av_cold void ff_interlace_init_x86(InterlaceContext *s, int depth)
{
int cpu_flags = av_get_cpu_flags();
if (depth > 8) {
if (EXTERNAL_SSE2(cpu_flags)) {
if (s->lowpass == VLPF_LIN)
s->lowpass_line = ff_lowpass_line_16_sse2;
else if (s->lowpass == VLPF_CMP)
s->lowpass_line = ff_lowpass_line_complex_12_sse2;
}
if (EXTERNAL_AVX(cpu_flags))
if (s->lowpass == VLPF_LIN)
s->lowpass_line = ff_lowpass_line_16_avx;
} else {
if (EXTERNAL_SSE2(cpu_flags)) {
if (s->lowpass == VLPF_LIN)
s->lowpass_line = ff_lowpass_line_sse2;
else if (s->lowpass == VLPF_CMP)
s->lowpass_line = ff_lowpass_line_complex_sse2;
}
if (EXTERNAL_AVX(cpu_flags))
if (s->lowpass == VLPF_LIN)
s->lowpass_line = ff_lowpass_line_avx;
}
}
void ff_hevcpred_init_x86(HEVCPredContext *c, const int bit_depth)
{
int mm_flags = av_get_cpu_flags();
if (bit_depth == 8) {
if (EXTERNAL_MMX(mm_flags)) {
if (EXTERNAL_MMXEXT(mm_flags)) {
if (EXTERNAL_SSE2(mm_flags)) {
}
if (EXTERNAL_SSSE3(mm_flags)) {
}
if (EXTERNAL_SSE4(mm_flags)) {
c->pred_planar[0]= pred_planar_0_8_sse;
c->pred_planar[1]= pred_planar_1_8_sse;
c->pred_planar[2]= pred_planar_2_8_sse;
c->pred_planar[3]= pred_planar_3_8_sse;
// c->pred_angular[0]= pred_angular_0_8_sse; //removed because too little data = bad performance
c->pred_angular[1]= pred_angular_1_8_sse;
c->pred_angular[2]= pred_angular_2_8_sse;
c->pred_angular[3]= pred_angular_3_8_sse;
}
if (EXTERNAL_AVX(mm_flags)) {
}
}
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMX(mm_flags)) {
if (EXTERNAL_MMXEXT(mm_flags)) {
if (EXTERNAL_SSE2(mm_flags)) {
}
if (EXTERNAL_SSE4(mm_flags)) {
}
if (EXTERNAL_AVX(mm_flags)) {
}
}
}
}
}
av_cold void ff_fft_init_x86(FFTContext *s)
{
int cpu_flags = av_get_cpu_flags();
#if ARCH_X86_32
if (EXTERNAL_AMD3DNOW(cpu_flags)) {
/* 3DNow! for K6-2/3 */
s->imdct_calc = ff_imdct_calc_3dnow;
s->imdct_half = ff_imdct_half_3dnow;
s->fft_calc = ff_fft_calc_3dnow;
}
if (EXTERNAL_AMD3DNOWEXT(cpu_flags)) {
/* 3DNowEx for K7 */
s->imdct_calc = ff_imdct_calc_3dnowext;
s->imdct_half = ff_imdct_half_3dnowext;
s->fft_calc = ff_fft_calc_3dnowext;
}
#endif
if (EXTERNAL_SSE(cpu_flags)) {
/* SSE for P3/P4/K8 */
s->imdct_calc = ff_imdct_calc_sse;
s->imdct_half = ff_imdct_half_sse;
s->fft_permute = ff_fft_permute_sse;
s->fft_calc = ff_fft_calc_sse;
s->fft_permutation = FF_FFT_PERM_SWAP_LSBS;
}
if (EXTERNAL_AVX(cpu_flags) && s->nbits >= 5) {
/* AVX for SB */
s->imdct_half = ff_imdct_half_avx;
s->fft_calc = ff_fft_calc_avx;
s->fft_permutation = FF_FFT_PERM_AVX;
}
}
void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
{
int mm_flags = av_get_cpu_flags();
#if HAVE_6REGS && HAVE_INLINE_ASM
if (INLINE_AMD3DNOWEXT(mm_flags)) {
fdsp->vector_fmul_window = vector_fmul_window_3dnowext;
}
if (INLINE_SSE(mm_flags)) {
fdsp->vector_fmul_window = vector_fmul_window_sse;
}
#endif
if (EXTERNAL_SSE(mm_flags)) {
fdsp->vector_fmul = ff_vector_fmul_sse;
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse;
fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse;
fdsp->vector_fmul_add = ff_vector_fmul_add_sse;
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
fdsp->scalarproduct_float = ff_scalarproduct_float_sse;
}
if (EXTERNAL_SSE2(mm_flags)) {
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2;
}
if (EXTERNAL_AVX(mm_flags)) {
fdsp->vector_fmul = ff_vector_fmul_avx;
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx;
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx;
fdsp->vector_fmul_add = ff_vector_fmul_add_avx;
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
}
}
av_cold void ff_h264chroma_init_x86(H264ChromaContext *c, int bit_depth)
{
int high_bit_depth = bit_depth > 8;
int cpu_flags = av_get_cpu_flags();
#if (HAVE_MMX_EXTERNAL == 1)
if (EXTERNAL_MMX(cpu_flags) && !high_bit_depth) {
c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_rnd_mmx;
c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmx;
}
#endif
#if (HAVE_AMD3DNOW_EXTERNAL == 1)
if (EXTERNAL_AMD3DNOW(cpu_flags) && !high_bit_depth) {
c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_rnd_3dnow;
c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_3dnow;
}
#endif
#if (HAVE_MMXEXT_EXTERNAL == 1)
if (EXTERNAL_MMXEXT(cpu_flags) && !high_bit_depth) {
c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_rnd_mmxext;
c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmxext;
c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_mmxext;
c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_mmxext;
}
if (EXTERNAL_MMXEXT(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_10_mmxext;
c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_10_mmxext;
c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_10_mmxext;
c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_10_mmxext;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_sse2;
c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_sse2;
}
#endif
#if (HAVE_SSSE3_EXTERNAL == 1)
if (EXTERNAL_SSSE3(cpu_flags) && !high_bit_depth) {
c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_rnd_ssse3;
c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_rnd_ssse3;
c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_ssse3;
c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_ssse3;
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(cpu_flags) && bit_depth > 8 && bit_depth <= 10) {
// AVX implies !cache64.
// TODO: Port cache(32|64) detection from x264.
c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_avx;
c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_avx;
}
#endif
}
av_cold void ff_interlace_init_x86(InterlaceContext *s)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSE2(cpu_flags))
s->lowpass_line = ff_lowpass_line_sse2;
if (EXTERNAL_AVX(cpu_flags))
s->lowpass_line = ff_lowpass_line_avx;
}
av_cold void ff_dct_init_x86(DCTContext *s)
{
int has_vectors = av_get_cpu_flags();
if (EXTERNAL_SSE(has_vectors))
s->dct32 = ff_dct32_float_sse;
if (EXTERNAL_SSE2(has_vectors))
s->dct32 = ff_dct32_float_sse2;
if (EXTERNAL_AVX(has_vectors))
s->dct32 = ff_dct32_float_avx;
}
av_cold void ff_v210enc_init_x86(V210EncContext *s)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSSE3(cpu_flags)) {
s->pack_line_8 = ff_v210_planar_pack_8_ssse3;
s->pack_line_10 = ff_v210_planar_pack_10_ssse3;
}
if (EXTERNAL_AVX(cpu_flags))
s->pack_line_8 = ff_v210_planar_pack_8_avx;
}
av_cold void ff_init_lls_x86(LLSModel2 *m)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSE2(cpu_flags)) {
m->update_lls = ff_update_lls_sse2;
if (m->indep_count >= 4)
m->evaluate_lls = ff_evaluate_lls_sse2;
}
if (EXTERNAL_AVX(cpu_flags)) {
m->update_lls = ff_update_lls_avx;
}
}
av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
{
int cpu_flags = av_get_cpu_flags();
#if (HAVE_AMD3DNOWEXT_EXTERNAL == 1)
if (EXTERNAL_AMD3DNOWEXT(cpu_flags)) {
fdsp->vector_fmul_window = ff_vector_fmul_window_3dnowext;
}
#endif
#if (HAVE_SSE_EXTERNAL == 1)
if (EXTERNAL_SSE(cpu_flags)) {
fdsp->vector_fmul = ff_vector_fmul_sse;
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse;
fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse;
fdsp->vector_fmul_window = ff_vector_fmul_window_sse;
fdsp->vector_fmul_add = ff_vector_fmul_add_sse;
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
fdsp->scalarproduct_float = ff_scalarproduct_float_sse;
fdsp->butterflies_float = ff_butterflies_float_sse;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(cpu_flags)) {
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_sse2;
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(cpu_flags)) {
fdsp->vector_fmul = ff_vector_fmul_avx;
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_avx;
fdsp->vector_dmul_scalar = ff_vector_dmul_scalar_avx;
fdsp->vector_fmul_add = ff_vector_fmul_add_avx;
fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
}
#endif
#if (HAVE_FMA3_EXTERNAL == 1)
if (EXTERNAL_FMA3(cpu_flags)) {
fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_fma3;
fdsp->vector_fmul_add = ff_vector_fmul_add_fma3;
}
#endif
}
av_cold void ff_proresdsp_init_x86(ProresDSPContext *dsp)
{
#if ARCH_X86_64
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSE2(cpu_flags)) {
dsp->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
dsp->idct_put = ff_prores_idct_put_10_sse2;
}
if (EXTERNAL_SSE4(cpu_flags)) {
dsp->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
dsp->idct_put = ff_prores_idct_put_10_sse4;
}
if (EXTERNAL_AVX(cpu_flags)) {
dsp->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
dsp->idct_put = ff_prores_idct_put_10_avx;
}
#endif /* ARCH_X86_64 */
}
av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
{
#if HAVE_YASM
int cpu_flags = av_get_cpu_flags();
#if ARCH_X86_32
if (EXTERNAL_SSE(cpu_flags)) {
s->synth_filter_float = synth_filter_sse;
}
#endif
if (EXTERNAL_SSE2(cpu_flags)) {
s->synth_filter_float = synth_filter_sse2;
}
if (EXTERNAL_AVX(cpu_flags)) {
s->synth_filter_float = synth_filter_avx;
}
if (EXTERNAL_FMA3(cpu_flags)) {
s->synth_filter_float = synth_filter_fma3;
}
#endif /* HAVE_YASM */
}
av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
{
#if HAVE_YASM
int cpu_flags = av_get_cpu_flags();
