
{
/* load src data in cache memory */
prefetch_load_int(src + src_stride);
/* use unaligned memory load and store */
__asm__ __volatile__ (
"ulw %[a0], 0(%[src]) \n\t"
"ulw %[a1], 4(%[src]) \n\t"
"ulw %[a2], 8(%[src]) \n\t"
"ulw %[a3], 12(%[src]) \n\t"
"sw %[a0], 0(%[dst]) \n\t"
"sw %[a1], 4(%[dst]) \n\t"
"sw %[a2], 8(%[dst]) \n\t"
"sw %[a3], 12(%[dst]) \n\t"
: [a0] "=&r" (a0), [a1] "=&r" (a1),
[a2] "=&r" (a2), [a3] "=&r" (a3)
: [src] "r" (src), [dst] "r" (dst)
);
src += src_stride;
dst += dst_stride;
}
}
__inline void vp8_copy_mem8x8_dspr2(
unsigned char *RESTRICT src,
int src_stride,
unsigned char *RESTRICT dst,
int dst_stride)
{
int r;
unsigned int a0, a1;
/* load src data in cache memory */
prefetch_load_int(src + src_stride);
for (r = 8; r--;)
{
/* use unaligned memory load and store */
__asm__ __volatile__ (
"ulw %[a0], 0(%[src]) \n\t"
"ulw %[a1], 4(%[src]) \n\t"
"sw %[a0], 0(%[dst]) \n\t"
"sw %[a1], 4(%[dst]) \n\t"
: [a0] "=&r" (a0), [a1] "=&r" (a1)
: [src] "r" (src), [dst] "r" (dst)
);
src += src_stride;
dst += dst_stride;
}
}
__inline void vp8_copy_mem8x4_dspr2(
unsigned char *RESTRICT src,
int src_stride,
unsigned char *RESTRICT dst,
int dst_stride)
{
int r;
unsigned int a0, a1;
/* load src data in cache memory */
prefetch_load_int(src + src_stride);
for (r = 4; r--;)
{
/* use unaligned memory load and store */
__asm__ __volatile__ (
"ulw %[a0], 0(%[src]) \n\t"
"ulw %[a1], 4(%[src]) \n\t"
"sw %[a0], 0(%[dst]) \n\t"
"sw %[a1], 4(%[dst]) \n\t"
: [a0] "=&r" (a0), [a1] "=&r" (a1)
: [src] "r" (src), [dst] "r" (dst)
);
src += src_stride;
dst += dst_stride;
}
}
#endif
......@@ -22,35 +22,42 @@ specialize vp8_dequantize_b mmx media neon
vp8_dequantize_b_media=vp8_dequantize_b_v6
prototype void vp8_dequant_idct_add "short *input, short *dq, unsigned char *output, int stride"
specialize vp8_dequant_idct_add mmx media neon
specialize vp8_dequant_idct_add mmx media neon dspr2
vp8_dequant_idct_add_media=vp8_dequant_idct_add_v6
vp8_dequant_idct_add_dspr2=vp8_dequant_idct_add_dspr2
prototype void vp8_dequant_idct_add_y_block "short *q, short *dq, unsigned char *dst, int stride, char *eobs"
specialize vp8_dequant_idct_add_y_block mmx sse2 media neon
specialize vp8_dequant_idct_add_y_block mmx sse2 media neon dspr2
vp8_dequant_idct_add_y_block_media=vp8_dequant_idct_add_y_block_v6
vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2
prototype void vp8_dequant_idct_add_uv_block "short *q, short *dq, unsigned char *dst_u, unsigned char *dst_v, int stride, char *eobs"
specialize vp8_dequant_idct_add_uv_block mmx sse2 media neon
specialize vp8_dequant_idct_add_uv_block mmx sse2 media neon dspr2
vp8_dequant_idct_add_uv_block_media=vp8_dequant_idct_add_uv_block_v6
vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2
#
# Loopfilter
#
prototype void vp8_loop_filter_mbv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_mbv mmx sse2 media neon
specialize vp8_loop_filter_mbv mmx sse2 media neon dspr2
vp8_loop_filter_mbv_media=vp8_loop_filter_mbv_armv6
vp8_loop_filter_mbv_dspr2=vp8_loop_filter_mbv_dspr2
prototype void vp8_loop_filter_bv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_bv mmx sse2 media neon
specialize vp8_loop_filter_bv mmx sse2 media neon dspr2
vp8_loop_filter_bv_media=vp8_loop_filter_bv_armv6
vp8_loop_filter_bv_dspr2=vp8_loop_filter_bv_dspr2
prototype void vp8_loop_filter_mbh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_mbh mmx sse2 media neon
specialize vp8_loop_filter_mbh mmx sse2 media neon dspr2
vp8_loop_filter_mbh_media=vp8_loop_filter_mbh_armv6
vp8_loop_filter_mbh_dspr2=vp8_loop_filter_mbh_dspr2
prototype void vp8_loop_filter_bh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
specialize vp8_loop_filter_bh mmx sse2 media neon
specialize vp8_loop_filter_bh mmx sse2 media neon dspr2
vp8_loop_filter_bh_media=vp8_loop_filter_bh_armv6
vp8_loop_filter_bh_dspr2=vp8_loop_filter_bh_dspr2
prototype void vp8_loop_filter_simple_mbv "unsigned char *y, int ystride, const unsigned char *blimit"
......@@ -90,37 +97,45 @@ vp8_loop_filter_simple_bh_neon=vp8_loop_filter_bhs_neon
#
#idct16
prototype void vp8_short_idct4x4llm "short *input, unsigned char *pred, int pitch, unsigned char *dst, int dst_stride"
specialize vp8_short_idct4x4llm mmx media neon
