aarch64 A55: chrRangeFromJpeg8_1920_c: 28835.2 (1.00x) chrRangeFromJpeg8_1920_neon: 5313.9 (5.43x) 5308.4 (5.43x) chrRangeToJpeg8_1920_c: 23074.7 (1.00x) chrRangeToJpeg8_1920_neon: 5551.3 (4.16x) 5549.2 (4.16x) lumRangeFromJpeg8_1920_c: 15389.7 (1.00x) lumRangeFromJpeg8_1920_neon: 3152.3 (4.88x) 3147.7 (4.89x) lumRangeToJpeg8_1920_c: 19227.8 (1.00x) lumRangeToJpeg8_1920_neon: 3628.7 (5.30x) 3630.2 (5.30x) aarch64 A76: chrRangeFromJpeg8_1920_c: 6324.4 (1.00x) chrRangeFromJpeg8_1920_neon: 2344.5 (2.70x) 2304.2 (2.74x) chrRangeToJpeg8_1920_c: 9656.0 (1.00x) chrRangeToJpeg8_1920_neon: 2824.2 (3.42x) 2794.2 (3.46x) lumRangeFromJpeg8_1920_c: 4422.0 (1.00x) lumRangeFromJpeg8_1920_neon: 1104.5 (4.00x) 1106.2 (4.00x) lumRangeToJpeg8_1920_c: 5949.1 (1.00x) lumRangeToJpeg8_1920_neon: 1329.8 (4.47x) 1328.2 (4.48x)
Unified · Split
| @@ -20,12 +20,13 @@ · 19 unmodified lines | ||
| 20 | 20 | |
| 21 | 21 | #include "libavutil/aarch64/asm.S" |
| 22 | 22 | |
| 23 | −.macro lumConvertRange name, fromto, mult, offset, shift | |
| 24 | −function ff_\name, export=1 | |
| 25 | − mov w3, #\mult | |
| 26 | − dup v25.4s, w3 | |
| 27 | − movz w3, #(\offset & 0xffff) | |
| 28 | − movk w3, #((\offset >> 16) & 0xffff), lsl #16 | |
| 23 | +.macro lumConvertRange fromto | |
| 24 | +function ff_lumRange\fromto\()Jpeg_neon, export=1 | |
| 25 | +// x0 int16_t *dst | |
| 26 | +// w1 int width | |
| 27 | +// w2 uint32_t coeff | |
| 28 | +// x3 int64_t offset | |
| 29 | + dup v25.4s, w2 | |
| 29 | 30 | dup v26.4s, w3 |
| 30 | 31 | 1: |
| 31 | 32 | ld1 {v0.8h}, [x0] |
| @@ -36,11 +37,11 @@ function ff_\name, export=1 · 4 unmodified lines | ||
| 36 | 37 | mla v16.4s, v20.4s, v25.4s |
| 37 | 38 | mla v18.4s, v22.4s, v25.4s |
| 38 | 39 | .ifc \fromto, To |
| 39 | − sqshrn v0.4h, v16.4s, #\shift | |
| 40 | − sqshrn2 v0.8h, v18.4s, #\shift | |
| 40 | + sqshrn v0.4h, v16.4s, 14 | |
| 41 | + sqshrn2 v0.8h, v18.4s, 14 | |
| 41 | 42 | .else |
| 42 | − shrn v0.4h, v16.4s, #\shift | |
| 43 | − shrn2 v0.8h, v18.4s, #\shift | |
| 43 | + shrn v0.4h, v16.4s, 14 | |
| 44 | + shrn2 v0.8h, v18.4s, 14 | |
| 44 | 45 | .endif |
| 45 | 46 | subs w1, w1, #8 |
| 46 | 47 | st1 {v0.8h}, [x0], #16 |
| @@ -49,13 +50,15 @@ function ff_\name, export=1 · 2 unmodified lines | ||
| 49 | 50 | endfunc |
| 50 | 51 | .endm |
| 51 | 52 | |
| 52 | −.macro chrConvertRange name, fromto, mult, offset, shift | |
| 53 | −function ff_\name, export=1 | |
| 54 | − mov w3, #\mult | |
| 53 | +.macro chrConvertRange fromto | |
| 54 | +function ff_chrRange\fromto\()Jpeg_neon, export=1 | |
| 55 | +// x0 int16_t *dstU | |
| 56 | +// x1 int16_t *dstV | |
| 57 | +// w2 int width | |
| 58 | +// w3 uint32_t coeff | |
| 59 | +// x4 int64_t offset | |
| 55 | 60 | dup v25.4s, w3 |
| 56 | − movz w3, #(\offset & 0xffff) | |
| 57 | − movk w3, #((\offset >> 16) & 0xffff), lsl #16 | |
| 58 | − dup v26.4s, w3 | |
| 61 | + dup v26.4s, w4 | |
| 59 | 62 | 1: |
