yading@10: /*
yading@10:  * Copyright (c) 2004 Romain Dolbeau <romain@dolbeau.org>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "config.h"
yading@10: #include "libavutil/attributes.h"
yading@10: #include "libavcodec/h264qpel.h"
yading@10: 
yading@10: #if HAVE_ALTIVEC
yading@10: #include "libavutil/cpu.h"
yading@10: #include "libavutil/intreadwrite.h"
yading@10: #include "libavutil/ppc/types_altivec.h"
yading@10: #include "libavutil/ppc/util_altivec.h"
yading@10: #include "dsputil_altivec.h"
yading@10: 
yading@10: #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s
yading@10: #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s)
yading@10: 
yading@10: #define OP_U8_ALTIVEC                          PUT_OP_U8_ALTIVEC
yading@10: #define PREFIX_h264_qpel16_h_lowpass_altivec   put_h264_qpel16_h_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_h_lowpass_num       altivec_put_h264_qpel16_h_lowpass_num
yading@10: #define PREFIX_h264_qpel16_v_lowpass_altivec   put_h264_qpel16_v_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_v_lowpass_num       altivec_put_h264_qpel16_v_lowpass_num
yading@10: #define PREFIX_h264_qpel16_hv_lowpass_altivec  put_h264_qpel16_hv_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_put_h264_qpel16_hv_lowpass_num
yading@10: #include "h264_qpel_template.c"
yading@10: #undef OP_U8_ALTIVEC
yading@10: #undef PREFIX_h264_qpel16_h_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_h_lowpass_num
yading@10: #undef PREFIX_h264_qpel16_v_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_v_lowpass_num
yading@10: #undef PREFIX_h264_qpel16_hv_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_hv_lowpass_num
yading@10: 
yading@10: #define OP_U8_ALTIVEC                          AVG_OP_U8_ALTIVEC
yading@10: #define PREFIX_h264_qpel16_h_lowpass_altivec   avg_h264_qpel16_h_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_h_lowpass_num       altivec_avg_h264_qpel16_h_lowpass_num
yading@10: #define PREFIX_h264_qpel16_v_lowpass_altivec   avg_h264_qpel16_v_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_v_lowpass_num       altivec_avg_h264_qpel16_v_lowpass_num
yading@10: #define PREFIX_h264_qpel16_hv_lowpass_altivec  avg_h264_qpel16_hv_lowpass_altivec
yading@10: #define PREFIX_h264_qpel16_hv_lowpass_num      altivec_avg_h264_qpel16_hv_lowpass_num
yading@10: #include "h264_qpel_template.c"
yading@10: #undef OP_U8_ALTIVEC
yading@10: #undef PREFIX_h264_qpel16_h_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_h_lowpass_num
yading@10: #undef PREFIX_h264_qpel16_v_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_v_lowpass_num
yading@10: #undef PREFIX_h264_qpel16_hv_lowpass_altivec
yading@10: #undef PREFIX_h264_qpel16_hv_lowpass_num
yading@10: 
yading@10: #define H264_MC(OPNAME, SIZE, CODETYPE) \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     ff_ ## OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: { \
yading@10:     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
yading@10:     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
yading@10:     put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: \
yading@10: static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, ptrdiff_t stride)\
yading@10: {\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\
yading@10:     DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\
yading@10:     put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\
yading@10:     put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\
yading@10:     OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\
yading@10: }\
yading@10: 
yading@10: static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
yading@10:                                     const uint8_t * src2, int dst_stride,
yading@10:                                     int src_stride1, int h)
yading@10: {
