From 95418634b1e8ff5167952727c1f26ae4d4d6d797 Mon Sep 17 00:00:00 2001 From: ZhouXiaoyong Date: Thu, 9 Jul 2015 14:34:21 +0800 Subject: [PATCH] avcodec: loongson optimized mpegvideo dct unquantize with mmi Signed-off-by: ZhouXiaoyong Signed-off-by: Michael Niedermayer --- libavcodec/mips/Makefile | 1 + libavcodec/mips/mpegvideo_init_mips.c | 14 ++ libavcodec/mips/mpegvideo_mips.h | 35 +++ libavcodec/mips/mpegvideo_mmi.c | 303 ++++++++++++++++++++++++++ 4 files changed, 353 insertions(+) create mode 100644 libavcodec/mips/mpegvideo_mips.h create mode 100644 libavcodec/mips/mpegvideo_mmi.c diff --git a/libavcodec/mips/Makefile b/libavcodec/mips/Makefile index c16c93a72f..9eb815a6d0 100644 --- a/libavcodec/mips/Makefile +++ b/libavcodec/mips/Makefile @@ -59,3 +59,4 @@ MSA-OBJS-$(CONFIG_ME_CMP) += mips/me_cmp_msa.o MMI-OBJS-$(CONFIG_H264DSP) += mips/h264dsp_mmi.o MMI-OBJS-$(CONFIG_H264CHROMA) += mips/h264chroma_mmi.o MMI-OBJS-$(CONFIG_H264PRED) += mips/h264pred_mmi.o +MMI-OBJS-$(CONFIG_MPEGVIDEO) += mips/mpegvideo_mmi.o diff --git a/libavcodec/mips/mpegvideo_init_mips.c b/libavcodec/mips/mpegvideo_init_mips.c index ee14b3198d..85a833cf33 100644 --- a/libavcodec/mips/mpegvideo_init_mips.c +++ b/libavcodec/mips/mpegvideo_init_mips.c @@ -19,6 +19,7 @@ */ #include "h263dsp_mips.h" +#include "mpegvideo_mips.h" #if HAVE_MSA static av_cold void dct_unquantize_init_msa(MpegEncContext *s) @@ -29,9 +30,22 @@ static av_cold void dct_unquantize_init_msa(MpegEncContext *s) } #endif // #if HAVE_MSA +#if HAVE_MMI +static av_cold void dct_unquantize_init_mmi(MpegEncContext *s) +{ + s->dct_unquantize_h263_intra = ff_dct_unquantize_h263_intra_mmi; + s->dct_unquantize_h263_inter = ff_dct_unquantize_h263_inter_mmi; + s->dct_unquantize_mpeg1_intra = ff_dct_unquantize_mpeg1_intra_mmi; + s->dct_unquantize_mpeg1_inter = ff_dct_unquantize_mpeg1_inter_mmi; +} +#endif /* HAVE_MMI */ + av_cold void ff_mpv_common_init_mips(MpegEncContext *s) { #if HAVE_MSA dct_unquantize_init_msa(s); #endif // #if HAVE_MSA +#if HAVE_MMI + dct_unquantize_init_mmi(s); +#endif /* HAVE_MMI */ } diff --git a/libavcodec/mips/mpegvideo_mips.h b/libavcodec/mips/mpegvideo_mips.h new file mode 100644 index 0000000000..dbcea6acde --- /dev/null +++ b/libavcodec/mips/mpegvideo_mips.h @@ -0,0 +1,35 @@ +/* + * Copyright (c) 2015 Zhou Xiaoyong + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef MPEGVIDEO_MIPS_H +#define MPEGVIDEO_MIPS_H + +#include "libavcodec/mpegvideo.h" + +void ff_dct_unquantize_h263_intra_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale); +void ff_dct_unquantize_h263_inter_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale); +void