#define init_fpel(idx1, idx2, sz, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_##type##sz##_##opt
#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_##opt; \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_##opt; \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_##opt
#define init_subpel2(idx, idxh, idxv, dir, type, opt) \
init_subpel1(0, idx, idxh, idxv, 64, dir, type, opt); \
init_subpel1(1, idx, idxh, idxv, 32, dir, type, opt); \
init_subpel1(2, idx, idxh, idxv, 16, dir, type, opt); \
init_subpel1(3, idx, idxh, idxv, 8, dir, type, opt); \
init_subpel1(4, idx, idxh, idxv, 4, dir, type, opt)
#define init_subpel3(idx, type, opt) \
init_subpel2(idx, 1, 1, hv, type, opt); \
init_subpel2(idx, 0, 1, v, type, opt); \
init_subpel2(idx, 1, 0, h, type, opt)
#define init_lpf(opt) do { \
if (ARCH_X86_64) { \
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_##opt; \
dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_##opt; \
dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_##opt; \
dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_##opt; \
dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_##opt; \
dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_##opt; \
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_##opt; \
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_##opt; \
} \
} while (0)
if (EXTERNAL_MMX(cpu_flags)) {
init_fpel(4, 0, 4, put, mmx);
init_fpel(3, 0, 8, put, mmx);
dsp->itxfm_add[4 /* lossless */][DCT_DCT] =
dsp->itxfm_add[4 /* lossless */][ADST_DCT] =
dsp->itxfm_add[4 /* lossless */][DCT_ADST] =
dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_mmx;
}
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_fpel(4, 1, 4, avg, mmxext);
init_fpel(3, 1, 8, avg, mmxext);
}
if (EXTERNAL_SSE(cpu_flags)) {
init_fpel(2, 0, 16, put, sse);
init_fpel(1, 0, 32, put, sse);
init_fpel(0, 0, 64, put, sse);
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_fpel(2, 1, 16, avg, sse2);
init_fpel(1, 1, 32, avg, sse2);
init_fpel(0, 1, 64, avg, sse2);
init_lpf(sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_subpel3(0, put, ssse3);
init_subpel3(1, avg, ssse3);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3;
if (ARCH_X86_64) {
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
}
init_lpf(ssse3);
}
if (EXTERNAL_AVX(cpu_flags)) {
if (ARCH_X86_64) {
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;
}
init_lpf(avx);
}
#undef init_fpel
#undef init_subpel1
#undef init_subpel2
#undef init_subpel3
#endif /* HAVE_YASM */
}
av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_MMX(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
0, 1, 8, "MMX", ff_conv_s32_to_s16_mmx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
6, 1, 4, "MMX", ff_conv_fltp_to_flt_6ch_mmx);
}
if (EXTERNAL_SSE(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
6, 1, 2, "SSE", ff_conv_fltp_to_s16_6ch_sse);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
2, 16, 8, "SSE", ff_conv_fltp_to_flt_2ch_sse);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
2, 16, 4, "SSE", ff_conv_flt_to_fltp_2ch_sse);
}
if (EXTERNAL_SSE2(cpu_flags)) {
if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
0, 16, 16, "SSE2", ff_conv_s32_to_s16_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
6, 16, 8, "SSE2", ff_conv_s16p_to_s16_6ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
6, 16, 4, "SSE2", ff_conv_fltp_to_s16_6ch_sse2);
} else {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
6, 1, 4, "SSE2SLOW", ff_conv_s16p_to_s16_6ch_sse2slow);
}
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S16,
0, 16, 8, "SSE2", ff_conv_s16_to_s32_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
0, 16, 8, "SSE2", ff_conv_s16_to_flt_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
0, 16, 8, "SSE2", ff_conv_s32_to_flt_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT,
0, 16, 16, "SSE2", ff_conv_flt_to_s16_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT,
0, 16, 16, "SSE2", ff_conv_flt_to_s32_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
2, 16, 16, "SSE2", ff_conv_s16p_to_s16_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
2, 16, 8, "SSE2", ff_conv_s16p_to_flt_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
6, 16, 4, "SSE2", ff_conv_s16p_to_flt_6ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
2, 16, 4, "SSE2", ff_conv_fltp_to_s16_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "SSE2", ff_conv_s16_to_s16p_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSE2", ff_conv_s16_to_s16p_6ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
2, 16, 8, "SSE2", ff_conv_s16_to_fltp_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSE2", ff_conv_s16_to_fltp_6ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
2, 16, 8, "SSE2", ff_conv_flt_to_s16p_2ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
6, 16, 4, "SSE2", ff_conv_flt_to_s16p_6ch_sse2);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
6, 16, 4, "SSE2", ff_conv_flt_to_fltp_6ch_sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
6, 16, 4, "SSSE3", ff_conv_s16p_to_flt_6ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
2, 16, 4, "SSSE3", ff_conv_fltp_to_s16_2ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "SSSE3", ff_conv_s16_to_s16p_2ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSSE3", ff_conv_s16_to_s16p_6ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
6, 16, 4, "SSSE3", ff_conv_s16_to_fltp_6ch_ssse3);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
6, 16, 4, "SSSE3", ff_conv_flt_to_s16p_6ch_ssse3);
}
if (EXTERNAL_SSE4(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
0, 16, 8, "SSE4", ff_conv_s16_to_flt_sse4);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
6, 16, 4, "SSE4", ff_conv_fltp_to_flt_6ch_sse4);
}
if (EXTERNAL_AVX_FAST(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
0, 32, 16, "AVX", ff_conv_s32_to_flt_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT,
0, 32, 32, "AVX", ff_conv_flt_to_s32_avx);
}
if (EXTERNAL_AVX(cpu_flags)) {
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
2, 16, 16, "AVX", ff_conv_s16p_to_s16_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
6, 16, 8, "AVX", ff_conv_s16p_to_s16_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
2, 16, 8, "AVX", ff_conv_s16p_to_flt_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
6, 16, 4, "AVX", ff_conv_s16p_to_flt_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
6, 16, 4, "AVX", ff_conv_fltp_to_s16_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
6, 16, 4, "AVX", ff_conv_fltp_to_flt_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
2, 16, 8, "AVX", ff_conv_s16_to_s16p_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
6, 16, 4, "AVX", ff_conv_s16_to_s16p_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
2, 16, 8, "AVX", ff_conv_s16_to_fltp_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
6, 16, 4, "AVX", ff_conv_s16_to_fltp_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
2, 16, 8, "AVX", ff_conv_flt_to_s16p_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
6, 16, 4, "AVX", ff_conv_flt_to_s16p_6ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
2, 16, 4, "AVX", ff_conv_flt_to_fltp_2ch_avx);
ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
6, 16, 4, "AVX", ff_conv_flt_to_fltp_6ch_avx);
}
}
av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
{
#if HAVE_YASM
int cpu_flags;
if (bpp == 10) {
ff_vp9dsp_init_10bpp_x86(dsp, bitexact);
return;
} else if (bpp == 12) {
ff_vp9dsp_init_12bpp_x86(dsp, bitexact);
return;
}
cpu_flags = av_get_cpu_flags();
#define init_lpf(opt) do { \
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_##opt; \
dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_##opt; \
dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_##opt; \
dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_##opt; \
dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_##opt; \
dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_##opt; \
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_##opt; \
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_##opt; \
} while (0)
#define init_ipred(sz, opt, t, e) \
dsp->intra_pred[TX_##sz##X##sz][e##_PRED] = ff_vp9_ipred_##t##_##sz##x##sz##_##opt
#define ff_vp9_ipred_hd_4x4_ssse3 ff_vp9_ipred_hd_4x4_mmxext
#define ff_vp9_ipred_vl_4x4_ssse3 ff_vp9_ipred_vl_4x4_mmxext
#define init_dir_tm_ipred(sz, opt) do { \
init_ipred(sz, opt, dl, DIAG_DOWN_LEFT); \
init_ipred(sz, opt, dr, DIAG_DOWN_RIGHT); \
init_ipred(sz, opt, hd, HOR_DOWN); \
init_ipred(sz, opt, vl, VERT_LEFT); \
init_ipred(sz, opt, hu, HOR_UP); \
init_ipred(sz, opt, tm, TM_VP8); \
init_ipred(sz, opt, vr, VERT_RIGHT); \
} while (0)
#define init_dir_tm_h_ipred(sz, opt) do { \
init_dir_tm_ipred(sz, opt); \
init_ipred(sz, opt, h, HOR); \
} while (0)
#define init_dc_ipred(sz, opt) do { \
init_ipred(sz, opt, dc, DC); \
init_ipred(sz, opt, dc_left, LEFT_DC); \
init_ipred(sz, opt, dc_top, TOP_DC); \
} while (0)
#define init_all_ipred(sz, opt) do { \
init_dc_ipred(sz, opt); \
