virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-02 03:06:28 +02:00
Code Issues Releases Activity
a543d075cd
FFmpeg/libavcodec/mips/hevcdsp_mmi.c

1146 lines
79 KiB
C
Raw Normal View History

avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
/*
* Copyright (c) 2019 Shiyou Yin (yinshiyou-hf@loongson.cn)
*
* 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
*/
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
#include "libavcodec/hevcdec.h"
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
#include "libavcodec/bit_depth_template.c"
#include "libavcodec/mips/hevcdsp_mips.h"
#include "libavutil/mips/mmiutils.h"
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
#define PUT_HEVC_QPEL_H(w, x_step, src_step, dst_step) \
void ff_hevc_put_hevc_qpel_h##w##_8_mmi(int16_t *dst, uint8_t *_src, \
ptrdiff_t _srcstride, \
int height, intptr_t mx, \
intptr_t my, int width) \
{ \
int x, y; \
pixel *src = (pixel*)_src - 3; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[15]; \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
uint64_t rtmp[1]; \
const int8_t *filter = ff_hevc_qpel_filters[mx - 1]; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
\
x = x_step; \
y = height; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
"punpcklbh %[ftmp7], %[ftmp3], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp3], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp4], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp5], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp6], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp6], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp6], %[ftmp7], %[ftmp8] \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[dst], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDU "%[src], %[src], %[stride] \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_h_8 with mmi. Optimize put_hevc_qpel_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(2.39x to 2.44x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:02 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [rtmp0]"=&r"(rtmp[0]), \
[src]"+&r"(src), [dst]"+&r"(dst), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
}
PUT_HEVC_QPEL_H(4, 1, -4, -8);
PUT_HEVC_QPEL_H(8, 2, -8, -16);
PUT_HEVC_QPEL_H(12, 3, -12, -24);
PUT_HEVC_QPEL_H(16, 4, -16, -32);
PUT_HEVC_QPEL_H(24, 6, -24, -48);
PUT_HEVC_QPEL_H(32, 8, -32, -64);
PUT_HEVC_QPEL_H(48, 12, -48, -96);
PUT_HEVC_QPEL_H(64, 16, -64, -128);
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
#define PUT_HEVC_QPEL_HV(w, x_step, src_step, dst_step) \
void ff_hevc_put_hevc_qpel_hv##w##_8_mmi(int16_t *dst, uint8_t *_src, \
ptrdiff_t _srcstride, \
int height, intptr_t mx, \
intptr_t my, int width) \
{ \
int x, y; \
const int8_t *filter; \
pixel *src = (pixel*)_src; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
int16_t tmp_array[(MAX_PB_SIZE + QPEL_EXTRA) * MAX_PB_SIZE]; \
int16_t *tmp = tmp_array; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[15]; \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
uint64_t rtmp[1]; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
\
src -= (QPEL_EXTRA_BEFORE * srcstride + 3); \
filter = ff_hevc_qpel_filters[mx - 1]; \
x = x_step; \
y = height + QPEL_EXTRA; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
"punpcklbh %[ftmp7], %[ftmp3], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp3], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp4], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp5], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp6], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp6], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp6], %[ftmp7], %[ftmp8] \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #dst_step " \n\t" \
PTR_ADDU "%[src], %[src], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [rtmp0]"=&r"(rtmp[0]), \
[src]"+&r"(src), [tmp]"+&r"(tmp), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
\
tmp = tmp_array + QPEL_EXTRA_BEFORE * 4 -12; \
filter = ff_hevc_qpel_filters[my - 1]; \
x = x_step; \
y = height; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
"li %[rtmp0], 0x06 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp5], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp6], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp7], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp8], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp9], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp10], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