yading@10:     int i;
yading@10:     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
yading@10: 
yading@10:     mask_ = vec_lvsl(0, src2);
yading@10: 
yading@10:     for (i = 0; i < h; i++) {
yading@10: 
yading@10:         tmp1 = vec_ld(i * src_stride1, src1);
yading@10:         mask = vec_lvsl(i * src_stride1, src1);
yading@10:         tmp2 = vec_ld(i * src_stride1 + 15, src1);
yading@10: 
yading@10:         a = vec_perm(tmp1, tmp2, mask);
yading@10: 
yading@10:         tmp1 = vec_ld(i * 16, src2);
yading@10:         tmp2 = vec_ld(i * 16 + 15, src2);
yading@10: 
yading@10:         b = vec_perm(tmp1, tmp2, mask_);
yading@10: 
yading@10:         tmp1 = vec_ld(0, dst);
yading@10:         mask = vec_lvsl(0, dst);
yading@10:         tmp2 = vec_ld(15, dst);
yading@10: 
yading@10:         d = vec_avg(a, b);
yading@10: 
yading@10:         edges = vec_perm(tmp2, tmp1, mask);
yading@10: 
yading@10:         align = vec_lvsr(0, dst);
yading@10: 
yading@10:         tmp2 = vec_perm(d, edges, align);
yading@10:         tmp1 = vec_perm(edges, d, align);
yading@10: 
yading@10:         vec_st(tmp2, 15, dst);
yading@10:         vec_st(tmp1, 0 , dst);
yading@10: 
yading@10:         dst += dst_stride;
yading@10:     }
yading@10: }
yading@10: 
yading@10: static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1,
yading@10:                                     const uint8_t * src2, int dst_stride,
yading@10:                                     int src_stride1, int h)
yading@10: {
yading@10:     int i;
yading@10:     vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align;
yading@10: 
yading@10:     mask_ = vec_lvsl(0, src2);
yading@10: 
yading@10:     for (i = 0; i < h; i++) {
yading@10: 
yading@10:         tmp1 = vec_ld(i * src_stride1, src1);
yading@10:         mask = vec_lvsl(i * src_stride1, src1);
yading@10:         tmp2 = vec_ld(i * src_stride1 + 15, src1);
yading@10: 
yading@10:         a = vec_perm(tmp1, tmp2, mask);
yading@10: 
yading@10:         tmp1 = vec_ld(i * 16, src2);
yading@10:         tmp2 = vec_ld(i * 16 + 15, src2);
yading@10: 
yading@10:         b = vec_perm(tmp1, tmp2, mask_);
yading@10: 
yading@10:         tmp1 = vec_ld(0, dst);
yading@10:         mask = vec_lvsl(0, dst);
yading@10:         tmp2 = vec_ld(15, dst);
yading@10: 
yading@10:         d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b));
yading@10: 
yading@10:         edges = vec_perm(tmp2, tmp1, mask);
yading@10: 
yading@10:         align = vec_lvsr(0, dst);
yading@10: 
yading@10:         tmp2 = vec_perm(d, edges, align);
yading@10:         tmp1 = vec_perm(edges, d, align);
yading@10: 
yading@10:         vec_st(tmp2, 15, dst);
yading@10:         vec_st(tmp1, 0 , dst);
yading@10: 
yading@10:         dst += dst_stride;
yading@10:     }
yading@10: }
yading@10: 
yading@10: /* Implemented but could be faster
yading@10: #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h)
yading@10: #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h)
yading@10:  */
yading@10: 
yading@10: H264_MC(put_, 16, altivec)
yading@10: H264_MC(avg_, 16, altivec)
yading@10: #endif /* HAVE_ALTIVEC */
yading@10: 
yading@10: av_cold void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth)
yading@10: {
yading@10: #if HAVE_ALTIVEC
yading@10:     const int high_bit_depth = bit_depth > 8;
yading@10: 
yading@10:     if (av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC) {
yading@10:     if (!high_bit_depth) {
yading@10: #define dspfunc(PFX, IDX, NUM) \
yading@10:         c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \
yading@10:         c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec
yading@10: 
yading@10:         dspfunc(put_h264_qpel, 0, 16);
yading@10:         dspfunc(avg_h264_qpel, 0, 16);
yading@10: #undef dspfunc
yading@10:     }
yading@10:     }
yading@10: #endif /* HAVE_ALTIVEC */
yading@10: }