| 60 | 63 | ld1 {v0.8h}, [x0] |
| 61 | 64 | ld1 {v1.8h}, [x1] |
| @@ -72,15 +75,15 @@ function ff_\name, export=1 · 10 unmodified lines | ||
| 72 | 75 | mla v18.4s, v22.4s, v25.4s |
| 73 | 76 | mla v19.4s, v23.4s, v25.4s |
| 74 | 77 | .ifc \fromto, To |
| 75 | − sqshrn v0.4h, v16.4s, #\shift | |
| 76 | − sqshrn v1.4h, v17.4s, #\shift | |
| 77 | − sqshrn2 v0.8h, v18.4s, #\shift | |
| 78 | − sqshrn2 v1.8h, v19.4s, #\shift | |
| 78 | + sqshrn v0.4h, v16.4s, 14 | |
| 79 | + sqshrn v1.4h, v17.4s, 14 | |
| 80 | + sqshrn2 v0.8h, v18.4s, 14 | |
| 81 | + sqshrn2 v1.8h, v19.4s, 14 | |
| 79 | 82 | .else |
| 80 | − shrn v0.4h, v16.4s, #\shift | |
| 81 | − shrn v1.4h, v17.4s, #\shift | |
| 82 | − shrn2 v0.8h, v18.4s, #\shift | |
| 83 | − shrn2 v1.8h, v19.4s, #\shift | |
| 83 | + shrn v0.4h, v16.4s, 14 | |
| 84 | + shrn v1.4h, v17.4s, 14 | |
| 85 | + shrn2 v0.8h, v18.4s, 14 | |
| 86 | + shrn2 v1.8h, v19.4s, 14 | |
| 84 | 87 | .endif |
| 85 | 88 | subs w2, w2, #8 |
| 86 | 89 | st1 {v0.8h}, [x0], #16 |
| @@ -90,7 +93,7 @@ function ff_\name, export=1 · 3 unmodified lines | ||
| 90 | 93 | endfunc |
| 91 | 94 | .endm |
| 92 | 95 | |
| 93 | −lumConvertRange lumRangeToJpeg_neon, To, 19077, -39057361, 14 | |
| 94 | −chrConvertRange chrRangeToJpeg_neon, To, 4663, -9289992, 12 | |
| 95 | −lumConvertRange lumRangeFromJpeg_neon, From, 14071, 33561947, 14 | |
| 96 | −chrConvertRange chrRangeFromJpeg_neon, From, 1799, 4081085, 11 | |
| 96 | +lumConvertRange To | |
| 97 | +chrConvertRange To | |
| 98 | +lumConvertRange From | |
| 99 | +chrConvertRange From | |
| @@ -218,17 +218,17 @@ 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); | |
| 222 | −void ff_chrRangeFromJpeg_neon(int16_t *dstU, int16_t *dstV, int width); | |
| 223 | −void ff_lumRangeToJpeg_neon(int16_t *dst, int width); | |
| 224 | −void ff_chrRangeToJpeg_neon(int16_t *dstU, int16_t *dstV, int width); | |
| 221 | +void ff_lumRangeFromJpeg_neon(int16_t *dst, int width, | |
| 222 | + uint32_t coeff, int64_t offset); | |
| 223 | +void ff_chrRangeFromJpeg_neon(int16_t *dstU, int16_t *dstV, int width, | |
| 224 | + 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); | |
| 225 | 229 | |
| 226 | 230 | av_cold void ff_sws_init_range_convert_aarch64(SwsInternal *c) |
| 227 | 231 | { |
| 228 | − /* This code is currently disabled because of changes in the base | |
| 229 | − * implementation of these functions. This code should be enabled | |
| 230 | − * again once those changes are ported to this architecture. */ | |
| 231 | −#if 0 | |
| 232 | 232 | int cpu_flags = av_get_cpu_flags(); |
| 233 | 233 | |
| 234 | 234 | if (have_neon(cpu_flags)) { |
| @@ -242,7 +242,6 @@ av_cold void ff_sws_init_range_convert_aarch64(SwsInternal *c) · 7 unmodified lines | ||
| 242 | 242 | } |
| 243 | 243 | } |
| 244 | 244 | } |
| 245 | −#endif | |
| 246 | 245 | } |
| 247 | 246 | |
| 248 | 247 | av_cold void ff_sws_init_swscale_aarch64(SwsInternal *c) |