PTR_ADDIU "%[tmp], %[tmp], -0x380 \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
TRANSPOSE_4H(%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
"pmaddhw %[ftmp11], %[ftmp3], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp7], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp4], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp8], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp3], %[ftmp4]) \
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \
"pmaddhw %[ftmp11], %[ftmp5], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp9], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp6], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp10], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp5], %[ftmp6]) \
"paddw %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"packsswh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USDC1(%[ftmp3], %[dst], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x08 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #dst_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x80 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_hv_8 with mmi. Optimize put_hevc_qpel_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11%(1.81x to 2.01x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:24 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]), \
[ftmp12]"=&f"(ftmp[12]), [ftmp13]"=&f"(ftmp[13]), \
[ftmp14]"=&f"(ftmp[14]), [rtmp0]"=&r"(rtmp[0]), \
[dst]"+&r"(dst), [tmp]"+&r"(tmp), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
}
PUT_HEVC_QPEL_HV(4, 1, -4, -8);
PUT_HEVC_QPEL_HV(8, 2, -8, -16);
PUT_HEVC_QPEL_HV(12, 3, -12, -24);
PUT_HEVC_QPEL_HV(16, 4, -16, -32);
PUT_HEVC_QPEL_HV(24, 6, -24, -48);
PUT_HEVC_QPEL_HV(32, 8, -32, -64);
PUT_HEVC_QPEL_HV(48, 12, -48, -96);
PUT_HEVC_QPEL_HV(64, 16, -64, -128);
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
#define PUT_HEVC_QPEL_BI_H(w, x_step, src_step, src2_step, dst_step) \
void ff_hevc_put_hevc_qpel_bi_h##w##_8_mmi(uint8_t *_dst, \
ptrdiff_t _dststride, \
uint8_t *_src, \
ptrdiff_t _srcstride, \
int16_t *src2, int height, \
intptr_t mx, intptr_t my, \
int width) \
{ \
int x, y; \
pixel *src = (pixel*)_src - 3; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
pixel *dst = (pixel *)_dst; \
ptrdiff_t dststride = _dststride / sizeof(pixel); \
const int8_t *filter = ff_hevc_qpel_filters[mx - 1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[20]; \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
uint64_t rtmp[1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
union av_intfloat64 shift; \
union av_intfloat64 offset; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
DECLARE_VAR_LOW32; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
shift.i = 7; \
offset.i = 64; \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
\
x = width >> 2; \
y = height; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
"punpcklhw %[offset], %[offset], %[offset] \n\t" \
"punpcklwd %[offset], %[offset], %[offset] \n\t" \
\
"1: \n\t" \
"li %[x], " #x_step " \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
"punpcklbh %[ftmp7], %[ftmp3], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp3], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp4], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp5], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp6], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp6], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp6], %[ftmp7], %[ftmp8] \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[offset] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[src2], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
"li %[rtmp0], 0x10 \n\t" \
"dmtc1 %[rtmp0], %[ftmp8] \n\t" \
"punpcklhw %[ftmp5], %[ftmp0], %[ftmp3] \n\t" \
"punpckhhw %[ftmp6], %[ftmp0], %[ftmp3] \n\t" \
"punpckhhw %[ftmp3], %[ftmp0], %[ftmp4] \n\t" \
"punpcklhw %[ftmp4], %[ftmp0], %[ftmp4] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp8] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[ftmp8] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp8] \n\t" \
"psraw %[ftmp4], %[ftmp4], %[ftmp8] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[ftmp4] \n\t" \
"paddw %[ftmp6], %[ftmp6], %[ftmp3] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[shift] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[shift] \n\t" \
"packsswh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"pcmpgth %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pand %[ftmp3], %[ftmp5], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
