Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(316)

Side by Side Diff: include/libyuv/macros_msa.h

Issue 2553403002: Add MSA optimized TransposeWx8_MSA and TransposeUVWx8_MSA functions (Closed)
Patch Set: Created 4 years ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
1 /* 1 /*
2 * Copyright 2016 The LibYuv Project Authors. All rights reserved. 2 * Copyright 2016 The LibYuv Project Authors. All rights reserved.
3 * 3 *
4 * Use of this source code is governed by a BSD-style license 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 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 6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may 7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree. 8 * be found in the AUTHORS file in the root of the source tree.
9 */ 9 */
10 10
(...skipping 32 matching lines...) Expand 10 before | Expand all | Expand 10 after
43 uint32 val0_m, val1_m; \ 43 uint32 val0_m, val1_m; \
44 uint64 val_m = 0; \ 44 uint64 val_m = 0; \
45 val0_m = LW(psrc_ld_m); \ 45 val0_m = LW(psrc_ld_m); \
46 val1_m = LW(psrc_ld_m + 4); \ 46 val1_m = LW(psrc_ld_m + 4); \
47 val_m = (uint64)(val1_m); /* NOLINT */ \ 47 val_m = (uint64)(val1_m); /* NOLINT */ \
48 val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \ 48 val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
49 val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \ 49 val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \
50 val_m; \ 50 val_m; \
51 }) 51 })
52 #endif // (__mips == 64) 52 #endif // (__mips == 64)
53
54 #define SW(val, pdst) \
55 ({ \
56 uint8_t* pdst_sw_m = (uint8_t*)(pdst); \
57 uint32_t val_m = (val); \
58 \
fbarchard1 2016/12/12 19:59:18 remove blank lines from macros.
manojkumar.bhosale 2016/12/13 08:52:22 Done.
59 asm volatile("sw %[val_m], %[pdst_sw_m] \n\t" \
60 \
61 : [pdst_sw_m] "=m"(*pdst_sw_m) \
62 : [val_m] "r"(val_m)); \
63 })
64
65 #if (__mips == 64)
66 #define SD(val, pdst) \
67 ({ \
68 uint8_t* pdst_sd_m = (uint8_t*)(pdst); \
69 uint64_t val_m = (val); \
70 \
71 asm volatile("sd %[val_m], %[pdst_sd_m] \n\t" \
72 \
73 : [pdst_sd_m] "=m"(*pdst_sd_m) \
74 : [val_m] "r"(val_m)); \
75 })
76 #else // !(__mips == 64)
77 #define SD(val, pdst) \
78 ({ \
79 uint8_t* pdst_sd_m = (uint8_t*)(pdst); \
80 uint32_t val0_m, val1_m; \
81 \
82 val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
83 val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
84 \
85 SW(val0_m, pdst_sd_m); \
86 SW(val1_m, pdst_sd_m + 4); \
87 })
88 #endif // !(__mips == 64)
53 #else // !(__mips_isa_rev >= 6) 89 #else // !(__mips_isa_rev >= 6)
54 #define LW(psrc) \ 90 #define LW(psrc) \
55 ({ \ 91 ({ \
56 uint8* psrc_lw_m = (uint8*)(psrc); /* NOLINT */ \ 92 uint8* psrc_lw_m = (uint8*)(psrc); /* NOLINT */ \
57 uint32 val_m; \ 93 uint32 val_m; \
58 asm volatile("ulw %[val_m], %[psrc_lw_m] \n\t" \ 94 asm volatile("ulw %[val_m], %[psrc_lw_m] \n\t" \
59 : [val_m] "=r"(val_m) \ 95 : [val_m] "=r"(val_m) \
60 : [psrc_lw_m] "m"(*psrc_lw_m)); \ 96 : [psrc_lw_m] "m"(*psrc_lw_m)); \
61 val_m; \ 97 val_m; \
62 }) 98 })
(...skipping 15 matching lines...) Expand all
78 uint32 val0_m, val1_m; \ 114 uint32 val0_m, val1_m; \
79 uint64 val_m = 0; \ 115 uint64 val_m = 0; \
80 val0_m = LW(psrc_ld_m); \ 116 val0_m = LW(psrc_ld_m); \
81 val1_m = LW(psrc_ld_m + 4); \ 117 val1_m = LW(psrc_ld_m + 4); \
82 val_m = (uint64)(val1_m); /* NOLINT */ \ 118 val_m = (uint64)(val1_m); /* NOLINT */ \
83 val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \ 119 val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
84 val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \ 120 val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \
85 val_m; \ 121 val_m; \
86 }) 122 })
87 #endif // (__mips == 64) 123 #endif // (__mips == 64)
124
125 #define SW(val, pdst) \
126 ({ \
127 uint8_t* pdst_sw_m = (uint8_t*)(pdst); \
128 uint32_t val_m = (val); \
129 \
130 asm volatile("usw %[val_m], %[pdst_sw_m] \n\t" \
fbarchard1 2016/12/12 19:59:18 remove \t ? I've never found it necessary.
manojkumar.bhosale 2016/12/13 08:52:22 Done.
131 \
132 : [pdst_sw_m] "=m"(*pdst_sw_m) \
133 : [val_m] "r"(val_m)); \
134 })
135
136 #define SD(val, pdst) \
137 ({ \
138 uint8_t* pdst_sd_m = (uint8_t*)(pdst); \
139 uint32_t val0_m, val1_m; \
140 \
141 val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
142 val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
143 \
144 SW(val0_m, pdst_sd_m); \
145 SW(val1_m, pdst_sd_m + 4); \
146 })
88 #endif // (__mips_isa_rev >= 6) 147 #endif // (__mips_isa_rev >= 6)
89 148
90 // TODO(fbarchard): Consider removing __VAR_ARGS versions. 149 // TODO(fbarchard): Consider removing __VAR_ARGS versions.
91 #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */ 150 #define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
92 #define LD_UB(...) LD_B(v16u8, __VA_ARGS__) 151 #define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
93 152
94 #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */ 153 #define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
95 #define ST_UB(...) ST_B(v16u8, __VA_ARGS__) 154 #define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
96 155
97 /* Description : Load two vectors with 16 'byte' sized elements 156 /* Description : Load two vectors with 16 'byte' sized elements
(...skipping 62 matching lines...) Expand 10 before | Expand all | Expand 10 after
160 #define ILVRL_B2(RTYPE, in0, in1, out0, out1) \ 219 #define ILVRL_B2(RTYPE, in0, in1, out0, out1) \
161 { \ 220 { \
162 out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \ 221 out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
163 out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \ 222 out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
164 } 223 }
165 #define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__) 224 #define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
166 225
167 #endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */ 226 #endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */
168 227
169 #endif // INCLUDE_LIBYUV_MACROS_MSA_H_ 228 #endif // INCLUDE_LIBYUV_MACROS_MSA_H_
OLDNEW
« no previous file with comments | « CMakeLists.txt ('k') | include/libyuv/rotate_row.h » ('j') | source/rotate.cc » ('J')

Powered by Google App Engine
This is Rietveld 408576698