init_dir_tm_h_ipred(sz, opt); \
} while (0)
if (EXTERNAL_MMX(cpu_flags)) {
init_fpel_func(4, 0, 4, put, , mmx);
init_fpel_func(3, 0, 8, put, , mmx);
if (!bitexact) {
dsp->itxfm_add[4 /* lossless */][DCT_DCT] =
dsp->itxfm_add[4 /* lossless */][ADST_DCT] =
dsp->itxfm_add[4 /* lossless */][DCT_ADST] =
dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_mmx;
}
init_ipred(8, mmx, v, VERT);
}
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_subpel2(4, 0, 4, put, 8, mmxext);
init_subpel2(4, 1, 4, avg, 8, mmxext);
init_fpel_func(4, 1, 4, avg, _8, mmxext);
init_fpel_func(3, 1, 8, avg, _8, mmxext);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_mmxext;
init_dc_ipred(4, mmxext);
init_dc_ipred(8, mmxext);
init_dir_tm_ipred(4, mmxext);
}
if (EXTERNAL_SSE(cpu_flags)) {
init_fpel_func(2, 0, 16, put, , sse);
init_fpel_func(1, 0, 32, put, , sse);
init_fpel_func(0, 0, 64, put, , sse);
init_ipred(16, sse, v, VERT);
init_ipred(32, sse, v, VERT);
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_subpel3_8to64(0, put, 8, sse2);
init_subpel3_8to64(1, avg, 8, sse2);
init_fpel_func(2, 1, 16, avg, _8, sse2);
init_fpel_func(1, 1, 32, avg, _8, sse2);
init_fpel_func(0, 1, 64, avg, _8, sse2);
init_lpf(sse2);
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_sse2;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_sse2;
init_dc_ipred(16, sse2);
init_dc_ipred(32, sse2);
init_dir_tm_h_ipred(8, sse2);
init_dir_tm_h_ipred(16, sse2);
init_dir_tm_h_ipred(32, sse2);
init_ipred(4, sse2, h, HOR);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_subpel3(0, put, 8, ssse3);
init_subpel3(1, avg, 8, ssse3);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
init_lpf(ssse3);
init_all_ipred(4, ssse3);
init_all_ipred(8, ssse3);
init_all_ipred(16, ssse3);
init_all_ipred(32, ssse3);
}
if (EXTERNAL_AVX(cpu_flags)) {
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;
init_lpf(avx);
init_dir_tm_h_ipred(8, avx);
init_dir_tm_h_ipred(16, avx);
init_dir_tm_h_ipred(32, avx);
}
if (EXTERNAL_AVX_FAST(cpu_flags)) {
init_fpel_func(1, 0, 32, put, , avx);
init_fpel_func(0, 0, 64, put, , avx);
init_ipred(32, avx, v, VERT);
}
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
init_fpel_func(1, 1, 32, avg, _8, avx2);
init_fpel_func(0, 1, 64, avg, _8, avx2);
if (ARCH_X86_64) {
#if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx2;
init_subpel3_32_64(0, put, 8, avx2);
init_subpel3_32_64(1, avg, 8, avx2);
#endif
}
init_dc_ipred(32, avx2);
init_ipred(32, avx2, h, HOR);
init_ipred(32, avx2, tm, TM_VP8);
}
#undef init_fpel
#undef init_subpel1
#undef init_subpel2
#undef init_subpel3
#endif /* HAVE_YASM */
}
av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
{
#if HAVE_YASM
int cpu_flags = av_get_cpu_flags();
#define init_fpel(idx1, idx2, sz, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_##type##sz##_##opt
#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_##opt; \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_##opt; \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_##opt
#define init_subpel2(idx1, idx2, sz, type, opt) \
init_subpel1(idx1, idx2, 1, 1, sz, hv, type, opt); \
init_subpel1(idx1, idx2, 0, 1, sz, v, type, opt); \
init_subpel1(idx1, idx2, 1, 0, sz, h, type, opt)
#define init_subpel3_32_64(idx, type, opt) \
init_subpel2(0, idx, 64, type, opt); \
init_subpel2(1, idx, 32, type, opt)
#define init_subpel3_8to64(idx, type, opt) \
init_subpel3_32_64(idx, type, opt); \
init_subpel2(2, idx, 16, type, opt); \
init_subpel2(3, idx, 8, type, opt)
#define init_subpel3(idx, type, opt) \
init_subpel3_8to64(idx, type, opt); \
init_subpel2(4, idx, 4, type, opt)
#define init_lpf(opt) do { \
if (ARCH_X86_64) { \
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_##opt; \
dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_##opt; \
dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_##opt; \
dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_##opt; \
dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_##opt; \
dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_##opt; \
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_##opt; \
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_##opt; \
} \
} while (0)
#define init_ipred(sz, opt, t, e) \
dsp->intra_pred[TX_##sz##X##sz][e##_PRED] = ff_vp9_ipred_##t##_##sz##x##sz##_##opt
#define ff_vp9_ipred_hd_4x4_ssse3 ff_vp9_ipred_hd_4x4_mmxext
#define ff_vp9_ipred_vl_4x4_ssse3 ff_vp9_ipred_vl_4x4_mmxext
#define init_dir_tm_ipred(sz, opt) do { \
init_ipred(sz, opt, dl, DIAG_DOWN_LEFT); \
init_ipred(sz, opt, dr, DIAG_DOWN_RIGHT); \
init_ipred(sz, opt, hd, HOR_DOWN); \
init_ipred(sz, opt, vl, VERT_LEFT); \
init_ipred(sz, opt, hu, HOR_UP); \
init_ipred(sz, opt, tm, TM_VP8); \
init_ipred(sz, opt, vr, VERT_RIGHT); \
} while (0)
#define init_dir_tm_h_ipred(sz, opt) do { \
init_dir_tm_ipred(sz, opt); \
init_ipred(sz, opt, h, HOR); \
} while (0)
#define init_dc_ipred(sz, opt) do { \
init_ipred(sz, opt, dc, DC); \
init_ipred(sz, opt, dc_left, LEFT_DC); \
init_ipred(sz, opt, dc_top, TOP_DC); \
} while (0)
#define init_all_ipred(sz, opt) do { \
init_dc_ipred(sz, opt); \
init_dir_tm_h_ipred(sz, opt); \
} while (0)
if (EXTERNAL_MMX(cpu_flags)) {
init_fpel(4, 0, 4, put, mmx);
init_fpel(3, 0, 8, put, mmx);
dsp->itxfm_add[4 /* lossless */][DCT_DCT] =
dsp->itxfm_add[4 /* lossless */][ADST_DCT] =
dsp->itxfm_add[4 /* lossless */][DCT_ADST] =
dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_mmx;
init_ipred(8, mmx, v, VERT);
}
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_subpel2(4, 0, 4, put, mmxext);
init_subpel2(4, 1, 4, avg, mmxext);
init_fpel(4, 1, 4, avg, mmxext);
init_fpel(3, 1, 8, avg, mmxext);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_mmxext;
init_dc_ipred(4, mmxext);
init_dc_ipred(8, mmxext);
init_dir_tm_ipred(4, mmxext);
}
if (EXTERNAL_SSE(cpu_flags)) {
init_fpel(2, 0, 16, put, sse);
init_fpel(1, 0, 32, put, sse);
init_fpel(0, 0, 64, put, sse);
init_ipred(16, sse, v, VERT);
init_ipred(32, sse, v, VERT);
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_subpel3_8to64(0, put, sse2);
init_subpel3_8to64(1, avg, sse2);
init_fpel(2, 1, 16, avg, sse2);
init_fpel(1, 1, 32, avg, sse2);
init_fpel(0, 1, 64, avg, sse2);
init_lpf(sse2);
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_sse2;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_sse2;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_sse2;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_sse2;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_sse2;
init_dc_ipred(16, sse2);
init_dc_ipred(32, sse2);
init_dir_tm_h_ipred(8, sse2);
init_dir_tm_h_ipred(16, sse2);
init_dir_tm_h_ipred(32, sse2);
init_ipred(4, sse2, h, HOR);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_subpel3(0, put, ssse3);
init_subpel3(1, avg, ssse3);
dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_DCT] = ff_vp9_idct_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][DCT_ADST] = ff_vp9_iadst_idct_4x4_add_ssse3;
dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_ssse3;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_ssse3;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
init_lpf(ssse3);
init_all_ipred(4, ssse3);
init_all_ipred(8, ssse3);
init_all_ipred(16, ssse3);
init_all_ipred(32, ssse3);
}
if (EXTERNAL_AVX(cpu_flags)) {
dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_DCT] = ff_vp9_idct_iadst_8x8_add_avx;
dsp->itxfm_add[TX_8X8][DCT_ADST] = ff_vp9_iadst_idct_8x8_add_avx;
dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx;
dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_DCT] = ff_vp9_idct_iadst_16x16_add_avx;
dsp->itxfm_add[TX_16X16][DCT_ADST] = ff_vp9_iadst_idct_16x16_add_avx;
dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx;
dsp->itxfm_add[TX_32X32][ADST_ADST] =
dsp->itxfm_add[TX_32X32][ADST_DCT] =
dsp->itxfm_add[TX_32X32][DCT_ADST] =
dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;
init_fpel(1, 0, 32, put, avx);
init_fpel(0, 0, 64, put, avx);
init_lpf(avx);
init_dir_tm_h_ipred(8, avx);
init_dir_tm_h_ipred(16, avx);
init_dir_tm_h_ipred(32, avx);
init_ipred(32, avx, v, VERT);
}
if (EXTERNAL_AVX2(cpu_flags)) {
init_fpel(1, 1, 32, avg, avx2);
init_fpel(0, 1, 64, avg, avx2);
if (ARCH_X86_64) {
#if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
init_subpel3_32_64(0, put, avx2);
init_subpel3_32_64(1, avg, avx2);
#endif
}
init_dc_ipred(32, avx2);
init_ipred(32, avx2, h, HOR);
init_ipred(32, avx2, tm, TM_VP8);
}
#undef init_fpel
#undef init_subpel1
#undef init_subpel2
#undef init_subpel3
#endif /* HAVE_YASM */
}
av_cold void ff_h264_pred_init_x86(H264PredContext *h, int codec_id,
const int bit_depth,
const int chroma_format_idc)
{
int mm_flags = av_get_cpu_flags();
if (bit_depth == 8) {
#if HAVE_MMX_EXTERNAL
if (EXTERNAL_MMX(mm_flags)) {
h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_8_mmx;
h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_mmx;
if (chroma_format_idc == 1) {
h->pred8x8 [VERT_PRED8x8 ] = ff_pred8x8_vertical_8_mmx;
h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_mmx;
}