"packushb %[ftmp3], %[ftmp3], %[ftmp3] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USWC1(%[ftmp3], %[dst], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x04 \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDIU "%[src2], %[src2], " #src2_step " \n\t" \
PTR_ADDU "%[src], %[src], %[src_stride] \n\t" \
PTR_ADDU "%[dst], %[dst], %[dst_stride] \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 RESTRICT_ASM_LOW32 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]), \
[ftmp12]"=&f"(ftmp[12]), [src2]"+&r"(src2), \
[dst]"+&r"(dst), [src]"+&r"(src), [y]"+&r"(y), [x]"=&r"(x), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[offset]"+&f"(offset.f), [rtmp0]"=&r"(rtmp[0]) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
: [src_stride]"r"(srcstride), [dst_stride]"r"(dststride), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[filter]"r"(filter), [shift]"f"(shift.f) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_h_8 with mmi. Optimize put_hevc_qpel_bi_h_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.1%(2.34x to 2.39x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:01 +02:00
: "memory" \
); \
}
PUT_HEVC_QPEL_BI_H(4, 1, -4, -8, -4);
PUT_HEVC_QPEL_BI_H(8, 2, -8, -16, -8);
PUT_HEVC_QPEL_BI_H(12, 3, -12, -24, -12);
PUT_HEVC_QPEL_BI_H(16, 4, -16, -32, -16);
PUT_HEVC_QPEL_BI_H(24, 6, -24, -48, -24);
PUT_HEVC_QPEL_BI_H(32, 8, -32, -64, -32);
PUT_HEVC_QPEL_BI_H(48, 12, -48, -96, -48);
PUT_HEVC_QPEL_BI_H(64, 16, -64, -128, -64);
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
#define PUT_HEVC_QPEL_BI_HV(w, x_step, src_step, src2_step, dst_step) \
void ff_hevc_put_hevc_qpel_bi_hv##w##_8_mmi(uint8_t *_dst, \
ptrdiff_t _dststride, \
uint8_t *_src, \
ptrdiff_t _srcstride, \
int16_t *src2, int height, \
intptr_t mx, intptr_t my, \
int width) \
{ \
int x, y; \
const int8_t *filter; \
pixel *src = (pixel*)_src; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
pixel *dst = (pixel *)_dst; \
ptrdiff_t dststride = _dststride / sizeof(pixel); \
int16_t tmp_array[(MAX_PB_SIZE + QPEL_EXTRA) * MAX_PB_SIZE]; \
int16_t *tmp = tmp_array; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[20]; \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
uint64_t rtmp[1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
union av_intfloat64 shift; \
union av_intfloat64 offset; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
DECLARE_VAR_LOW32; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
shift.i = 7; \
offset.i = 64; \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
\
src -= (QPEL_EXTRA_BEFORE * srcstride + 3); \
filter = ff_hevc_qpel_filters[mx - 1]; \
x = width >> 2; \
y = height + QPEL_EXTRA; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
"punpcklbh %[ftmp7], %[ftmp3], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp3], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp4], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp5], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp6], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp6], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp6], %[ftmp7], %[ftmp8] \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #src2_step " \n\t" \
PTR_ADDU "%[src], %[src], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [rtmp0]"=&r"(rtmp[0]), \
[src]"+&r"(src), [tmp]"+&r"(tmp), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
\
tmp = tmp_array; \
filter = ff_hevc_qpel_filters[my - 1]; \
x = width >> 2; \
y = height; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
"li %[rtmp0], 0x06 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpcklwd %[offset], %[offset], %[offset] \n\t" \
\
"1: \n\t" \
"li %[x], " #x_step " \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp5], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp6], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp7], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp8], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp9], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp10], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
PTR_ADDIU "%[tmp], %[tmp], -0x380 \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
TRANSPOSE_4H(%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
"pmaddhw %[ftmp11], %[ftmp3], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp7], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp4], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp8], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp3], %[ftmp4]) \
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \
"pmaddhw %[ftmp11], %[ftmp5], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp9], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp6], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp10], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp5], %[ftmp6]) \
"paddw %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"packsswh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[src2], 0x00) \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp7], %[ftmp7], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
"li %[rtmp0], 0x10 \n\t" \
"dmtc1 %[rtmp0], %[ftmp8] \n\t" \
"punpcklhw %[ftmp5], %[ftmp7], %[ftmp3] \n\t" \
"punpckhhw %[ftmp6], %[ftmp7], %[ftmp3] \n\t" \
"punpckhhw %[ftmp3], %[ftmp7], %[ftmp4] \n\t" \
"punpcklhw %[ftmp4], %[ftmp7], %[ftmp4] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp8] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[ftmp8] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp8] \n\t" \
"psraw %[ftmp4], %[ftmp4], %[ftmp8] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[ftmp4] \n\t" \
"paddw %[ftmp6], %[ftmp6], %[ftmp3] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[offset] \n\t" \
"paddw %[ftmp6], %[ftmp6], %[offset] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[shift] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[shift] \n\t" \
"packsswh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"pcmpgth %[ftmp7], %[ftmp5], %[ftmp7] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pand %[ftmp3], %[ftmp5], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
"packushb %[ftmp3], %[ftmp3], %[ftmp3] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USWC1(%[ftmp3], %[dst], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x08 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x04 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
PTR_ADDIU "%[src2], %[src2], " #src2_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #src2_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x80 \n\t" \
PTR_ADDU "%[dst], %[dst], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 RESTRICT_ASM_LOW32 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]), \
[ftmp12]"=&f"(ftmp[12]), [ftmp13]"=&f"(ftmp[13]), \
[ftmp14]"=&f"(ftmp[14]), [src2]"+&r"(src2), \
[dst]"+&r"(dst), [tmp]"+&r"(tmp), [y]"+&r"(y), [x]"=&r"(x), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[offset]"+&f"(offset.f), [rtmp0]"=&r"(rtmp[0]) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
: [filter]"r"(filter), [stride]"r"(dststride), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[shift]"f"(shift.f) \
avcodec/mips: [loongson] optimize put_hevc_qpel_bi_hv_8 with mmi. Optimize put_hevc_qpel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 11.4%(2.01x to 2.24x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-21 12:10:25 +02:00
: "memory" \
); \
}
PUT_HEVC_QPEL_BI_HV(4, 1, -4, -8, -4);
PUT_HEVC_QPEL_BI_HV(8, 2, -8, -16, -8);
PUT_HEVC_QPEL_BI_HV(12, 3, -12, -24, -12);
PUT_HEVC_QPEL_BI_HV(16, 4, -16, -32, -16);
PUT_HEVC_QPEL_BI_HV(24, 6, -24, -48, -24);
PUT_HEVC_QPEL_BI_HV(32, 8, -32, -64, -32);
PUT_HEVC_QPEL_BI_HV(48, 12, -48, -96, -48);
PUT_HEVC_QPEL_BI_HV(64, 16, -64, -128, -64);
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
#define PUT_HEVC_EPEL_BI_HV(w, x_step, src_step, src2_step, dst_step) \
void ff_hevc_put_hevc_epel_bi_hv##w##_8_mmi(uint8_t *_dst, \
ptrdiff_t _dststride, \
uint8_t *_src, \
ptrdiff_t _srcstride, \
int16_t *src2, int height, \
intptr_t mx, intptr_t my, \
int width) \
{ \
int x, y; \
pixel *src = (pixel *)_src; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
pixel *dst = (pixel *)_dst; \
ptrdiff_t dststride = _dststride / sizeof(pixel); \
const int8_t *filter = ff_hevc_epel_filters[mx - 1]; \
int16_t tmp_array[(MAX_PB_SIZE + EPEL_EXTRA) * MAX_PB_SIZE]; \
int16_t *tmp = tmp_array; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[12]; \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
uint64_t rtmp[1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
union av_intfloat64 shift; \
union av_intfloat64 offset; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
DECLARE_VAR_LOW32; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
shift.i = 7; \
offset.i = 64; \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
\
src -= (EPEL_EXTRA_BEFORE * srcstride + 1); \
x = width >> 2; \
y = height + EPEL_EXTRA; \
__asm__ volatile( \
MMI_LWC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
"punpcklbh %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
"pmullh %[ftmp2], %[ftmp2], %[ftmp1] \n\t" \
