diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index f13f3e0ea8..d58ec84c64 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -31,7 +31,8 @@ OBJS-$(CONFIG_VP6_DECODER) += x86/vp56dsp_init.o OBJS-$(CONFIG_VP8_DECODER) += x86/vp8dsp_init.o OBJS-$(CONFIG_XMM_CLOBBER_TEST) += x86/w64xmmtest.o -MMX-OBJS-$(CONFIG_DSPUTIL) += x86/dsputil_mmx.o \ +MMX-OBJS-$(CONFIG_DSPUTIL) += x86/dsputil_init.o \ + x86/dsputil_mmx.o \ x86/fdct.o \ x86/fpel_mmx.o \ x86/idct_mmx_xvid.o \ diff --git a/libavcodec/x86/dsputil_init.c b/libavcodec/x86/dsputil_init.c new file mode 100644 index 0000000000..2122874cf3 --- /dev/null +++ b/libavcodec/x86/dsputil_init.c @@ -0,0 +1,718 @@ +/* + * 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 "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/x86/asm.h" +#include "libavcodec/dsputil.h" +#include "libavcodec/simple_idct.h" +#include "dsputil_mmx.h" +#include "idct_xvid.h" + +void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, + int dstStride, int src1Stride, int h); +void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, + uint8_t *src2, int dstStride, + int src1Stride, int h); +void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, + int dstStride, int src1Stride, int h); +void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, + int dstStride, int src1Stride, int h); +void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, + int dstStride, int src1Stride, int h); +void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, + int dstStride, int src1Stride, int h); +void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, int h); +void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, int h); +void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, + int h); +void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, int h); +void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, int h); +void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride, + int h); +void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, + int dstStride, int srcStride); +#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext +#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext + +void ff_h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale); +void ff_h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale); + +int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2, + int order); +int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, + int order); +int32_t ff_scalarproduct_and_madd_int16_mmxext(int16_t *v1, const int16_t *v2, + const int16_t *v3, + int order, int mul); +int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, + const int16_t *v3, + int order, int mul); +int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, + const int16_t *v3, + int order, int mul); + +void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); +void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, + const int16_t *window, unsigned int len); + +void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w); +void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w); + +void ff_add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top, + const uint8_t *diff, int w, + int *left, int *left_top); +int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, + int w, int left); +int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, + int w, int left); + +void ff_vector_clip_int32_mmx (int32_t *dst, const int32_t *src, + int32_t min, int32_t max, unsigned int len); +void ff_vector_clip_int32_sse2 (int32_t *dst, const int32_t *src, + int32_t min, int32_t max, unsigned int len); +void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src, + int32_t min, int32_t max, unsigned int len); +void ff_vector_clip_int32_sse4 (int32_t *dst, const int32_t *src, + int32_t min, int32_t max, unsigned int len); + +#if HAVE_YASM + +PIXELS16(static, ff_avg, , , _mmxext) +PIXELS16(static, ff_put, , , _mmxext) + +#define QPEL_OP(OPNAME, RND, MMX) \ +static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[8]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \ + stride, 8); \ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \ + stride, stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, \ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[8]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \ + stride, 8); \ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride, \ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[8]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \ + 8, stride); \ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \ + stride, stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, \ + stride, stride); \ +} \ + \ +static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[8]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \ + 8, stride); \ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half) + 64; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ + ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ + stride, 8, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, \ + 8, stride, 9); \ + ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[8 + 9]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ + stride, 9); \ + ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[9]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ + stride, 9); \ + ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ + stride, 8); \ +} \ + \ +static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[32]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \ + stride, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \ + stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, \ + stride, stride, 16);\ +} \ + \ +static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[32]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \ + stride, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half, \ + stride, stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[32]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \ + stride); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \ + stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, \ + stride, stride); \ +} \ + \ +static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t temp[32]; \ + uint8_t * const half = (uint8_t*)temp; \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \ + stride); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, \ + stride, stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[16 * 2 + 17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half) + 256; \ + uint8_t * const halfHV = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ + 16, 16); \ + ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ + stride, 16, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ + stride, 17); \ + ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ + stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ + stride, 17); \ + ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ + stride, 16); \ +} \ + \ +static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \ + ptrdiff_t stride) \ +{ \ + uint64_t half[17 * 2]; \ + uint8_t * const halfH = ((uint8_t*)half); \ + ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ + stride, 17); \ + ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ + stride, 16); \ +} + +QPEL_OP(put_, _, mmxext) +QPEL_OP(avg_, _, mmxext) +QPEL_OP(put_no_rnd_, _no_rnd_, mmxext) +#endif /* HAVE_YASM */ + +#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \ + do { \ + c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \ + c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \ + } while (0) + +static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_MMX_INLINE + const int high_bit_depth = avctx->bits_per_raw_sample > 8; + + c->put_pixels_clamped = ff_put_pixels_clamped_mmx; + c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx; + c->add_pixels_clamped = ff_add_pixels_clamped_mmx; + + if (!high_bit_depth) { + c->clear_block = ff_clear_block_mmx; + c->clear_blocks = ff_clear_blocks_mmx; + c->draw_edges = ff_draw_edges_mmx; + + switch (avctx->idct_algo) { + case FF_IDCT_AUTO: + case FF_IDCT_SIMPLEMMX: + c->idct_put = ff_simple_idct_put_mmx; + c->idct_add = ff_simple_idct_add_mmx; + c->idct = ff_simple_idct_mmx; + c->idct_permutation_type = FF_SIMPLE_IDCT_PERM; + break; + case FF_IDCT_XVIDMMX: + c->idct_put = ff_idct_xvid_mmx_put; + c->idct_add = ff_idct_xvid_mmx_add; + c->idct = ff_idct_xvid_mmx; + break; + } + } + + c->gmc = ff_gmc_mmx; + + c->add_bytes = ff_add_bytes_mmx; +#endif /* HAVE_MMX_INLINE */ + +#if HAVE_MMX_EXTERNAL + if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { + c->h263_v_loop_filter = ff_h263_v_loop_filter_mmx; + c->h263_h_loop_filter = ff_h263_h_loop_filter_mmx; + } + + c->vector_clip_int32 = ff_vector_clip_int32_mmx; +#endif /* HAVE_MMX_EXTERNAL */ +} + +static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_MMXEXT_INLINE + const int high_bit_depth = avctx->bits_per_raw_sample > 8; + + if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) { + c->idct_put = ff_idct_xvid_mmxext_put; + c->idct_add = ff_idct_xvid_mmxext_add; + c->idct = ff_idct_xvid_mmxext; + } +#endif /* HAVE_MMXEXT_INLINE */ + +#if HAVE_MMXEXT_EXTERNAL + SET_QPEL_FUNCS(avg_qpel, 0, 16, mmxext, ); + SET_QPEL_FUNCS(avg_qpel, 1, 8, mmxext, ); + + SET_QPEL_FUNCS(put_qpel, 0, 16, mmxext, ); + SET_QPEL_FUNCS(put_qpel, 1, 8, mmxext, ); + SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, ); + SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmxext, ); + + /* slower than cmov version on AMD */ + if (!(mm_flags & AV_CPU_FLAG_3DNOW)) + c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext; + + c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext; + c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext; + + if (avctx->flags & CODEC_FLAG_BITEXACT) { + c->apply_window_int16 = ff_apply_window_int16_mmxext; + } else { + c->apply_window_int16 = ff_apply_window_int16_round_mmxext; + } +#endif /* HAVE_MMXEXT_EXTERNAL */ +} + +static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_SSE_INLINE + const int high_bit_depth = avctx->bits_per_raw_sample > 8; + + if (!high_bit_depth) { + if (!