vpx_convolve_msa.h revision 7ce0a1d1337c01056ba24006efab21f00e179e04
1/* 2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11#ifndef VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_ 12#define VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_ 13 14#include "vpx_dsp/mips/macros_msa.h" 15#include "vpx_dsp/vpx_filter.h" 16 17extern const uint8_t mc_filt_mask_arr[16 * 3]; 18 19#define FILT_8TAP_DPADD_S_H(vec0, vec1, vec2, vec3, \ 20 filt0, filt1, filt2, filt3) ({ \ 21 v8i16 tmp0, tmp1; \ 22 \ 23 tmp0 = __msa_dotp_s_h((v16i8)vec0, (v16i8)filt0); \ 24 tmp0 = __msa_dpadd_s_h(tmp0, (v16i8)vec1, (v16i8)filt1); \ 25 tmp1 = __msa_dotp_s_h((v16i8)vec2, (v16i8)filt2); \ 26 tmp1 = __msa_dpadd_s_h(tmp1, (v16i8)vec3, (v16i8)filt3); \ 27 tmp0 = __msa_adds_s_h(tmp0, tmp1); \ 28 \ 29 tmp0; \ 30}) 31 32#define HORIZ_8TAP_FILT(src0, src1, mask0, mask1, mask2, mask3, \ 33 filt_h0, filt_h1, filt_h2, filt_h3) ({ \ 34 v16i8 vec0_m, vec1_m, vec2_m, vec3_m; \ 35 v8i16 hz_out_m; \ 36 \ 37 VSHF_B4_SB(src0, src1, mask0, mask1, mask2, mask3, \ 38 vec0_m, vec1_m, vec2_m, vec3_m); \ 39 hz_out_m = FILT_8TAP_DPADD_S_H(vec0_m, vec1_m, vec2_m, vec3_m, \ 40 filt_h0, filt_h1, filt_h2, filt_h3); \ 41 \ 42 hz_out_m = __msa_srari_h(hz_out_m, FILTER_BITS); \ 43 hz_out_m = __msa_sat_s_h(hz_out_m, 7); \ 44 \ 45 hz_out_m; \ 46}) 47 48#define HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, \ 49 mask0, mask1, mask2, mask3, \ 50 filt0, filt1, filt2, filt3, \ 51 out0, out1) { \ 52 v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ 53 v8i16 res0_m, res1_m, res2_m, res3_m; \ 54 \ 55 VSHF_B2_SB(src0, src1, src2, src3, mask0, mask0, vec0_m, vec1_m); \ 56 DOTP_SB2_SH(vec0_m, vec1_m, filt0, filt0, res0_m, res1_m); \ 57 VSHF_B2_SB(src0, src1, src2, src3, mask1, mask1, vec2_m, vec3_m); \ 58 DPADD_SB2_SH(vec2_m, vec3_m, filt1, filt1, res0_m, res1_m); \ 59 VSHF_B2_SB(src0, src1, src2, src3, mask2, mask2, vec4_m, vec5_m); \ 60 DOTP_SB2_SH(vec4_m, vec5_m, filt2, filt2, res2_m, res3_m); \ 61 VSHF_B2_SB(src0, src1, src2, src3, mask3, mask3, vec6_m, vec7_m); \ 62 DPADD_SB2_SH(vec6_m, vec7_m, filt3, filt3, res2_m, res3_m); \ 63 ADDS_SH2_SH(res0_m, res2_m, res1_m, res3_m, out0, out1); \ 64} 65 66#define HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, \ 67 mask0, mask1, mask2, mask3, \ 68 filt0, filt1, filt2, filt3, \ 69 out0, out1, out2, out3) { \ 70 v16i8 vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ 71 v8i16 res0_m, res1_m, res2_m, res3_m, res4_m, res5_m, res6_m, res7_m; \ 72 \ 73 VSHF_B2_SB(src0, src0, src1, src1, mask0, mask0, vec0_m, vec1_m); \ 74 VSHF_B2_SB(src2, src2, src3, src3, mask0, mask0, vec2_m, vec3_m); \ 75 DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt0, filt0, filt0, filt0, \ 76 res0_m, res1_m, res2_m, res3_m); \ 77 VSHF_B2_SB(src0, src0, src1, src1, mask2, mask2, vec0_m, vec1_m); \ 78 VSHF_B2_SB(src2, src2, src3, src3, mask2, mask2, vec2_m, vec3_m); \ 79 DOTP_SB4_SH(vec0_m, vec1_m, vec2_m, vec3_m, filt2, filt2, filt2, filt2, \ 80 res4_m, res5_m, res6_m, res7_m); \ 81 VSHF_B2_SB(src0, src0, src1, src1, mask1, mask1, vec4_m, vec5_m); \ 82 VSHF_B2_SB(src2, src2, src3, src3, mask1, mask1, vec6_m, vec7_m); \ 83 DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt1, filt1, filt1, filt1, \ 84 res0_m, res1_m, res2_m, res3_m); \ 85 VSHF_B2_SB(src0, src0, src1, src1, mask3, mask3, vec4_m, vec5_m); \ 86 VSHF_B2_SB(src2, src2, src3, src3, mask3, mask3, vec6_m, vec7_m); \ 87 DPADD_SB4_SH(vec4_m, vec5_m, vec6_m, vec7_m, filt3, filt3, filt3, filt3, \ 88 res4_m, res5_m, res6_m, res7_m); \ 89 ADDS_SH4_SH(res0_m, res4_m, res1_m, res5_m, res2_m, res6_m, res3_m, \ 90 res7_m, out0, out1, out2, out3); \ 91} 92 93#define PCKEV_XORI128_AVG_ST_UB(in0, in1, dst, pdst) { \ 94 v16u8 tmp_m; \ 95 \ 96 tmp_m = PCKEV_XORI128_UB(in1, in0); \ 97 tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \ 98 ST_UB(tmp_m, (pdst)); \ 99} 100 101#define PCKEV_AVG_ST_UB(in0, in1, dst, pdst) { \ 102 v16u8 tmp_m; \ 103 \ 104 tmp_m = (v16u8)__msa_pckev_b((v16i8)in0, (v16i8)in1); \ 105 tmp_m = __msa_aver_u_b(tmp_m, (v16u8)dst); \ 106 ST_UB(tmp_m, (pdst)); \ 107} 108 109#define PCKEV_AVG_ST8x4_UB(in1, dst0, in2, dst1, in3, dst2, in4, dst3, \ 110 pdst, stride) { \ 111 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \ 112 uint8_t *pdst_m = (uint8_t *)(pdst); \ 113 \ 114 PCKEV_B2_UB(in2, in1, in4, in3, tmp0_m, tmp1_m); \ 115 PCKEV_D2_UB(dst1, dst0, dst3, dst2, tmp2_m, tmp3_m); \ 116 AVER_UB2_UB(tmp0_m, tmp2_m, tmp1_m, tmp3_m, tmp0_m, tmp1_m); \ 117 ST8x4_UB(tmp0_m, tmp1_m, pdst_m, stride); \ 118} 119#endif /* VPX_DSP_MIPS_VPX_CONVOLVE_MSA_H_ */ 120