"punpcklbh %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp3], %[ftmp3], %[ftmp1] \n\t" \
"punpcklbh %[ftmp4], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp4], %[ftmp4], %[ftmp1] \n\t" \
"punpcklbh %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp5], %[ftmp5], %[ftmp1] \n\t" \
TRANSPOSE_4H(%[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5], \
%[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9]) \
"paddh %[ftmp2], %[ftmp2], %[ftmp3] \n\t" \
"paddh %[ftmp4], %[ftmp4], %[ftmp5] \n\t" \
"paddh %[ftmp2], %[ftmp2], %[ftmp4] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp2], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #src2_step " \n\t" \
PTR_ADDU "%[src], %[src], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[rtmp0]"=&r"(rtmp[0]), \
[src]"+&r"(src), [tmp]"+&r"(tmp), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
\
tmp = tmp_array; \
filter = ff_hevc_epel_filters[my - 1]; \
x = width >> 2; \
y = height; \
__asm__ volatile( \
MMI_LWC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"li %[rtmp0], 0x06 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpcklwd %[offset], %[offset], %[offset] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp2], %[ftmp2], %[ftmp2] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
\
"1: \n\t" \
"li %[x], " #x_step " \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp5], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp6], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
PTR_ADDIU "%[tmp], %[tmp], -0x180 \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"pmaddhw %[ftmp7], %[ftmp3], %[ftmp1] \n\t" \
"pmaddhw %[ftmp8], %[ftmp4], %[ftmp1] \n\t" \
TRANSPOSE_2W(%[ftmp7], %[ftmp8], %[ftmp3], %[ftmp4]) \
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \
"pmaddhw %[ftmp7], %[ftmp5], %[ftmp1] \n\t" \
"pmaddhw %[ftmp8], %[ftmp6], %[ftmp1] \n\t" \
TRANSPOSE_2W(%[ftmp7], %[ftmp8], %[ftmp5], %[ftmp6]) \
"paddw %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"packsswh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[tmp], 0x02) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
"li %[rtmp0], 0x10 \n\t" \
"dmtc1 %[rtmp0], %[ftmp8] \n\t" \
"punpcklhw %[ftmp5], %[ftmp2], %[ftmp3] \n\t" \
"punpckhhw %[ftmp6], %[ftmp2], %[ftmp3] \n\t" \
"punpckhhw %[ftmp3], %[ftmp2], %[ftmp4] \n\t" \
"punpcklhw %[ftmp4], %[ftmp2], %[ftmp4] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp8] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[ftmp8] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp8] \n\t" \
"psraw %[ftmp4], %[ftmp4], %[ftmp8] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[ftmp4] \n\t" \
"paddw %[ftmp6], %[ftmp6], %[ftmp3] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[offset] \n\t" \
"paddw %[ftmp6], %[ftmp6], %[offset] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[shift] \n\t" \
"psraw %[ftmp6], %[ftmp6], %[shift] \n\t" \
"packsswh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"pcmpgth %[ftmp7], %[ftmp5], %[ftmp2] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pand %[ftmp3], %[ftmp5], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
"packushb %[ftmp3], %[ftmp3], %[ftmp3] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USWC1(%[ftmp3], %[dst], 0x0) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x08 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x04 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
PTR_ADDIU "%[src2], %[src2], " #src2_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #src2_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x80 \n\t" \
PTR_ADDU "%[dst], %[dst], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_LOW32 RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [src2]"+&r"(src2), \
[dst]"+&r"(dst), [tmp]"+&r"(tmp), [y]"+&r"(y), [x]"=&r"(x), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[offset]"+&f"(offset.f), [rtmp0]"=&r"(rtmp[0]) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
: [filter]"r"(filter), [stride]"r"(dststride), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[shift]"f"(shift.f) \
avcodec/mips: [loongson] optimize put_hevc_epel_bi_hv_8 with mmi. Optimize put_hevc_epel_bi_hv_8 with mmi in the case width=4/8/12/16/24/32. This optimization improved HEVC decoding performance 1.7%(2.30x to 2.34x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:05:00 +02:00
: "memory" \
); \
}
PUT_HEVC_EPEL_BI_HV(4, 1, -4, -8, -4);
PUT_HEVC_EPEL_BI_HV(8, 2, -8, -16, -8);
PUT_HEVC_EPEL_BI_HV(12, 3, -12, -24, -12);
PUT_HEVC_EPEL_BI_HV(16, 4, -16, -32, -16);
