loopfilter_4_msa.c 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#include "vpx_dsp/mips/loopfilter_msa.h" 12 13void vpx_lpf_horizontal_4_msa(uint8_t *src, int32_t pitch, 14 const uint8_t *b_limit_ptr, 15 const uint8_t *limit_ptr, 16 const uint8_t *thresh_ptr, 17 int32_t count) { 18 uint64_t p1_d, p0_d, q0_d, q1_d; 19 v16u8 mask, hev, flat, thresh, b_limit, limit; 20 v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p1_out, p0_out, q0_out, q1_out; 21 22 (void)count; 23 24 /* load vector elements */ 25 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3); 26 27 thresh = (v16u8)__msa_fill_b(*thresh_ptr); 28 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr); 29 limit = (v16u8)__msa_fill_b(*limit_ptr); 30 31 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, 32 hev, mask, flat); 33 VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out); 34 35 p1_d = __msa_copy_u_d((v2i64)p1_out, 0); 36 p0_d = __msa_copy_u_d((v2i64)p0_out, 0); 37 q0_d = __msa_copy_u_d((v2i64)q0_out, 0); 38 q1_d = __msa_copy_u_d((v2i64)q1_out, 0); 39 SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch); 40} 41 42void vpx_lpf_horizontal_4_dual_msa(uint8_t *src, int32_t pitch, 43 const uint8_t *b_limit0_ptr, 44 const uint8_t *limit0_ptr, 45 const uint8_t *thresh0_ptr, 46 const uint8_t *b_limit1_ptr, 47 const uint8_t *limit1_ptr, 48 const uint8_t *thresh1_ptr) { 49 v16u8 mask, hev, flat, thresh0, b_limit0, limit0, thresh1, b_limit1, limit1; 50 v16u8 p3, p2, p1, p0, q3, q2, q1, q0; 51 52 /* load vector elements */ 53 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3); 54 55 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr); 56 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr); 57 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0); 58 59 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr); 60 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr); 61 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0); 62 63 limit0 = (v16u8)__msa_fill_b(*limit0_ptr); 64 limit1 = (v16u8)__msa_fill_b(*limit1_ptr); 65 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0); 66 67 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, 68 hev, mask, flat); 69 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1); 70 71 ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch); 72} 73 74void vpx_lpf_vertical_4_msa(uint8_t *src, int32_t pitch, 75 const uint8_t *b_limit_ptr, 76 const uint8_t *limit_ptr, 77 const uint8_t *thresh_ptr, 78 int32_t count) { 79 v16u8 mask, hev, flat, limit, thresh, b_limit; 80 v16u8 p3, p2, p1, p0, q3, q2, q1, q0; 81 v8i16 vec0, vec1, vec2, vec3; 82 83 (void)count; 84 85 LD_UB8((src - 4), pitch, p3, p2, p1, p0, q0, q1, q2, q3); 86 87 thresh = (v16u8)__msa_fill_b(*thresh_ptr); 88 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr); 89 limit = (v16u8)__msa_fill_b(*limit_ptr); 90 91 TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3, 92 p3, p2, p1, p0, q0, q1, q2, q3); 93 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, 94 hev, mask, flat); 95 VP9_LPF_FILTER4_8W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1); 96 ILVR_B2_SH(p0, p1, q1, q0, vec0, vec1); 97 ILVRL_H2_SH(vec1, vec0, vec2, vec3); 98 99 src -= 2; 100 ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch); 101 src += 4 * pitch; 102 ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch); 103} 104 105void vpx_lpf_vertical_4_dual_msa(uint8_t *src, int32_t pitch, 106 const uint8_t *b_limit0_ptr, 107 const uint8_t *limit0_ptr, 108 const uint8_t *thresh0_ptr, 109 const uint8_t *b_limit1_ptr, 110 const uint8_t *limit1_ptr, 111 const uint8_t *thresh1_ptr) { 112 v16u8 mask, hev, flat; 113 v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1; 114 v16u8 p3, p2, p1, p0, q3, q2, q1, q0; 115 v16u8 row0, row1, row2, row3, row4, row5, row6, row7; 116 v16u8 row8, row9, row10, row11, row12, row13, row14, row15; 117 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5; 118 119 LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7); 120 LD_UB8(src - 4 + (8 * pitch), pitch, 121 row8, row9, row10, row11, row12, row13, row14, row15); 122 123 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, 124 row8, row9, row10, row11, row12, row13, row14, row15, 125 p3, p2, p1, p0, q0, q1, q2, q3); 126 127 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr); 128 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr); 129 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0); 130 131 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr); 132 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr); 133 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0); 134 135 limit0 = (v16u8)__msa_fill_b(*limit0_ptr); 136 limit1 = (v16u8)__msa_fill_b(*limit1_ptr); 137 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0); 138 139 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, 140 hev, mask, flat); 141 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1); 142 ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1); 143 ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3); 144 ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1); 145 ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5); 146 147 src -= 2; 148 149 ST4x8_UB(tmp2, tmp3, src, pitch); 150 src += (8 * pitch); 151 ST4x8_UB(tmp4, tmp5, src, pitch); 152} 153