if (codec_id == AV_CODEC_ID_VP8) {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_tm_vp8_8_mmx;
h->pred8x8 [PLANE_PRED8x8 ] = ff_pred8x8_tm_vp8_8_mmx;
h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_8_mmx;
} else {
if (chroma_format_idc == 1)
h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_8_mmx;
if (codec_id == AV_CODEC_ID_SVQ3) {
if (mm_flags & AV_CPU_FLAG_CMOV)
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_8_mmx;
} else if (codec_id == AV_CODEC_ID_RV40) {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_8_mmx;
} else {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_8_mmx;
}
}
}
#endif
#if HAVE_MMXEXT_EXTERNAL
if (EXTERNAL_MMXEXT(mm_flags)) {
h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_mmxext;
h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_mmxext;
if (chroma_format_idc == 1)
h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_mmxext;
h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_8_mmxext;
h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_8_mmxext;
h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_8_mmxext;
h->pred8x8l [VERT_PRED ] = ff_pred8x8l_vertical_8_mmxext;
h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_8_mmxext;
h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_8_mmxext;
h->pred8x8l [HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_8_mmxext;
h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_8_mmxext;
h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_8_mmxext;
h->pred4x4 [DIAG_DOWN_RIGHT_PRED ] = ff_pred4x4_down_right_8_mmxext;
h->pred4x4 [VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_8_mmxext;
h->pred4x4 [HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_8_mmxext;
h->pred4x4 [DC_PRED ] = ff_pred4x4_dc_8_mmxext;
if (codec_id == AV_CODEC_ID_VP8 || codec_id == AV_CODEC_ID_H264) {
h->pred4x4 [DIAG_DOWN_LEFT_PRED] = ff_pred4x4_down_left_8_mmxext;
}
if (codec_id == AV_CODEC_ID_SVQ3 || codec_id == AV_CODEC_ID_H264) {
h->pred4x4 [VERT_LEFT_PRED ] = ff_pred4x4_vertical_left_8_mmxext;
}
if (codec_id != AV_CODEC_ID_RV40) {
h->pred4x4 [HOR_UP_PRED ] = ff_pred4x4_horizontal_up_8_mmxext;
}
if (codec_id == AV_CODEC_ID_SVQ3 || codec_id == AV_CODEC_ID_H264) {
if (chroma_format_idc == 1) {
h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_8_mmxext;
h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_8_mmxext;
}
}
if (codec_id == AV_CODEC_ID_VP8) {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_tm_vp8_8_mmxext;
h->pred8x8 [DC_PRED8x8 ] = ff_pred8x8_dc_rv40_8_mmxext;
h->pred8x8 [PLANE_PRED8x8 ] = ff_pred8x8_tm_vp8_8_mmxext;
h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_8_mmxext;
h->pred4x4 [VERT_PRED ] = ff_pred4x4_vertical_vp8_8_mmxext;
} else {
if (chroma_format_idc == 1)
h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_8_mmxext;
if (codec_id == AV_CODEC_ID_SVQ3) {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_svq3_8_mmxext;
} else if (codec_id == AV_CODEC_ID_RV40) {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_rv40_8_mmxext;
} else {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_plane_h264_8_mmxext;
}
}
}
#endif
#if HAVE_SSE_EXTERNAL
if (EXTERNAL_SSE(mm_flags)) {
h->pred16x16[VERT_PRED8x8] = ff_pred16x16_vertical_8_sse;
}
#endif
#if HAVE_SSE2_EXTERNAL
if (EXTERNAL_SSE2(mm_flags)) {
h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_sse2;
h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_8_sse2;
h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_8_sse2;
h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_8_sse2;
h->pred8x8l [VERT_LEFT_PRED ] = ff_pred8x8l_vertical_left_8_sse2;
h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_8_sse2;
if (codec_id == AV_CODEC_ID_VP8) {
h->pred16x16[PLANE_PRED8x8 ] = ff_pred16x16_tm_vp8_8_sse2;
h->pred8x8 [PLANE_PRED8x8 ] = ff_pred8x8_tm_vp8_8_sse2;
} else {
if (chroma_format_idc == 1)
h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_8_sse2;
if (codec_id == AV_CODEC_ID_SVQ3) {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_8_sse2;
} else if (codec_id == AV_CODEC_ID_RV40) {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_8_sse2;
} else {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_8_sse2;
}
}
}
#endif
#if HAVE_SSSE3_EXTERNAL
if (EXTERNAL_SSSE3(mm_flags)) {
h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_8_ssse3;
h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_8_ssse3;
if (chroma_format_idc == 1)
h->pred8x8 [HOR_PRED8x8 ] = ff_pred8x8_horizontal_8_ssse3;
h->pred8x8l [TOP_DC_PRED ] = ff_pred8x8l_top_dc_8_ssse3;
h->pred8x8l [DC_PRED ] = ff_pred8x8l_dc_8_ssse3;
h->pred8x8l [HOR_PRED ] = ff_pred8x8l_horizontal_8_ssse3;
h->pred8x8l [VERT_PRED ] = ff_pred8x8l_vertical_8_ssse3;
h->pred8x8l [DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_8_ssse3;
h->pred8x8l [DIAG_DOWN_RIGHT_PRED ] = ff_pred8x8l_down_right_8_ssse3;
h->pred8x8l [VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_8_ssse3;
h->pred8x8l [VERT_LEFT_PRED ] = ff_pred8x8l_vertical_left_8_ssse3;
h->pred8x8l [HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_8_ssse3;
h->pred8x8l [HOR_DOWN_PRED ] = ff_pred8x8l_horizontal_down_8_ssse3;
if (codec_id == AV_CODEC_ID_VP8) {
h->pred8x8 [PLANE_PRED8x8 ] = ff_pred8x8_tm_vp8_8_ssse3;
h->pred4x4 [TM_VP8_PRED ] = ff_pred4x4_tm_vp8_8_ssse3;
} else {
if (chroma_format_idc == 1)
h->pred8x8 [PLANE_PRED8x8] = ff_pred8x8_plane_8_ssse3;
if (codec_id == AV_CODEC_ID_SVQ3) {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_svq3_8_ssse3;
} else if (codec_id == AV_CODEC_ID_RV40) {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_rv40_8_ssse3;
} else {
h->pred16x16[PLANE_PRED8x8] = ff_pred16x16_plane_h264_8_ssse3;
}
}
}
#endif
} else if (bit_depth == 10) {
#if HAVE_MMXEXT_EXTERNAL
if (EXTERNAL_MMXEXT(mm_flags)) {
h->pred4x4[DC_PRED ] = ff_pred4x4_dc_10_mmxext;
h->pred4x4[HOR_UP_PRED ] = ff_pred4x4_horizontal_up_10_mmxext;
if (chroma_format_idc == 1)
h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_10_mmxext;
h->pred8x8l[DC_128_PRED ] = ff_pred8x8l_128_dc_10_mmxext;
h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_10_mmxext;
h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_10_mmxext;
h->pred16x16[DC_128_PRED8x8 ] = ff_pred16x16_128_dc_10_mmxext;
h->pred16x16[LEFT_DC_PRED8x8 ] = ff_pred16x16_left_dc_10_mmxext;
h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_10_mmxext;
h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_10_mmxext;
}
#endif
#if HAVE_SSE2_EXTERNAL
if (EXTERNAL_SSE2(mm_flags)) {
h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_sse2;
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_sse2;
h->pred4x4[VERT_LEFT_PRED ] = ff_pred4x4_vertical_left_10_sse2;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_sse2;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_sse2;
if (chroma_format_idc == 1) {
h->pred8x8[DC_PRED8x8 ] = ff_pred8x8_dc_10_sse2;
h->pred8x8[TOP_DC_PRED8x8 ] = ff_pred8x8_top_dc_10_sse2;
h->pred8x8[PLANE_PRED8x8 ] = ff_pred8x8_plane_10_sse2;
h->pred8x8[VERT_PRED8x8 ] = ff_pred8x8_vertical_10_sse2;
h->pred8x8[HOR_PRED8x8 ] = ff_pred8x8_horizontal_10_sse2;
}
h->pred8x8l[VERT_PRED ] = ff_pred8x8l_vertical_10_sse2;
h->pred8x8l[HOR_PRED ] = ff_pred8x8l_horizontal_10_sse2;
h->pred8x8l[DC_PRED ] = ff_pred8x8l_dc_10_sse2;
h->pred8x8l[DC_128_PRED ] = ff_pred8x8l_128_dc_10_sse2;
h->pred8x8l[TOP_DC_PRED ] = ff_pred8x8l_top_dc_10_sse2;
h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_sse2;
h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_sse2;
h->pred8x8l[VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_10_sse2;
h->pred8x8l[HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_10_sse2;
h->pred16x16[DC_PRED8x8 ] = ff_pred16x16_dc_10_sse2;
h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_10_sse2;
h->pred16x16[DC_128_PRED8x8 ] = ff_pred16x16_128_dc_10_sse2;
h->pred16x16[LEFT_DC_PRED8x8 ] = ff_pred16x16_left_dc_10_sse2;
h->pred16x16[VERT_PRED8x8 ] = ff_pred16x16_vertical_10_sse2;
h->pred16x16[HOR_PRED8x8 ] = ff_pred16x16_horizontal_10_sse2;
}
#endif
#if HAVE_SSSE3_EXTERNAL
if (EXTERNAL_SSSE3(mm_flags)) {
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_ssse3;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_ssse3;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_ssse3;
h->pred8x8l[HOR_PRED ] = ff_pred8x8l_horizontal_10_ssse3;
h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_ssse3;
h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_ssse3;
h->pred8x8l[VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_10_ssse3;
h->pred8x8l[HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_10_ssse3;
}
#endif
#if HAVE_AVX_EXTERNAL
if (EXTERNAL_AVX(mm_flags)) {
h->pred4x4[DIAG_DOWN_LEFT_PRED ] = ff_pred4x4_down_left_10_avx;
h->pred4x4[DIAG_DOWN_RIGHT_PRED] = ff_pred4x4_down_right_10_avx;
h->pred4x4[VERT_LEFT_PRED ] = ff_pred4x4_vertical_left_10_avx;
h->pred4x4[VERT_RIGHT_PRED ] = ff_pred4x4_vertical_right_10_avx;
h->pred4x4[HOR_DOWN_PRED ] = ff_pred4x4_horizontal_down_10_avx;
h->pred8x8l[VERT_PRED ] = ff_pred8x8l_vertical_10_avx;
h->pred8x8l[HOR_PRED ] = ff_pred8x8l_horizontal_10_avx;
