yading@10: /* yading@10: * Copyright (c) 2009 David Conrad 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 yading@10: yading@10: #include "libavutil/attributes.h" yading@10: #include "libavutil/cpu.h" yading@10: #include "libavutil/x86/cpu.h" yading@10: #include "libavutil/x86/asm.h" yading@10: #include "libavcodec/avcodec.h" yading@10: #include "libavcodec/dsputil.h" yading@10: #include "libavcodec/vp3dsp.h" yading@10: #include "config.h" yading@10: yading@10: void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, int16_t *block); yading@10: void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, int16_t *block); yading@10: yading@10: void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, int16_t *block); yading@10: void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, int16_t *block); yading@10: yading@10: void ff_vp3_idct_dc_add_mmxext(uint8_t *dest, int line_size, yading@10: int16_t *block); yading@10: yading@10: void ff_vp3_v_loop_filter_mmxext(uint8_t *src, int stride, yading@10: int *bounding_values); yading@10: void ff_vp3_h_loop_filter_mmxext(uint8_t *src, int stride, yading@10: int *bounding_values); yading@10: yading@10: #if HAVE_INLINE_ASM yading@10: yading@10: #define MOVQ_BFE(regd) \ yading@10: __asm__ volatile ( \ yading@10: "pcmpeqd %%"#regd", %%"#regd" \n\t" \ yading@10: "paddb %%"#regd", %%"#regd" \n\t" ::) yading@10: yading@10: #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \ yading@10: "movq "#rega", "#regr" \n\t" \ yading@10: "movq "#regc", "#regp" \n\t" \ yading@10: "pand "#regb", "#regr" \n\t" \ yading@10: "pand "#regd", "#regp" \n\t" \ yading@10: "pxor "#rega", "#regb" \n\t" \ yading@10: "pxor "#regc", "#regd" \n\t" \ yading@10: "pand %%mm6, "#regb" \n\t" \ yading@10: "pand %%mm6, "#regd" \n\t" \ yading@10: "psrlq $1, "#regb" \n\t" \ yading@10: "psrlq $1, "#regd" \n\t" \ yading@10: "paddb "#regb", "#regr" \n\t" \ yading@10: "paddb "#regd", "#regp" \n\t" yading@10: yading@10: static void put_vp_no_rnd_pixels8_l2_mmx(uint8_t *dst, const uint8_t *a, const uint8_t *b, ptrdiff_t stride, int h) yading@10: { yading@10: // START_TIMER yading@10: MOVQ_BFE(mm6); yading@10: __asm__ volatile( yading@10: "1: \n\t" yading@10: "movq (%1), %%mm0 \n\t" yading@10: "movq (%2), %%mm1 \n\t" yading@10: "movq (%1,%4), %%mm2 \n\t" yading@10: "movq (%2,%4), %%mm3 \n\t" yading@10: PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) yading@10: "movq %%mm4, (%3) \n\t" yading@10: "movq %%mm5, (%3,%4) \n\t" yading@10: yading@10: "movq (%1,%4,2), %%mm0 \n\t" yading@10: "movq (%2,%4,2), %%mm1 \n\t" yading@10: "movq (%1,%5), %%mm2 \n\t" yading@10: "movq (%2,%5), %%mm3 \n\t" yading@10: "lea (%1,%4,4), %1 \n\t" yading@10: "lea (%2,%4,4), %2 \n\t" yading@10: PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5) yading@10: "movq %%mm4, (%3,%4,2) \n\t" yading@10: "movq %%mm5, (%3,%5) \n\t" yading@10: "lea (%3,%4,4), %3 \n\t" yading@10: "subl $4, %0 \n\t" yading@10: "jnz 1b \n\t" yading@10: :"+r"(h), "+r"(a), "+r"(b), "+r"(dst) yading@10: :"r"((x86_reg)stride), "r"((x86_reg)3L*stride) yading@10: :"memory"); yading@10: // STOP_TIMER("put_vp_no_rnd_pixels8_l2_mmx") yading@10: } yading@10: #endif /* HAVE_INLINE_ASM */ yading@10: yading@10: av_cold void ff_vp3dsp_init_x86(VP3DSPContext *c, int flags) yading@10: { yading@10: int cpuflags = av_get_cpu_flags(); yading@10: yading@10: #if HAVE_INLINE_ASM yading@10: c->put_no_rnd_pixels_l2 = put_vp_no_rnd_pixels8_l2_mmx; yading@10: #endif /* HAVE_INLINE_ASM */ yading@10: yading@10: #if ARCH_X86_32 yading@10: if (EXTERNAL_MMX(cpuflags)) { yading@10: c->idct_put = ff_vp3_idct_put_mmx; yading@10: c->idct_add = ff_vp3_idct_add_mmx; yading@10: } yading@10: #endif yading@10: yading@10: if (EXTERNAL_MMXEXT(cpuflags)) { yading@10: c->idct_dc_add = ff_vp3_idct_dc_add_mmxext; yading@10: yading@10: if (!(flags & CODEC_FLAG_BITEXACT)) { yading@10: c->v_loop_filter = ff_vp3_v_loop_filter_mmxext; yading@10: c->h_loop_filter = ff_vp3_h_loop_filter_mmxext; yading@10: } yading@10: } yading@10: yading@10: if (EXTERNAL_SSE2(cpuflags)) { yading@10: c->idct_put = ff_vp3_idct_put_sse2; yading@10: c->idct_add = ff_vp3_idct_add_sse2; yading@10: } yading@10: }