@@ -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
ldp w12, w11, [x5] // w12: ry, w11: gy
@@ -68,7 +78,11 @@
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
@@ -78,9 +92,15 @@
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
@@ -94,6 +114,14 @@
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
@@ -136,7 +164,21 @@ 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
+ add w2, w2, w4 // w2 = r1 + r2
+
+ ldrb w4, [x3, #\off_g1] // w4: g1
+ ldrb w7, [x3, #\off_g2] // w7: g2
+ add w4, w4, w7 // w4 = g1 + g2
+
+ ldrb w7, [x3, #\off_b1] // w7: b1
+ ldrb w8, [x3, #\off_b2] // w8: b2
+ add w7, w7, w8 // w7 = b1 + b2
+.endm
+
+.macro rgbToUV_half bgr, element=3, alpha_first=0
cmp w5, #0 // check width > 0
b.le 3f
@@ -149,9 +191,15 @@ endfunc
.else
ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [x3]
.endif
+ .if \alpha_first
+ uaddlp v21.8h, v19.16b
+ uaddlp v20.8h, v18.16b
+ uaddlp v19.8h, v17.16b
+ .else
uaddlp v19.8h, v16.16b // v19: r
uaddlp v20.8h, v17.16b // v20: g
uaddlp v21.8h, v18.16b // v21: b
+ .endif
rgb_to_yuv_product v19, v20, v21, v22, v23, v16, v0, v1, v2, #10
rgb_to_yuv_product v19, v20, v21, v24, v25, v17, v3, v4, v5, #10
@@ -163,27 +211,15 @@ endfunc
b.ge 1b
cbz w5, 3f
2:
- ldrb w2, [x3] // w2: r1
- ldrb w4, [x3, #\element] // w4: r2
- add w2, w2, w4 // w2 = r1 + r2
-
+.if \alpha_first
+ rgb_load_add_half 1, 5, 2, 6, 3, 7
+.else
.if \element == 3
- ldrb w4, [x3, #1] // w4: g1
- ldrb w7, [x3, #4] // w7: g2
- add w4, w4, w7 // w4 = g1 + g2
-
- ldrb w7, [x3, #2] // w7: b1
- ldrb w8, [x3, #5] // w8: b2
- add w7, w7, w8 // w7 = b1 + b2
+ rgb_load_add_half 0, 3, 1, 4, 2, 5
.else
- ldrb w4, [x3, #1] // w4: g1
- ldrb w7, [x3, #5] // w7: g2
- add w4, w4, w7 // w4 = g1 + g2
-
- ldrb w7, [x3, #2] // w7: b1
- ldrb w8, [x3, #6] // w8: b2
- add w7, w7, w8 // w7 = b1 + b2
+ rgb_load_add_half 0, 4, 1, 5, 2, 6
.endif
+.endif
smaddl x8, w2, w10, x9 // dst_u = ru * r + const_offset
smaddl x8, w4, w11, x8 // dst_u += gu * g
@@ -204,6 +240,14 @@ endfunc
ret
.endm
+function ff_argb32ToUV_half_neon, export=1
+ rgbToUV_half bgr=0, element=4, alpha_first=1
+endfunc
+
+function ff_abgr32ToUV_half_neon, export=1
+ rgbToUV_half bgr=1, element=4, alpha_first=1
+endfunc
+
function ff_rgb24ToUV_half_neon, export=1
rgbToUV_half bgr=0
endfunc
@@ -220,7 +264,7 @@ function ff_bgra32ToUV_half_neon, export=1
rgbToUV_half bgr=1, element=4
endfunc
-.macro rgbToUV bgr, element=3
+.macro rgbToUV bgr, element=3, alpha_first=0
cmp w5, #0 // check width > 0
b.le 3f
@@ -228,7 +272,11 @@ endfunc
rgb_load_uv_coeff half=0, bgr=\bgr
b.lt 2f
1:
+ .if \alpha_first
+ argb_to_yuv_load_rgb x3
+ .else
rgb_to_yuv_load_rgb x3, \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
rgb_to_yuv_product v19, v20, v21, v25, v26, v18, v3, v4, v5, #9
@@ -241,9 +289,15 @@ endfunc
b.ge 1b
cbz w5, 3f
2:
+ .if \alpha_first
+ ldrb w16, [x3, #1] // w16: r
+ ldrb w17, [x3, #2] // w17: g
+ ldrb w4, [x3, #3] // w4: b
+ .else
ldrb w16, [x3] // w16: r
ldrb w17, [x3, #1] // w17: g
ldrb w4, [x3, #2] // w4: b
+ .endif
smaddl x8, w16, w10, x9 // x8 = ru * r + const_offset
smaddl x8, w17, w11, x8 // x8 += gu * g
@@ -263,6 +317,14 @@ endfunc
ret
.endm
+function ff_argb32ToUV_neon, export=1
+ rgbToUV bgr=0, element=4, alpha_first=1
+endfunc
+
+function ff_abgr32ToUV_neon, export=1
+ rgbToUV bgr=1, element=4, alpha_first=1
+endfunc
+
function ff_rgb24ToUV_neon, export=1
rgbToUV bgr=0
endfunc
@@ -211,6 +211,8 @@ void ff_##name##ToUV_half_neon(uint8_t *, uint8_t *, const uint8_t *, \
const uint8_t *, const uint8_t *, int w, \
uint32_t *coeffs, void *)
+NEON_INPUT(abgr32);
+NEON_INPUT(argb32);
NEON_INPUT(bgr24);
NEON_INPUT(bgra32);
NEON_INPUT(rgb24);
@@ -248,6 +250,21 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c)
c->yuv2planeX = ff_yuv2planeX_8_neon;
}
switch (c->srcFormat) {
+ case AV_PIX_FMT_ABGR:
+ c->lumToYV12 = ff_abgr32ToY_neon;
+ if (c->chrSrcHSubSample)
+ c->chrToYV12 = ff_abgr32ToUV_half_neon;
+ else
+ c->chrToYV12 = ff_abgr32ToUV_neon;
+ break;
+
+ case AV_PIX_FMT_ARGB:
+ c->lumToYV12 = ff_argb32ToY_neon;
+ if (c->chrSrcHSubSample)
+ c->chrToYV12 = ff_argb32ToUV_half_neon;
+ else
+ c->chrToYV12 = ff_argb32ToUV_neon;
+ break;
case AV_PIX_FMT_BGR24:
c->lumToYV12 = ff_bgr24ToY_neon;
if (c->chrSrcHSubSample)
From: Zhao Zhili <zhilizhao@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(-)