ff_dct_unquantize_mpeg1_intra_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale); +void ff_dct_unquantize_mpeg1_inter_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale); + +#endif /* MPEGVIDEO_MIPS_H */ diff --git a/libavcodec/mips/mpegvideo_mmi.c b/libavcodec/mips/mpegvideo_mmi.c new file mode 100644 index 0000000000..b3d58bd32c --- /dev/null +++ b/libavcodec/mips/mpegvideo_mmi.c @@ -0,0 +1,303 @@ +/* + * Loongson SIMD optimized mpegvideo + * + * Copyright (c) 2015 Loongson Technology Corporation Limited + * Copyright (c) 2015 Zhou Xiaoyong + * Zhang Shuangshuang + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "mpegvideo_mips.h" + +void ff_dct_unquantize_h263_intra_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale) +{ + int64_t level, qmul, qadd, nCoeffs; + + qmul = qscale << 1; + assert(s->block_last_index[n]>=0 || s->h263_aic); + + if (!s->h263_aic) { + if (n<4) + level = block[0] * s->y_dc_scale; + else + level = block[0] * s->c_dc_scale; + qadd = (qscale-1) | 1; + } else { + qadd = 0; + level = block[0]; + } + + if(s->ac_pred) + nCoeffs = 63; + else + nCoeffs = s->inter_scantable.raster_end[s->block_last_index[n]]; + + __asm__ volatile ( + "xor $f12, $f12, $f12 \r\n" + "lwc1 $f12, %1 \n\r" + "xor $f10, $f10, $f10 \r\n" + "lwc1 $f10, %2 \r\n" + "xor $f14, $f14, $f14 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "packsswh $f10, $f10, $f10 \r\n" + "packsswh $f10, $f10, $f10 \r\n" + "psubh $f14, $f14, $f10 \r\n" + "xor $f8, $f8, $f8 \r\n" + ".p2align 4 \r\n" + "1: \r\n" + "daddu $8, %0, %3 \r\n" + "gsldlc1 $f0, 7($8) \r\n" + "gsldrc1 $f0, 0($8) \r\n" + "gsldlc1 $f2, 15($8) \r\n" + "gsldrc1 $f2, 8($8) \r\n" + "mov.d $f4, $f0 \r\n" + "mov.d $f6, $f2 \r\n" + "pmullh $f0, $f0, $f12 \r\n" + "pmullh $f2, $f2, $f12 \r\n" + "pcmpgth $f4, $f4, $f8 \r\n" + "pcmpgth $f6, $f6, $f8 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "paddh $f0, $f0, $f14 \r\n" + "paddh $f2, $f2, $f14 \r\n" + "xor $f4, $f4, $f0 \r\n" + "xor $f6, $f6, $f2 \r\n" + "pcmpeqh $f0, $f0, $f14 \r\n" + "pcmpeqh $f2, $f2, $f14 \r\n" + "pandn $f0, $f0, $f4 \r\n" + "pandn $f2, $f2, $f6 \r\n" + "gssdlc1 $f0, 7($8) \r\n" + "gssdrc1 $f0, 0($8) \r\n" + "gssdlc1 $f2, 15($8) \r\n" + "gssdrc1 $f2, 8($8) \r\n" + "addi %3, %3, 16 \r\n" + "blez %3, 1b \r\n" + ::"r"(block+nCoeffs),"m"(qmul),"m"(qadd),"r"(2*(-nCoeffs)) + :"$8","memory" + ); + + block[0] = level; +} + +void ff_dct_unquantize_h263_inter_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale) +{ + int64_t qmul, qadd, nCoeffs; + + qmul = qscale << 1; + qadd = (qscale - 1) | 1; + assert(s->block_last_index[n]>=0 || s->h263_aic); + nCoeffs = s->inter_scantable.raster_end[s->block_last_index[n]]; + + __asm__ volatile ( + "xor $f12, $f12, $f12 \r\n" + "lwc1 $f12, %1 \r\n" + "xor $f10, $f10, $f10 \r\n" + "lwc1 $f10, %2 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "xor $f14, $f14, $f14 \r\n" + "packsswh $f10, $f10, $f10 \r\n" + "packsswh $f10, $f10, $f10 \r\n" + "psubh $f14, $f14, $f10 \r\n" + "xor $f8, $f8, $f8 \r\n" + ".p2align 4 \r\n" + "1: \r\n" + "daddu $8, %0, %3 \r\n" + "gsldlc1 $f0, 7($8) \r\n" + "gsldrc1 $f0, 0($8) \r\n" + "gsldlc1 $f2, 15($8) \r\n" + "gsldrc1 $f2, 8($8) \r\n" + "mov.d $f4, $f0 \r\n" + "mov.d $f6, $f2 \r\n" + "pmullh $f0, $f0, $f12 \r\n" + "pmullh $f2, $f2, $f12 \r\n" + "pcmpgth $f4, $f4, $f8 \r\n" + "pcmpgth $f6, $f6, $f8 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "paddh $f0, $f0, $f14 \r\n" + "paddh $f2, $f2, $f14 \r\n" + "xor $f4, $f4, $f0 \r\n" + "xor $f6, $f6, $f2 \r\n" + "pcmpeqh $f0, $f0, $f14 \r\n" + "pcmpeqh $f2, $f2, $f14 \r\n" + "pandn $f0, $f0, $f4 \r\n" + "pandn $f2, $f2, $f6 \r\n" + "gssdlc1 $f0, 7($8) \r\n" + "gssdrc1 $f0, 0($8) \r\n" + "gssdlc1 $f2, 15($8) \r\n" + "gssdrc1 $f2, 8($8) \r\n" + "addi %3, %3, 16 \r\n" + "blez %3, 1b \r\n" + ::"r"(block+nCoeffs),"m"(qmul),"m"(qadd),"r"(2*(-nCoeffs)) + : "$8","memory" + ); +} + +void ff_dct_unquantize_mpeg1_intra_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale) +{ + int64_t nCoeffs; + const uint16_t *quant_matrix; + int block0; + + assert(s->block_last_index[n]>=0); + nCoeffs = s->intra_scantable.raster_end[s->block_last_index[n]] + 1; + + if (n<4) + block0 = block[0] * s->y_dc_scale; + else + block0 = block[0] * s->c_dc_scale; + + /* XXX: only mpeg1 */ + quant_matrix = s->intra_matrix; + + __asm__ volatile ( + "pcmpeqh $f14, $f14, $f14 \r\n" + "dli $10, 15 \r\n" + "dmtc1 $10, $f16 \r\n" + "xor $f12, $f12, $f12 \r\n" + "lwc1 $f12, %2 \r\n" + "psrlh $f14, $f14, $f16 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "or $8, %3, $0 \r\n" + ".p2align 4 \r\n" + "1: \r\n" + "gsldxc1 $f0, 0($8, %0) \r\n" + "gsldxc1 $f2, 8($8, %0) \r\n" + "mov.d $f16, $f0 \r\n" + "mov.d $f18, $f2 \r\n" + "gsldxc1 $f8, 0($8, %1) \r\n" + "gsldxc1 $f10, 8($8, %1) \r\n" + "pmullh $f8, $f8, $f12 \r\n" + "pmullh $f10, $f10, $f12 \r\n" + "xor $f4, $f4, $f4 \r\n" + "xor $f6, $f6, $f6 \r\n" + "pcmpgth $f4, $f4, $f0 \r\n" + "pcmpgth $f6, $f6, $f2 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "psubh $f0, $f0, $f4 \r\n" + "psubh $f2, $f2, $f6 \r\n" + "pmullh $f0, $f0, $f8 \r\n" + "pmullh $f2, $f2, $f10 \r\n" + "xor $f8, $f8, $f8 \r\n" + "xor $f10, $f10, $f10 \r\n" + "pcmpeqh $f8, $f8, $f16 \r\n" + "pcmpeqh $f10, $f10, $f18 \r\n" + "dli $10, 3 \r\n" + "dmtc1 $10, $f16 \r\n" + "psrah $f0, $f0, $f16 \r\n" + "psrah $f2, $f2, $f16 \r\n" + "psubh $f0, $f0, $f14 \r\n" + "psubh $f2, $f2, $f14 \r\n" + "or $f0, $f0, $f14 \r\n" + "or $f2, $f2, $f14 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "psubh $f0, $f0, $f4 \r\n" + "psubh $f2, $f2, $f6 \r\n" + "pandn $f8, $f8, $f0 \r\n" + "pandn $f10, $f10, $f2 \r\n" + "gssdxc1 $f8, 0($8, %0) \r\n" + "gssdxc1 $f10, 8($8, %0) \r\n" + "addi $8, $8, 16 \r\n" + "bltz $8, 1b \r\n" + ::"r"(block+nCoeffs),"r"(quant_matrix+nCoeffs),"m"(qscale), + "g"(-2*nCoeffs) + : "$8","$10","memory" + ); + + block[0] = block0; +} + +void ff_dct_unquantize_mpeg1_inter_mmi(MpegEncContext *s, int16_t *block, + int n, int qscale) +{ + int64_t nCoeffs; + const uint16_t *quant_matrix; + + assert(s->block_last_index[n] >= 0); + nCoeffs = s->intra_scantable.raster_end[s->block_last_index[n]] + 1; + quant_matrix = s->inter_matrix; + + __asm__ volatile ( + "pcmpeqh $f14, $f14, $f14 \r\n" + "dli $10, 15 \r\n" + "dmtc1 $10, $f16 \r\n" + "xor $f12, $f12, $f12 \r\n" + "lwc1 $f12, %2 \r\n" + "psrlh $f14, $f14, $f16 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "packsswh $f12, $f12, $f12 \r\n" + "or $8, %3, $0 \r\n" + ".p2align 4 \r\n" + "1: \r\n" + "gsldxc1 $f0, 0($8, %0) \r\n" + "gsldxc1 $f2, 8($8, %0) \r\n" + "mov.d $f16, $f0 \r\n" + "mov.d $f18, $f2 \r\n" + "gsldxc1 $f8, 0($8, %1) \r\n" + "gsldxc1 $f10, 8($8, %1) \r\n" + "pmullh $f8, $f8, $f12 \r\n" + "pmullh $f10, $f10, $f12 \r\n" + "xor $f4, $f4, $f4 \r\n" + "xor $f6, $f6, $f6 \r\n" + "pcmpgth $f4, $f4, $f0 \r\n" + "pcmpgth $f6, $f6, $f2 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "psubh $f0, $f0, $f4 \r\n" + "psubh $f2, $f2, $f6 \r\n" + "paddh $f0, $f0, $f0 \r\n" + "paddh $f2, $f2, $f2 \r\n" + "paddh $f0, $f0, $f14 \r\n" + "paddh $f2, $f2, $f14 \r\n" + "pmullh $f0, $f0, $f8 \r\n" + "pmullh $f2, $f2, $f10 \r\n" + "xor $f8, $f8, $f8 \r\n" + "xor $f10, $f10, $f10 \r\n" + "pcmpeqh $f8, $f8, $f16 \r\n" + "pcmpeqh $f10, $f10, $f18 \r\n" + "dli $10, 4 \r\n" + "dmtc1 $10, $f16 \r\n" + "psrah $f0, $f0, $f16 \r\n" + "psrah $f2, $f2, $f16 \r\n" + "psubh $f0, $f0, $f14 \r\n" + "psubh $f2, $f2, $f14 \r\n" + "or $f0, $f0, $f14 \r\n" + "or $f2, $f2, $f14 \r\n" + "xor $f0, $f0, $f4 \r\n" + "xor $f2, $f2, $f6 \r\n" + "psubh $f0, $f0, $f4 \r\n" + "psubh $f2, $f2, $f6 \r\n" + "pandn $f8, $f8, $f0 \r\n" + "pandn $f10, $f10, $f2 \r\n" + "gssdxc1 $f8, 0($8, %0) \r\n" + "gssdxc1 $f10, 8($8, %0) \r\n" + "addi $8, $8, 16 \r\n" + "bltz $8, 1b \r\n" + ::"r"(block+nCoeffs),"r"(quant_matrix+nCoeffs),"m"(qscale), + "g"(-2*nCoeffs) + :"$8","$10","memory" + ); +}