From: Zhao Zhili <zhiliz...@tencent.com>
Test on Apple M1 with kperf:
abgr_to_uv_8_c: 19.4
abgr_to_uv_8_neon: 29.9
abgr_to_uv_128_c: 146.4
abgr_to_uv_128_neon: 85.1
abgr_to_uv_1080_c: 1162.6
abgr_to_uv_1080_neon: 819.6
abgr_to_uv_1920_c: 2063.6
abgr_to_uv_1920_neon: 1435.1
abgr_to_uv_half_8_c: 16.4
abgr_to_uv_half_8_neon: 35.6
abgr_to_uv_half_128_c: 108.6
abgr_to_uv_half_128_neon: 75.4
abgr_to_uv_half_1080_c: 883.4
abgr_to_uv_half_1080_neon: 460.6
abgr_to_uv_half_1920_c: 1553.6
abgr_to_uv_half_1920_neon: 817.6
abgr_to_y_8_c: 6.1
abgr_to_y_8_neon: 40.6
abgr_to_y_128_c: 99.9
abgr_to_y_128_neon: 67.4
abgr_to_y_1080_c: 735.9
abgr_to_y_1080_neon: 534.6
abgr_to_y_1920_c: 1279.4
abgr_to_y_1920_neon: 932.6
Disable vectorization in clang:
abgr_to_uv_8_c: 20.9
abgr_to_uv_8_neon: 44.1
abgr_to_uv_128_c: 555.6
abgr_to_uv_128_neon: 105.9
abgr_to_uv_1080_c: 4773.1
abgr_to_uv_1080_neon: 846.1
abgr_to_uv_1920_c: 8499.9
abgr_to_uv_1920_neon: 1436.6
abgr_to_uv_half_8_c: 13.9
abgr_to_uv_half_8_neon: 167.9
abgr_to_uv_half_128_c: 338.9
abgr_to_uv_half_128_neon: 60.6
abgr_to_uv_half_1080_c: 2906.9
abgr_to_uv_half_1080_neon: 466.1
abgr_to_uv_half_1920_c: 5084.4
abgr_to_uv_half_1920_neon: 832.9
abgr_to_y_8_c: 9.6
abgr_to_y_8_neon: 18.6
abgr_to_y_128_c: 372.4
abgr_to_y_128_neon: 61.4
abgr_to_y_1080_c: 3168.9
abgr_to_y_1080_neon: 513.9
abgr_to_y_1920_c: 6187.6
abgr_to_y_1920_neon: 918.9
---
libswscale/aarch64/input.S | 104 ++++++++++++++++++++++++++++-------
libswscale/aarch64/swscale.c | 17 ++++++
2 files changed, 100 insertions(+), 21 deletions(-)
diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S
index 37f1158504..b4e827dd50 100644
--- a/libswscale/aarch64/input.S
+++ b/libswscale/aarch64/input.S
@@ -34,6 +34,16 @@
uxtl2 v24.8h, v18.16b // v24: b
.endm
+.macro argb_to_yuv_load_rgb src
+ ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [\src]
+ uxtl v21.8h, v19.8b // v21: b
+ uxtl2 v24.8h, v19.16b // v24: b
+ uxtl v19.8h, v17.8b // v19: r
+ uxtl v20.8h, v18.8b // v20: g
+ uxtl2 v22.8h, v17.16b // v22: r
+ uxtl2 v23.8h, v18.16b // v23: g
+.endm
+
.macro rgb_to_yuv_product r, g, b, dst1, dst2, dst, coef0, coef1, coef2,
right_shift
mov \dst1\().16b, v6.16b // dst1 =
const_offset
mov \dst2\().16b, v6.16b // dst2 =
const_offset
@@ -47,7 +57,7 @@
sqshrn2 \dst\().8h, \dst2\().4s, \right_shift // dst_higher_half =
dst2 >> right_shift
.endm
-.macro rgbToY bgr, element=3
+.macro rgbToY bgr, element=3, alpha_first=0
cmp w4, #0 // check width > 0
.if \bgr
ldr w12, [x5] // w12: ry
@@ -71,7 +81,11 @@
dup v2.8h, w12
b.lt 2f
1:
+ .if \alpha_first
+ argb_to_yuv_load_rgb x1
+ .else
rgb_to_yuv_load_rgb x1, \element
+ .endif
rgb_to_yuv_product v19, v20, v21, v25, v26, v16, v0, v1, v2, #9
rgb_to_yuv_product v22, v23, v24, v27, v28, v17, v0, v1, v2, #9
sub w4, w4, #16 // width -= 16
@@ -85,9 +99,15 @@
b.ge 1b
cbz x4, 3f
2:
+ .if \alpha_first
+ ldrb w13, [x1, #1] // w13: r
+ ldrb w14, [x1, #2] // w14: g
+ ldrb w15, [x1, #3] // w15: b
+ .else
ldrb w13, [x1] // w13: r
ldrb w14, [x1, #1] // w14: g
ldrb w15, [x1, #2] // w15: b
+ .endif
smaddl x13, w13, w10, x9 // x13 = ry * r + const_offset
smaddl x13, w14, w11, x13 // x13 += gy * g
@@ -101,6 +121,14 @@
ret
.endm
+function ff_argb32ToY_neon, export=1
+ rgbToY bgr=0, element=4, alpha_first=1
+endfunc
+
+function ff_abgr32ToY_neon, export=1
+ rgbToY bgr=1, element=4, alpha_first=1
+endfunc
+
function ff_rgb24ToY_neon, export=1
rgbToY bgr=0
endfunc
@@ -146,7 +174,21 @@ endfunc
dup v6.4s, w9
.endm
-.macro rgbToUV_half bgr, element=3
+.macro rgb_load_add_half off_r1, off_r2, off_g1, off_g2, off_b1, off_b2
+ ldrb w2, [x3, \off_r1] // w2: r1
+ ldrb w4, [x3, \off_r2] // w4: r2
