| Message ID | tencent_7724B883967D4D3DA02FCDC3CF86D52C2707@qq.com |
|---|---|
| State | New |
| Headers | show |
| Series | [FFmpeg-devel,v3,1/3] swscale/aarch64: Add bgr24 to yuv | expand |
| Context | Check | Description |
|---|---|---|
| yinshiyou/make_loongarch64 | success | Make finished |
| yinshiyou/make_fate_loongarch64 | success | Make fate finished |
| andriy/make_x86 | success | Make finished |
| andriy/make_fate_x86 | success | Make fate finished |
On Mon, 24 Jun 2024, Zhao Zhili wrote: > From: Zhao Zhili <zhilizhao@tencent.com> > > Test on Apple M1 with kperf > : -O3 : -O3 -fno-vectorize > bgra_to_uv_8_c : 13.4 : 27.5 > bgra_to_uv_8_neon : 37.4 : 41.7 > bgra_to_uv_128_c : 155.9 : 550.2 > bgra_to_uv_128_neon : 91.7 : 92.7 > bgra_to_uv_1080_c : 1173.2 : 4558.2 > bgra_to_uv_1080_neon : 822.7 : 809.5 > bgra_to_uv_1920_c : 2078.2 : 8115.2 > bgra_to_uv_1920_neon : 1437.7 : 1438.7 > bgra_to_uv_half_8_c : 17.9 : 14.2 > bgra_to_uv_half_8_neon : 37.4 : 10.5 > bgra_to_uv_half_128_c : 103.9 : 326.0 > bgra_to_uv_half_128_neon : 73.9 : 68.7 > bgra_to_uv_half_1080_c : 850.2 : 3732.0 > bgra_to_uv_half_1080_neon : 484.2 : 490.0 > bgra_to_uv_half_1920_c : 1479.2 : 4942.7 > bgra_to_uv_half_1920_neon : 824.2 : 824.7 > bgra_to_y_8_c : 8.2 : 29.5 > bgra_to_y_8_neon : 18.2 : 32.7 > bgra_to_y_128_c : 101.4 : 361.5 > bgra_to_y_128_neon : 74.9 : 73.7 > bgra_to_y_1080_c : 739.4 : 3018.0 > bgra_to_y_1080_neon : 613.4 : 544.2 > bgra_to_y_1920_c : 1298.7 : 5326.0 > bgra_to_y_1920_neon : 918.7 : 934.2 > --- > libswscale/aarch64/input.S | 91 ++++++++++++++++++++++++++++++------ > libswscale/aarch64/swscale.c | 16 +++++++ > 2 files changed, 94 insertions(+), 13 deletions(-) > > diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S > index 2cfec4cb6a..6d2c6034bb 100644 > --- a/libswscale/aarch64/input.S > +++ b/libswscale/aarch64/input.S > @@ -20,8 +20,12 @@ > > #include "libavutil/aarch64/asm.S" > > -.macro rgb_to_yuv_load_rgb src > +.macro rgb_to_yuv_load_rgb src, element=3 > + .if \element == 3 > ld3 { v16.16b, v17.16b, v18.16b }, [\src] > + .else > + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [\src] > + .endif > uxtl v19.8h, v16.8b // v19: r > uxtl v20.8h, v17.8b // v20: g > uxtl v21.8h, v18.8b // v21: b > @@ -51,7 +55,8 @@ function ff_bgr24ToY_neon, export=1 > ret > endfunc > > -function ff_rgb24ToY_neon, export=1 > +.macro rgbToY_neon fmt, element > +function ff_\fmt\()ToY_neon, export=1 > cmp w4, #0 // check width > 0 > ldp w10, w11, [x5] // w10: ry, w11: gy > ldr w12, [x5, #8] // w12: by > @@ -67,11 +72,11 @@ function ff_rgb24ToY_neon, export=1 > dup v2.8h, w12 > b.lt 2f > 1: > - rgb_to_yuv_load_rgb x1 > + rgb_to_yuv_load_rgb x1, \element > 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 > - add x1, x1, #48 // src += 48 > + add x1, x1, #(16*\element) > cmp w4, #16 // width >= 16 ? > stp q16, q17, [x0], #32 // store to dst > b.ge 1b > @@ -86,12 +91,25 @@ function ff_rgb24ToY_neon, export=1 > smaddl x13, w15, w12, x13 // x13 += by * b > asr w13, w13, #9 // x13 >>= 9 > sub w4, w4, #1 // width-- > - add x1, x1, #3 // src += 3 > + add x1, x1, #\element > strh w13, [x0], #2 // store to dst > cbnz w4, 2b > 3: > ret > endfunc > +.endm > + > +rgbToY_neon fmt=rgb24, element=3 > + > +function ff_bgra32ToY_neon, export=1 > + cmp w4, #0 // check width > 0 > + ldp w12, w11, [x5] // w12: ry, w11: gy > + ldr w10, [x5, #8] // w10: by > + b.gt 4f > + ret > +endfunc > + > +rgbToY_neon fmt=rgba32, element=4 It is extremely obscure to jump to a local label (4f) that is defined by the following macro. I think this would be much more readable if you'd include the bgr(a) version in the macro, so the reference to 4f is near to the actual label it refers to. > .macro rgb_set_uv_coeff half > .if \half > @@ -120,7 +138,8 @@ function ff_bgr24ToUV_half_neon, export=1 > b 4f > endfunc > > -function ff_rgb24ToUV_half_neon, export=1 > +.macro rgbToUV_half_neon fmt, element > +function ff_\fmt\()ToUV_half_neon, export=1 > cmp w5, #0 // check width > 0 > b.le 3f > > @@ -132,7 +151,11 @@ function ff_rgb24ToUV_half_neon, export=1 > rgb_set_uv_coeff half=1 > b.lt 2f > 1: > + .if \element == 3 > ld3 { v16.16b, v17.16b, v18.16b }, [x3] > + .else > + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [x3] > + .endif > uaddlp v19.8h, v16.16b // v19: r > uaddlp v20.8h, v17.16b // v20: g > uaddlp v21.8h, v18.16b // v21: b > @@ -140,7 +163,7 @@ function ff_rgb24ToUV_half_neon, export=1 > 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 > sub w5, w5, #8 // width -= 8 > - add x3, x3, #48 // src += 48 > + add x3, x3, #(16*\element) > cmp w5, #8 // width >= 8 ? > str q16, [x0], #16 // store dst_u > str q17, [x1], #16 // store dst_v > @@ -148,9 +171,10 @@ function ff_rgb24ToUV_half_neon, export=1 > cbz w5, 3f > 2: > ldrb w2, [x3] // w2: r1 > - ldrb w4, [x3, #3] // w4: r2 > + ldrb w4, [x3, #\element] // w4: r2 > add w2, w2, w4 // w2 = r1 + r2 > > + .if \element == 3 > ldrb w4, [x3, #1] // w4: g1 > ldrb w7, [x3, #4] // w7: g2 > add w4, w4, w7 // w4 = g1 + g2 > @@ -158,6 +182,15 @@ function ff_rgb24ToUV_half_neon, export=1 > ldrb w7, [x3, #2] // w7: b1 > ldrb w8, [x3, #5] // w8: b2 > add w7, w7, w8 // w7 = b1 + b2 > + .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 > + .endif > > smaddl x8, w2, w10, x9 // dst_u = ru * r + const_offset > smaddl x8, w4, w11, x8 // dst_u += gu * g > @@ -170,12 +203,28 @@ function ff_rgb24ToUV_half_neon, export=1 > smaddl x8, w7, w15, x8 // dst_v += bv * b > asr x8, x8, #10 // dst_v >>= 10 > sub w5, w5, #1 > - add x3, x3, #6 // src += 6 > + ldrb w4, [x3, #1] // w4: g1 > + add x3, x3, #(2*\element) Is the new ldrb a typo/copypaste mistake here? // Martin
