From c166148409fe8f0dbccef2fe684286a40ba1e37d Mon Sep 17 00:00:00 2001 From: Diego Biurrun Date: Mon, 30 Dec 2013 19:56:07 +0100 Subject: dsputil: Move pix_sum, pix_norm1, shrink function pointers to mpegvideoenc --- libavcodec/arm/Makefile | 2 + libavcodec/arm/dsputil_armv6.S | 55 ---------------- libavcodec/arm/dsputil_init_armv6.c | 6 -- libavcodec/arm/mpegvideoencdsp_armv6.S | 76 ++++++++++++++++++++++ libavcodec/arm/mpegvideoencdsp_init_arm.c | 38 +++++++++++ libavcodec/dnxhdenc.c | 5 +- libavcodec/dsputil.c | 79 ----------------------- libavcodec/dsputil.h | 6 -- libavcodec/motion_est.c | 5 +- libavcodec/mpegvideo_enc.c | 31 +++++---- libavcodec/mpegvideoencdsp.c | 83 ++++++++++++++++++++++++ libavcodec/mpegvideoencdsp.h | 9 +++ libavcodec/ppc/Makefile | 1 + libavcodec/ppc/dsputil_altivec.c | 60 ----------------- libavcodec/ppc/mpegvideoencdsp.c | 103 ++++++++++++++++++++++++++++++ libavcodec/svq1enc.c | 1 + libavcodec/x86/Makefile | 1 + libavcodec/x86/dsputilenc.asm | 69 -------------------- libavcodec/x86/dsputilenc_mmx.c | 4 -- libavcodec/x86/mpegvideoencdsp.asm | 95 +++++++++++++++++++++++++++ libavcodec/x86/mpegvideoencdsp_init.c | 11 +++- 21 files changed, 444 insertions(+), 296 deletions(-) create mode 100644 libavcodec/arm/mpegvideoencdsp_armv6.S create mode 100644 libavcodec/arm/mpegvideoencdsp_init_arm.c create mode 100644 libavcodec/ppc/mpegvideoencdsp.c create mode 100644 libavcodec/x86/mpegvideoencdsp.asm diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile index 3a3e244c4d..6c90e1374d 100644 --- a/libavcodec/arm/Makefile +++ b/libavcodec/arm/Makefile @@ -21,6 +21,7 @@ OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_arm.o \ arm/simple_idct_arm.o OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_init_arm.o OBJS-$(CONFIG_MPEGVIDEO) += arm/mpegvideo_arm.o +OBJS-$(CONFIG_MPEGVIDEOENC) += arm/mpegvideoencdsp_init_arm.o OBJS-$(CONFIG_NEON_CLOBBER_TEST) += arm/neontest.o OBJS-$(CONFIG_VIDEODSP) += arm/videodsp_init_arm.o OBJS-$(CONFIG_VP3DSP) += arm/vp3dsp_init_arm.o @@ -60,6 +61,7 @@ ARMV6-OBJS-$(CONFIG_IDCTDSP) += arm/idctdsp_init_armv6.o \ arm/idctdsp_armv6.o \ arm/simple_idct_armv6.o ARMV6-OBJS-$(CONFIG_MPEGAUDIODSP) += arm/mpegaudiodsp_fixed_armv6.o +ARMV6-OBJS-$(CONFIG_MPEGVIDEOENC) += arm/mpegvideoencdsp_armv6.o ARMV6-OBJS-$(CONFIG_MLP_DECODER) += arm/mlpdsp_armv6.o ARMV6-OBJS-$(CONFIG_VP7_DECODER) += arm/vp8_armv6.o \ diff --git a/libavcodec/arm/dsputil_armv6.S b/libavcodec/arm/dsputil_armv6.S index b89171ff94..8ff0e4a754 100644 --- a/libavcodec/arm/dsputil_armv6.S +++ b/libavcodec/arm/dsputil_armv6.S @@ -297,58 +297,3 @@ function ff_sse16_armv6, export=1 pop {r4-r9, pc} endfunc - -function ff_pix_norm1_armv6, export=1 - push {r4-r6, lr} - mov r12, #16 - mov lr, #0 -1: - ldm r0, {r2-r5} - uxtb16 r6, r2 - uxtb16 r2, r2, ror #8 - smlad lr, r6, r6, lr - uxtb16 r6, r3 - smlad lr, r2, r2, lr - uxtb16 r3, r3, ror #8 - smlad lr, r6, r6, lr - uxtb16 r6, r4 - smlad lr, r3, r3, lr - uxtb16 r4, r4, ror #8 - smlad lr, r6, r6, lr - uxtb16 r6, r5 - smlad lr, r4, r4, lr - uxtb16 r5, r5, ror #8 - smlad lr, r6, r6, lr - subs r12, r12, #1 - add r0, r0, r1 - smlad lr, r5, r5, lr - bgt 1b - - mov r0, lr - pop {r4-r6, pc} -endfunc - -function ff_pix_sum_armv6, export=1 - push {r4-r7, lr} - mov r12, #16 - mov r2, #0 - mov r3, #0 - mov lr, #0 - ldr r4, [r0] -1: - subs r12, r12, #1 - ldr r5, [r0, #4] - usada8 r2, r4, lr, r2 - ldr r6, [r0, #8] - usada8 r3, r5, lr, r3 - ldr r7, [r0, #12] - usada8 r2, r6, lr, r2 - beq 2f - ldr_pre r4, r0, r1 - usada8 r3, r7, lr, r3 - bgt 1b -2: - usada8 r3, r7, lr, r3 - add r0, r2, r3 - pop {r4-r7, pc} -endfunc diff --git a/libavcodec/arm/dsputil_init_armv6.c