yading@10: /*
yading@10:  * VC3/DNxHD SIMD functions
yading@10:  * Copyright (c) 2007 Baptiste Coudurier <baptiste dot coudurier at smartjog dot com>
yading@10:  *
yading@10:  * VC-3 encoder funded by the British Broadcasting Corporation
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 "libavutil/attributes.h"
yading@10: #include "libavutil/x86/asm.h"
yading@10: #include "libavcodec/dnxhdenc.h"
yading@10: 
yading@10: #if HAVE_SSE2_INLINE
yading@10: 
yading@10: static void get_pixels_8x4_sym_sse2(int16_t *block, const uint8_t *pixels, int line_size)
yading@10: {
yading@10:     __asm__ volatile(
yading@10:         "pxor %%xmm5,      %%xmm5       \n\t"
yading@10:         "movq (%0),        %%xmm0       \n\t"
yading@10:         "add  %2,          %0           \n\t"
yading@10:         "movq (%0),        %%xmm1       \n\t"
yading@10:         "movq (%0, %2),    %%xmm2       \n\t"
yading@10:         "movq (%0, %2,2),  %%xmm3       \n\t"
yading@10:         "punpcklbw %%xmm5, %%xmm0       \n\t"
yading@10:         "punpcklbw %%xmm5, %%xmm1       \n\t"
yading@10:         "punpcklbw %%xmm5, %%xmm2       \n\t"
yading@10:         "punpcklbw %%xmm5, %%xmm3       \n\t"
yading@10:         "movdqa %%xmm0,      (%1)       \n\t"
yading@10:         "movdqa %%xmm1,    16(%1)       \n\t"
yading@10:         "movdqa %%xmm2,    32(%1)       \n\t"
yading@10:         "movdqa %%xmm3,    48(%1)       \n\t"
yading@10:         "movdqa %%xmm3 ,   64(%1)       \n\t"
yading@10:         "movdqa %%xmm2 ,   80(%1)       \n\t"
yading@10:         "movdqa %%xmm1 ,   96(%1)       \n\t"
yading@10:         "movdqa %%xmm0,   112(%1)       \n\t"
yading@10:         : "+r" (pixels)
yading@10:         : "r" (block), "r" ((x86_reg)line_size)
yading@10:     );
yading@10: }
yading@10: 
yading@10: #endif /* HAVE_SSE2_INLINE */
yading@10: 
yading@10: av_cold void ff_dnxhdenc_init_x86(DNXHDEncContext *ctx)
yading@10: {
yading@10: #if HAVE_SSE2_INLINE
yading@10:     if (av_get_cpu_flags() & AV_CPU_FLAG_SSE2) {
yading@10:         if (ctx->cid_table->bit_depth == 8)
yading@10:             ctx->get_pixels_8x4_sym = get_pixels_8x4_sym_sse2;
yading@10:     }
yading@10: #endif /* HAVE_SSE2_INLINE */
yading@10: }