> On Jun 24, 2024, at 19:55, Martin Storsjö <martin@martin.st> wrote: > > On Mon, 24 Jun 2024, Zhao Zhili wrote: > >> From: Zhao Zhili <zhilizhao@tencent.com> >> >> Test on Apple M1 with kperf >> : -O3 : -O3 -fno-vectorize >> bgra_to_uv_8_c : 13.4 : 27.5 >> bgra_to_uv_8_neon : 37.4 : 41.7 >> bgra_to_uv_128_c : 155.9 : 550.2 >> bgra_to_uv_128_neon : 91.7 : 92.7 >> bgra_to_uv_1080_c : 1173.2 : 4558.2 >> bgra_to_uv_1080_neon : 822.7 : 809.5 >> bgra_to_uv_1920_c : 2078.2 : 8115.2 >> bgra_to_uv_1920_neon : 1437.7 : 1438.7 >> bgra_to_uv_half_8_c : 17.9 : 14.2 >> bgra_to_uv_half_8_neon : 37.4 : 10.5 >> bgra_to_uv_half_128_c : 103.9 : 326.0 >> bgra_to_uv_half_128_neon : 73.9 : 68.7 >> bgra_to_uv_half_1080_c : 850.2 : 3732.0 >> bgra_to_uv_half_1080_neon : 484.2 : 490.0 >> bgra_to_uv_half_1920_c : 1479.2 : 4942.7 >> bgra_to_uv_half_1920_neon : 824.2 : 824.7 >> bgra_to_y_8_c : 8.2 : 29.5 >> bgra_to_y_8_neon : 18.2 : 32.7 >> bgra_to_y_128_c : 101.4 : 361.5 >> bgra_to_y_128_neon : 74.9 : 73.7 >> bgra_to_y_1080_c : 739.4 : 3018.0 >> bgra_to_y_1080_neon : 613.4 : 544.2 >> bgra_to_y_1920_c : 1298.7 : 5326.0 >> bgra_to_y_1920_neon : 918.7 : 934.2 >> --- >> libswscale/aarch64/input.S | 91 ++++++++++++++++++++++++++++++------ >> libswscale/aarch64/swscale.c | 16 +++++++ >> 2 files changed, 94 insertions(+), 13 deletions(-) >> >> diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S >> index 2cfec4cb6a..6d2c6034bb 100644 >> --- a/libswscale/aarch64/input.S >> +++ b/libswscale/aarch64/input.S >> @@ -20,8 +20,12 @@ >> >> #include "libavutil/aarch64/asm.S" >> >> -.macro rgb_to_yuv_load_rgb src >> +.macro rgb_to_yuv_load_rgb src, element=3 >> + .if \element == 3 >> ld3 { v16.16b, v17.16b, v18.16b }, [\src] >> + .else >> + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [\src] >> + .endif >> uxtl v19.8h, v16.8b // v19: r >> uxtl v20.8h, v17.8b // v20: g >> uxtl v21.8h, v18.8b // v21: b >> @@ -51,7 +55,8 @@ function ff_bgr24ToY_neon, export=1 >> ret >> endfunc >> >> -function ff_rgb24ToY_neon, export=1 >> +.macro rgbToY_neon fmt, element >> +function ff_\fmt\()ToY_neon, export=1 >> cmp w4, #0 // check width > 0 >> ldp w10, w11, [x5] // w10: ry, w11: gy >> ldr w12, [x5, #8] // w12: by >> @@ -67,11 +72,11 @@ function ff_rgb24ToY_neon, export=1 >> dup v2.8h, w12 >> b.lt 2f >> 1: >> - rgb_to_yuv_load_rgb x1 >> + rgb_to_yuv_load_rgb x1, \element >> 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 >> - add x1, x1, #48 // src += 48 >> + add x1, x1, #(16*\element) >> cmp w4, #16 // width >= 16 ? >> stp q16, q17, [x0], #32 // store to dst >> b.ge 1b >> @@ -86,12 +91,25 @@ function ff_rgb24ToY_neon, export=1 >> smaddl x13, w15, w12, x13 // x13 += by * b >> asr w13, w13, #9 // x13 >>= 9 >> sub w4, w4, #1 // width-- >> - add x1, x1, #3 // src += 3 >> + add x1, x1, #\element >> strh w13, [x0], #2 // store to dst >> cbnz w4, 2b >> 3: >> ret >> endfunc >> +.endm >> + >> +rgbToY_neon