h->pred8x8l[DC_PRED ] = ff_pred8x8l_dc_10_avx;
h->pred8x8l[TOP_DC_PRED ] = ff_pred8x8l_top_dc_10_avx;
h->pred8x8l[DIAG_DOWN_RIGHT_PRED] = ff_pred8x8l_down_right_10_avx;
h->pred8x8l[DIAG_DOWN_LEFT_PRED ] = ff_pred8x8l_down_left_10_avx;
h->pred8x8l[VERT_RIGHT_PRED ] = ff_pred8x8l_vertical_right_10_avx;
h->pred8x8l[HOR_UP_PRED ] = ff_pred8x8l_horizontal_up_10_avx;
}
#endif
}
}
av_cold void INIT_FUNC(VP9DSPContext *dsp, int bitexact)
{
#if HAVE_X86ASM
int cpu_flags = av_get_cpu_flags();
#define init_lpf_8_func(idx1, idx2, dir, wd, bpp, opt) \
dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt
#define init_lpf_16_func(idx, dir, bpp, opt) \
dsp->loop_filter_16[idx] = loop_filter_##dir##_16_##bpp##_##opt
#define init_lpf_mix2_func(idx1, idx2, idx3, dir, wd1, wd2, bpp, opt) \
dsp->loop_filter_mix2[idx1][idx2][idx3] = loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt
#define init_lpf_funcs(bpp, opt) \
init_lpf_8_func(0, 0, h, 4, bpp, opt); \
init_lpf_8_func(0, 1, v, 4, bpp, opt); \
init_lpf_8_func(1, 0, h, 8, bpp, opt); \
init_lpf_8_func(1, 1, v, 8, bpp, opt); \
init_lpf_8_func(2, 0, h, 16, bpp, opt); \
init_lpf_8_func(2, 1, v, 16, bpp, opt); \
init_lpf_16_func(0, h, bpp, opt); \
init_lpf_16_func(1, v, bpp, opt); \
init_lpf_mix2_func(0, 0, 0, h, 4, 4, bpp, opt); \
init_lpf_mix2_func(0, 1, 0, h, 4, 8, bpp, opt); \
init_lpf_mix2_func(1, 0, 0, h, 8, 4, bpp, opt); \
init_lpf_mix2_func(1, 1, 0, h, 8, 8, bpp, opt); \
init_lpf_mix2_func(0, 0, 1, v, 4, 4, bpp, opt); \
init_lpf_mix2_func(0, 1, 1, v, 4, 8, bpp, opt); \
init_lpf_mix2_func(1, 0, 1, v, 8, 4, bpp, opt); \
init_lpf_mix2_func(1, 1, 1, v, 8, 8, bpp, opt)
#define init_itx_func(idxa, idxb, typea, typeb, size, bpp, opt) \
dsp->itxfm_add[idxa][idxb] = \
cat(ff_vp9_##typea##_##typeb##_##size##x##size##_add_, bpp, _##opt);
#define init_itx_func_one(idx, typea, typeb, size, bpp, opt) \
init_itx_func(idx, DCT_DCT, typea, typeb, size, bpp, opt); \
init_itx_func(idx, ADST_DCT, typea, typeb, size, bpp, opt); \
init_itx_func(idx, DCT_ADST, typea, typeb, size, bpp, opt); \
init_itx_func(idx, ADST_ADST, typea, typeb, size, bpp, opt)
#define init_itx_funcs(idx, size, bpp, opt) \
init_itx_func(idx, DCT_DCT, idct, idct, size, bpp, opt); \
init_itx_func(idx, ADST_DCT, idct, iadst, size, bpp, opt); \
init_itx_func(idx, DCT_ADST, iadst, idct, size, bpp, opt); \
init_itx_func(idx, ADST_ADST, iadst, iadst, size, bpp, opt); \
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_ipred_func(tm, TM_VP8, 4, BPC, mmxext);
if (!bitexact) {
init_itx_func_one(4 /* lossless */, iwht, iwht, 4, BPC, mmxext);
#if BPC == 10
init_itx_func(TX_4X4, DCT_DCT, idct, idct, 4, 10, mmxext);
#endif
}
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_subpel3(0, put, BPC, sse2);
init_subpel3(1, avg, BPC, sse2);
init_lpf_funcs(BPC, sse2);
init_8_16_32_ipred_funcs(tm, TM_VP8, BPC, sse2);
#if BPC == 10
if (!bitexact) {
init_itx_func(TX_4X4, ADST_DCT, idct, iadst, 4, 10, sse2);
init_itx_func(TX_4X4, DCT_ADST, iadst, idct, 4, 10, sse2);
init_itx_func(TX_4X4, ADST_ADST, iadst, iadst, 4, 10, sse2);
}
#else
init_itx_funcs(TX_4X4, 4, 12, sse2);
#endif
init_itx_funcs(TX_8X8, 8, BPC, sse2);
init_itx_funcs(TX_16X16, 16, BPC, sse2);
init_itx_func_one(TX_32X32, idct, idct, 32, BPC, sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_lpf_funcs(BPC, ssse3);
#if BPC == 10
if (!bitexact) {
init_itx_funcs(TX_4X4, 4, BPC, ssse3);
}
#endif
}
if (EXTERNAL_AVX(cpu_flags)) {
init_lpf_funcs(BPC, avx);
}
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
#if HAVE_AVX2_EXTERNAL
init_subpel3_32_64(0, put, BPC, avx2);
init_subpel3_32_64(1, avg, BPC, avx2);
init_subpel2(2, 0, 16, put, BPC, avx2);
init_subpel2(2, 1, 16, avg, BPC, avx2);
#endif
}
#endif /* HAVE_X86ASM */
ff_vp9dsp_init_16bpp_x86(dsp);
}
void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
{
int cpu_flags = av_get_cpu_flags();
#define SET_LUMA_FUNCS(tabname, funcname, depth, cf) \
c->tabname[0] = funcname ## _4_ ## depth ## _ ## cf; \
c->tabname[1] = funcname ## _8_ ## depth ## _ ## cf; \
c->tabname[2] = funcname ## _12_ ## depth ## _ ## cf; \
c->tabname[3] = funcname ## _16_ ## depth ## _ ## cf; \
c->tabname[4] = funcname ## _24_ ## depth ## _ ## cf; \
c->tabname[5] = funcname ## _32_ ## depth ## _ ## cf; \
c->tabname[6] = funcname ## _48_ ## depth ## _ ## cf; \
c->tabname[7] = funcname ## _64_ ## depth ## _ ## cf;
#define SET_CHROMA_FUNCS(tabname, funcname, depth, cf) \
c->tabname[1] = funcname ## _4_ ## depth ## _ ## cf; \
c->tabname[3] = funcname ## _8_ ## depth ## _ ## cf; \
c->tabname[4] = funcname ## _12_ ## depth ## _ ## cf; \
c->tabname[5] = funcname ## _16_ ## depth ## _ ## cf; \
c->tabname[6] = funcname ## _24_ ## depth ## _ ## cf; \
c->tabname[7] = funcname ## _32_ ## depth ## _ ## cf;
#define SET_QPEL_FUNCS(v, h, depth, cf, name) SET_LUMA_FUNCS (put_hevc_qpel[v][h], name, depth, cf)
#define SET_EPEL_FUNCS(v, h, depth, cf, name) SET_CHROMA_FUNCS(put_hevc_epel[v][h], name, depth, cf)
if (bit_depth == 8) {
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct_4x4_dc_8_mmxext;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_8_sse2;
c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_sse2;
c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_sse2;
c->idct[0] = ff_hevc_idct_4x4_8_sse2;
c->idct[1] = ff_hevc_idct_8x8_8_sse2;
SET_QPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
SET_EPEL_FUNCS(0, 0, 8, sse2, ff_hevc_get_pixels);
SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 8, sse2);
SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 8, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 8, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 8, sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
SET_QPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_qpel_h);
SET_QPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_qpel_v);
SET_EPEL_FUNCS(0, 1, 8, ssse3, ff_hevc_epel_h);
SET_EPEL_FUNCS(1, 0, 8, ssse3, ff_hevc_epel_v);
}
if (EXTERNAL_AVX(cpu_flags)) {
c->idct[0] = ff_hevc_idct_4x4_8_avx;
c->idct[1] = ff_hevc_idct_8x8_8_avx;
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct_4x4_dc_10_mmxext;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_mmxext;
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
c->idct_dc[1] = ff_hevc_idct_8x8_dc_10_sse2;
c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_sse2;
c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_sse2;
c->idct[0] = ff_hevc_idct_4x4_10_sse2;
c->idct[1] = ff_hevc_idct_8x8_10_sse2;
SET_QPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
SET_EPEL_FUNCS(0, 0, 10, sse2, ff_hevc_get_pixels);
SET_LUMA_FUNCS(put_unweighted_pred, ff_hevc_put_unweighted_pred, 10, sse2);
SET_LUMA_FUNCS(put_unweighted_pred_avg, ff_hevc_put_unweighted_pred_avg, 10, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_chroma, ff_hevc_put_unweighted_pred, 10, sse2);
SET_CHROMA_FUNCS(put_unweighted_pred_avg_chroma, ff_hevc_put_unweighted_pred_avg, 10, sse2);
}
if (EXTERNAL_AVX(cpu_flags)) {
c->idct[0] = ff_hevc_idct_4x4_10_avx;
c->idct[1] = ff_hevc_idct_8x8_10_avx;
}
}
#if ARCH_X86_64
if (bit_depth == 8) {
if (EXTERNAL_SSE2(cpu_flags)) {
c->idct[2] = ff_hevc_idct_16x16_8_sse2;
c->idct[3] = ff_hevc_idct_32x32_8_sse2;
}
if (EXTERNAL_SSSE3(cpu_flags)) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
}
if (EXTERNAL_SSE4(cpu_flags)) {
SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 8, sse4);
SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 8, sse4);
SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 8, sse4);
SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 8, sse4);
}
if (EXTERNAL_AVX(cpu_flags)) {
#if HAVE_AVX_EXTERNAL
SET_QPEL_FUNCS(1, 1, 8, avx, hevc_qpel_hv);
SET_EPEL_FUNCS(1, 1, 8, avx, hevc_epel_hv);
#endif /* HAVE_AVX_EXTERNAL */
c->idct[2] = ff_hevc_idct_16x16_8_avx;
c->idct[3] = ff_hevc_idct_32x32_8_avx;
}
if (EXTERNAL_AVX2(cpu_flags)) {
c->idct_dc[2] = ff_hevc_idct_16x16_dc_8_avx2;
c->idct_dc[3] = ff_hevc_idct_32x32_dc_8_avx2;
}
} else if (bit_depth == 10) {
if (EXTERNAL_SSE2(cpu_flags)) {
c->idct[2] = ff_hevc_idct_16x16_10_sse2;
c->idct[3] = ff_hevc_idct_32x32_10_sse2;
}
if (EXTERNAL_SSSE3(cpu_flags)) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3;
}
if (EXTERNAL_SSE4(cpu_flags)) {
SET_LUMA_FUNCS(weighted_pred, ff_hevc_put_weighted_pred, 10, sse4);
SET_CHROMA_FUNCS(weighted_pred_chroma, ff_hevc_put_weighted_pred, 10, sse4);
SET_LUMA_FUNCS(weighted_pred_avg, ff_hevc_put_weighted_pred_avg, 10, sse4);
SET_CHROMA_FUNCS(weighted_pred_avg_chroma, ff_hevc_put_weighted_pred_avg, 10, sse4);
}
if (EXTERNAL_AVX(cpu_flags)) {
#if HAVE_AVX_EXTERNAL
SET_QPEL_FUNCS(0, 1, 10, avx, ff_hevc_qpel_h);
SET_QPEL_FUNCS(1, 0, 10, avx, ff_hevc_qpel_v);
SET_QPEL_FUNCS(1, 1, 10, avx, hevc_qpel_hv);
SET_EPEL_FUNCS(0, 1, 10, avx, ff_hevc_epel_h);
SET_EPEL_FUNCS(1, 0, 10, avx, ff_hevc_epel_v);
SET_EPEL_FUNCS(1, 1, 10, avx, hevc_epel_hv);
#endif /* HAVE_AVX_EXTERNAL */
c->idct[2] = ff_hevc_idct_16x16_10_avx;
c->idct[3] = ff_hevc_idct_32x32_10_avx;
}
if (EXTERNAL_AVX2(cpu_flags)) {
c->idct_dc[2] = ff_hevc_idct_16x16_dc_10_avx2;