(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)) { + /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */ + c->clear_block = ff_clear_block_sse; + c->clear_blocks = ff_clear_blocks_sse; + } + } + + c->vector_clipf = ff_vector_clipf_sse; +#endif /* HAVE_SSE_INLINE */ +} + +static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_SSE2_INLINE + const int high_bit_depth = avctx->bits_per_raw_sample > 8; + + if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) { + c->idct_put = ff_idct_xvid_sse2_put; + c->idct_add = ff_idct_xvid_sse2_add; + c->idct = ff_idct_xvid_sse2; + c->idct_permutation_type = FF_SSE2_IDCT_PERM; + } +#endif /* HAVE_SSE2_INLINE */ + +#if HAVE_SSE2_EXTERNAL + c->scalarproduct_int16 = ff_scalarproduct_int16_sse2; + c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2; + if (mm_flags & AV_CPU_FLAG_ATOM) { + c->vector_clip_int32 = ff_vector_clip_int32_int_sse2; + } else { + c->vector_clip_int32 = ff_vector_clip_int32_sse2; + } + if (avctx->flags & CODEC_FLAG_BITEXACT) { + c->apply_window_int16 = ff_apply_window_int16_sse2; + } else if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) { + c->apply_window_int16 = ff_apply_window_int16_round_sse2; + } + c->bswap_buf = ff_bswap32_buf_sse2; +#endif /* HAVE_SSE2_EXTERNAL */ +} + +static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_SSSE3_EXTERNAL + c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3; + if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe + c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4; + + if (mm_flags & AV_CPU_FLAG_ATOM) + c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; + else + c->apply_window_int16 = ff_apply_window_int16_ssse3; + if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) // cachesplit + c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3; + c->bswap_buf = ff_bswap32_buf_ssse3; +#endif /* HAVE_SSSE3_EXTERNAL */ +} + +static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx, + int mm_flags) +{ +#if HAVE_SSE4_EXTERNAL + c->vector_clip_int32 = ff_vector_clip_int32_sse4; +#endif /* HAVE_SSE4_EXTERNAL */ +} + +av_cold void ff_dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx) +{ + int mm_flags = av_get_cpu_flags(); + +#if HAVE_7REGS && HAVE_INLINE_ASM + if (mm_flags & AV_CPU_FLAG_CMOV) + c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_cmov; +#endif + + if (mm_flags & AV_CPU_FLAG_MMX) + dsputil_init_mmx(c, avctx, mm_flags); + + if (mm_flags & AV_CPU_FLAG_MMXEXT) + dsputil_init_mmxext(c, avctx, mm_flags); + + if (mm_flags & AV_CPU_FLAG_SSE) + dsputil_init_sse(c, avctx, mm_flags); + + if (mm_flags & AV_CPU_FLAG_SSE2) + dsputil_init_sse2(c, avctx, mm_flags); + + if (mm_flags & AV_CPU_FLAG_SSSE3) + dsputil_init_ssse3(c, avctx, mm_flags); + + if (mm_flags & AV_CPU_FLAG_SSE4) + dsputil_init_sse4(c, avctx, mm_flags); + + if (CONFIG_ENCODERS) + ff_dsputilenc_init_mmx(c, avctx); +} diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c index 89da17890a..832ee5acfe 100644 --- a/libavcodec/x86/dsputil_mmx.c +++ b/libavcodec/x86/dsputil_mmx.c @@ -22,68 +22,14 @@ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ -#include "libavutil/attributes.h" +#include "config.h" #include "libavutil/cpu.h" #include "libavutil/x86/asm.h" -#include "libavcodec/dsputil.h" -#include "libavcodec/h264dsp.h" -#include "libavcodec/mpegvideo.h" -#include "libavcodec/simple_idct.h" #include "constants.h" #include "dsputil_mmx.h" -#include "idct_xvid.h" - -//#undef NDEBUG -//#include - -void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, - int dstStride, int src1Stride, int h); -void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, - uint8_t *src2, int dstStride, - int src1Stride, int h); -void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, - int dstStride, int src1Stride, int h); -void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, - int dstStride, int src1Stride, int h); -void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, - int dstStride, int src1Stride, int h); -void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2, - int dstStride, int src1Stride, int h); -void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, int h); -void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, int h); -void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, - int h); -void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, int h); -void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, int h); -void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride, - int h); -void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src, - int dstStride, int srcStride); -#define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext -#define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext - #if HAVE_INLINE_ASM -/***********************************/ -/* standard MMX */ - void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_size) { @@ -219,7 +165,7 @@ void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, } #define CLEAR_BLOCKS(name, n) \ -static void name(int16_t *blocks) \ +void name(int16_t *blocks) \ { \ __asm__ volatile ( \ "pxor %%mm7, %%mm7 \n\t" \ @@ -236,10 +182,10 @@ static void name(int16_t *blocks) \ : "%"REG_a \ ); \ } -CLEAR_BLOCKS(clear_blocks_mmx, 6) -CLEAR_BLOCKS(clear_block_mmx, 1) +CLEAR_BLOCKS(ff_clear_blocks_mmx, 6) +CLEAR_BLOCKS(ff_clear_block_mmx, 1) -static void clear_block_sse(int16_t *block) +void ff_clear_block_sse(int16_t *block) { __asm__ volatile ( "xorps %%xmm0, %%xmm0 \n" @@ -256,7 +202,7 @@ static void clear_block_sse(int16_t *block) ); } -static void clear_blocks_sse(int16_t *blocks) +void ff_clear_blocks_sse(int16_t *blocks) { __asm__ volatile ( "xorps %%xmm0, %%xmm0 \n" @@ -278,7 +224,7 @@ static void clear_blocks_sse(int16_t *blocks) ); } -static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) +void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) { x86_reg i = 0; __asm__ volatile ( @@ -304,9 +250,9 @@ static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) } #if HAVE_7REGS -static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, - const uint8_t *diff, int w, - int *left, int *left_top) +void ff_add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, + const uint8_t *diff, int w, + int *left, int *left_top) { x86_reg w2 = -w; x86_reg x; @@ -343,8 +289,8 @@ static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, /* Draw the edges of width 'w' of an image of size width, height * this MMX version can only handle w == 8 || w == 16. */ -static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, - int w, int h, int sides) +void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, + int w, int h, int sides) { uint8_t *ptr, *last_line; int i; @@ -436,406 +382,11 @@ static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, } } } -#endif /* HAVE_INLINE_ASM */ - - -#if HAVE_YASM - -PIXELS16(static, ff_avg, , , _mmxext) -PIXELS16(static, ff_put, , , _mmxext) - -#define QPEL_OP(OPNAME, RND, MMX) \ -static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[8]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \ - stride, 8); \ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \ - stride, stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, \ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[8]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \ - stride, 8); \ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride, \ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[8]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \ - 8, stride); \ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half, \ - stride, stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, \ - stride, stride); \ -} \ - \ -static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[8]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, \ - 8, stride); \ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half) + 64; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\ - ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, \ - stride, 8, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, \ - 8, stride, 9); \ - ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[8 + 9]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \ - stride, 9); \ - ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[9]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \ - stride, 9); \ - ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, \ - stride, 8); \ -} \ - \ -static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[32]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \ - stride, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \ - stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, \ - stride, stride, 16);\ -} \ - \ -static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[32]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \ - stride, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half, \ - stride, stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[32]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \ - stride); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, \ - stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, \ - stride, stride); \ -} \ - \ -static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t temp[32]; \ - uint8_t * const half = (uint8_t*)temp; \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \ - stride); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, \ - stride, stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[16 * 2 + 17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half) + 256; \ - uint8_t * const halfHV = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \ - 16, 16); \ - ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, \ - stride, 16, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \ - stride, 17); \ - ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ - stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \ - stride, 17); \ - ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ - stride, 16); \ -} \ - \ -static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \ - ptrdiff_t stride) \ -{ \ - uint64_t half[17 * 2]; \ - uint8_t * const halfH = ((uint8_t*)half); \ - ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \ - stride, 17); \ - ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, \ - stride, 16); \ -} -QPEL_OP(put_, _, mmxext) -QPEL_OP(avg_, _, mmxext) -QPEL_OP(put_no_rnd_, _no_rnd_, mmxext) -#endif /* HAVE_YASM */ - - -#if HAVE_INLINE_ASM -static void gmc_mmx(uint8_t *dst, uint8_t *src, - int stride, int h, int ox, int oy, - int dxx, int dxy, int dyx, int dyy, - int shift, int r, int width, int height) +void ff_gmc_mmx(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height) { const int w = 8; const int ix = ox >> (16 + shift); @@ -948,8 +499,8 @@ static void gmc_mmx(uint8_t *dst, uint8_t *src, } } -static void vector_clipf_sse(float *dst, const float *src, - float min, float max, int len) +void ff_vector_clipf_sse(float *dst, const float *src, + float min, float max, int len) { x86_reg i = (len - 16) * 4; __asm__ volatile ( @@ -983,262 +534,3 @@ static void vector_clipf_sse(float *dst, const float *src, } #endif /* HAVE_INLINE_ASM */ - -void ff_h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale); -void ff_h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale); - -int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2, - int order); -int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2, - int order); -int32_t ff_scalarproduct_and_madd_int16_mmxext(int16_t *v1, const int16_t *v2, - const int16_t *v3, - int order, int mul); -int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2, - const int16_t *v3, - int order, int mul); -int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2, - const int16_t *v3, - int order, int mul); - -void ff_apply_window_int16_round_mmxext(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_round_sse2(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_mmxext(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_sse2(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_ssse3(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); -void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input, - const int16_t *window, unsigned int len); - -void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w); -void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w); - -void ff_add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top, - const uint8_t *diff, int w, - int *left, int *left_top); -int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src, - int w, int left); -int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src, - int w, int left); - -void ff_vector_clip_int32_mmx (int32_t *dst, const int32_t *src, - int32_t min, int32_t max, unsigned int len); -void ff_vector_clip_int32_sse2 (int32_t *dst, const int32_t *src, - int32_t min, int32_t max, unsigned int len); -void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src, - int32_t min, int32_t max, unsigned int len); -void ff_vector_clip_int32_sse4 (int32_t *dst, const int32_t *src, - int32_t min, int32_t max, unsigned int len); - -#define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \ - do { \ - c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \ - c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \ - } while (0) - -static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_MMX_INLINE - const int high_bit_depth = avctx->bits_per_raw_sample > 8; - - c->put_pixels_clamped = ff_put_pixels_clamped_mmx; - c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx; - c->add_pixels_clamped = ff_add_pixels_clamped_mmx; - - if (!high_bit_depth) { - c->clear_block = clear_block_mmx; - c->clear_blocks = clear_blocks_mmx; - c->draw_edges = draw_edges_mmx; - - switch (avctx->idct_algo) { - case FF_IDCT_AUTO: - case FF_IDCT_SIMPLEMMX: - c->idct_put = ff_simple_idct_put_mmx; - c->idct_add = ff_simple_idct_add_mmx; - c->idct = ff_simple_idct_mmx; - c->idct_permutation_type = FF_SIMPLE_IDCT_PERM; - break; - case FF_IDCT_XVIDMMX: - c->idct_put = ff_idct_xvid_mmx_put; - c->idct_add = ff_idct_xvid_mmx_add; - c->idct = ff_idct_xvid_mmx; - break; - } - } - - c->gmc = gmc_mmx; - - c->add_bytes = add_bytes_mmx; -#endif /* HAVE_MMX_INLINE */ - -#if HAVE_MMX_EXTERNAL - if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) { - c->h263_v_loop_filter = ff_h263_v_loop_filter_mmx; - c->h263_h_loop_filter = ff_h263_h_loop_filter_mmx; - } - - c->vector_clip_int32 = ff_vector_clip_int32_mmx; -#endif /* HAVE_MMX_EXTERNAL */ -} - -static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_MMXEXT_INLINE - const int high_bit_depth = avctx->bits_per_raw_sample > 8; - - if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) { - c->idct_put = ff_idct_xvid_mmxext_put; - c->idct_add = ff_idct_xvid_mmxext_add; - c->idct = ff_idct_xvid_mmxext; - } -#endif /* HAVE_MMXEXT_INLINE */ - -#if HAVE_MMXEXT_EXTERNAL - SET_QPEL_FUNCS(avg_qpel, 0, 16, mmxext, ); - SET_QPEL_FUNCS(avg_qpel, 1, 8, mmxext, ); - - SET_QPEL_FUNCS(put_qpel, 0, 16, mmxext, ); - SET_QPEL_FUNCS(put_qpel, 1, 8, mmxext, ); - SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, ); - SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmxext, ); - - /* slower than cmov version on AMD */ - if (!