fmt=rgb24, element=3 >> + >> +function ff_bgra32ToY_neon, export=1 >> + cmp w4, #0 // check width > 0 >> + ldp w12, w11, [x5] // w12: ry, w11: gy >> + ldr w10, [x5, #8] // w10: by >> + b.gt 4f >> + ret >> +endfunc >> + >> +rgbToY_neon fmt=rgba32, element=4 > > It is extremely obscure to jump to a local label (4f) that is defined by the following macro. I think this would be much more readable if you'd include the bgr(a) version in the macro, so the reference to 4f is near to the actual label it refers to. Good idea, it saved a lot of typing. Fixed in v4. > >> .macro rgb_set_uv_coeff half >> .if \half >> @@ -120,7 +138,8 @@ function ff_bgr24ToUV_half_neon, export=1 >> b 4f >> endfunc >> >> -function ff_rgb24ToUV_half_neon, export=1 >> +.macro rgbToUV_half_neon fmt, element >> +function ff_\fmt\()ToUV_half_neon, export=1 >> cmp w5, #0 // check width > 0 >> b.le 3f >> >> @@ -132,7 +151,11 @@ function ff_rgb24ToUV_half_neon, export=1 >> rgb_set_uv_coeff half=1 >> b.lt 2f >> 1: >> + .if \element == 3 >> ld3 { v16.16b, v17.16b, v18.16b }, [x3] >> + .else >> + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [x3] >> + .endif >> uaddlp v19.8h, v16.16b // v19: r >> uaddlp v20.8h, v17.16b // v20: g >> uaddlp v21.8h, v18.16b // v21: b >> @@ -140,7 +163,7 @@ function ff_rgb24ToUV_half_neon, export=1 >> 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 >> sub w5, w5, #8 // width -= 8 >> - add x3, x3, #48 // src += 48 >> + add x3, x3, #(16*\element) >> cmp w5, #8 // width >= 8 ? >> str q16, [x0], #16 // store dst_u >> str q17, [x1], #16 // store dst_v >> @@ -148,9 +171,10 @@ function ff_rgb24ToUV_half_neon, export=1 >> cbz w5, 3f >> 2: >> ldrb w2, [x3] // w2: r1 >> - ldrb w4, [x3, #3] // w4: r2 >> + ldrb w4, [x3, #\element] // w4: r2 >> add w2, w2, w4 // w2 = r1 + r2 >> >> + .if \element == 3 >> ldrb w4, [x3, #1] // w4: g1 >> ldrb w7, [x3, #4] // w7: g2 >> add w4, w4, w7 // w4 = g1 + g2 >> @@ -158,6 +182,15 @@ function ff_rgb24ToUV_half_neon, export=1 >> ldrb w7, [x3, #2] // w7: b1 >> ldrb w8, [x3, #5] // w8: b2 >> add w7, w7, w8 // w7 = b1 + b2 >> + .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 >> + .endif >> >> smaddl x8, w2, w10, x9 // dst_u = ru * r + const_offset >> smaddl x8, w4, w11, x8 // dst_u += gu * g >> @@ -170,12 +203,28 @@ function ff_rgb24ToUV_half_neon, export=1 >> smaddl x8, w7, w15, x8 // dst_v += bv * b >> asr x8, x8, #10 // dst_v >>= 10 >> sub w5, w5, #1 >> - add x3, x3, #6 // src += 6 >> + ldrb w4, [x3, #1] // w4: g1 >> + add x3, x3, #(2*\element) > > Is the new ldrb a typo/copypaste mistake here? Yes, it’s a copypaste mistake. Fixed in v4. > > // Martin > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S index 2cfec4cb6a..6d2c6034bb 100644 --- a/libswscale/aarch64/input.S +++ b/libswscale/aarch64/input.S @@ -20,8 +20,12 @@ #include "libavutil/aarch64/asm.S" -.macro rgb_to_yuv_load_rgb