b/libavcodec/arm/dsputil_init_armv6.c index fab5e0d232..85ee8e1db4 100644 --- a/libavcodec/arm/dsputil_init_armv6.c +++ b/libavcodec/arm/dsputil_init_armv6.c @@ -43,9 +43,6 @@ int ff_pix_abs8_armv6(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2, int ff_sse16_armv6(MpegEncContext *s, uint8_t *blk1, uint8_t *blk2, int line_size, int h); -int ff_pix_norm1_armv6(uint8_t *pix, int line_size); -int ff_pix_sum_armv6(uint8_t *pix, int line_size); - av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, unsigned high_bit_depth) { @@ -63,7 +60,4 @@ av_cold void ff_dsputil_init_armv6(DSPContext *c, AVCodecContext *avctx, c->sad[1] = ff_pix_abs8_armv6; c->sse[0] = ff_sse16_armv6; - - c->pix_norm1 = ff_pix_norm1_armv6; - c->pix_sum = ff_pix_sum_armv6; } diff --git a/libavcodec/arm/mpegvideoencdsp_armv6.S b/libavcodec/arm/mpegvideoencdsp_armv6.S new file mode 100644 index 0000000000..99db501b25 --- /dev/null +++ b/libavcodec/arm/mpegvideoencdsp_armv6.S @@ -0,0 +1,76 @@ +/* + * Copyright (c) 2009 Mans Rullgard + * + * This file is part of Libav. + * + * Libav is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * Libav is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "libavutil/arm/asm.S" + +function ff_pix_norm1_armv6, export=1 + push {r4-r6, lr} + mov r12, #16 + mov lr, #0 +1: + ldm r0, {r2-r5} + uxtb16 r6, r2 + uxtb16 r2, r2, ror #8 + smlad lr, r6, r6, lr + uxtb16 r6, r3 + smlad lr, r2, r2, lr + uxtb16 r3, r3, ror #8 + smlad lr, r6, r6, lr + uxtb16 r6, r4 + smlad lr, r3, r3, lr + uxtb16 r4, r4, ror #8 + smlad lr, r6, r6, lr + uxtb16 r6, r5 + smlad lr, r4, r4, lr + uxtb16 r5, r5, ror #8 + smlad lr, r6, r6, lr + subs r12, r12, #1 + add r0, r0, r1 + smlad lr, r5, r5, lr + bgt 1b + + mov r0, lr + pop {r4-r6, pc} +endfunc + +function ff_pix_sum_armv6, export=1 + push {r4-r7, lr} + mov r12, #16 + mov r2, #0 + mov r3, #0 + mov lr, #0 + ldr r4, [r0] +1: + subs r12, r12, #1 + ldr r5, [r0, #4] + usada8 r2, r4, lr, r2 + ldr r6, [r0, #8] + usada8 r3, r5, lr, r3 + ldr r7, [r0, #12] + usada8 r2, r6, lr, r2 + beq 2f + ldr_pre r4, r0, r1 + usada8 r3, r7, lr, r3 + bgt 1b +2: + usada8 r3, r7, lr, r3 + add r0, r2, r3 + pop {r4-r7, pc} +endfunc diff --git a/libavcodec/arm/mpegvideoencdsp_init_arm.c b/libavcodec/arm/mpegvideoencdsp_init_arm.c new file mode 100644 index 0000000000..ab9ba3e1be --- /dev/null +++ b/libavcodec/arm/mpegvideoencdsp_init_arm.c @@ -0,0 +1,38 @@ +/* + * This file is part of Libav. + * + * Libav is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * Libav is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include + +#include "libavutil/cpu.h" +#include "libavutil/arm/cpu.h" +#include "libavcodec/avcodec.h" +#include "libavcodec/mpegvideoencdsp.h" + +int ff_pix_norm1_armv6(uint8_t *pix, int line_size); +int ff_pix_sum_armv6(uint8_t *pix, int line_size); + +av_cold void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c, + AVCodecContext *avctx) +{ + int cpu_flags = av_get_cpu_flags(); + + if (have_armv6(cpu_flags)) { + c->pix_norm1 = ff_pix_norm1_armv6; + c->pix_sum = ff_pix_sum_armv6; + } +} diff --git a/libavcodec/dnxhdenc.c b/libavcodec/dnxhdenc.c index 223791acbf..f3c291fa7c 100644 --- a/libavcodec/dnxhdenc.c +++ b/libavcodec/dnxhdenc.c @@ -309,6 +309,7 @@ static av_cold int dnxhd_encode_init(AVCodecContext *avctx) ff_blockdsp_init(&ctx->bdsp, avctx); ff_dsputil_init(&ctx->m.dsp, avctx); ff_idctdsp_init(&ctx->m.idsp, avctx); + ff_mpegvideoencdsp_init(&ctx->m.mpvencdsp, avctx); ff_dct_common_init(&ctx->m); if (!ctx->m.dct_quantize) ctx->m.dct_quantize = ff_dct_quantize_c; @@ -719,8 +720,8 @@ static int dnxhd_mb_var_thread(AVCodecContext *avctx, void *arg, int varc; if (!partial_last_row && mb_x * 16 <= avctx->width - 16) { - sum = ctx->m.dsp.pix_sum(pix, ctx->m.linesize); - varc = ctx->m.dsp.pix_norm1(pix, ctx->m.linesize); + sum = ctx->m.mpvencdsp.pix_sum(pix, ctx->m.linesize); + varc = ctx->m.mpvencdsp.pix_norm1(pix, ctx->m.linesize); } else { int bw = FFMIN(avctx->width - 16 * mb_x, 16); int bh = FFMIN((avctx->height >> ctx->interlaced) - 16 * mb_y, 16); diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index 68f5120c41..8c4c670630 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -26,15 +26,12 @@ */ #include "libavutil/attributes.h" -#include "libavutil/imgutils.h" #include "avcodec.h" #include "copy_block.h" #include "dct.h" #include "dsputil.h" #include "simple_idct.h" #include "faandct.h" -#include "imgconvert.h" -#include "mathops.h" #include "mpegvideo.h" #include "config.h" @@ -47,74 +44,6 @@ uint32_t ff_square_tab[512] = { 0, }; #define BIT_DEPTH 8 #include "dsputilenc_template.c" -static int pix_sum_c(uint8_t *pix, int line_size) -{ - int s = 0, i, j; - - for (i = 0; i < 16; i++) { - for (j = 0; j < 16; j += 8) { - s += pix[0]; - s += pix[1]; - s += pix[2]; - s += pix[3]; - s += pix[4]; - s += pix[5]; - s += pix[6]; - s += pix[7]; - pix += 8; - } - pix += line_size - 16; - } - return s; -} - -static int pix_norm1_c(uint8_t *pix, int line_size) -{ - int s = 0, i, j; - uint32_t *sq = ff_square_tab + 256; - - for (i = 0; i < 16; i++) { - for (j = 0; j < 16; j += 8) { -#if 0 - s += sq[pix[0]]; - s += sq[pix[1]]; - s += sq[pix[2]]; - s += sq[pix[3]]; - s += sq[pix[4]]; - s += sq[pix[5]]; - s += sq[pix[6]]; - s += sq[pix[7]]; -#else -#if HAVE_FAST_64BIT - register uint64_t x = *(uint64_t *) pix; - s += sq[x & 0xff]; - s += sq[(x >> 8) & 0xff]; - s += sq[(x >> 16) & 0xff]; - s += sq[(x >> 24) & 0xff]; - s += sq[(x >> 32) & 0xff]; - s += sq[(x >> 40) & 0xff]; - s += sq[(x >> 48) & 0xff]; - s += sq[(x >> 56) & 0xff]; -#else - register uint32_t x = *(uint32_t *) pix; - s += sq[x & 0xff]; - s += sq[(x >> 8) & 0xff]; - s += sq[(x >> 16) & 0xff]; - s += sq[(x >> 24) & 0xff]; - x = *(uint32_t *) (pix + 4); - s += sq[x & 0xff]; - s += sq[(x >> 8) & 0xff]; - s += sq[(x >> 16) & 0xff]; - s += sq[(x >> 24) & 0xff]; -#endif -#endif - pix += 8; - } - pix += line_size - 16; - } - return s; -} - static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h) { @@ -1055,9 +984,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx) c->sum_abs_dctelem = sum_abs_dctelem_c; - c->pix_sum = pix_sum_c; - c->pix_norm1 = pix_norm1_c; - /* TODO [0] 16 [1] 8 */ c->pix_abs[0][0] = pix_abs16_c; c->pix_abs[0][1] = pix_abs16_x2_c; @@ -1097,11 +1023,6 @@ av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx) c->nsse[0] = nsse16_c; c->nsse[1] = nsse8_c; - c->shrink[0] = av_image_copy_plane; - c->shrink[1] = ff_shrink22; - c->shrink[2] = ff_shrink44; - c->shrink[3] = ff_shrink88; - c->draw_edges = draw_edges_8_c; switch (avctx->bits_per_raw_sample) { diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index f8b7b0d892..284973e138 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -58,9 +58,6 @@ typedef struct DSPContext { int stride); int (*sum_abs_dctelem)(int16_t *block /* align 16 */); - int (*pix_sum)(uint8_t *pix, int line_size); - int (*pix_norm1)(uint8_t *pix, int line_size); - me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ me_cmp_func sse[6]; me_cmp_func hadamard8_diff[6]; @@ -92,9 +89,6 @@ typedef struct DSPContext { #define EDGE_WIDTH 16 #define EDGE_TOP 1 #define EDGE_BOTTOM 2 - - void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, - int src_wrap, int width, int height); } DSPContext; void ff_dsputil_static_init(void); diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c