(mm_flags & AV_CPU_FLAG_3DNOW)) - c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext; - - c->scalarproduct_int16 = ff_scalarproduct_int16_mmxext; - c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext; - - if (avctx->flags & CODEC_FLAG_BITEXACT) { - c->apply_window_int16 = ff_apply_window_int16_mmxext; - } else { - c->apply_window_int16 = ff_apply_window_int16_round_mmxext; - } -#endif /* HAVE_MMXEXT_EXTERNAL */ -} - -static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_SSE_INLINE - const int high_bit_depth = avctx->bits_per_raw_sample > 8; - - if (!high_bit_depth) { - if (!(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)) { - /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */ - c->clear_block = clear_block_sse; - c->clear_blocks = clear_blocks_sse; - } - } - - c->vector_clipf = vector_clipf_sse; -#endif /* HAVE_SSE_INLINE */ -} - -static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_SSE2_INLINE - const int high_bit_depth = avctx->bits_per_raw_sample > 8; - - if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX) { - c->idct_put = ff_idct_xvid_sse2_put; - c->idct_add = ff_idct_xvid_sse2_add; - c->idct = ff_idct_xvid_sse2; - c->idct_permutation_type = FF_SSE2_IDCT_PERM; - } -#endif /* HAVE_SSE2_INLINE */ - -#if HAVE_SSE2_EXTERNAL - c->scalarproduct_int16 = ff_scalarproduct_int16_sse2; - c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2; - if (mm_flags & AV_CPU_FLAG_ATOM) { - c->vector_clip_int32 = ff_vector_clip_int32_int_sse2; - } else { - c->vector_clip_int32 = ff_vector_clip_int32_sse2; - } - if (avctx->flags & CODEC_FLAG_BITEXACT) { - c->apply_window_int16 = ff_apply_window_int16_sse2; - } else if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) { - c->apply_window_int16 = ff_apply_window_int16_round_sse2; - } - c->bswap_buf = ff_bswap32_buf_sse2; -#endif /* HAVE_SSE2_EXTERNAL */ -} - -static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_SSSE3_EXTERNAL - c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3; - if (mm_flags & AV_CPU_FLAG_SSE4) // not really sse4, just slow on Conroe - c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4; - - if (mm_flags & AV_CPU_FLAG_ATOM) - c->apply_window_int16 = ff_apply_window_int16_ssse3_atom; - else - c->apply_window_int16 = ff_apply_window_int16_ssse3; - if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW))) // cachesplit - c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3; - c->bswap_buf = ff_bswap32_buf_ssse3; -#endif /* HAVE_SSSE3_EXTERNAL */ -} - -static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx, - int mm_flags) -{ -#if HAVE_SSE4_EXTERNAL - c->vector_clip_int32 = ff_vector_clip_int32_sse4; -#endif /* HAVE_SSE4_EXTERNAL */ -} - -av_cold void ff_dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx) -{ - int mm_flags = av_get_cpu_flags(); - -#if HAVE_7REGS && HAVE_INLINE_ASM - if (mm_flags & AV_CPU_FLAG_CMOV) - c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov; -#endif - - if (mm_flags & AV_CPU_FLAG_MMX) - dsputil_init_mmx(c, avctx, mm_flags); - - if (mm_flags & AV_CPU_FLAG_MMXEXT) - dsputil_init_mmxext(c, avctx, mm_flags); - - if (mm_flags & AV_CPU_FLAG_SSE) - dsputil_init_sse(c, avctx, mm_flags); - - if (mm_flags & AV_CPU_FLAG_SSE2) - dsputil_init_sse2(c, avctx, mm_flags); - - if (mm_flags & AV_CPU_FLAG_SSSE3) - dsputil_init_ssse3(c, avctx, mm_flags); - - if (mm_flags & AV_CPU_FLAG_SSE4) - dsputil_init_sse4(c, avctx, mm_flags); - - if (CONFIG_ENCODERS) - ff_dsputilenc_init_mmx(c, avctx); -} diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h index 388916b895..c8615b2472 100644 --- a/libavcodec/x86/dsputil_mmx.h +++ b/libavcodec/x86/dsputil_mmx.h @@ -110,6 +110,27 @@ void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_s void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_size); void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, int line_size); +void ff_clear_block_mmx(int16_t *block); +void ff_clear_block_sse(int16_t *block); +void ff_clear_blocks_mmx(int16_t *blocks); +void ff_clear_blocks_sse(int16_t *blocks); + +void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w); + +void ff_add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top, + const uint8_t *diff, int w, + int *left, int *left_top); + +void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, + int w, int h, int sides); + +void ff_gmc_mmx(uint8_t *dst, uint8_t *src, + int stride, int h, int ox, int oy, + int dxx, int dxy, int dyx, int dyy, + int shift, int r, int width, int height); + +void ff_vector_clipf_sse(float *dst, const float *src, + float min, float max, int len); void ff_avg_pixels8_mmx(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h);