nengel@2: /* nengel@2: * Copyright (c) 2004 Romain Dolbeau nengel@2: * nengel@2: * This file is part of FFmpeg. nengel@2: * nengel@2: * FFmpeg is free software; you can redistribute it and/or nengel@2: * modify it under the terms of the GNU Lesser General Public nengel@2: * License as published by the Free Software Foundation; either nengel@2: * version 2.1 of the License, or (at your option) any later version. nengel@2: * nengel@2: * FFmpeg is distributed in the hope that it will be useful, nengel@2: * but WITHOUT ANY WARRANTY; without even the implied warranty of nengel@2: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU nengel@2: * Lesser General Public License for more details. nengel@2: * nengel@2: * You should have received a copy of the GNU Lesser General Public nengel@2: * License along with FFmpeg; if not, write to the Free Software nengel@2: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA nengel@2: */ nengel@2: nengel@2: #include "libavcodec/dsputil.h" nengel@2: #include "libavcodec/h264_data.h" nengel@2: #include "libavcodec/h264_dsp.h" nengel@2: nengel@2: #include "dsputil_ppc.h" nengel@2: #include "dsputil_altivec.h" nengel@2: #include "util_altivec.h" nengel@2: #include "types_altivec.h" nengel@2: nengel@2: #define PUT_OP_U8_ALTIVEC(d, s, dst) d = s nengel@2: #define AVG_OP_U8_ALTIVEC(d, s, dst) d = vec_avg(dst, s) nengel@2: nengel@2: #define OP_U8_ALTIVEC PUT_OP_U8_ALTIVEC nengel@2: #define PREFIX_h264_chroma_mc8_altivec put_h264_chroma_mc8_altivec nengel@2: #define PREFIX_no_rnd_vc1_chroma_mc8_altivec put_no_rnd_vc1_chroma_mc8_altivec nengel@2: #define PREFIX_h264_chroma_mc8_num altivec_put_h264_chroma_mc8_num nengel@2: #define PREFIX_h264_qpel16_h_lowpass_altivec put_h264_qpel16_h_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_h_lowpass_num altivec_put_h264_qpel16_h_lowpass_num nengel@2: #define PREFIX_h264_qpel16_v_lowpass_altivec put_h264_qpel16_v_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_v_lowpass_num altivec_put_h264_qpel16_v_lowpass_num nengel@2: #define PREFIX_h264_qpel16_hv_lowpass_altivec put_h264_qpel16_hv_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_hv_lowpass_num altivec_put_h264_qpel16_hv_lowpass_num nengel@2: #include "h264_template_altivec.c" nengel@2: #undef OP_U8_ALTIVEC nengel@2: #undef PREFIX_h264_chroma_mc8_altivec nengel@2: #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec nengel@2: #undef PREFIX_h264_chroma_mc8_num nengel@2: #undef PREFIX_h264_qpel16_h_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_h_lowpass_num nengel@2: #undef PREFIX_h264_qpel16_v_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_v_lowpass_num nengel@2: #undef PREFIX_h264_qpel16_hv_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_hv_lowpass_num nengel@2: nengel@2: #define OP_U8_ALTIVEC AVG_OP_U8_ALTIVEC nengel@2: #define PREFIX_h264_chroma_mc8_altivec avg_h264_chroma_mc8_altivec nengel@2: #define PREFIX_no_rnd_vc1_chroma_mc8_altivec avg_no_rnd_vc1_chroma_mc8_altivec nengel@2: #define PREFIX_h264_chroma_mc8_num altivec_avg_h264_chroma_mc8_num nengel@2: #define PREFIX_h264_qpel16_h_lowpass_altivec avg_h264_qpel16_h_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_h_lowpass_num altivec_avg_h264_qpel16_h_lowpass_num nengel@2: #define PREFIX_h264_qpel16_v_lowpass_altivec avg_h264_qpel16_v_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_v_lowpass_num altivec_avg_h264_qpel16_v_lowpass_num nengel@2: #define PREFIX_h264_qpel16_hv_lowpass_altivec avg_h264_qpel16_hv_lowpass_altivec nengel@2: #define PREFIX_h264_qpel16_hv_lowpass_num altivec_avg_h264_qpel16_hv_lowpass_num nengel@2: #include "h264_template_altivec.c" nengel@2: #undef OP_U8_ALTIVEC nengel@2: #undef PREFIX_h264_chroma_mc8_altivec nengel@2: #undef PREFIX_no_rnd_vc1_chroma_mc8_altivec nengel@2: #undef PREFIX_h264_chroma_mc8_num nengel@2: #undef PREFIX_h264_qpel16_h_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_h_lowpass_num nengel@2: #undef PREFIX_h264_qpel16_v_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_v_lowpass_num nengel@2: #undef PREFIX_h264_qpel16_hv_lowpass_altivec nengel@2: #undef PREFIX_h264_qpel16_hv_lowpass_num nengel@2: nengel@2: #define H264_MC(OPNAME, SIZE, CODETYPE) \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## CODETYPE (uint8_t *dst, uint8_t *src, int stride){\ nengel@2: OPNAME ## pixels ## SIZE ## _ ## CODETYPE(dst, src, stride, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){ \ nengel@2: DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(dst, src, stride, stride);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+1, half, stride, stride, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src, half, stride, stride, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(dst, src, stride, stride);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, half)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(half, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, src+stride, half, stride, stride, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ nengel@2: OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(dst, tmp, src, stride, SIZE, stride);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfH)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ nengel@2: put_h264_qpel ## SIZE ## _h_lowpass_ ## CODETYPE(halfH, src + stride, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfH, halfHV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: \ nengel@2: static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## CODETYPE(uint8_t *dst, uint8_t *src, int stride){\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, uint8_t, halfHV)[SIZE*SIZE];\ nengel@2: DECLARE_ALIGNED(16, int16_t, tmp)[SIZE*(SIZE+8)];\ nengel@2: put_h264_qpel ## SIZE ## _v_lowpass_ ## CODETYPE(halfV, src+1, SIZE, stride);\ nengel@2: put_h264_qpel ## SIZE ## _hv_lowpass_ ## CODETYPE(halfHV, tmp, src, SIZE, SIZE, stride);\ nengel@2: OPNAME ## pixels ## SIZE ## _l2_ ## CODETYPE(dst, halfV, halfHV, stride, SIZE, SIZE);\ nengel@2: }\ nengel@2: nengel@2: static inline void put_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, nengel@2: const uint8_t * src2, int dst_stride, nengel@2: int src_stride1, int h) nengel@2: { nengel@2: int i; nengel@2: vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; nengel@2: nengel@2: mask_ = vec_lvsl(0, src2); nengel@2: nengel@2: for (i = 0; i < h; i++) { nengel@2: nengel@2: tmp1 = vec_ld(i * src_stride1, src1); nengel@2: mask = vec_lvsl(i * src_stride1, src1); nengel@2: tmp2 = vec_ld(i * src_stride1 + 15, src1); nengel@2: nengel@2: a = vec_perm(tmp1, tmp2, mask); nengel@2: nengel@2: tmp1 = vec_ld(i * 16, src2); nengel@2: tmp2 = vec_ld(i * 16 + 15, src2); nengel@2: nengel@2: b = vec_perm(tmp1, tmp2, mask_); nengel@2: nengel@2: tmp1 = vec_ld(0, dst); nengel@2: mask = vec_lvsl(0, dst); nengel@2: tmp2 = vec_ld(15, dst); nengel@2: nengel@2: d = vec_avg(a, b); nengel@2: nengel@2: edges = vec_perm(tmp2, tmp1, mask); nengel@2: nengel@2: align = vec_lvsr(0, dst); nengel@2: nengel@2: tmp2 = vec_perm(d, edges, align); nengel@2: tmp1 = vec_perm(edges, d, align); nengel@2: nengel@2: vec_st(tmp2, 15, dst); nengel@2: vec_st(tmp1, 0 , dst); nengel@2: nengel@2: dst += dst_stride; nengel@2: } nengel@2: } nengel@2: nengel@2: static inline void avg_pixels16_l2_altivec( uint8_t * dst, const uint8_t * src1, nengel@2: const uint8_t * src2, int dst_stride, nengel@2: int src_stride1, int h) nengel@2: { nengel@2: int i; nengel@2: vec_u8 a, b, d, tmp1, tmp2, mask, mask_, edges, align; nengel@2: nengel@2: mask_ = vec_lvsl(0, src2); nengel@2: nengel@2: for (i = 0; i < h; i++) { nengel@2: nengel@2: tmp1 = vec_ld(i * src_stride1, src1); nengel@2: mask = vec_lvsl(i * src_stride1, src1); nengel@2: tmp2 = vec_ld(i * src_stride1 + 15, src1); nengel@2: nengel@2: a = vec_perm(tmp1, tmp2, mask); nengel@2: nengel@2: tmp1 = vec_ld(i * 16, src2); nengel@2: tmp2 = vec_ld(i * 16 + 15, src2); nengel@2: nengel@2: b = vec_perm(tmp1, tmp2, mask_); nengel@2: nengel@2: tmp1 = vec_ld(0, dst); nengel@2: mask = vec_lvsl(0, dst); nengel@2: tmp2 = vec_ld(15, dst); nengel@2: nengel@2: d = vec_avg(vec_perm(tmp1, tmp2, mask), vec_avg(a, b)); nengel@2: nengel@2: edges = vec_perm(tmp2, tmp1, mask); nengel@2: nengel@2: align = vec_lvsr(0, dst); nengel@2: nengel@2: tmp2 = vec_perm(d, edges, align); nengel@2: tmp1 = vec_perm(edges, d, align); nengel@2: nengel@2: vec_st(tmp2, 15, dst); nengel@2: vec_st(tmp1, 0 , dst); nengel@2: nengel@2: dst += dst_stride; nengel@2: } nengel@2: } nengel@2: nengel@2: /* Implemented but could be faster nengel@2: #define put_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) put_pixels16_l2(d,s1,s2,ds,s1s,16,h) nengel@2: #define avg_pixels16_l2_altivec(d,s1,s2,ds,s1s,h) avg_pixels16_l2(d,s1,s2,ds,s1s,16,h) nengel@2: */ nengel@2: nengel@2: H264_MC(put_, 16, altivec) nengel@2: H264_MC(avg_, 16, altivec) nengel@2: nengel@2: nengel@2: /**************************************************************************** nengel@2: * IDCT transform: nengel@2: ****************************************************************************/ nengel@2: nengel@2: #define VEC_1D_DCT(vb0,vb1,vb2,vb3,va0,va1,va2,va3) \ nengel@2: /* 1st stage */ \ nengel@2: vz0 = vec_add(vb0,vb2); /* temp[0] = Y[0] + Y[2] */ \ nengel@2: vz1 = vec_sub(vb0,vb2); /* temp[1] = Y[0] - Y[2] */ \ nengel@2: vz2 = vec_sra(vb1,vec_splat_u16(1)); \ nengel@2: vz2 = vec_sub(vz2,vb3); /* temp[2] = Y[1].1/2 - Y[3] */ \ nengel@2: vz3 = vec_sra(vb3,vec_splat_u16(1)); \ nengel@2: vz3 = vec_add(vb1,vz3); /* temp[3] = Y[1] + Y[3].1/2 */ \ nengel@2: /* 2nd stage: output */ \ nengel@2: va0 = vec_add(vz0,vz3); /* x[0] = temp[0] + temp[3] */ \ nengel@2: va1 = vec_add(vz1,vz2); /* x[1] = temp[1] + temp[2] */ \ nengel@2: va2 = vec_sub(vz1,vz2); /* x[2] = temp[1] - temp[2] */ \ nengel@2: va3 = vec_sub(vz0,vz3) /* x[3] = temp[0] - temp[3] */ nengel@2: nengel@2: #define VEC_TRANSPOSE_4(a0,a1,a2,a3,b0,b1,b2,b3) \ nengel@2: b0 = vec_mergeh( a0, a0 ); \ nengel@2: b1 = vec_mergeh( a1, a0 ); \ nengel@2: b2 = vec_mergeh( a2, a0 ); \ nengel@2: b3 = vec_mergeh( a3, a0 ); \ nengel@2: a0 = vec_mergeh( b0, b2 ); \ nengel@2: a1 = vec_mergel( b0, b2 ); \ nengel@2: a2 = vec_mergeh( b1, b3 ); \ nengel@2: a3 = vec_mergel( b1, b3 ); \ nengel@2: b0 = vec_mergeh( a0, a2 ); \ nengel@2: b1 = vec_mergel( a0, a2 ); \ nengel@2: b2 = vec_mergeh( a1, a3 ); \ nengel@2: b3 = vec_mergel( a1, a3 ) nengel@2: nengel@2: #define VEC_LOAD_U8_ADD_S16_STORE_U8(va) \ nengel@2: vdst_orig = vec_ld(0, dst); \ nengel@2: vdst = vec_perm(vdst_orig, zero_u8v, vdst_mask); \ nengel@2: vdst_ss = (vec_s16) vec_mergeh(zero_u8v, vdst); \ nengel@2: va = vec_add(va, vdst_ss); \ nengel@2: va_u8 = vec_packsu(va, zero_s16v); \ nengel@2: va_u32 = vec_splat((vec_u32)va_u8, 0); \ nengel@2: vec_ste(va_u32, element, (uint32_t*)dst); nengel@2: nengel@2: static void ff_h264_idct_add_altivec(uint8_t *dst, DCTELEM *block, int stride) nengel@2: { nengel@2: vec_s16 va0, va1, va2, va3; nengel@2: vec_s16 vz0, vz1, vz2, vz3; nengel@2: vec_s16 vtmp0, vtmp1, vtmp2, vtmp3; nengel@2: vec_u8 va_u8; nengel@2: vec_u32 va_u32; nengel@2: vec_s16 vdst_ss; nengel@2: const vec_u16 v6us = vec_splat_u16(6); nengel@2: vec_u8 vdst, vdst_orig; nengel@2: vec_u8 vdst_mask = vec_lvsl(0, dst); nengel@2: int element = ((unsigned long)dst & 0xf) >> 2; nengel@2: LOAD_ZERO; nengel@2: nengel@2: block[0] += 32; /* add 32 as a DC-level for rounding */ nengel@2: nengel@2: vtmp0 = vec_ld(0,block); nengel@2: vtmp1 = vec_sld(vtmp0, vtmp0, 8); nengel@2: vtmp2 = vec_ld(16,block); nengel@2: vtmp3 = vec_sld(vtmp2, vtmp2, 8); nengel@2: nengel@2: VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); nengel@2: VEC_TRANSPOSE_4(va0,va1,va2,va3,vtmp0,vtmp1,vtmp2,vtmp3); nengel@2: VEC_1D_DCT(vtmp0,vtmp1,vtmp2,vtmp3,va0,va1,va2,va3); nengel@2: nengel@2: va0 = vec_sra(va0,v6us); nengel@2: va1 = vec_sra(va1,v6us); nengel@2: va2 = vec_sra(va2,v6us); nengel@2: va3 = vec_sra(va3,v6us); nengel@2: nengel@2: VEC_LOAD_U8_ADD_S16_STORE_U8(va0); nengel@2: dst += stride; nengel@2: VEC_LOAD_U8_ADD_S16_STORE_U8(va1); nengel@2: dst += stride; nengel@2: VEC_LOAD_U8_ADD_S16_STORE_U8(va2); nengel@2: dst += stride; nengel@2: VEC_LOAD_U8_ADD_S16_STORE_U8(va3); nengel@2: } nengel@2: nengel@2: #define IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, d0, d1, d2, d3, d4, d5, d6, d7) {\ nengel@2: /* a0 = SRC(0) + SRC(4); */ \ nengel@2: vec_s16 a0v = vec_add(s0, s4); \ nengel@2: /* a2 = SRC(0) - SRC(4); */ \ nengel@2: vec_s16 a2v = vec_sub(s0, s4); \ nengel@2: /* a4 = (SRC(2)>>1) - SRC(6); */ \ nengel@2: vec_s16 a4v = vec_sub(vec_sra(s2, onev), s6); \ nengel@2: /* a6 = (SRC(6)>>1) + SRC(2); */ \ nengel@2: vec_s16 a6v = vec_add(vec_sra(s6, onev), s2); \ nengel@2: /* b0 = a0 + a6; */ \ nengel@2: vec_s16 b0v = vec_add(a0v, a6v); \ nengel@2: /* b2 = a2 + a4; */ \ nengel@2: vec_s16 b2v = vec_add(a2v, a4v); \ nengel@2: /* b4 = a2 - a4; */ \ nengel@2: vec_s16 b4v = vec_sub(a2v, a4v); \ nengel@2: /* b6 = a0 - a6; */ \ nengel@2: vec_s16 b6v = vec_sub(a0v, a6v); \ nengel@2: /* a1 = SRC(5) - SRC(3) - SRC(7) - (SRC(7)>>1); */ \ nengel@2: /* a1 = (SRC(5)-SRC(3)) - (SRC(7) + (SRC(7)>>1)); */ \ nengel@2: vec_s16 a1v = vec_sub( vec_sub(s5, s3), vec_add(s7, vec_sra(s7, onev)) ); \ nengel@2: /* a3 = SRC(7) + SRC(1) - SRC(3) - (SRC(3)>>1); */ \ nengel@2: /* a3 = (SRC(7)+SRC(1)) - (SRC(3) + (SRC(3)>>1)); */ \ nengel@2: vec_s16 a3v = vec_sub( vec_add(s7, s1), vec_add(s3, vec_sra(s3, onev)) );\ nengel@2: /* a5 = SRC(7) - SRC(1) + SRC(5) + (SRC(5)>>1); */ \ nengel@2: /* a5 = (SRC(7)-SRC(1)) + SRC(5) + (SRC(5)>>1); */ \ nengel@2: vec_s16 a5v = vec_add( vec_sub(s7, s1), vec_add(s5, vec_sra(s5, onev)) );\ nengel@2: /* a7 = SRC(5)+SRC(3) + SRC(1) + (SRC(1)>>1); */ \ nengel@2: vec_s16 a7v = vec_add( vec_add(s5, s3), vec_add(s1, vec_sra(s1, onev)) );\ nengel@2: /* b1 = (a7>>2) + a1; */ \ nengel@2: vec_s16 b1v = vec_add( vec_sra(a7v, twov), a1v); \ nengel@2: /* b3 = a3 + (a5>>2); */ \ nengel@2: vec_s16 b3v = vec_add(a3v, vec_sra(a5v, twov)); \ nengel@2: /* b5 = (a3>>2) - a5; */ \ nengel@2: vec_s16 b5v = vec_sub( vec_sra(a3v, twov), a5v); \ nengel@2: /* b7 = a7 - (a1>>2); */ \ nengel@2: vec_s16 b7v = vec_sub( a7v, vec_sra(a1v, twov)); \ nengel@2: /* DST(0, b0 + b7); */ \ nengel@2: d0 = vec_add(b0v, b7v); \ nengel@2: /* DST(1, b2 + b5); */ \ nengel@2: d1 = vec_add(b2v, b5v); \ nengel@2: /* DST(2, b4 + b3); */ \ nengel@2: d2 = vec_add(b4v, b3v); \ nengel@2: /* DST(3, b6 + b1); */ \ nengel@2: d3 = vec_add(b6v, b1v); \ nengel@2: /* DST(4, b6 - b1); */ \ nengel@2: d4 = vec_sub(b6v, b1v); \ nengel@2: /* DST(5, b4 - b3); */ \ nengel@2: d5 = vec_sub(b4v, b3v); \ nengel@2: /* DST(6, b2 - b5); */ \ nengel@2: d6 = vec_sub(b2v, b5v); \ nengel@2: /* DST(7, b0 - b7); */ \ nengel@2: d7 = vec_sub(b0v, b7v); \ nengel@2: } nengel@2: nengel@2: #define ALTIVEC_STORE_SUM_CLIP(dest, idctv, perm_ldv, perm_stv, sel) { \ nengel@2: /* unaligned load */ \ nengel@2: vec_u8 hv = vec_ld( 0, dest ); \ nengel@2: vec_u8 lv = vec_ld( 7, dest ); \ nengel@2: vec_u8 dstv = vec_perm( hv, lv, (vec_u8)perm_ldv ); \ nengel@2: vec_s16 idct_sh6 = vec_sra(idctv, sixv); \ nengel@2: vec_u16 dst16 = (vec_u16)vec_mergeh(zero_u8v, dstv); \ nengel@2: vec_s16 idstsum = vec_adds(idct_sh6, (vec_s16)dst16); \ nengel@2: vec_u8 idstsum8 = vec_packsu(zero_s16v, idstsum); \ nengel@2: vec_u8 edgehv; \ nengel@2: /* unaligned store */ \ nengel@2: vec_u8 bodyv = vec_perm( idstsum8, idstsum8, perm_stv );\ nengel@2: vec_u8 edgelv = vec_perm( sel, zero_u8v, perm_stv ); \ nengel@2: lv = vec_sel( lv, bodyv, edgelv ); \ nengel@2: vec_st( lv, 7, dest ); \ nengel@2: hv = vec_ld( 0, dest ); \ nengel@2: edgehv = vec_perm( zero_u8v, sel, perm_stv ); \ nengel@2: hv = vec_sel( hv, bodyv, edgehv ); \ nengel@2: vec_st( hv, 0, dest ); \ nengel@2: } nengel@2: nengel@2: static void ff_h264_idct8_add_altivec( uint8_t *dst, DCTELEM *dct, int stride ) { nengel@2: vec_s16 s0, s1, s2, s3, s4, s5, s6, s7; nengel@2: vec_s16 d0, d1, d2, d3, d4, d5, d6, d7; nengel@2: vec_s16 idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7; nengel@2: nengel@2: vec_u8 perm_ldv = vec_lvsl(0, dst); nengel@2: vec_u8 perm_stv = vec_lvsr(8, dst); nengel@2: nengel@2: const vec_u16 onev = vec_splat_u16(1); nengel@2: const vec_u16 twov = vec_splat_u16(2); nengel@2: const vec_u16 sixv = vec_splat_u16(6); nengel@2: nengel@2: const vec_u8 sel = (vec_u8) {0,0,0,0,0,0,0,0,-1,-1,-1,-1,-1,-1,-1,-1}; nengel@2: LOAD_ZERO; nengel@2: nengel@2: dct[0] += 32; // rounding for the >>6 at the end nengel@2: nengel@2: s0 = vec_ld(0x00, (int16_t*)dct); nengel@2: s1 = vec_ld(0x10, (int16_t*)dct); nengel@2: s2 = vec_ld(0x20, (int16_t*)dct); nengel@2: s3 = vec_ld(0x30, (int16_t*)dct); nengel@2: s4 = vec_ld(0x40, (int16_t*)dct); nengel@2: s5 = vec_ld(0x50, (int16_t*)dct); nengel@2: s6 = vec_ld(0x60, (int16_t*)dct); nengel@2: s7 = vec_ld(0x70, (int16_t*)dct); nengel@2: nengel@2: IDCT8_1D_ALTIVEC(s0, s1, s2, s3, s4, s5, s6, s7, nengel@2: d0, d1, d2, d3, d4, d5, d6, d7); nengel@2: nengel@2: TRANSPOSE8( d0, d1, d2, d3, d4, d5, d6, d7 ); nengel@2: nengel@2: IDCT8_1D_ALTIVEC(d0, d1, d2, d3, d4, d5, d6, d7, nengel@2: idct0, idct1, idct2, idct3, idct4, idct5, idct6, idct7); nengel@2: nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[0*stride], idct0, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[1*stride], idct1, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[2*stride], idct2, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[3*stride], idct3, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[4*stride], idct4, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[5*stride], idct5, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[6*stride], idct6, perm_ldv, perm_stv, sel); nengel@2: ALTIVEC_STORE_SUM_CLIP(&dst[7*stride], idct7, perm_ldv, perm_stv, sel); nengel@2: } nengel@2: nengel@2: static av_always_inline void h264_idct_dc_add_internal(uint8_t *dst, DCTELEM *block, int stride, int size) nengel@2: { nengel@2: vec_s16 dc16; nengel@2: vec_u8 dcplus, dcminus, v0, v1, v2, v3, aligner; nengel@2: LOAD_ZERO; nengel@2: DECLARE_ALIGNED(16, int, dc); nengel@2: int i; nengel@2: nengel@2: dc = (block[0] + 32) >> 6; nengel@2: dc16 = vec_splat((vec_s16) vec_lde(0, &dc), 1); nengel@2: nengel@2: if (size == 4) nengel@2: dc16 = vec_sld(dc16, zero_s16v, 8); nengel@2: dcplus = vec_packsu(dc16, zero_s16v); nengel@2: dcminus = vec_packsu(vec_sub(zero_s16v, dc16), zero_s16v); nengel@2: nengel@2: aligner = vec_lvsr(0, dst); nengel@2: dcplus = vec_perm(dcplus, dcplus, aligner); nengel@2: dcminus = vec_perm(dcminus, dcminus, aligner); nengel@2: nengel@2: for (i = 0; i < size; i += 4) { nengel@2: v0 = vec_ld(0, dst+0*stride); nengel@2: v1 = vec_ld(0, dst+1*stride); nengel@2: v2 = vec_ld(0, dst+2*stride); nengel@2: v3 = vec_ld(0, dst+3*stride); nengel@2: nengel@2: v0 = vec_adds(v0, dcplus); nengel@2: v1 = vec_adds(v1, dcplus); nengel@2: v2 = vec_adds(v2, dcplus); nengel@2: v3 = vec_adds(v3, dcplus); nengel@2: nengel@2: v0 = vec_subs(v0, dcminus); nengel@2: v1 = vec_subs(v1, dcminus); nengel@2: v2 = vec_subs(v2, dcminus); nengel@2: v3 = vec_subs(v3, dcminus); nengel@2: nengel@2: vec_st(v0, 0, dst+0*stride); nengel@2: vec_st(v1, 0, dst+1*stride); nengel@2: vec_st(v2, 0, dst+2*stride); nengel@2: vec_st(v3, 0, dst+3*stride); nengel@2: nengel@2: dst += 4*stride; nengel@2: } nengel@2: } nengel@2: nengel@2: static void h264_idct_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride) nengel@2: { nengel@2: h264_idct_dc_add_internal(dst, block, stride, 4); nengel@2: } nengel@2: nengel@2: static void ff_h264_idct8_dc_add_altivec(uint8_t *dst, DCTELEM *block, int stride) nengel@2: { nengel@2: h264_idct_dc_add_internal(dst, block, stride, 8); nengel@2: } nengel@2: nengel@2: static void ff_h264_idct_add16_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ nengel@2: int i; nengel@2: for(i=0; i<16; i++){ nengel@2: int nnz = nnzc[ scan8[i] ]; nengel@2: if(nnz){ nengel@2: if(nnz==1 && block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); nengel@2: else ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); nengel@2: } nengel@2: } nengel@2: } nengel@2: nengel@2: static void ff_h264_idct_add16intra_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ nengel@2: int i; nengel@2: for(i=0; i<16; i++){ nengel@2: if(nnzc[ scan8[i] ]) ff_h264_idct_add_altivec(dst + block_offset[i], block + i*16, stride); nengel@2: else if(block[i*16]) h264_idct_dc_add_altivec(dst + block_offset[i], block + i*16, stride); nengel@2: } nengel@2: } nengel@2: nengel@2: static void ff_h264_idct8_add4_altivec(uint8_t *dst, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ nengel@2: int i; nengel@2: for(i=0; i<16; i+=4){ nengel@2: int nnz = nnzc[ scan8[i] ]; nengel@2: if(nnz){ nengel@2: if(nnz==1 && block[i*16]) ff_h264_idct8_dc_add_altivec(dst + block_offset[i], block + i*16, stride); nengel@2: else ff_h264_idct8_add_altivec (dst + block_offset[i], block + i*16, stride); nengel@2: } nengel@2: } nengel@2: } nengel@2: nengel@2: static void ff_h264_idct_add8_altivec(uint8_t **dest, const int *block_offset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]){ nengel@2: int i; nengel@2: for(i=16; i<16+8; i++){ nengel@2: if(nnzc[ scan8[i] ]) nengel@2: ff_h264_idct_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride); nengel@2: else if(block[i*16]) nengel@2: h264_idct_dc_add_altivec(dest[(i&4)>>2] + block_offset[i], block + i*16, stride); nengel@2: } nengel@2: } nengel@2: nengel@2: #define transpose4x16(r0, r1, r2, r3) { \ nengel@2: register vec_u8 r4; \ nengel@2: register vec_u8 r5; \ nengel@2: register vec_u8 r6; \ nengel@2: register vec_u8 r7; \ nengel@2: \ nengel@2: r4 = vec_mergeh(r0, r2); /*0, 2 set 0*/ \ nengel@2: r5 = vec_mergel(r0, r2); /*0, 2 set 1*/ \ nengel@2: r6 = vec_mergeh(r1, r3); /*1, 3 set 0*/ \ nengel@2: r7 = vec_mergel(r1, r3); /*1, 3 set 1*/ \ nengel@2: \ nengel@2: r0 = vec_mergeh(r4, r6); /*all set 0*/ \ nengel@2: r1 = vec_mergel(r4, r6); /*all set 1*/ \ nengel@2: r2 = vec_mergeh(r5, r7); /*all set 2*/ \ nengel@2: r3 = vec_mergel(r5, r7); /*all set 3*/ \ nengel@2: } nengel@2: nengel@2: static inline void write16x4(uint8_t *dst, int dst_stride, nengel@2: register vec_u8 r0, register vec_u8 r1, nengel@2: register vec_u8 r2, register vec_u8 r3) { nengel@2: DECLARE_ALIGNED(16, unsigned char, result)[64]; nengel@2: uint32_t *src_int = (uint32_t *)result, *dst_int = (uint32_t *)dst; nengel@2: int int_dst_stride = dst_stride/4; nengel@2: nengel@2: vec_st(r0, 0, result); nengel@2: vec_st(r1, 16, result); nengel@2: vec_st(r2, 32, result); nengel@2: vec_st(r3, 48, result); nengel@2: /* FIXME: there has to be a better way!!!! */ nengel@2: *dst_int = *src_int; nengel@2: *(dst_int+ int_dst_stride) = *(src_int + 1); nengel@2: *(dst_int+ 2*int_dst_stride) = *(src_int + 2); nengel@2: *(dst_int+ 3*int_dst_stride) = *(src_int + 3); nengel@2: *(dst_int+ 4*int_dst_stride) = *(src_int + 4); nengel@2: *(dst_int+ 5*int_dst_stride) = *(src_int + 5); nengel@2: *(dst_int+ 6*int_dst_stride) = *(src_int + 6); nengel@2: *(dst_int+ 7*int_dst_stride) = *(src_int + 7); nengel@2: *(dst_int+ 8*int_dst_stride) = *(src_int + 8); nengel@2: *(dst_int+ 9*int_dst_stride) = *(src_int + 9); nengel@2: *(dst_int+10*int_dst_stride) = *(src_int + 10); nengel@2: *(dst_int+11*int_dst_stride) = *(src_int + 11); nengel@2: *(dst_int+12*int_dst_stride) = *(src_int + 12); nengel@2: *(dst_int+13*int_dst_stride) = *(src_int + 13); nengel@2: *(dst_int+14*int_dst_stride) = *(src_int + 14); nengel@2: *(dst_int+15*int_dst_stride) = *(src_int + 15); nengel@2: } nengel@2: nengel@2: /** \brief performs a 6x16 transpose of data in src, and stores it to dst nengel@2: \todo FIXME: see if we can't spare some vec_lvsl() by them factorizing nengel@2: out of unaligned_load() */ nengel@2: #define readAndTranspose16x6(src, src_stride, r8, r9, r10, r11, r12, r13) {\ nengel@2: register vec_u8 r0 = unaligned_load(0, src); \ nengel@2: register vec_u8 r1 = unaligned_load( src_stride, src); \ nengel@2: register vec_u8 r2 = unaligned_load(2* src_stride, src); \ nengel@2: register vec_u8 r3 = unaligned_load(3* src_stride, src); \ nengel@2: register vec_u8 r4 = unaligned_load(4* src_stride, src); \ nengel@2: register vec_u8 r5 = unaligned_load(5* src_stride, src); \ nengel@2: register vec_u8 r6 = unaligned_load(6* src_stride, src); \ nengel@2: register vec_u8 r7 = unaligned_load(7* src_stride, src); \ nengel@2: register vec_u8 r14 = unaligned_load(14*src_stride, src); \ nengel@2: register vec_u8 r15 = unaligned_load(15*src_stride, src); \ nengel@2: \ nengel@2: r8 = unaligned_load( 8*src_stride, src); \ nengel@2: r9 = unaligned_load( 9*src_stride, src); \ nengel@2: r10 = unaligned_load(10*src_stride, src); \ nengel@2: r11 = unaligned_load(11*src_stride, src); \ nengel@2: r12 = unaligned_load(12*src_stride, src); \ nengel@2: r13 = unaligned_load(13*src_stride, src); \ nengel@2: \ nengel@2: /*Merge first pairs*/ \ nengel@2: r0 = vec_mergeh(r0, r8); /*0, 8*/ \ nengel@2: r1 = vec_mergeh(r1, r9); /*1, 9*/ \ nengel@2: r2 = vec_mergeh(r2, r10); /*2,10*/ \ nengel@2: r3 = vec_mergeh(r3, r11); /*3,11*/ \ nengel@2: r4 = vec_mergeh(r4, r12); /*4,12*/ \ nengel@2: r5 = vec_mergeh(r5, r13); /*5,13*/ \ nengel@2: r6 = vec_mergeh(r6, r14); /*6,14*/ \ nengel@2: r7 = vec_mergeh(r7, r15); /*7,15*/ \ nengel@2: \ nengel@2: /*Merge second pairs*/ \ nengel@2: r8 = vec_mergeh(r0, r4); /*0,4, 8,12 set 0*/ \ nengel@2: r9 = vec_mergel(r0, r4); /*0,4, 8,12 set 1*/ \ nengel@2: r10 = vec_mergeh(r1, r5); /*1,5, 9,13 set 0*/ \ nengel@2: r11 = vec_mergel(r1, r5); /*1,5, 9,13 set 1*/ \ nengel@2: r12 = vec_mergeh(r2, r6); /*2,6,10,14 set 0*/ \ nengel@2: r13 = vec_mergel(r2, r6); /*2,6,10,14 set 1*/ \ nengel@2: r14 = vec_mergeh(r3, r7); /*3,7,11,15 set 0*/ \ nengel@2: r15 = vec_mergel(r3, r7); /*3,7,11,15 set 1*/ \ nengel@2: \ nengel@2: /*Third merge*/ \ nengel@2: r0 = vec_mergeh(r8, r12); /*0,2,4,6,8,10,12,14 set 0*/ \ nengel@2: r1 = vec_mergel(r8, r12); /*0,2,4,6,8,10,12,14 set 1*/ \ nengel@2: r2 = vec_mergeh(r9, r13); /*0,2,4,6,8,10,12,14 set 2*/ \ nengel@2: r4 = vec_mergeh(r10, r14); /*1,3,5,7,9,11,13,15 set 0*/ \ nengel@2: r5 = vec_mergel(r10, r14); /*1,3,5,7,9,11,13,15 set 1*/ \ nengel@2: r6 = vec_mergeh(r11, r15); /*1,3,5,7,9,11,13,15 set 2*/ \ nengel@2: /* Don't need to compute 3 and 7*/ \ nengel@2: \ nengel@2: /*Final merge*/ \ nengel@2: r8 = vec_mergeh(r0, r4); /*all set 0*/ \ nengel@2: r9 = vec_mergel(r0, r4); /*all set 1*/ \ nengel@2: r10 = vec_mergeh(r1, r5); /*all set 2*/ \ nengel@2: r11 = vec_mergel(r1, r5); /*all set 3*/ \ nengel@2: r12 = vec_mergeh(r2, r6); /*all set 4*/ \ nengel@2: r13 = vec_mergel(r2, r6); /*all set 5*/ \ nengel@2: /* Don't need to compute 14 and 15*/ \ nengel@2: \ nengel@2: } nengel@2: nengel@2: // out: o = |x-y| < a nengel@2: static inline vec_u8 diff_lt_altivec ( register vec_u8 x, nengel@2: register vec_u8 y, nengel@2: register vec_u8 a) { nengel@2: nengel@2: register vec_u8 diff = vec_subs(x, y); nengel@2: register vec_u8 diffneg = vec_subs(y, x); nengel@2: register vec_u8 o = vec_or(diff, diffneg); /* |x-y| */ nengel@2: o = (vec_u8)vec_cmplt(o, a); nengel@2: return o; nengel@2: } nengel@2: nengel@2: static inline vec_u8 h264_deblock_mask ( register vec_u8 p0, nengel@2: register vec_u8 p1, nengel@2: register vec_u8 q0, nengel@2: register vec_u8 q1, nengel@2: register vec_u8 alpha, nengel@2: register vec_u8 beta) { nengel@2: nengel@2: register vec_u8 mask; nengel@2: register vec_u8 tempmask; nengel@2: nengel@2: mask = diff_lt_altivec(p0, q0, alpha); nengel@2: tempmask = diff_lt_altivec(p1, p0, beta); nengel@2: mask = vec_and(mask, tempmask); nengel@2: tempmask = diff_lt_altivec(q1, q0, beta); nengel@2: mask = vec_and(mask, tempmask); nengel@2: nengel@2: return mask; nengel@2: } nengel@2: nengel@2: // out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0) nengel@2: static inline vec_u8 h264_deblock_q1(register vec_u8 p0, nengel@2: register vec_u8 p1, nengel@2: register vec_u8 p2, nengel@2: register vec_u8 q0, nengel@2: register vec_u8 tc0) { nengel@2: nengel@2: register vec_u8 average = vec_avg(p0, q0); nengel@2: register vec_u8 temp; nengel@2: register vec_u8 uncliped; nengel@2: register vec_u8 ones; nengel@2: register vec_u8 max; nengel@2: register vec_u8 min; nengel@2: register vec_u8 newp1; nengel@2: nengel@2: temp = vec_xor(average, p2); nengel@2: average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */ nengel@2: ones = vec_splat_u8(1); nengel@2: temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */ nengel@2: uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */ nengel@2: max = vec_adds(p1, tc0); nengel@2: min = vec_subs(p1, tc0); nengel@2: newp1 = vec_max(min, uncliped); nengel@2: newp1 = vec_min(max, newp1); nengel@2: return newp1; nengel@2: } nengel@2: nengel@2: #define h264_deblock_p0_q0(p0, p1, q0, q1, tc0masked) { \ nengel@2: \ nengel@2: const vec_u8 A0v = vec_sl(vec_splat_u8(10), vec_splat_u8(4)); \ nengel@2: \ nengel@2: register vec_u8 pq0bit = vec_xor(p0,q0); \ nengel@2: register vec_u8 q1minus; \ nengel@2: register vec_u8 p0minus; \ nengel@2: register vec_u8 stage1; \ nengel@2: register vec_u8 stage2; \ nengel@2: register vec_u8 vec160; \ nengel@2: register vec_u8 delta; \ nengel@2: register vec_u8 deltaneg; \ nengel@2: \ nengel@2: q1minus = vec_nor(q1, q1); /* 255 - q1 */ \ nengel@2: stage1 = vec_avg(p1, q1minus); /* (p1 - q1 + 256)>>1 */ \ nengel@2: stage2 = vec_sr(stage1, vec_splat_u8(1)); /* (p1 - q1 + 256)>>2 = 64 + (p1 - q1) >> 2 */ \ nengel@2: p0minus = vec_nor(p0, p0); /* 255 - p0 */ \ nengel@2: stage1 = vec_avg(q0, p0minus); /* (q0 - p0 + 256)>>1 */ \ nengel@2: pq0bit = vec_and(pq0bit, vec_splat_u8(1)); \ nengel@2: stage2 = vec_avg(stage2, pq0bit); /* 32 + ((q0 - p0)&1 + (p1 - q1) >> 2 + 1) >> 1 */ \ nengel@2: stage2 = vec_adds(stage2, stage1); /* 160 + ((p0 - q0) + (p1 - q1) >> 2 + 1) >> 1 */ \ nengel@2: vec160 = vec_ld(0, &A0v); \ nengel@2: deltaneg = vec_subs(vec160, stage2); /* -d */ \ nengel@2: delta = vec_subs(stage2, vec160); /* d */ \ nengel@2: deltaneg = vec_min(tc0masked, deltaneg); \ nengel@2: delta = vec_min(tc0masked, delta); \ nengel@2: p0 = vec_subs(p0, deltaneg); \ nengel@2: q0 = vec_subs(q0, delta); \ nengel@2: p0 = vec_adds(p0, delta); \ nengel@2: q0 = vec_adds(q0, deltaneg); \ nengel@2: } nengel@2: nengel@2: #define h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0) { \ nengel@2: DECLARE_ALIGNED(16, unsigned char, temp)[16]; \ nengel@2: register vec_u8 alphavec; \ nengel@2: register vec_u8 betavec; \ nengel@2: register vec_u8 mask; \ nengel@2: register vec_u8 p1mask; \ nengel@2: register vec_u8 q1mask; \ nengel@2: register vector signed char tc0vec; \ nengel@2: register vec_u8 finaltc0; \ nengel@2: register vec_u8 tc0masked; \ nengel@2: register vec_u8 newp1; \ nengel@2: register vec_u8 newq1; \ nengel@2: \ nengel@2: temp[0] = alpha; \ nengel@2: temp[1] = beta; \ nengel@2: alphavec = vec_ld(0, temp); \ nengel@2: betavec = vec_splat(alphavec, 0x1); \ nengel@2: alphavec = vec_splat(alphavec, 0x0); \ nengel@2: mask = h264_deblock_mask(p0, p1, q0, q1, alphavec, betavec); /*if in block */ \ nengel@2: \ nengel@2: *((int *)temp) = *((int *)tc0); \ nengel@2: tc0vec = vec_ld(0, (signed char*)temp); \ nengel@2: tc0vec = vec_mergeh(tc0vec, tc0vec); \ nengel@2: tc0vec = vec_mergeh(tc0vec, tc0vec); \ nengel@2: mask = vec_and(mask, vec_cmpgt(tc0vec, vec_splat_s8(-1))); /* if tc0[i] >= 0 */ \ nengel@2: finaltc0 = vec_and((vec_u8)tc0vec, mask); /* tc = tc0 */ \ nengel@2: \ nengel@2: p1mask = diff_lt_altivec(p2, p0, betavec); \ nengel@2: p1mask = vec_and(p1mask, mask); /* if ( |p2 - p0| < beta) */ \ nengel@2: tc0masked = vec_and(p1mask, (vec_u8)tc0vec); \ nengel@2: finaltc0 = vec_sub(finaltc0, p1mask); /* tc++ */ \ nengel@2: newp1 = h264_deblock_q1(p0, p1, p2, q0, tc0masked); \ nengel@2: /*end if*/ \ nengel@2: \ nengel@2: q1mask = diff_lt_altivec(q2, q0, betavec); \ nengel@2: q1mask = vec_and(q1mask, mask); /* if ( |q2 - q0| < beta ) */\ nengel@2: tc0masked = vec_and(q1mask, (vec_u8)tc0vec); \ nengel@2: finaltc0 = vec_sub(finaltc0, q1mask); /* tc++ */ \ nengel@2: newq1 = h264_deblock_q1(p0, q1, q2, q0, tc0masked); \ nengel@2: /*end if*/ \ nengel@2: \ nengel@2: h264_deblock_p0_q0(p0, p1, q0, q1, finaltc0); \ nengel@2: p1 = newp1; \ nengel@2: q1 = newq1; \ nengel@2: } nengel@2: nengel@2: static void h264_v_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { nengel@2: nengel@2: if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) >= 0) { nengel@2: register vec_u8 p2 = vec_ld(-3*stride, pix); nengel@2: register vec_u8 p1 = vec_ld(-2*stride, pix); nengel@2: register vec_u8 p0 = vec_ld(-1*stride, pix); nengel@2: register vec_u8 q0 = vec_ld(0, pix); nengel@2: register vec_u8 q1 = vec_ld(stride, pix); nengel@2: register vec_u8 q2 = vec_ld(2*stride, pix); nengel@2: h264_loop_filter_luma_altivec(p2, p1, p0, q0, q1, q2, alpha, beta, tc0); nengel@2: vec_st(p1, -2*stride, pix); nengel@2: vec_st(p0, -1*stride, pix); nengel@2: vec_st(q0, 0, pix); nengel@2: vec_st(q1, stride, pix); nengel@2: } nengel@2: } nengel@2: nengel@2: static void h264_h_loop_filter_luma_altivec(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0) { nengel@2: nengel@2: register vec_u8 line0, line1, line2, line3, line4, line5; nengel@2: if ((tc0[0] & tc0[1] & tc0[2] & tc0[3]) < 0) nengel@2: return; nengel@2: readAndTranspose16x6(pix-3, stride, line0, line1, line2, line3, line4, line5); nengel@2: h264_loop_filter_luma_altivec(line0, line1, line2, line3, line4, line5, alpha, beta, tc0); nengel@2: transpose4x16(line1, line2, line3, line4); nengel@2: write16x4(pix-2, stride, line1, line2, line3, line4); nengel@2: } nengel@2: nengel@2: static av_always_inline nengel@2: void weight_h264_WxH_altivec(uint8_t *block, int stride, int log2_denom, int weight, int offset, int w, int h) nengel@2: { nengel@2: int y, aligned; nengel@2: vec_u8 vblock; nengel@2: vec_s16 vtemp, vweight, voffset, v0, v1; nengel@2: vec_u16 vlog2_denom; nengel@2: DECLARE_ALIGNED(16, int32_t, temp)[4]; nengel@2: LOAD_ZERO; nengel@2: nengel@2: offset <<= log2_denom; nengel@2: if(log2_denom) offset += 1<<(log2_denom-1); nengel@2: temp[0] = log2_denom; nengel@2: temp[1] = weight; nengel@2: temp[2] = offset; nengel@2: nengel@2: vtemp = (vec_s16)vec_ld(0, temp); nengel@2: vlog2_denom = (vec_u16)vec_splat(vtemp, 1); nengel@2: vweight = vec_splat(vtemp, 3); nengel@2: voffset = vec_splat(vtemp, 5); nengel@2: aligned = !((unsigned long)block & 0xf); nengel@2: nengel@2: for (y=0; yput_h264_chroma_pixels_tab[0] = put_h264_chroma_mc8_altivec; nengel@2: c->avg_h264_chroma_pixels_tab[0] = avg_h264_chroma_mc8_altivec; nengel@2: nengel@2: #define dspfunc(PFX, IDX, NUM) \ nengel@2: c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_altivec; \ nengel@2: c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_altivec nengel@2: nengel@2: dspfunc(put_h264_qpel, 0, 16); nengel@2: dspfunc(avg_h264_qpel, 0, 16); nengel@2: #undef dspfunc nengel@2: } nengel@2: nengel@2: void ff_h264dsp_init_ppc(H264DSPContext *c){ nengel@2: c->h264_idct_dc_add= h264_idct_dc_add_altivec; nengel@2: c->h264_idct_add = ff_h264_idct_add_altivec; nengel@2: c->h264_idct_add8 = ff_h264_idct_add8_altivec; nengel@2: c->h264_idct_add16 = ff_h264_idct_add16_altivec; nengel@2: c->h264_idct_add16intra = ff_h264_idct_add16intra_altivec; nengel@2: nengel@2: c->h264_idct8_dc_add = ff_h264_idct8_dc_add_altivec; nengel@2: c->h264_idct8_add = ff_h264_idct8_add_altivec; nengel@2: c->h264_idct8_add4 = ff_h264_idct8_add4_altivec; nengel@2: c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_altivec; nengel@2: c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_altivec; nengel@2: nengel@2: c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels16x16_altivec; nengel@2: c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels16x8_altivec; nengel@2: c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels8x16_altivec; nengel@2: c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels8x8_altivec; nengel@2: c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels8x4_altivec; nengel@2: c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels16x16_altivec; nengel@2: c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels16x8_altivec; nengel@2: c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels8x16_altivec; nengel@2: c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels8x8_altivec; nengel@2: c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels8x4_altivec; nengel@2: }