specialize vp8_short_idct4x4llm mmx media neon dspr2
vp8_short_idct4x4llm_media=vp8_short_idct4x4llm_v6_dual
vp8_short_idct4x4llm_dspr2=vp8_short_idct4x4llm_dspr2
#iwalsh1
prototype void vp8_short_inv_walsh4x4_1 "short *input, short *output"
specialize vp8_short_inv_walsh4x4_1 dspr2
vp8_short_inv_walsh4x4_1_dspr2=vp8_short_inv_walsh4x4_1_dspr2
# no asm yet
#iwalsh16
prototype void vp8_short_inv_walsh4x4 "short *input, short *output"
specialize vp8_short_inv_walsh4x4 mmx sse2 media neon
specialize vp8_short_inv_walsh4x4 mmx sse2 media neon dspr2
vp8_short_inv_walsh4x4_media=vp8_short_inv_walsh4x4_v6
vp8_short_inv_walsh4x4_dspr2=vp8_short_inv_walsh4x4_dspr2
#idct1_scalar_add
prototype void vp8_dc_only_idct_add "short input, unsigned char *pred, int pred_stride, unsigned char *dst, int dst_stride"
specialize vp8_dc_only_idct_add mmx media neon
specialize vp8_dc_only_idct_add mmx media neon dspr2
vp8_dc_only_idct_add_media=vp8_dc_only_idct_add_v6
vp8_dc_only_idct_add_dspr2=vp8_dc_only_idct_add_dspr2
#
# RECON
#
prototype void vp8_copy_mem16x16 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
specialize vp8_copy_mem16x16 mmx sse2 media neon
specialize vp8_copy_mem16x16 mmx sse2 media neon dspr2
vp8_copy_mem16x16_media=vp8_copy_mem16x16_v6
vp8_copy_mem16x16_dspr2=vp8_copy_mem16x16_dspr2
prototype void vp8_copy_mem8x8 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
specialize vp8_copy_mem8x8 mmx media neon
specialize vp8_copy_mem8x8 mmx media neon dspr2
vp8_copy_mem8x8_media=vp8_copy_mem8x8_v6
vp8_copy_mem8x8_dspr2=vp8_copy_mem8x8_dspr2
prototype void vp8_copy_mem8x4 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
specialize vp8_copy_mem8x4 mmx media neon
specialize vp8_copy_mem8x4 mmx media neon dspr2
vp8_copy_mem8x4_media=vp8_copy_mem8x4_v6
vp8_copy_mem8x4_dspr2=vp8_copy_mem8x4_dspr2
prototype void vp8_build_intra_predictors_mby_s "struct macroblockd *x, unsigned char * yabove_row, unsigned char * yleft, int left_stride, unsigned char * ypred_ptr, int y_stride"
specialize vp8_build_intra_predictors_mby_s sse2 ssse3
......@@ -177,20 +192,24 @@ fi
# Subpixel
#
prototype void vp8_sixtap_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_sixtap_predict16x16 mmx sse2 ssse3 media neon
specialize vp8_sixtap_predict16x16 mmx sse2 ssse3 media neon dspr2
vp8_sixtap_predict16x16_media=vp8_sixtap_predict16x16_armv6
vp8_sixtap_predict16x16_dspr2=vp8_sixtap_predict16x16_dspr2
prototype void vp8_sixtap_predict8x8 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_sixtap_predict8x8 mmx sse2 ssse3 media neon
specialize vp8_sixtap_predict8x8 mmx sse2 ssse3 media neon dspr2
vp8_sixtap_predict8x8_media=vp8_sixtap_predict8x8_armv6
vp8_sixtap_predict8x8_dspr2=vp8_sixtap_predict8x8_dspr2
prototype void vp8_sixtap_predict8x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_sixtap_predict8x4 mmx sse2 ssse3 media neon
specialize vp8_sixtap_predict8x4 mmx sse2 ssse3 media neon dspr2
vp8_sixtap_predict8x4_media=vp8_sixtap_predict8x4_armv6
vp8_sixtap_predict8x4_dspr2=vp8_sixtap_predict8x4_dspr2
prototype void vp8_sixtap_predict4x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_sixtap_predict4x4 mmx ssse3 media neon
specialize vp8_sixtap_predict4x4 mmx ssse3 media neon dspr2
vp8_sixtap_predict4x4_media=vp8_sixtap_predict4x4_armv6
vp8_sixtap_predict4x4_dspr2=vp8_sixtap_predict4x4_dspr2
prototype void vp8_bilinear_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon
......
......@@ -119,6 +119,14 @@ ifeq ($(ARCH_X86_64),yes)
VP8_COMMON_SRCS-$(HAVE_SSE2) += common/x86/loopfilter_block_sse2.asm
endif
# common (c)
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/idctllm_dspr2.c
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/filter_dspr2.c
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/loopfilter_filters_dspr2.c
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/reconinter_dspr2.c
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/idct_blk_dspr2.c
VP8_COMMON_SRCS-$(HAVE_DSPR2) += common/mips/dspr2/dequantize_dspr2.c
# common (c)
VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/filter_arm.c
VP8_COMMON_SRCS-$(ARCH_ARM) += common/arm/loopfilter_arm.c
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment