aarch64 A55: chrRangeFromJpeg16_1920_c: 32684.2 chrRangeFromJpeg16_1920_neon: 8431.2 (3.88x) chrRangeToJpeg16_1920_c: 24996.8 chrRangeToJpeg16_1920_neon: 9395.0 (2.66x) lumRangeFromJpeg16_1920_c: 17305.2 lumRangeFromJpeg16_1920_neon: 4586.5 (3.77x) lumRangeToJpeg16_1920_c: 21144.8 lumRangeToJpeg16_1920_neon: 5069.8 (4.17x) aarch64 A76: chrRangeFromJpeg16_1920_c: 11523.8 chrRangeFromJpeg16_1920_neon: 3367.5 (3.42x) chrRangeToJpeg16_1920_c: 11655.2 chrRangeToJpeg16_1920_neon: 4087.2 (2.85x) lumRangeFromJpeg16_1920_c: 5762.0 lumRangeFromJpeg16_1920_neon: 1815.8 (3.17x) lumRangeToJpeg16_1920_c: 5946.2 lumRangeToJpeg16_1920_neon: 2148.2 (2.77x)
Unified · Split
| @@ -20,12 +20,42 @@ · 19 unmodified lines | ||
| 20 | 20 | |
| 21 | 21 | #include "libavutil/aarch64/asm.S" |
| 22 | 22 | |
| 23 | −.macro lumConvertRange fromto | |
| 24 | −function ff_lumRange\fromto\()Jpeg_neon, export=1 | |
| 23 | +.macro lumConvertRange fromto, bit_depth | |
| 24 | +function ff_lumRange\fromto\()Jpeg\bit_depth\()_neon, export=1 | |
| 25 | 25 | // x0 int16_t *dst |
| 26 | 26 | // w1 int width |
| 27 | 27 | // w2 uint32_t coeff |
| 28 | 28 | // x3 int64_t offset |
| 29 | +.if \bit_depth == 16 | |
| 30 | +.ifc \fromto, To | |
| 31 | + movi v25.4s, #1 | |
| 32 | + movi v24.4s, #1<<3, lsl #16 | |
| 33 | + sub v24.4s, v24.4s, v25.4s | |
| 34 | +.endif | |
| 35 | + dup v25.4s, w2 | |
| 36 | + dup v26.2d, x3 | |
| 37 | +1: | |
| 38 | + ld1 {v0.4s, v1.4s}, [x0] | |
| 39 | + mov v16.16b, v26.16b | |
| 40 | + mov v17.16b, v26.16b | |
| 41 | + mov v18.16b, v26.16b | |
| 42 | + mov v19.16b, v26.16b | |
| 43 | + smlal v16.2d, v0.2s, v25.2s | |
| 44 | + smlal2 v17.2d, v0.4s, v25.4s | |
| 45 | + smlal v18.2d, v1.2s, v25.2s | |
| 46 | + smlal2 v19.2d, v1.4s, v25.4s | |
| 47 | + shrn v0.2s, v16.2d, 18 | |
| 48 | + shrn2 v0.4s, v17.2d, 18 | |
| 49 | + shrn v1.2s, v18.2d, 18 | |
| 50 | + shrn2 v1.4s, v19.2d, 18 | |
| 51 | + subs w1, w1, #8 | |
| 52 | +.ifc \fromto, To | |
| 53 | + smin v0.4s, v0.4s, v24.4s | |
| 54 | + smin v1.4s, v1.4s, v24.4s | |
| 55 | +.endif | |
| 56 | + st1 {v0.4s, v1.4s}, [x0], #32 | |
| 57 | + b.gt 1b | |
| 58 | +.else | |
| 29 | 59 | dup v25.4s, w2 |
| 30 | 60 | dup v26.4s, w3 |
| 31 | 61 | 1: |
| @@ -46,17 +76,64 @@ function ff_lumRange\fromto\()Jpeg_neon, export=1 · 14 unmodified lines | ||
| 46 | 76 | subs w1, w1, #8 |
| 47 | 77 | st1 {v0.8h}, [x0], #16 |
| 48 | 78 | b.gt 1b |
| 79 | +.endif | |
| 49 | 80 | ret |
| 50 | 81 | endfunc |
| 51 | 82 | .endm |
| 52 | 83 | |
| 53 | −.macro chrConvertRange fromto | |
| 54 | −function ff_chrRange\fromto\()Jpeg_neon, export=1 | |
| 84 | +.macro chrConvertRange fromto, bit_depth | |
| 85 | +function ff_chrRange\fromto\()Jpeg\bit_depth\()_neon, export=1 | |
| 55 | 86 | // x0 int16_t *dstU |
| 56 | 87 | // x1 int16_t *dstV |
| 57 | 88 | // w2 int width |
| 58 | 89 | // w3 uint32_t coeff |
| 59 | 90 | // x4 int64_t offset |
| 91 | +.if \bit_depth == 16 | |
| 92 | +.ifc \fromto, To | |
| 93 | + movi v25.4s, #1 | |
| 94 | + movi v24.4s, #1<<3, lsl #16 | |
| 95 | + sub v24.4s, v24.4s, v25.4s | |
| 96 | +.endif | |
| 97 | + dup v25.4s, w3 | |
| 98 | + dup v26.2d, x4 | |
| 99 | +1: | |
| 100 | + ld1 {v0.4s, v1.4s}, [x0] | |
| 101 | + ld1 {v2.4s, v3.4s}, [x1] | |
| 102 | + mov v16.16b, v26.16b | |
| 103 | + mov v17.16b, v26.16b | |
| 104 | + mov v18.16b, v26.16b | |
| 105 | + mov v19.16b, v26.16b | |
| 106 | + mov v20.16b, v26.16b | |
| 107 | + mov v21.16b, v26.16b | |
| 108 | + mov v22.16b, v26.16b | |
| 109 | + mov v23.16b, v26.16b | |
| 110 | + smlal v16.2d, v0.2s, v25.2s | |
| 111 | + smlal2 v17.2d, v0.4s, v25.4s | |
| 112 | + smlal v18.2d, v1.2s, v25.2s | |
| 113 | + smlal2 v19.2d, v1.4s, v25.4s | |
| 114 | + smlal v20.2d, v2.2s, v25.2s | |
| 115 | + smlal2 v21.2d, v2.4s, v25.4s | |
| 116 | + smlal v22.2d, v3.2s, v25.2s | |
| 117 | + smlal2 v23.2d, v3.4s, v25.4s | |
| 118 | + shrn v0.2s, v16.2d, 18 | |
| 119 | + shrn2 v0.4s, v17.2d, 18 | |
| 120 | + shrn v1.2s, v18.2d, 18 | |
| 121 | + shrn2 v1.4s, v19.2d, 18 | |
| 122 | + shrn v2.2s, v20.2d, 18 | |
| 123 | + shrn2 v2.4s, v21.2d, 18 | |
| 124 | + shrn v3.2s, v22.2d, 18 | |
| 125 | + shrn2 v3.4s, v23.2d, 18 | |
| 126 | + subs w2, w2, #8 | |
| 127 | +.ifc \fromto, To | |
| 128 | + smin v0.4s, v0.4s, v24.4s | |
| 129 | + smin v1.4s, v1.4s, v24.4s | |
| 130 | + smin v2.4s, v2.4s, v24.4s | |
| 131 | + smin v3.4s, v3.4s, v24.4s | |
| 132 | +.endif | |
| 133 | + st1 {v0.4s, v1.4s}, [x0], #32 | |
| 134 | + st1 {v2.4s, v3.4s}, [x1], #32 | |
| 135 | + b.gt 1b | |
| 136 | +.else | |
| 60 | 137 | dup v25.4s, w3 |
| 61 | 138 | dup v26.4s, w4 |
| 62 | 139 | 1: |
| @@ -89,11 +166,16 @@ function ff_chrRange\fromto\()Jpeg_neon, export=1 · 26 unmodified lines | ||
| 89 | 166 | st1 {v0.8h}, [x0], #16 |
| 90 | 167 | st1 {v1.8h}, [x1], #16 |
| 91 | 168 | b.gt 1b |
| 169 | +.endif | |
| 92 | 170 | ret |
| 93 | 171 | endfunc |
| 94 | 172 | .endm |
| 95 | 173 | |
| 96 | −lumConvertRange To | |
| 97 | −chrConvertRange To | |
| 98 | −lumConvertRange From | |
| 99 | −chrConvertRange From | |
| 174 | +lumConvertRange To, 8 | |
| 175 | +lumConvertRange To, 16 | |
| 176 | +chrConvertRange To, 8 | |
| 177 | +chrConvertRange To, 16 | |
| 178 | +lumConvertRange From, 8 | |
| 179 | +lumConvertRange From, 16 | |
| 180 | +chrConvertRange From, 8 | |
| 181 | +chrConvertRange From, 16 | |
| @@ -218,14 +218,22 @@ NEON_INPUT(bgra32); · 217 unmodified lines | ||
| 218 | 218 | NEON_INPUT(rgb24); |
| 219 | 219 | NEON_INPUT(rgba32); |
| 220 | 220 | |
| 221 | −void ff_lumRangeFromJpeg_neon(int16_t *dst, int width, | |
| 221 | +void ff_lumRangeFromJpeg8_neon(int16_t *dst, int width, | |
| 222 | + uint32_t coeff, int64_t offset); | |
| 223 | +void ff_chrRangeFromJpeg8_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 224 | + uint32_t coeff, int64_t offset); | |
| 225 | +void ff_lumRangeToJpeg8_neon(int16_t *dst, int width, | |
| 226 | + uint32_t coeff, int64_t offset); | |
| 227 | +void ff_chrRangeToJpeg8_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 228 | + uint32_t coeff, int64_t offset); | |
| 229 | +void ff_lumRangeFromJpeg16_neon(int16_t *dst, int width, | |
| 230 | + uint32_t coeff, int64_t offset); | |
| 231 | +void ff_chrRangeFromJpeg16_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 232 | + uint32_t coeff, int64_t offset); | |
| 233 | +void ff_lumRangeToJpeg16_neon(int16_t *dst, int width, | |
| 222 | 234 | uint32_t coeff, int64_t offset); |
| 223 | −void ff_chrRangeFromJpeg_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 235 | +void ff_chrRangeToJpeg16_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 224 | 236 | uint32_t coeff, int64_t offset); |
| 225 | −void ff_lumRangeToJpeg_neon(int16_t *dst, int width, | |
| 226 | − uint32_t coeff, int64_t offset); | |
| 227 | −void ff_chrRangeToJpeg_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 228 | − uint32_t coeff, int64_t offset); | |
| 229 | 237 | |
| 230 | 238 | av_cold void ff_sws_init_range_convert_aarch64(SwsInternal *c) |
| 231 | 239 | { |
| @@ -234,11 +242,19 @@ av_cold void ff_sws_init_range_convert_aarch64(SwsInternal *c) · 2 unmodified lines | ||
| 234 | 242 | if (have_neon(cpu_flags)) { |
| 235 | 243 | if (c->dstBpc <= 14) { |
| 236 | 244 | if (c->opts.src_range) { |
| 237 | − c->lumConvertRange = ff_lumRangeFromJpeg_neon; | |
| 238 | − c->chrConvertRange = ff_chrRangeFromJpeg_neon; | |
| 245 | + c->lumConvertRange = ff_lumRangeFromJpeg8_neon; | |
| 246 | + c->chrConvertRange = ff_chrRangeFromJpeg8_neon; | |
| 239 | 247 | } else { |
| 240 | − c->lumConvertRange = ff_lumRangeToJpeg_neon; | |
| 241 | − c->chrConvertRange = ff_chrRangeToJpeg_neon; | |
| 248 | + c->lumConvertRange = ff_lumRangeToJpeg8_neon; | |
| 249 | + c->chrConvertRange = ff_chrRangeToJpeg8_neon; | |
| 250 | + } | |
| 251 | + } else { | |
| 252 | + if (c->opts.src_range) { | |
| 253 | + c->lumConvertRange = ff_lumRangeFromJpeg16_neon; | |
| 254 | + c->chrConvertRange = ff_chrRangeFromJpeg16_neon; | |
| 255 | + } else { | |
| 256 | + c->lumConvertRange = ff_lumRangeToJpeg16_neon; | |
| 257 | + c->chrConvertRange = ff_chrRangeToJpeg16_neon; | |
| 242 | 258 | } |
| 243 | 259 | } |
| 244 | 260 | } |