Commit a0fc780a authored by Janne Grunau's avatar Janne Grunau

arm64: int32_to_float_fmul neon asm

3% faster dts decoding on a cortex-a57.

                                 cortex-a57   cortex-a53
int32_to_float_fmul_array8_c:    1270.9       4475.6
int32_to_float_fmul_array8_neon:  328.6        569.2
int32_to_float_fmul_scalar_c:     928.5       4119.6
int32_to_float_fmul_scalar_neon:  309.1        524.1
parent 705f5e5e
OBJS-$(CONFIG_DCA_DECODER) += aarch64/dcadsp_init.o OBJS-$(CONFIG_DCA_DECODER) += aarch64/dcadsp_init.o
OBJS-$(CONFIG_FFT) += aarch64/fft_init_aarch64.o OBJS-$(CONFIG_FFT) += aarch64/fft_init_aarch64.o
OBJS-$(CONFIG_FMTCONVERT) += aarch64/fmtconvert_init.o
OBJS-$(CONFIG_H264CHROMA) += aarch64/h264chroma_init_aarch64.o OBJS-$(CONFIG_H264CHROMA) += aarch64/h264chroma_init_aarch64.o
OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_init_aarch64.o OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_init_aarch64.o
OBJS-$(CONFIG_H264PRED) += aarch64/h264pred_init.o OBJS-$(CONFIG_H264PRED) += aarch64/h264pred_init.o
...@@ -19,6 +20,7 @@ ARMV8-OBJS-$(CONFIG_VIDEODSP) += aarch64/videodsp.o ...@@ -19,6 +20,7 @@ ARMV8-OBJS-$(CONFIG_VIDEODSP) += aarch64/videodsp.o
NEON-OBJS-$(CONFIG_DCA_DECODER) += aarch64/dcadsp_neon.o \ NEON-OBJS-$(CONFIG_DCA_DECODER) += aarch64/dcadsp_neon.o \
aarch64/synth_filter_neon.o aarch64/synth_filter_neon.o
NEON-OBJS-$(CONFIG_FFT) += aarch64/fft_neon.o NEON-OBJS-$(CONFIG_FFT) += aarch64/fft_neon.o
NEON-OBJS-$(CONFIG_FMTCONVERT) += aarch64/fmtconvert_neon.o
NEON-OBJS-$(CONFIG_H264CHROMA) += aarch64/h264cmc_neon.o NEON-OBJS-$(CONFIG_H264CHROMA) += aarch64/h264cmc_neon.o
NEON-OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_neon.o \ NEON-OBJS-$(CONFIG_H264DSP) += aarch64/h264dsp_neon.o \
aarch64/h264idct_neon.o aarch64/h264idct_neon.o
......
/*
* ARM optimized Format Conversion Utils
*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include <stdint.h>
#include "libavutil/attributes.h"
#include "libavutil/aarch64/cpu.h"
#include "libavcodec/avcodec.h"
#include "libavcodec/fmtconvert.h"
void ff_int32_to_float_fmul_array8_neon(FmtConvertContext *c, float *dst,
const int32_t *src, const float *mul,
int len);
void ff_int32_to_float_fmul_scalar_neon(float *dst, const int32_t *src,
float mul, int len);
av_cold void ff_fmt_convert_init_aarch64(FmtConvertContext *c,
AVCodecContext *avctx)
{
int cpu_flags = av_get_cpu_flags();
if (have_neon(cpu_flags)) {
c->int32_to_float_fmul_array8 = ff_int32_to_float_fmul_array8_neon;
c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_neon;
}
}
/*
* ARM NEON optimised Format Conversion Utils
* Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
* Copyright (c) 2015 Janne Grunau <janne-libav@jannau.net>
*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#include "libavutil/aarch64/asm.S"
function ff_int32_to_float_fmul_scalar_neon, export=1
ld1 {v1.4s,v2.4s}, [x1], #32
scvtf v1.4s, v1.4s
scvtf v2.4s, v2.4s
1:
subs w2, w2, #8
fmul v3.4s, v1.4s, v0.s[0]
fmul v4.4s, v2.4s, v0.s[0]
b.le 2f
ld1 {v1.4s,v2.4s}, [x1], #32
st1 {v3.4s,v4.4s}, [x0], #32
scvtf v1.4s, v1.4s
scvtf v2.4s, v2.4s
b 1b
2:
st1 {v3.4s,v4.4s}, [x0]
ret
endfunc
function ff_int32_to_float_fmul_array8_neon, export=1
lsr w4, w4, #3
subs w5, w4, #1
b.eq 1f
2:
ld1 {v0.4s,v1.4s}, [x2], #32
ld1 {v2.4s,v3.4s}, [x2], #32
scvtf v0.4s, v0.4s
scvtf v1.4s, v1.4s
ld1 {v16.2s}, [x3], #8
scvtf v2.4s, v2.4s
scvtf v3.4s, v3.4s
fmul v4.4s, v0.4s, v16.s[0]
fmul v5.4s, v1.4s, v16.s[0]
fmul v6.4s, v2.4s, v16.s[1]
fmul v7.4s, v3.4s, v16.s[1]
st1 {v4.4s,v5.4s}, [x1], #32
st1 {v6.4s,v7.4s}, [x1], #32
subs w5, w5, #2
b.gt 2b
b.eq 1f
ret
1:
ld1 {v0.4s,v1.4s}, [x2]
ld1 {v16.s}[0], [x3]
scvtf v0.4s, v0.4s
scvtf v1.4s, v1.4s
fmul v4.4s, v0.4s, v16.s[0]
fmul v5.4s, v1.4s, v16.s[0]
st1 {v4.4s,v5.4s}, [x1]
ret
endfunc
...@@ -46,7 +46,12 @@ av_cold void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx) ...@@ -46,7 +46,12 @@ av_cold void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx)
c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_c; c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_c;
c->int32_to_float_fmul_array8 = int32_to_float_fmul_array8_c; c->int32_to_float_fmul_array8 = int32_to_float_fmul_array8_c;
if (ARCH_ARM) ff_fmt_convert_init_arm(c, avctx); if (ARCH_AARCH64)
if (ARCH_PPC) ff_fmt_convert_init_ppc(c, avctx); ff_fmt_convert_init_aarch64(c, avctx);
if (ARCH_X86) ff_fmt_convert_init_x86(c, avctx); if (ARCH_ARM)
ff_fmt_convert_init_arm(c, avctx);
if (ARCH_PPC)
ff_fmt_convert_init_ppc(c, avctx);
if (ARCH_X86)
ff_fmt_convert_init_x86(c, avctx);
} }
...@@ -58,6 +58,7 @@ typedef struct FmtConvertContext { ...@@ -58,6 +58,7 @@ typedef struct FmtConvertContext {
void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx); void ff_fmt_convert_init(FmtConvertContext *c, AVCodecContext *avctx);
void ff_fmt_convert_init_aarch64(FmtConvertContext *c, AVCodecContext *avctx);
void ff_fmt_convert_init_arm(FmtConvertContext *c, AVCodecContext *avctx); void ff_fmt_convert_init_arm(FmtConvertContext *c, AVCodecContext *avctx);
void ff_fmt_convert_init_ppc(FmtConvertContext *c, AVCodecContext *avctx); void ff_fmt_convert_init_ppc(FmtConvertContext *c, AVCodecContext *avctx);
void ff_fmt_convert_init_x86(FmtConvertContext *c, AVCodecContext *avctx); void ff_fmt_convert_init_x86(FmtConvertContext *c, AVCodecContext *avctx);
......
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