index 9677e04b3e..a8a93b73e8 100644 --- a/libavcodec/motion_est.c +++ b/libavcodec/motion_est.c @@ -881,8 +881,9 @@ void ff_estimate_p_frame_motion(MpegEncContext * s, /* intra / predictive decision */ pix = c->src[0][0]; - sum = s->dsp.pix_sum(pix, s->linesize); - varc = s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500; + sum = s->mpvencdsp.pix_sum(pix, s->linesize); + varc = s->mpvencdsp.pix_norm1(pix, s->linesize) - + (((unsigned) sum * sum) >> 8) + 500; pic->mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8; pic->mb_var [s->mb_stride * mb_y + mb_x] = (varc+128)>>8; diff --git a/libavcodec/mpegvideo_enc.c b/libavcodec/mpegvideo_enc.c index f95a76b837..12440bb67a 100644 --- a/libavcodec/mpegvideo_enc.c +++ b/libavcodec/mpegvideo_enc.c @@ -895,7 +895,7 @@ static int get_intra_count(MpegEncContext *s, uint8_t *src, int offset = x + y * stride; int sad = s->dsp.sad[0](NULL, src + offset, ref + offset, stride, 16); - int mean = (s->dsp.pix_sum(src + offset, stride) + 128) >> 8; + int mean = (s->mpvencdsp.pix_sum(src + offset, stride) + 128) >> 8; int sae = get_sae(src + offset, mean, stride); acc += sae + 500 < sad; @@ -1138,15 +1138,21 @@ static int estimate_best_b_count(MpegEncContext *s) pre_input.f->data[2] += INPLACE_OFFSET; } - s->dsp.shrink[scale](s->tmp_frames[i]->data[0], s->tmp_frames[i]->linesize[0], - pre_input.f->data[0], pre_input.f->linesize[0], - c->width, c->height); - s->dsp.shrink[scale](s->tmp_frames[i]->data[1], s->tmp_frames[i]->linesize[1], - pre_input.f->data[1], pre_input.f->linesize[1], - c->width >> 1, c->height >> 1); - s->dsp.shrink[scale](s->tmp_frames[i]->data[2], s->tmp_frames[i]->linesize[2], - pre_input.f->data[2], pre_input.f->linesize[2], - c->width >> 1, c->height >> 1); + s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[0], + s->tmp_frames[i]->linesize[0], + pre_input.f->data[0], + pre_input.f->linesize[0], + c->width, c->height); + s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[1], + s->tmp_frames[i]->linesize[1], + pre_input.f->data[1], + pre_input.f->linesize[1], + c->width >> 1, c->height >> 1); + s->mpvencdsp.shrink[scale](s->tmp_frames[i]->data[2], + s->tmp_frames[i]->linesize[2], + pre_input.f->data[2], + pre_input.f->linesize[2], + c->width >> 1, c->height >> 1); } } @@ -2420,9 +2426,10 @@ static int mb_var_thread(AVCodecContext *c, void *arg){ int yy = mb_y * 16; uint8_t *pix = s->new_picture.f->data[0] + (yy * s->linesize) + xx; int varc; - int sum = s->dsp.pix_sum(pix, s->linesize); + int sum = s->mpvencdsp.pix_sum(pix, s->linesize); - varc = (s->dsp.pix_norm1(pix, s->linesize) - (((unsigned)sum*sum)>>8) + 500 + 128)>>8; + varc = (s->mpvencdsp.pix_norm1(pix, s->linesize) - + (((unsigned) sum * sum) >> 8) + 500 + 128) >> 8; s->current_picture.mb_var [s->mb_stride * mb_y + mb_x] = varc; s->current_picture.mb_mean[s->mb_stride * mb_y + mb_x] = (sum+128)>>8; diff --git a/libavcodec/mpegvideoencdsp.c b/libavcodec/mpegvideoencdsp.c index 16ff1f2db4..ee6327ded9 100644 --- a/libavcodec/mpegvideoencdsp.c +++ b/libavcodec/mpegvideoencdsp.c @@ -21,7 +21,10 @@ #include "config.h" #include "libavutil/attributes.h" +#include "libavutil/imgutils.h" #include "avcodec.h" +#include "dsputil.h" +#include "imgconvert.h" #include "mpegvideoencdsp.h" static int try_8x8basis_c(int16_t rem[64], int16_t weight[64], @@ -53,12 +56,92 @@ static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale) (BASIS_SHIFT - RECON_SHIFT); } +static int pix_sum_c(uint8_t *pix, int line_size) +{ + int s = 0, i, j; + + for (i = 0; i < 16; i++) { + for (j = 0; j < 16; j += 8) { + s += pix[0]; + s += pix[1]; + s += pix[2]; + s += pix[3]; + s += pix[4]; + s += pix[5]; + s += pix[6]; + s += pix[7]; + pix += 8; + } + pix += line_size - 16; + } + return s; +} + +static int pix_norm1_c(uint8_t *pix, int line_size) +{ + int s = 0, i, j; + uint32_t *sq = ff_square_tab + 256; + + for (i = 0; i < 16; i++) { + for (j = 0; j < 16; j += 8) { +#if 0 + s += sq[pix[0]]; + s += sq[pix[1]]; + s += sq[pix[2]]; + s += sq[pix[3]]; + s += sq[pix[4]]; + s += sq[pix[5]]; + s += sq[pix[6]]; + s += sq[pix[7]]; +#else +#if HAVE_FAST_64BIT + register uint64_t x = *(uint64_t *) pix; + s += sq[x & 0xff]; + s += sq[(x >> 8) & 0xff]; + s += sq[(x >> 16) & 0xff]; + s += sq[(x >> 24) & 0xff]; + s += sq[(x >> 32) & 0xff]; + s += sq[(x >> 40) & 0xff]; + s += sq[(x >> 48) & 0xff]; + s += sq[(x >> 56) & 0xff]; +#else + register uint32_t x = *(uint32_t *) pix; + s += sq[x & 0xff]; + s += sq[(x >> 8) & 0xff]; + s += sq[(x >> 16) & 0xff]; + s += sq[(x >> 24) & 0xff]; + x = *(uint32_t *) (pix + 4); + s += sq[x & 0xff]; + s += sq[(x >> 8) & 0xff]; + s += sq[(x >> 16) & 0xff]; + s += sq[(x >> 24) & 0xff]; +#endif +#endif + pix += 8; + } + pix += line_size - 16; + } + return s; +} + av_cold void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, AVCodecContext *avctx) { c->try_8x8basis = try_8x8basis_c; c->add_8x8basis = add_8x8basis_c; + c->shrink[0] = av_image_copy_plane; + c->shrink[1] = ff_shrink22; + c->shrink[2] = ff_shrink44; + c->shrink[3] = ff_shrink88; + + c->pix_sum = pix_sum_c; + c->pix_norm1 = pix_norm1_c; + + if (ARCH_ARM) + ff_mpegvideoencdsp_init_arm(c, avctx); + if (ARCH_PPC) + ff_mpegvideoencdsp_init_ppc(c, avctx); if (ARCH_X86) ff_mpegvideoencdsp_init_x86(c, avctx); } diff --git a/libavcodec/mpegvideoencdsp.h b/libavcodec/mpegvideoencdsp.h index 2e8427d0a2..dcbeb3cbae 100644 --- a/libavcodec/mpegvideoencdsp.h +++ b/libavcodec/mpegvideoencdsp.h @@ -31,10 +31,19 @@ typedef struct MpegvideoEncDSPContext { int16_t basis[64], int scale); void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); + int (*pix_sum)(uint8_t *pix, int line_size); + int (*pix_norm1)(uint8_t *pix, int line_size); + + void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, + int src_wrap, int width, int height); } MpegvideoEncDSPContext; void ff_mpegvideoencdsp_init(MpegvideoEncDSPContext *c, AVCodecContext *avctx); +void ff_mpegvideoencdsp_init_arm(MpegvideoEncDSPContext *c, + AVCodecContext *avctx); +void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c, + AVCodecContext *avctx); void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, AVCodecContext *avctx); diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile index ee0c18c09e..256c081dc1 100644 --- a/libavcodec/ppc/Makefile +++ b/libavcodec/ppc/Makefile @@ -13,6 +13,7 @@ OBJS-$(CONFIG_IDCTDSP) += ppc/idctdsp.o OBJS-$(CONFIG_MPEGAUDIODSP) += ppc/mpegaudiodsp_altivec.o OBJS-$(CONFIG_MPEGVIDEO) += ppc/mpegvideo_altivec.o \ ppc/mpegvideodsp.o +OBJS-$(CONFIG_MPEGVIDEOENC) += ppc/mpegvideoencdsp.o OBJS-$(CONFIG_VIDEODSP) += ppc/videodsp_ppc.o OBJS-$(CONFIG_VP3DSP) += ppc/vp3dsp_altivec.o diff --git a/libavcodec/ppc/dsputil_altivec.c b/libavcodec/ppc/dsputil_altivec.c index c3f90e91c4..efb4a18535 100644 --- a/libavcodec/ppc/dsputil_altivec.c +++ b/libavcodec/ppc/dsputil_altivec.c @@ -308,34 +308,6 @@ static int sad8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, return s; } -static int pix_norm1_altivec(uint8_t *pix, int line_size) -{ - int i, s = 0; - const vector unsigned int zero = - (const vector unsigned int) vec_splat_u32(0); - vector unsigned char perm = vec_lvsl(0, pix); - vector unsigned int sv = (vector unsigned int) vec_splat_u32(0); - vector signed int sum; - - for (i = 0; i < 16; i++) { - /* Read the potentially unaligned pixels. */ - vector unsigned char pixl = vec_ld(0, pix); - vector unsigned char pixr = vec_ld(15, pix); - vector unsigned char pixv = vec_perm(pixl, pixr, perm); - - /* Square the values, and add them to our sum. */ - sv = vec_msum(pixv, pixv, sv); - - pix += line_size; - } - /* Sum up the four partial sums, and put the result into s. */ - sum = vec_sums((vector signed int) sv, (vector signed int) zero); - sum = vec_splat(sum, 3); - vec_ste(sum, 0, &s); - - return s; -} - /* Sum of Squared Errors for an 8x8 block, AltiVec-enhanced. * It's the sad8_altivec code above w/ squaring added. */ static int sse8_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, @@ -430,35 +402,6 @@ static int sse16_altivec(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, return s; } -static int pix_sum_altivec(uint8_t *pix, int line_size) -{ - int i, s; - const vector unsigned int zero = - (const vector unsigned int) vec_splat_u32(0); - vector unsigned char perm = vec_lvsl(0, pix); - vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); - vector signed int sumdiffs; - - for (i = 0; i < 16; i++) { - /* Read the potentially unaligned 16 pixels into t1. */ - vector unsigned char pixl = vec_ld(0, pix); - vector unsigned char pixr = vec_ld(15, pix); - vector unsigned char t1 = vec_perm(pixl, pixr, perm); - - /* Add each 4 pixel group together and put 4 results into sad. */ - sad = vec_sum4s(t1, sad); - - pix += line_size; - } - - /* Sum up the four partial sums, and put the result into s. */ - sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); - sumdiffs = vec_splat(sumdiffs, 3); - vec_ste(sumdiffs, 0, &s); - - return s; -} - static void get_pixels_altivec(int16_t *restrict block, const uint8_t *pixels, int line_size) { @@ -911,9 +854,6 @@ av_cold void ff_dsputil_init_altivec(DSPContext *c, AVCodecContext *avctx, c->sse[0] = sse16_altivec; c->sse[1] = sse8_altivec; - c->pix_norm1 = pix_norm1_altivec; - c->pix_sum = pix_sum_altivec; - c->diff_pixels = diff_pixels_altivec; if (!high_bit_depth) { diff --git a/libavcodec/ppc/mpegvideoencdsp.c b/libavcodec/ppc/mpegvideoencdsp.c new file mode 100644 index 0000000000..b5348e6abd --- /dev/null +++ b/libavcodec/ppc/mpegvideoencdsp.c @@ -0,0 +1,103 @@ +/* + * This file is part of Libav. + * + * Libav is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * Libav is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with Libav; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#include "config.h" +#include +#if HAVE_ALTIVEC_H +#include +#endif + +#include "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/ppc/cpu.h" +#include "libavutil/ppc/types_altivec.h" +#include "libavutil/ppc/util_altivec.h" +#include "libavcodec/mpegvideoencdsp.h" + +#if HAVE_ALTIVEC + +static int pix_norm1_altivec(uint8_t *pix, int line_size) +{ + int i, s = 0; + const vector unsigned int zero = + (const vector unsigned int) vec_splat_u32(0); + vector unsigned char perm = vec_lvsl(0, pix); + vector unsigned int sv = (vector unsigned int) vec_splat_u32(0); + vector signed int sum; + + for (i = 0; i < 16; i++) { + /* Read the potentially unaligned pixels. */ + vector unsigned char pixl = vec_ld(0, pix); + vector unsigned char pixr = vec_ld(15, pix); + vector unsigned char pixv = vec_perm(pixl, pixr, perm); + + /* Square the values, and add them to our sum. */ + sv = vec_msum(pixv, pixv, sv); + + pix += line_size; + } + /* Sum up the four partial sums, and put the result into s. */ + sum = vec_sums((vector signed int) sv, (vector signed int) zero); + sum = vec_splat(sum, 3); + vec_ste(sum, 0, &s); + + return s; +} + +static int pix_sum_altivec(uint8_t *pix, int line_size) +{ + int i, s; + const vector unsigned