PUT_HEVC_EPEL_BI_HV(24, 6, -24, -48, -24);
PUT_HEVC_EPEL_BI_HV(32, 8, -32, -64, -32);
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
#define PUT_HEVC_PEL_BI_PIXELS(w, x_step, src_step, dst_step, src2_step) \
void ff_hevc_put_hevc_pel_bi_pixels##w##_8_mmi(uint8_t *_dst, \
ptrdiff_t _dststride, \
uint8_t *_src, \
ptrdiff_t _srcstride, \
int16_t *src2, int height, \
intptr_t mx, intptr_t my, \
int width) \
{ \
int x, y; \
pixel *src = (pixel *)_src; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
pixel *dst = (pixel *)_dst; \
ptrdiff_t dststride = _dststride / sizeof(pixel); \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[12]; \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
uint64_t rtmp[1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
union av_intfloat64 shift; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
shift.i = 7; \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
\
y = height; \
x = width >> 3; \
__asm__ volatile( \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
"li %[rtmp0], 0x06 \n\t" \
"dmtc1 %[rtmp0], %[ftmp1] \n\t" \
"li %[rtmp0], 0x10 \n\t" \
"dmtc1 %[rtmp0], %[ftmp10] \n\t" \
"li %[rtmp0], 0x40 \n\t" \
"dmtc1 %[rtmp0], %[offset] \n\t" \
"punpcklhw %[offset], %[offset], %[offset] \n\t" \
"punpcklwd %[offset], %[offset], %[offset] \n\t" \
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp5], %[src], 0x00) \
MMI_ULDC1(%[ftmp2], %[src2], 0x00) \
MMI_ULDC1(%[ftmp3], %[src2], 0x08) \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
"punpcklbh %[ftmp4], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"psllh %[ftmp4], %[ftmp4], %[ftmp1] \n\t" \
"psllh %[ftmp5], %[ftmp5], %[ftmp1] \n\t" \
"paddh %[ftmp4], %[ftmp4], %[offset] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[offset] \n\t" \
"punpcklhw %[ftmp6], %[ftmp4], %[ftmp0] \n\t" \
"punpckhhw %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpcklhw %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"punpckhhw %[ftmp9], %[ftmp5], %[ftmp0] \n\t" \
"punpcklhw %[ftmp4], %[ftmp0], %[ftmp3] \n\t" \
"punpckhhw %[ftmp5], %[ftmp0], %[ftmp3] \n\t" \
"punpckhhw %[ftmp3], %[ftmp0], %[ftmp2] \n\t" \
"punpcklhw %[ftmp2], %[ftmp0], %[ftmp2] \n\t" \
"psraw %[ftmp2], %[ftmp2], %[ftmp10] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp10] \n\t" \
"psraw %[ftmp4], %[ftmp4], %[ftmp10] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp10] \n\t" \
"paddw %[ftmp2], %[ftmp2], %[ftmp6] \n\t" \
"paddw %[ftmp3], %[ftmp3], %[ftmp7] \n\t" \
"paddw %[ftmp4], %[ftmp4], %[ftmp8] \n\t" \
"paddw %[ftmp5], %[ftmp5], %[ftmp9] \n\t" \
"psraw %[ftmp2], %[ftmp2], %[shift] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[shift] \n\t" \
"psraw %[ftmp4], %[ftmp4], %[shift] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[shift] \n\t" \
"packsswh %[ftmp2], %[ftmp2], %[ftmp3] \n\t" \
"packsswh %[ftmp4], %[ftmp4], %[ftmp5] \n\t" \
"pcmpgth %[ftmp3], %[ftmp2], %[ftmp0] \n\t" \
"pcmpgth %[ftmp5], %[ftmp4], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pand %[ftmp2], %[ftmp2], %[ftmp3] \n\t" \
"pand %[ftmp4], %[ftmp4], %[ftmp5] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
"packushb %[ftmp2], %[ftmp2], %[ftmp4] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USDC1(%[ftmp2], %[dst], 0x0) \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x08 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x08 \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x10 \n\t" \
"bnez %[x], 2b \n\t" \
\
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDIU "%[src2], %[src2], " #src2_step " \n\t" \
"li %[x], " #x_step " \n\t" \
"daddi %[y], %[y], -0x01 \n\t" \
PTR_ADDU "%[src], %[src], %[srcstride] \n\t" \
PTR_ADDU "%[dst], %[dst], %[dststride] \n\t" \
PTR_ADDIU "%[src2], %[src2], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [offset]"=&f"(ftmp[11]), \
[src2]"+&r"(src2), [dst]"+&r"(dst), [src]"+&r"(src), \
[x]"+&r"(x), [y]"+&r"(y), [rtmp0]"=&r"(rtmp[0]) \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
: [dststride]"r"(dststride), [shift]"f"(shift.f), \
avcodec/mips: [loongson] optimize put_hevc_pel_bi_pixels_8 with mmi. Optimize put_hevc_pel_bi_pixels_8 with mmi in the case width=8/16/24/32/48/64. This optimization improved HEVC decoding performance 2%(1.77x to 1.81x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-01-18 08:42:20 +02:00
[srcstride]"r"(srcstride) \
: "memory" \
); \
} \
PUT_HEVC_PEL_BI_PIXELS(8, 1, -8, -8, -16);
PUT_HEVC_PEL_BI_PIXELS(16, 2, -16, -16, -32);
PUT_HEVC_PEL_BI_PIXELS(24, 3, -24, -24, -48);