c->idct_dc[3] = ff_hevc_idct_32x32_dc_10_avx2;
}
}
#endif /* ARCH_X86_64 */
}
void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth)
{
int cpu_flags = av_get_cpu_flags();
if (bit_depth == 8) {
#if (HAVE_MMXEXT_EXTERNAL == 1)
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct4x4_dc_8_mmxext;
c->idct_dc[1] = ff_hevc_idct8x8_dc_8_mmxext;
c->transform_add[0] = ff_hevc_transform_add4_8_mmxext;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_sse2;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_sse2;
}
c->idct_dc[1] = ff_hevc_idct8x8_dc_8_sse2;
c->idct_dc[2] = ff_hevc_idct16x16_dc_8_sse2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_8_sse2;
c->transform_add[1] = ff_hevc_transform_add8_8_sse2;
c->transform_add[2] = ff_hevc_transform_add16_8_sse2;
c->transform_add[3] = ff_hevc_transform_add32_8_sse2;
}
#endif
#if (HAVE_SSSE3_EXTERNAL == 1)
if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
}
#endif
#if (HAVE_SSE4_EXTERNAL == 1)
if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 8, sse4);
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_avx;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_avx;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_avx;
}
c->transform_add[1] = ff_hevc_transform_add8_8_avx;
c->transform_add[2] = ff_hevc_transform_add16_8_avx;
c->transform_add[3] = ff_hevc_transform_add32_8_avx;
}
#endif
#if (HAVE_AVX2_EXTERNAL == 1)
if (EXTERNAL_AVX2(cpu_flags)) {
c->idct_dc[2] = ff_hevc_idct16x16_dc_8_avx2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_8_avx2;
c->transform_add[3] = ff_hevc_transform_add32_8_avx2;
}
#endif
} else if (bit_depth == 10) {
#if (HAVE_MMXEXT_EXTERNAL == 1)
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->transform_add[0] = ff_hevc_transform_add4_10_mmxext;
c->idct_dc[0] = ff_hevc_idct4x4_dc_10_mmxext;
c->idct_dc[1] = ff_hevc_idct8x8_dc_10_mmxext;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_sse2;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_sse2;
}
c->idct_dc[1] = ff_hevc_idct8x8_dc_10_sse2;
c->idct_dc[2] = ff_hevc_idct16x16_dc_10_sse2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_10_sse2;
c->transform_add[1] = ff_hevc_transform_add8_10_sse2;
c->transform_add[2] = ff_hevc_transform_add16_10_sse2;
c->transform_add[3] = ff_hevc_transform_add32_10_sse2;
}
#endif
#if (HAVE_SSSE3_EXTERNAL == 1)
if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3;
}
#endif
#if (HAVE_SSE4_EXTERNAL == 1)
if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 10, sse4);
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_avx;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_avx;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_avx;
}
}
#endif
#if (HAVE_AVX2_EXTERNAL == 1)
if (EXTERNAL_AVX2(cpu_flags)) {
c->idct_dc[2] = ff_hevc_idct16x16_dc_10_avx2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_10_avx2;
c->transform_add[2] = ff_hevc_transform_add16_10_avx2;
c->transform_add[3] = ff_hevc_transform_add32_10_avx2;
}
#endif
} else if (bit_depth == 12) {
#if (HAVE_MMXEXT_EXTERNAL == 1)
if (EXTERNAL_MMXEXT(cpu_flags)) {
c->idct_dc[0] = ff_hevc_idct4x4_dc_12_mmxext;
c->idct_dc[1] = ff_hevc_idct8x8_dc_12_mmxext;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_12_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_12_sse2;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_sse2;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_12_sse2;
}
c->idct_dc[1] = ff_hevc_idct8x8_dc_12_sse2;
c->idct_dc[2] = ff_hevc_idct16x16_dc_12_sse2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_12_sse2;
}
#endif
#if (HAVE_SSSE3_EXTERNAL == 1)
if (EXTERNAL_SSSE3(cpu_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_12_ssse3;
}
#endif
#if (HAVE_SSE4_EXTERNAL == 1)
if (EXTERNAL_SSE4(cpu_flags) && ARCH_X86_64) {
EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 12, sse4);
EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h, 12, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v, 12, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 12, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 12, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 12, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 12, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 12, sse4);
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(cpu_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_12_avx;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_12_avx;
if (ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_12_avx;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_12_avx;
}
}
#endif
#if (HAVE_AVX2_EXTERNAL == 1)
if (EXTERNAL_AVX2(cpu_flags)) {
c->idct_dc[2] = ff_hevc_idct16x16_dc_12_avx2;
c->idct_dc[3] = ff_hevc_idct32x32_dc_12_avx2;
}
#endif
}
}
void swri_resample_dsp_x86_init(ResampleContext *c)
{
int av_unused mm_flags = av_get_cpu_flags();
switch(c->format){
case AV_SAMPLE_FMT_S16P:
#if (ARCH_X86_32 == 1) && (HAVE_MMXEXT_EXTERNAL == 1)
if (ARCH_X86_32 && EXTERNAL_MMXEXT(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_int16_mmxext
: ff_resample_common_int16_mmxext;
}
#endif
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_int16_sse2
: ff_resample_common_int16_sse2;
}
#endif
#if (HAVE_XOP_EXTERNAL == 1)
if (EXTERNAL_XOP(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_int16_xop
: ff_resample_common_int16_xop;
}
#endif
break;
case AV_SAMPLE_FMT_FLTP:
#if (HAVE_SSE_EXTERNAL == 1)
if (EXTERNAL_SSE(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_float_sse
: ff_resample_common_float_sse;
}
#endif
#if (HAVE_AVX_EXTERNAL == 1)
if (EXTERNAL_AVX(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_float_avx
: ff_resample_common_float_avx;
}
#endif
#if (HAVE_FMA3_EXTERNAL == 1)
if (EXTERNAL_FMA3(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_float_fma3
: ff_resample_common_float_fma3;
}
#endif
#if (HAVE_FMA4_EXTERNAL == 1)
if (EXTERNAL_FMA4(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_float_fma4
: ff_resample_common_float_fma4;
}
#endif
break;
case AV_SAMPLE_FMT_DBLP:
#if (HAVE_SSE2_EXTERNAL == 1)
if (EXTERNAL_SSE2(mm_flags)) {
c->dsp.resample = c->linear ? ff_resample_linear_double_sse2
: ff_resample_common_double_sse2;
}
#endif
break;
}
}
void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
{
int mm_flags = av_get_cpu_flags();
if (bit_depth == 8) {
if (EXTERNAL_MMXEXT(mm_flags)) {
c->transform_dc_add[0] = ff_hevc_idct4_dc_add_8_mmxext;
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_8_mmxext;
}
if (EXTERNAL_SSE2(mm_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_8_sse2;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_8_sse2;
}
if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
}
if (EXTERNAL_SSE4(mm_flags) && ARCH_X86_64) {
EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v, 8, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 8, sse4);
}
if (EXTERNAL_AVX2(mm_flags)) {
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_8_avx2;
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMXEXT(mm_flags)) {
c->transform_dc_add[0] = ff_hevc_idct4_dc_add_10_mmxext;
}
if (EXTERNAL_SSE2(mm_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_10_sse2;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_10_sse2;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_10_sse2;
}
if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_10_ssse3;
}
if (EXTERNAL_SSE4(mm_flags) && ARCH_X86_64) {
EPEL_LINKS(c->put_hevc_epel, 0, 0, pel_pixels, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 0, 1, epel_h, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 0, epel_v, 10, sse4);
EPEL_LINKS(c->put_hevc_epel, 1, 1, epel_hv, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 0, pel_pixels, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 10, sse4);
}
if (EXTERNAL_AVX(mm_flags)) {
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_10_avx;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_10_avx;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_10_avx;
}
if (EXTERNAL_AVX2(mm_flags)) {
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_10_avx2;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_10_avx2;
}
}
}
av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
const int chroma_format_idc)
{
#if HAVE_YASM
int mm_flags = av_get_cpu_flags();
if (chroma_format_idc == 1 && EXTERNAL_MMXEXT(mm_flags))
c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
if (bit_depth == 8) {
if (EXTERNAL_MMX(mm_flags)) {
c->h264_idct_dc_add =
c->h264_idct_add = ff_h264_idct_add_8_mmx;
c->h264_idct8_dc_add =
c->h264_idct8_add = ff_h264_idct8_add_8_mmx;
c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
if (chroma_format_idc == 1)
c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
if (mm_flags & AV_CPU_FLAG_CMOV)
c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
if (EXTERNAL_MMXEXT(mm_flags)) {
c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmxext;
c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
c->h264_idct_add16 = ff_h264_idct_add16_8_mmxext;
c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmxext;
if (chroma_format_idc == 1)
c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmxext;
c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
if (chroma_format_idc == 1) {
c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmxext;
c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
}
#if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_mmxext;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmxext;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmxext;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
#endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
if (EXTERNAL_SSE2(mm_flags)) {
c->h264_idct8_add = ff_h264_idct8_add_8_sse2;
c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
if (chroma_format_idc == 1)
c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2;
c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
}
if (EXTERNAL_SSSE3(mm_flags)) {
c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
}
if (EXTERNAL_AVX(mm_flags)) {
c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
}
}
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMX(mm_flags)) {
if (EXTERNAL_MMXEXT(mm_flags)) {
#if ARCH_X86_32
c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmxext;
c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmxext;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmxext;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
#endif /* ARCH_X86_32 */
c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
if (EXTERNAL_SSE2(mm_flags)) {
c->h264_idct_add = ff_h264_idct_add_10_sse2;
c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
if (chroma_format_idc == 1)
c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
#if HAVE_ALIGNED_STACK
c->h264_idct8_add = ff_h264_idct8_add_10_sse2;
c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
#endif /* HAVE_ALIGNED_STACK */
c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2;
c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
#if HAVE_ALIGNED_STACK
c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
#endif /* HAVE_ALIGNED_STACK */
}
if (EXTERNAL_SSE4(mm_flags)) {
c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
}
if (EXTERNAL_AVX(mm_flags)) {
c->h264_idct_dc_add =
c->h264_idct_add = ff_h264_idct_add_10_avx;
c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
if (chroma_format_idc == 1)
c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
#if HAVE_ALIGNED_STACK
c->h264_idct8_add = ff_h264_idct8_add_10_avx;
c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
#endif /* HAVE_ALIGNED_STACK */
c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx;
c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
#if HAVE_ALIGNED_STACK
c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
#endif /* HAVE_ALIGNED_STACK */
}
}
}
}
#endif
}
av_cold void INIT_FUNC(VP9DSPContext *dsp)
{
#if HAVE_YASM
int cpu_flags = av_get_cpu_flags();
#define init_lpf_8_func(idx1, idx2, dir, wd, bpp, opt) \
dsp->loop_filter_8[idx1][idx2] = ff_vp9_loop_filter_##dir##_##wd##_##bpp##_##opt
#define init_lpf_16_func(idx, dir, bpp, opt) \
dsp->loop_filter_16[idx] = loop_filter_##dir##_16_##bpp##_##opt
#define init_lpf_mix2_func(idx1, idx2, idx3, dir, wd1, wd2, bpp, opt) \
dsp->loop_filter_mix2[idx1][idx2][idx3] = loop_filter_##dir##_##wd1##wd2##_##bpp##_##opt
#define init_lpf_funcs(bpp, opt) \
init_lpf_8_func(0, 0, h, 4, bpp, opt); \
init_lpf_8_func(0, 1, v, 4, bpp, opt); \
init_lpf_8_func(1, 0, h, 8, bpp, opt); \
init_lpf_8_func(1, 1, v, 8, bpp, opt); \
init_lpf_8_func(2, 0, h, 16, bpp, opt); \
init_lpf_8_func(2, 1, v, 16, bpp, opt); \
init_lpf_16_func(0, h, bpp, opt); \
init_lpf_16_func(1, v, bpp, opt); \
init_lpf_mix2_func(0, 0, 0, h, 4, 4, bpp, opt); \
init_lpf_mix2_func(0, 1, 0, h, 4, 8, bpp, opt); \
init_lpf_mix2_func(1, 0, 0, h, 8, 4, bpp, opt); \
init_lpf_mix2_func(1, 1, 0, h, 8, 8, bpp, opt); \
init_lpf_mix2_func(0, 0, 1, v, 4, 4, bpp, opt); \
init_lpf_mix2_func(0, 1, 1, v, 4, 8, bpp, opt); \
init_lpf_mix2_func(1, 0, 1, v, 8, 4, bpp, opt); \
init_lpf_mix2_func(1, 1, 1, v, 8, 8, bpp, opt)
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_ipred_func(tm, TM_VP8, 4, BPC, mmxext);
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_subpel3(0, put, BPC, sse2);
init_subpel3(1, avg, BPC, sse2);
init_lpf_funcs(BPC, sse2);
init_8_16_32_ipred_funcs(tm, TM_VP8, BPC, sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_lpf_funcs(BPC, ssse3);
}
if (EXTERNAL_AVX(cpu_flags)) {
init_lpf_funcs(BPC, avx);
}
if (EXTERNAL_AVX2(cpu_flags)) {
#if HAVE_AVX2_EXTERNAL
init_subpel3_32_64(0, put, BPC, avx2);
init_subpel3_32_64(1, avg, BPC, avx2);
init_subpel2(2, 0, 16, put, BPC, avx2);
init_subpel2(2, 1, 16, avg, BPC, avx2);
#endif
}
#endif /* HAVE_YASM */
ff_vp9dsp_init_16bpp_x86(dsp);
}
av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
{
#if HAVE_X86ASM
int cpu_flags = av_get_cpu_flags();
#define init_fpel(idx1, idx2, sz, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \
dsp->mc[idx1][FILTER_8TAP_SHARP ][idx2][0][0] = \
dsp->mc[idx1][FILTER_BILINEAR ][idx2][0][0] = ff_vp9_ ## type ## sz ## _ ## opt
#define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \
dsp->mc[idx1][FILTER_8TAP_SMOOTH][idx2][idxh][idxv] = type ## _8tap_smooth_ ## sz ## dir ## _ ## opt; \
dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type ## _8tap_regular_ ## sz ## dir ## _ ## opt; \
dsp->mc[idx1][FILTER_8TAP_SHARP][idx2][idxh][idxv] = type ## _8tap_sharp_ ## sz ## dir ## _ ## opt
#define init_subpel2(idx1, idx2, sz, type, opt) \
init_subpel1(idx1, idx2, 1, 1, sz, hv, type, opt); \
init_subpel1(idx1, idx2, 0, 1, sz, v, type, opt); \
init_subpel1(idx1, idx2, 1, 0, sz, h, type, opt)
#define init_subpel3_32_64(idx, type, opt) \
init_subpel2(0, idx, 64, type, opt); \
init_subpel2(1, idx, 32, type, opt)
#define init_subpel3_8to64(idx, type, opt) \
init_subpel3_32_64(idx, type, opt); \
init_subpel2(2, idx, 16, type, opt); \
init_subpel2(3, idx, 8, type, opt)
#define init_subpel3(idx, type, opt) \
init_subpel3_8to64(idx, type, opt); \
init_subpel2(4, idx, 4, type, opt)
#define init_lpf(opt) do { \
dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_##opt; \
dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_##opt; \
dsp->loop_filter_mix2[0][0][0] = ff_vp9_loop_filter_h_44_16_##opt; \
dsp->loop_filter_mix2[0][0][1] = ff_vp9_loop_filter_v_44_16_##opt; \
dsp->loop_filter_mix2[0][1][0] = ff_vp9_loop_filter_h_48_16_##opt; \
dsp->loop_filter_mix2[0][1][1] = ff_vp9_loop_filter_v_48_16_##opt; \
dsp->loop_filter_mix2[1][0][0] = ff_vp9_loop_filter_h_84_16_##opt; \
dsp->loop_filter_mix2[1][0][1] = ff_vp9_loop_filter_v_84_16_##opt; \
dsp->loop_filter_mix2[1][1][0] = ff_vp9_loop_filter_h_88_16_##opt; \
dsp->loop_filter_mix2[1][1][1] = ff_vp9_loop_filter_v_88_16_##opt; \
} while (0)
if (EXTERNAL_MMX(cpu_flags)) {
init_fpel(4, 0, 4, put, mmx);
init_fpel(3, 0, 8, put, mmx);
}
if (EXTERNAL_MMXEXT(cpu_flags)) {
init_subpel2(4, 0, 4, put, mmxext);
init_subpel2(4, 1, 4, avg, mmxext);
init_fpel(4, 1, 4, avg, mmxext);
init_fpel(3, 1, 8, avg, mmxext);
}
if (EXTERNAL_SSE(cpu_flags)) {
init_fpel(2, 0, 16, put, sse);
init_fpel(1, 0, 32, put, sse);
init_fpel(0, 0, 64, put, sse);
}
if (EXTERNAL_SSE2(cpu_flags)) {
init_subpel3_8to64(0, put, sse2);
init_subpel3_8to64(1, avg, sse2);
init_fpel(2, 1, 16, avg, sse2);
init_fpel(1, 1, 32, avg, sse2);
init_fpel(0, 1, 64, avg, sse2);
init_lpf(sse2);
}
if (EXTERNAL_SSSE3(cpu_flags)) {
init_subpel3(0, put, ssse3);
init_subpel3(1, avg, ssse3);
init_lpf(ssse3);
}
if (EXTERNAL_AVX(cpu_flags)) {
init_fpel(1, 0, 32, put, avx);
init_fpel(0, 0, 64, put, avx);
init_lpf(avx);
}
if (EXTERNAL_AVX2(cpu_flags)) {
init_fpel(1, 1, 32, avg, avx2);
init_fpel(0, 1, 64, avg, avx2);
#if ARCH_X86_64 && HAVE_AVX2_EXTERNAL
init_subpel3_32_64(0, put, avx2);
init_subpel3_32_64(1, avg, avx2);
#endif /* ARCH_X86_64 && HAVE_AVX2_EXTERNAL */
}
#undef init_fpel
#undef init_subpel1
#undef init_subpel2
#undef init_subpel3
#endif /* HAVE_X86ASM */
}
av_cold void swri_audio_convert_init_x86(struct AudioConvert *ac,
enum AVSampleFormat out_fmt,
enum AVSampleFormat in_fmt,
int channels){
int mm_flags = av_get_cpu_flags();
ac->simd_f= NULL;
//FIXME add memcpy case
#define MULTI_CAPS_FUNC(flag, cap) \
if (EXTERNAL_##flag(mm_flags)) {\
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S16 || out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_S16P)\
ac->simd_f = ff_int16_to_int32_a_ ## cap;\
if( out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_S32 || out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_S32P)\
ac->simd_f = ff_int32_to_int16_a_ ## cap;\
}
MULTI_CAPS_FUNC(MMX, mmx)
MULTI_CAPS_FUNC(SSE2, sse2)
if(EXTERNAL_MMX(mm_flags)) {
if(channels == 6) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_FLTP || out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_6ch_float_to_float_a_mmx;
}
}
if(EXTERNAL_SSE(mm_flags)) {
if(channels == 6) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_FLTP || out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_6ch_float_to_float_a_sse;
}
}