int zero = + (const vector unsigned int) vec_splat_u32(0); + vector unsigned char perm = vec_lvsl(0, pix); + vector unsigned int sad = (vector unsigned int) vec_splat_u32(0); + vector signed int sumdiffs; + + for (i = 0; i < 16; i++) { + /* Read the potentially unaligned 16 pixels into t1. */ + vector unsigned char pixl = vec_ld(0, pix); + vector unsigned char pixr = vec_ld(15, pix); + vector unsigned char t1 = vec_perm(pixl, pixr, perm); + + /* Add each 4 pixel group together and put 4 results into sad. */ + sad = vec_sum4s(t1, sad); + + pix += line_size; + } + + /* Sum up the four partial sums, and put the result into s. */ + sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero); + sumdiffs = vec_splat(sumdiffs, 3); + vec_ste(sumdiffs, 0, &s); + + return s; +} + +#endif /* HAVE_ALTIVEC */ + +av_cold void ff_mpegvideoencdsp_init_ppc(MpegvideoEncDSPContext *c, + AVCodecContext *avctx) +{ +#if HAVE_ALTIVEC + if (!PPC_ALTIVEC(av_get_cpu_flags())) + return; + + c->pix_norm1 = pix_norm1_altivec; + c->pix_sum = pix_sum_altivec; +#endif /* HAVE_ALTIVEC */ +} diff --git a/libavcodec/svq1enc.c b/libavcodec/svq1enc.c index bdb6f0fd54..5e3cc55efe 100644 --- a/libavcodec/svq1enc.c +++ b/libavcodec/svq1enc.c @@ -511,6 +511,7 @@ static av_cold int svq1_encode_init(AVCodecContext *avctx) ff_dsputil_init(&s->dsp, avctx); ff_hpeldsp_init(&s->hdsp, avctx->flags); + ff_mpegvideoencdsp_init(&s->m.mpvencdsp, avctx); avctx->coded_frame = av_frame_alloc(); s->current_picture = av_frame_alloc(); diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index bfc13737d1..701aad7916 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -92,6 +92,7 @@ YASM-OBJS-$(CONFIG_HPELDSP) += x86/fpel.o \ x86/hpeldsp.o YASM-OBJS-$(CONFIG_HUFFYUVDSP) += x86/huffyuvdsp.o YASM-OBJS-$(CONFIG_MPEGAUDIODSP) += x86/imdct36.o +YASM-OBJS-$(CONFIG_MPEGVIDEOENC) += x86/mpegvideoencdsp.o YASM-OBJS-$(CONFIG_QPELDSP) += x86/qpeldsp.o \ x86/fpel.o \ x86/qpel.o diff --git a/libavcodec/x86/dsputilenc.asm b/libavcodec/x86/dsputilenc.asm index f676ba63fd..d996962eac 100644 --- a/libavcodec/x86/dsputilenc.asm +++ b/libavcodec/x86/dsputilenc.asm @@ -418,72 +418,3 @@ cglobal diff_pixels, 4,5 add r4, 16 jne .loop REP_RET - -INIT_MMX mmx -; int ff_pix_sum16_mmx(uint8_t *pix, int line_size) -cglobal pix_sum16, 2, 3 - movsxdifnidn r1, r1d - mov r2, r1 - neg r2 - shl r2, 4 - sub r0, r2 - pxor m7, m7 - pxor m6, m6 -.loop: - mova m0, [r0+r2+0] - mova m1, [r0+r2+0] - mova m2, [r0+r2+8] - mova m3, [r0+r2+8] - punpcklbw m0, m7 - punpckhbw m1, m7 - punpcklbw m2, m7 - punpckhbw m3, m7 - paddw m1, m0 - paddw m3, m2 - paddw m3, m1 - paddw m6, m3 - add r2, r1 - js .loop - mova m5, m6 - psrlq m6, 32 - paddw m6, m5 - mova m5, m6 - psrlq m6, 16 - paddw m6, m5 - movd eax, m6 - and eax, 0xffff - RET - -INIT_MMX mmx -; int ff_pix_norm1_mmx(uint8_t *pix, int line_size) -cglobal pix_norm1, 2, 4 - movsxdifnidn r1, r1d - mov r2, 16 - pxor m0, m0 - pxor m7, m7 -.loop: - mova m2, [r0+0] - mova m3, [r0+8] - mova m1, m2 - punpckhbw m1, m0 - punpcklbw m2, m0 - mova m4, m3 - punpckhbw m3, m0 - punpcklbw m4, m0 - pmaddwd m1, m1 - pmaddwd m2, m2 - pmaddwd m3, m3 - pmaddwd m4, m4 - paddd m2, m1 - paddd m4, m3 - paddd m7, m2 - add r0, r1 - paddd m7, m4 - dec r2 - jne .loop - mova m1, m7 - psrlq m7, 32 - paddd m1, m7 - movd eax, m1 - RET - diff --git a/libavcodec/x86/dsputilenc_mmx.c b/libavcodec/x86/dsputilenc_mmx.c index 7af091365f..563543ee7d 100644 --- a/libavcodec/x86/dsputilenc_mmx.c +++ b/libavcodec/x86/dsputilenc_mmx.c @@ -35,8 +35,6 @@ void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size); void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size); void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2, int