PUT_HEVC_PEL_BI_PIXELS(32, 4, -32, -32, -64);
PUT_HEVC_PEL_BI_PIXELS(48, 6, -48, -48, -96);
PUT_HEVC_PEL_BI_PIXELS(64, 8, -64, -64, -128);
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
#define PUT_HEVC_QPEL_UNI_HV(w, x_step, src_step, dst_step, tmp_step) \
void ff_hevc_put_hevc_qpel_uni_hv##w##_8_mmi(uint8_t *_dst, \
ptrdiff_t _dststride, \
uint8_t *_src, \
ptrdiff_t _srcstride, \
int height, \
intptr_t mx, intptr_t my, \
int width) \
{ \
int x, y; \
const int8_t *filter; \
pixel *src = (pixel*)_src; \
ptrdiff_t srcstride = _srcstride / sizeof(pixel); \
pixel *dst = (pixel *)_dst; \
ptrdiff_t dststride = _dststride / sizeof(pixel); \
int16_t tmp_array[(MAX_PB_SIZE + QPEL_EXTRA) * MAX_PB_SIZE]; \
int16_t *tmp = tmp_array; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
double ftmp[20]; \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
uint64_t rtmp[1]; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
union av_intfloat64 shift; \
union av_intfloat64 offset; \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
DECLARE_VAR_ALL64; \
DECLARE_VAR_LOW32; \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
shift.i = 6; \
offset.i = 32; \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
\
src -= (QPEL_EXTRA_BEFORE * srcstride + 3); \
filter = ff_hevc_qpel_filters[mx - 1]; \
x = width >> 2; \
y = height + QPEL_EXTRA; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
\
"1: \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[src], 0x00) \
MMI_ULDC1(%[ftmp4], %[src], 0x01) \
MMI_ULDC1(%[ftmp5], %[src], 0x02) \
MMI_ULDC1(%[ftmp6], %[src], 0x03) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
"punpcklbh %[ftmp7], %[ftmp3], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp3], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp3], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp4], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp4], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp4], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp5], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp5], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp5], %[ftmp7], %[ftmp8] \n\t" \
"punpcklbh %[ftmp7], %[ftmp6], %[ftmp0] \n\t" \
"punpckhbh %[ftmp8], %[ftmp6], %[ftmp0] \n\t" \
"pmullh %[ftmp7], %[ftmp7], %[ftmp1] \n\t" \
"pmullh %[ftmp8], %[ftmp8], %[ftmp2] \n\t" \
"paddh %[ftmp6], %[ftmp7], %[ftmp8] \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10]) \
"paddh %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"paddh %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USDC1(%[ftmp3], %[tmp], 0x0) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[src], %[src], 0x04 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
"li %[x], " #x_step " \n\t" \
PTR_ADDIU "%[src], %[src], " #src_step " \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #tmp_step " \n\t" \
PTR_ADDU "%[src], %[src], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [rtmp0]"=&r"(rtmp[0]), \
[src]"+&r"(src), [tmp]"+&r"(tmp), [y]"+&r"(y), \
[x]"+&r"(x) \
: [filter]"r"(filter), [stride]"r"(srcstride) \
: "memory" \
); \
\
tmp = tmp_array; \
filter = ff_hevc_qpel_filters[my - 1]; \
x = width >> 2; \
y = height; \
__asm__ volatile( \
MMI_LDC1(%[ftmp1], %[filter], 0x00) \
"li %[rtmp0], 0x08 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpckhbh %[ftmp2], %[ftmp0], %[ftmp1] \n\t" \
"punpcklbh %[ftmp1], %[ftmp0], %[ftmp1] \n\t" \
"psrah %[ftmp1], %[ftmp1], %[ftmp0] \n\t" \
"psrah %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
"li %[rtmp0], 0x06 \n\t" \
"dmtc1 %[rtmp0], %[ftmp0] \n\t" \
"punpcklhw %[offset], %[offset], %[offset] \n\t" \
"punpcklwd %[offset], %[offset], %[offset] \n\t" \
\
"1: \n\t" \
"li %[x], " #x_step " \n\t" \
"2: \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp3], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp4], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp5], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp6], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp7], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp8], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp9], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_ULDC1(%[ftmp10], %[tmp], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
PTR_ADDIU "%[tmp], %[tmp], -0x380 \n\t" \
TRANSPOSE_4H(%[ftmp3], %[ftmp4], %[ftmp5], %[ftmp6], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
TRANSPOSE_4H(%[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10], \
%[ftmp11], %[ftmp12], %[ftmp13], %[ftmp14]) \
