yading@10: /*
yading@10:  * Copyright (C) 2002-2012 Michael Niedermayer
yading@10:  * Copyright (C) 2012 Ronald S. Bultje
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "config.h"
yading@10: #include "libavutil/attributes.h"
yading@10: #include "libavutil/avassert.h"
yading@10: #include "libavutil/common.h"
yading@10: #include "libavutil/cpu.h"
yading@10: #include "libavutil/mem.h"
yading@10: #include "libavutil/x86/asm.h"
yading@10: #include "libavcodec/videodsp.h"
yading@10: 
yading@10: #if HAVE_YASM
yading@10: typedef void emu_edge_core_func(uint8_t *buf, const uint8_t *src,
yading@10:                                 x86_reg linesize, x86_reg start_y,
yading@10:                                 x86_reg end_y, x86_reg block_h,
yading@10:                                 x86_reg start_x, x86_reg end_x,
yading@10:                                 x86_reg block_w);
yading@10: extern emu_edge_core_func ff_emu_edge_core_mmx;
yading@10: extern emu_edge_core_func ff_emu_edge_core_sse;
yading@10: 
yading@10: static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src,
yading@10:                                               ptrdiff_t linesize_arg,
yading@10:                                               int block_w, int block_h,
yading@10:                                               int src_x, int src_y,
yading@10:                                               int w, int h,
yading@10:                                               emu_edge_core_func *core_fn)
yading@10: {
yading@10:     int start_y, start_x, end_y, end_x, src_y_add = 0;
yading@10:     int linesize = linesize_arg;
yading@10: 
yading@10:     if(!w || !h)
yading@10:         return;
yading@10: 
yading@10:     if (src_y >= h) {
yading@10:         src -= src_y*linesize;
yading@10:         src_y_add = h - 1;
yading@10:         src_y     = h - 1;
yading@10:     } else if (src_y <= -block_h) {
yading@10:         src -= src_y*linesize;
yading@10:         src_y_add = 1 - block_h;
yading@10:         src_y     = 1 - block_h;
yading@10:     }
yading@10:     if (src_x >= w) {
yading@10:         src   += w - 1 - src_x;
yading@10:         src_x  = w - 1;
yading@10:     } else if (src_x <= -block_w) {
yading@10:         src   += 1 - block_w - src_x;
yading@10:         src_x  = 1 - block_w;
yading@10:     }
yading@10: 
yading@10:     start_y = FFMAX(0, -src_y);
yading@10:     start_x = FFMAX(0, -src_x);
yading@10:     end_y   = FFMIN(block_h, h-src_y);
yading@10:     end_x   = FFMIN(block_w, w-src_x);
yading@10:     av_assert2(start_x < end_x && block_w > 0);
yading@10:     av_assert2(start_y < end_y && block_h > 0);
yading@10: 
yading@10:     // fill in the to-be-copied part plus all above/below
yading@10:     src += (src_y_add + start_y) * linesize + start_x;
yading@10:     buf += start_x;
yading@10:     core_fn(buf, src, linesize, start_y, end_y,
yading@10:             block_h, start_x, end_x, block_w);
yading@10: }
yading@10: 
yading@10: #if ARCH_X86_32
yading@10: static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
yading@10:                                              ptrdiff_t linesize,
yading@10:                                              int block_w, int block_h,
yading@10:                                              int src_x, int src_y, int w, int h)
yading@10: {
yading@10:     emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
yading@10:                      w, h, &ff_emu_edge_core_mmx);
yading@10: }
yading@10: #endif
yading@10: 
yading@10: static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
yading@10:                                              ptrdiff_t linesize,
yading@10:                                              int block_w, int block_h,
yading@10:                                              int src_x, int src_y, int w, int h)
yading@10: {
yading@10:     emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
yading@10:                      w, h, &ff_emu_edge_core_sse);
yading@10: }
yading@10: #endif /* HAVE_YASM */
yading@10: 
yading@10: void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h);
yading@10: void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h);
yading@10: 
yading@10: av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
yading@10: {
yading@10: #if HAVE_YASM
yading@10:     int mm_flags = av_get_cpu_flags();
yading@10: 
yading@10: #if ARCH_X86_32
yading@10:     if (bpc <= 8 && mm_flags & AV_CPU_FLAG_MMX) {
yading@10:         ctx->emulated_edge_mc = emulated_edge_mc_mmx;
yading@10:     }
yading@10:     if (mm_flags & AV_CPU_FLAG_3DNOW) {
yading@10:         ctx->prefetch = ff_prefetch_3dnow;
yading@10:     }
yading@10: #endif /* ARCH_X86_32 */
yading@10:     if (mm_flags & AV_CPU_FLAG_MMXEXT) {
yading@10:         ctx->prefetch = ff_prefetch_mmxext;
yading@10:     }
yading@10:     if (bpc <= 8 && mm_flags & AV_CPU_FLAG_SSE) {
yading@10:         ctx->emulated_edge_mc = emulated_edge_mc_sse;
yading@10:     }
yading@10: #endif /* HAVE_YASM */
yading@10: }