stride); -int ff_pix_sum16_mmx(uint8_t *pix, int line_size); -int ff_pix_norm1_mmx(uint8_t *pix, int line_size); #if HAVE_INLINE_ASM @@ -831,8 +829,6 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx, if (!high_bit_depth) c->get_pixels = ff_get_pixels_mmx; c->diff_pixels = ff_diff_pixels_mmx; - c->pix_sum = ff_pix_sum16_mmx; - c->pix_norm1 = ff_pix_norm1_mmx; } if (EXTERNAL_SSE2(cpu_flags)) diff --git a/libavcodec/x86/mpegvideoencdsp.asm b/libavcodec/x86/mpegvideoencdsp.asm new file mode 100644 index 0000000000..9326ee776d --- /dev/null +++ b/libavcodec/x86/mpegvideoencdsp.asm @@ -0,0 +1,95 @@ +;***************************************************************************** +;* SIMD-optimized MPEG encoding functions +;***************************************************************************** +;* Copyright (c) 2000, 2001 Fabrice Bellard +;* Copyright (c) 2002-2004 Michael Niedermayer +;* +;* This file is part of Libav. +;* +;* Libav is free software; you can redistribute it and/or +;* modify it under the terms of the GNU Lesser General Public +;* License as published by the Free Software Foundation; either +;* version 2.1 of the License, or (at your option) any later version. +;* +;* Libav is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +;* Lesser General Public License for more details. +;* +;* You should have received a copy of the GNU Lesser General Public +;* License along with Libav; if not, write to the Free Software +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +;***************************************************************************** + +%include "libavutil/x86/x86util.asm" + +SECTION .text + +INIT_MMX mmx +; int ff_pix_sum16_mmx(uint8_t *pix, int line_size) +cglobal pix_sum16, 2, 3 + movsxdifnidn r1, r1d + mov r2, r1 + neg r2 + shl r2, 4 + sub r0, r2 + pxor m7, m7 + pxor m6, m6 +.loop: + mova m0, [r0+r2+0] + mova m1, [r0+r2+0] + mova m2, [r0+r2+8] + mova m3, [r0+r2+8] + punpcklbw m0, m7 + punpckhbw m1, m7 + punpcklbw m2, m7 + punpckhbw m3, m7 + paddw m1, m0 + paddw m3, m2 + paddw m3, m1 + paddw m6, m3 + add r2, r1 + js .loop + mova m5, m6 + psrlq m6, 32 + paddw m6, m5 + mova m5, m6 + psrlq m6, 16 + paddw m6, m5 + movd eax, m6 + and eax, 0xffff + RET + +INIT_MMX mmx +; int ff_pix_norm1_mmx(uint8_t *pix, int line_size) +cglobal pix_norm1, 2, 4 + movsxdifnidn r1, r1d + mov r2, 16 + pxor m0, m0 + pxor m7, m7 +.loop: + mova m2, [r0+0] + mova m3, [r0+8] + mova m1, m2 + punpckhbw m1, m0 + punpcklbw m2, m0 + mova m4, m3 + punpckhbw m3, m0 + punpcklbw m4, m0 + pmaddwd m1, m1 + pmaddwd m2, m2 + pmaddwd m3, m3 + pmaddwd m4, m4 + paddd m2, m1 + paddd m4, m3 + paddd m7, m2 + add r0, r1 + paddd m7, m4 + dec r2 + jne .loop + mova m1, m7 + psrlq m7, 32 + paddd m1, m7 + movd eax, m1 + RET + diff --git a/libavcodec/x86/mpegvideoencdsp_init.c b/libavcodec/x86/mpegvideoencdsp_init.c index db2c37f43e..4ef2f34961 100644 --- a/libavcodec/x86/mpegvideoencdsp_init.c +++ b/libavcodec/x86/mpegvideoencdsp_init.c @@ -22,6 +22,9 @@ #include "libavcodec/avcodec.h" #include "libavcodec/mpegvideoencdsp.h" +int ff_pix_sum16_mmx(uint8_t *pix, int line_size); +int ff_pix_norm1_mmx(uint8_t *pix, int line_size); + #if HAVE_INLINE_ASM #define PHADDD(a, t) \ @@ -95,9 +98,15 @@ av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c, AVCodecContext *avctx) { -#if HAVE_INLINE_ASM int cpu_flags = av_get_cpu_flags(); + if (EXTERNAL_MMX(cpu_flags)) { + c->pix_sum = ff_pix_sum16_mmx; + c->pix_norm1 = ff_pix_norm1_mmx; + } + +#if HAVE_INLINE_ASM + if (INLINE_MMX(cpu_flags)) { if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { c->try_8x8basis = try_8x8basis_mmx; -- cgit v1.2.3