"pmaddhw %[ftmp11], %[ftmp3], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp7], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp4], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp8], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp3], %[ftmp4]) \
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t" \
"psraw %[ftmp3], %[ftmp3], %[ftmp0] \n\t" \
"pmaddhw %[ftmp11], %[ftmp5], %[ftmp1] \n\t" \
"pmaddhw %[ftmp12], %[ftmp9], %[ftmp2] \n\t" \
"pmaddhw %[ftmp13], %[ftmp6], %[ftmp1] \n\t" \
"pmaddhw %[ftmp14], %[ftmp10], %[ftmp2] \n\t" \
"paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t" \
"paddw %[ftmp13], %[ftmp13], %[ftmp14] \n\t" \
TRANSPOSE_2W(%[ftmp11], %[ftmp13], %[ftmp5], %[ftmp6]) \
"paddw %[ftmp5], %[ftmp5], %[ftmp6] \n\t" \
"psraw %[ftmp5], %[ftmp5], %[ftmp0] \n\t" \
"packsswh %[ftmp3], %[ftmp3], %[ftmp5] \n\t" \
"paddh %[ftmp3], %[ftmp3], %[offset] \n\t" \
"psrah %[ftmp3], %[ftmp3], %[shift] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pxor %[ftmp7], %[ftmp7], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
"pcmpgth %[ftmp7], %[ftmp3], %[ftmp7] \n\t" \
libavcodec/mips: Fix specification of instruction name 1.'xor,or,and' to 'pxor,por,pand'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. 2.'dsrl,srl' to 'ssrld,ssrlw'. In the case of operating FPR, gcc supports both of them, clang only supports the second type. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-05-28 04:04:39 +02:00
"pand %[ftmp3], %[ftmp3], %[ftmp7] \n\t" \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
"packushb %[ftmp3], %[ftmp3], %[ftmp3] \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
MMI_USWC1(%[ftmp3], %[dst], 0x00) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
\
"daddi %[x], %[x], -0x01 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x08 \n\t" \
PTR_ADDIU "%[dst], %[dst], 0x04 \n\t" \
"bnez %[x], 2b \n\t" \
\
"daddi %[y], %[y], -0x01 \n\t" \
PTR_ADDIU "%[tmp], %[tmp], " #tmp_step " \n\t" \
PTR_ADDIU "%[dst], %[dst], " #dst_step " \n\t" \
PTR_ADDU "%[dst], %[dst], %[stride] \n\t" \
PTR_ADDIU "%[tmp], %[tmp], 0x80 \n\t" \
"bnez %[y], 1b \n\t" \
avcodec/mips: Use MMI marcos to replace Loongson3 instructions Loongson3's extention instructions (prefixed with gs) are widely used in our MMI codebase. However, these instructions are not avilable on Loongson-2E/F while MMI code should work on these processors. Previously we introduced mmiutils marcos to provide backward compactbility but newly commited code didn't follow that. In this patch I revised the codebase and converted all these instructions into MMI marcos to get Loongson2 supproted again. Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-07-23 07:53:42 +02:00
: RESTRICT_ASM_ALL64 RESTRICT_ASM_LOW32 \
[ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]), \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
[ftmp2]"=&f"(ftmp[2]), [ftmp3]"=&f"(ftmp[3]), \
[ftmp4]"=&f"(ftmp[4]), [ftmp5]"=&f"(ftmp[5]), \
[ftmp6]"=&f"(ftmp[6]), [ftmp7]"=&f"(ftmp[7]), \
[ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]), \
[ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]), \
[ftmp12]"=&f"(ftmp[12]), [ftmp13]"=&f"(ftmp[13]), \
[ftmp14]"=&f"(ftmp[14]), \
[dst]"+&r"(dst), [tmp]"+&r"(tmp), [y]"+&r"(y), [x]"=&r"(x), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[offset]"+&f"(offset.f), [rtmp0]"=&r"(rtmp[0]) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
: [filter]"r"(filter), [stride]"r"(dststride), \
libavcodec/mips: Fix build errors reported by clang Clang is more strict on the type of asm operands, float or double type variable should use constraint 'f', integer variable should use constraint 'r'. Signed-off-by: Jin Bo <jinbo@loongson.cn> Reviewed-by: yinshiyou-hf@loongson.cn Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2021-06-01 08:22:09 +02:00
[shift]"f"(shift.f) \
avcodec/mips: [loongson] optimize put_hevc_qpel_uni_hv_8 with mmi. Optimize put_hevc_qpel_uni_hv_8 with mmi in the case width=4/8/12/16/24/32/48/64. This optimization improved HEVC decoding performance 2.7%(2.24x to 2.30x, tested on loongson 3A3000). Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2019-02-01 08:04:59 +02:00
: "memory" \
); \
}
PUT_HEVC_QPEL_UNI_HV(4, 1, -4, -4, -8);
PUT_HEVC_QPEL_UNI_HV(8, 2, -8, -8, -16);
PUT_HEVC_QPEL_UNI_HV(12, 3, -12, -12, -24);
PUT_HEVC_QPEL_UNI_HV(16, 4, -16, -16, -32);
PUT_HEVC_QPEL_UNI_HV(24, 6, -24, -24, -48);
PUT_HEVC_QPEL_UNI_HV(32, 8, -32, -32, -64);
PUT_HEVC_QPEL_UNI_HV(48, 12, -48, -48, -96);
PUT_HEVC_QPEL_UNI_HV(64, 16, -64, -64, -128);
Reference in New Issue Copy Permalink