if(EXTERNAL_SSE2(mm_flags)) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32 || out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_int32_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S16 || out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S16P)
ac->simd_f = ff_int16_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_float_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_float_to_int16_a_sse2;
if(channels == 2) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_FLTP || out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_2ch_int32_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_S16P)
ac->simd_f = ff_pack_2ch_int16_to_int16_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S16P)
ac->simd_f = ff_pack_2ch_int16_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_2ch_int32_to_int16_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_S32)
ac->simd_f = ff_unpack_2ch_int32_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_int16_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_S32)
ac->simd_f = ff_unpack_2ch_int32_to_int16_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_2ch_int32_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_pack_2ch_float_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S16P)
ac->simd_f = ff_pack_2ch_int16_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16 && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_pack_2ch_float_to_int16_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S32)
ac->simd_f = ff_unpack_2ch_int32_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_FLT)
ac->simd_f = ff_unpack_2ch_float_to_int32_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_FLT)
ac->simd_f = ff_unpack_2ch_float_to_int16_a_sse2;
}
if(channels == 6) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_6ch_int32_to_float_a_sse2;
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_pack_6ch_float_to_int32_a_sse2;
}
}
if(EXTERNAL_SSSE3(mm_flags)) {
if(channels == 2) {
if( out_fmt == AV_SAMPLE_FMT_S16P && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_int16_a_ssse3;
if( out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_int32_a_ssse3;
if( out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S16)
ac->simd_f = ff_unpack_2ch_int16_to_float_a_ssse3;
}
}
if(EXTERNAL_AVX(mm_flags)) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32 || out_fmt == AV_SAMPLE_FMT_FLTP && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_int32_to_float_a_avx;
if(channels == 6) {
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_FLTP || out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_6ch_float_to_float_a_avx;
if( out_fmt == AV_SAMPLE_FMT_FLT && in_fmt == AV_SAMPLE_FMT_S32P)
ac->simd_f = ff_pack_6ch_int32_to_float_a_avx;
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_pack_6ch_float_to_int32_a_avx;
}
}
if(EXTERNAL_AVX2(mm_flags)) {
if( out_fmt == AV_SAMPLE_FMT_S32 && in_fmt == AV_SAMPLE_FMT_FLT || out_fmt == AV_SAMPLE_FMT_S32P && in_fmt == AV_SAMPLE_FMT_FLTP)
ac->simd_f = ff_float_to_int32_a_avx2;
}
}
void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
{
int mm_flags = av_get_cpu_flags();
if (bit_depth == 8) {
if (EXTERNAL_MMX(mm_flags)) {
/*if (mm_flags & AV_CPU_FLAG_CMOV)
c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx; */
if (EXTERNAL_MMXEXT(mm_flags)) {
#if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
/* MMEXT optimizations */
#endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
if (EXTERNAL_SSE2(mm_flags))
{
#if 0
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
#endif
}
if (EXTERNAL_SSSE3(mm_flags)) {
c->transform_4x4_luma_add = ff_hevc_transform_4x4_luma_add_8_sse4;
c->transform_add[0] = ff_hevc_transform_4x4_add_8_sse4;
c->transform_add[1] = ff_hevc_transform_8x8_add_8_sse4;
c->transform_add[2] = ff_hevc_transform_16x16_add_8_sse4;
c->transform_add[3] = ff_hevc_transform_32x32_add_8_sse4;
c->put_unweighted_pred = ff_hevc_put_unweighted_pred_8_sse;
c->put_hevc_qpel[0][0] = ff_hevc_put_hevc_qpel_pixels_8_sse;
c->put_hevc_qpel[0][1] = ff_hevc_put_hevc_qpel_h_1_8_sse;
c->put_hevc_qpel[0][2] = ff_hevc_put_hevc_qpel_h_2_8_sse;
c->put_hevc_qpel[0][3] = ff_hevc_put_hevc_qpel_h_3_8_sse;
c->put_hevc_qpel[1][0] = ff_hevc_put_hevc_qpel_v_1_8_sse;
c->put_hevc_qpel[2][0] = ff_hevc_put_hevc_qpel_v_2_8_sse;
c->put_hevc_qpel[3][0] = ff_hevc_put_hevc_qpel_v_3_8_sse;
//#if ARCH_X86_64
#if 0
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
c->hevc_h_loop_filter_luma = ff_hevc_h_loop_filter_luma_8_ssse3;
#endif
}
if (EXTERNAL_SSE4(mm_flags)) {
#if GCC_VERSION > MIN_GCC_VERSION_MC || __APPLE__
c->put_weighted_pred_avg = ff_hevc_put_weighted_pred_avg_8_sse;
c->weighted_pred = ff_hevc_weighted_pred_8_sse;
c->weighted_pred_avg = ff_hevc_weighted_pred_avg_8_sse;
c->put_hevc_epel[0][0] = ff_hevc_put_hevc_epel_pixels_8_sse;
c->put_hevc_epel[0][1] = ff_hevc_put_hevc_epel_h_8_sse;
c->put_hevc_epel[1][0] = ff_hevc_put_hevc_epel_v_8_sse;
c->put_hevc_epel[1][1] = ff_hevc_put_hevc_epel_hv_8_sse;
#endif
c->sao_edge_filter[0] = ff_hevc_sao_edge_filter_0_8_sse;
c->sao_edge_filter[1] = ff_hevc_sao_edge_filter_1_8_sse;
c->sao_edge_filter[2] = ff_hevc_sao_edge_filter_2_8_sse;
c->sao_edge_filter[3] = ff_hevc_sao_edge_filter_3_8_sse;
c->sao_band_filter[0] = ff_hevc_sao_band_filter_0_8_sse;
c->sao_band_filter[1] = ff_hevc_sao_band_filter_1_8_sse;
c->sao_band_filter[2] = ff_hevc_sao_band_filter_2_8_sse;
c->sao_band_filter[3] = ff_hevc_sao_band_filter_3_8_sse;
c->put_hevc_qpel[1][1] = ff_hevc_put_hevc_qpel_h_1_v_1_sse;
c->put_hevc_qpel[1][2] = ff_hevc_put_hevc_qpel_h_2_v_1_sse;
c->put_hevc_qpel[1][3] = ff_hevc_put_hevc_qpel_h_3_v_1_sse;
c->put_hevc_qpel[2][1] = ff_hevc_put_hevc_qpel_h_1_v_2_sse;
c->put_hevc_qpel[2][2] = ff_hevc_put_hevc_qpel_h_2_v_2_sse;
c->put_hevc_qpel[2][3] = ff_hevc_put_hevc_qpel_h_3_v_2_sse;
c->put_hevc_qpel[3][1] = ff_hevc_put_hevc_qpel_h_1_v_3_sse;
c->put_hevc_qpel[3][2] = ff_hevc_put_hevc_qpel_h_2_v_3_sse;
c->put_hevc_qpel[3][3] = ff_hevc_put_hevc_qpel_h_3_v_3_sse;
}
if (EXTERNAL_AVX(mm_flags)) {
}
}
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMX(mm_flags)) {
if (EXTERNAL_MMXEXT(mm_flags)) {
#if ARCH_X86_32
#endif /* ARCH_X86_32 */
if (EXTERNAL_SSE2(mm_flags)) {
#if HAVE_ALIGNED_STACK
/*stuff that requires aligned stack */
#endif /* HAVE_ALIGNED_STACK */
}
if (EXTERNAL_SSE4(mm_flags)) {
c->transform_4x4_luma_add = ff_hevc_transform_4x4_luma_add_10_sse4;
c->transform_add[0] = ff_hevc_transform_4x4_add_10_sse4;
c->transform_add[1] = ff_hevc_transform_8x8_add_10_sse4;
c->transform_add[2] = ff_hevc_transform_16x16_add_10_sse4;
c->transform_add[3] = ff_hevc_transform_32x32_add_10_sse4;
c->put_hevc_epel[0][0] = ff_hevc_put_hevc_epel_pixels_10_sse;
c->put_hevc_epel[0][1] = ff_hevc_put_hevc_epel_h_10_sse;
c->put_hevc_epel[1][0] = ff_hevc_put_hevc_epel_v_10_sse;
c->put_hevc_epel[1][1] = ff_hevc_put_hevc_epel_hv_10_sse;
c->put_hevc_qpel[0][0] = ff_hevc_put_hevc_qpel_pixels_10_sse;
c->put_hevc_qpel[0][1] = ff_hevc_put_hevc_qpel_h_1_10_sse;
c->put_hevc_qpel[1][0] = ff_hevc_put_hevc_qpel_v_1_10_sse;
c->put_hevc_qpel[2][0] = ff_hevc_put_hevc_qpel_v_2_10_sse;
c->put_hevc_qpel[3][0] = ff_hevc_put_hevc_qpel_v_3_10_sse;
}
if (EXTERNAL_AVX(mm_flags)) {
}
}
}
}
}
av_cold void ff_idctdsp_init_x86(IDCTDSPContext *c, AVCodecContext *avctx,
unsigned high_bit_depth)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_MMX(cpu_flags)) {
c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
if (!high_bit_depth &&
avctx->lowres == 0 &&
(avctx->idct_algo == FF_IDCT_AUTO ||
avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
avctx->idct_algo == 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->perm_type = FF_IDCT_PERM_SIMPLE;
}
}
if (EXTERNAL_SSE2(cpu_flags)) {
c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
c->put_pixels_clamped = ff_put_pixels_clamped_sse2;
c->add_pixels_clamped = ff_add_pixels_clamped_sse2;
if (!high_bit_depth &&
avctx->lowres == 0 &&
(avctx->idct_algo == FF_IDCT_AUTO ||
avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
c->idct_put = ff_simple_idct_put_sse2;
c->idct_add = ff_simple_idct_add_sse2;
c->perm_type = FF_IDCT_PERM_SIMPLE;
}
}
if (ARCH_X86_64 && avctx->lowres == 0) {
if (avctx->bits_per_raw_sample == 10 &&
(avctx->idct_algo == FF_IDCT_AUTO ||
avctx->idct_algo == FF_IDCT_SIMPLEAUTO ||
avctx->idct_algo == FF_IDCT_SIMPLE)) {
if (EXTERNAL_SSE2(cpu_flags)) {
c->idct_put = ff_simple_idct10_put_sse2;
c->idct_add = NULL;
c->idct = ff_simple_idct10_sse2;
c->perm_type = FF_IDCT_PERM_TRANSPOSE;
}
if (EXTERNAL_AVX(cpu_flags)) {
c->idct_put = ff_simple_idct10_put_avx;
c->idct_add = NULL;
c->idct = ff_simple_idct10_avx;
c->perm_type = FF_IDCT_PERM_TRANSPOSE;
}
}
if (avctx->bits_per_raw_sample == 12 &&
(avctx->idct_algo == FF_IDCT_AUTO ||
avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
if (EXTERNAL_SSE2(cpu_flags)) {
c->idct_put = ff_simple_idct12_put_sse2;
c->idct_add = NULL;
c->idct = ff_simple_idct12_sse2;
c->perm_type = FF_IDCT_PERM_TRANSPOSE;
}
if (EXTERNAL_AVX(cpu_flags)) {
c->idct_put = ff_simple_idct12_put_avx;
c->idct_add = NULL;
c->idct = ff_simple_idct12_avx;
c->perm_type = FF_IDCT_PERM_TRANSPOSE;
}
}
}
}