src +.macro rgb_to_yuv_load_rgb src, element=3 + .if \element == 3 ld3 { v16.16b, v17.16b, v18.16b }, [\src] + .else + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [\src] + .endif uxtl v19.8h, v16.8b // v19: r uxtl v20.8h, v17.8b // v20: g uxtl v21.8h, v18.8b // v21: b @@ -51,7 +55,8 @@ function ff_bgr24ToY_neon, export=1 ret endfunc -function ff_rgb24ToY_neon, export=1 +.macro rgbToY_neon fmt, element +function ff_\fmt\()ToY_neon, export=1 cmp w4, #0 // check width > 0 ldp w10, w11, [x5] // w10: ry, w11: gy ldr w12, [x5, #8] // w12: by @@ -67,11 +72,11 @@ function ff_rgb24ToY_neon, export=1 dup v2.8h, w12 b.lt 2f 1: - rgb_to_yuv_load_rgb x1 + rgb_to_yuv_load_rgb x1, \element 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 - add x1, x1, #48 // src += 48 + add x1, x1, #(16*\element) cmp w4, #16 // width >= 16 ? stp q16, q17, [x0], #32 // store to dst b.ge 1b @@ -86,12 +91,25 @@ function ff_rgb24ToY_neon, export=1 smaddl x13, w15, w12, x13 // x13 += by * b asr w13, w13, #9 // x13 >>= 9 sub w4, w4, #1 // width-- - add x1, x1, #3 // src += 3 + add x1, x1, #\element strh w13, [x0], #2 // store to dst cbnz w4, 2b 3: ret endfunc +.endm + +rgbToY_neon fmt=rgb24, element=3 + +function ff_bgra32ToY_neon, export=1 + cmp w4, #0 // check width > 0 + ldp w12, w11, [x5] // w12: ry, w11: gy + ldr w10, [x5, #8] // w10: by + b.gt 4f + ret +endfunc + +rgbToY_neon fmt=rgba32, element=4 .macro rgb_set_uv_coeff half .if \half @@ -120,7 +138,8 @@ function ff_bgr24ToUV_half_neon, export=1 b 4f endfunc -function ff_rgb24ToUV_half_neon, export=1 +.macro rgbToUV_half_neon fmt, element +function ff_\fmt\()ToUV_half_neon, export=1 cmp w5, #0 // check width > 0 b.le 3f @@ -132,7 +151,11 @@ function ff_rgb24ToUV_half_neon, export=1 rgb_set_uv_coeff half=1 b.lt 2f 1: + .if \element == 3 ld3 { v16.16b, v17.16b, v18.16b }, [x3] + .else + ld4 { v16.16b, v17.16b, v18.16b, v19.16b }, [x3] + .endif uaddlp v19.8h, v16.16b // v19: r uaddlp v20.8h, v17.16b // v20: g uaddlp v21.8h, v18.16b // v21: b @@ -140,7 +163,7 @@ function ff_rgb24ToUV_half_neon, export=1 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 sub w5, w5, #8 // width -= 8 - add x3, x3, #48 // src += 48 + add x3, x3, #(16*\element) cmp w5, #8 // width >= 8 ? str q16, [x0], #16 // store dst_u str q17, [x1], #16 // store dst_v @@ -148,9 +171,10 @@ function ff_rgb24ToUV_half_neon, export=1 cbz w5, 3f 2: ldrb w2, [x3] // w2: r1 - ldrb w4, [x3, #3] // w4: r2 + ldrb w4, [x3, #\element] // w4: r2 add w2, w2, w4 // w2 = r1 + r2 + .if \element == 3 ldrb w4, [x3, #1] // w4: g1 ldrb w7, [x3, #4] // w7: g2 add w4, w4, w7 // w4 = g1 + g2 @@ -158,6 +182,15 @@ function ff_rgb24ToUV_half_neon, export=1 ldrb w7, [x3, #2] // w7: b1 ldrb w8, [x3, #5] // w8: b2 add w7, w7, w8 // w7 = b1 + b2 + .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 + .endif smaddl x8, w2, w10, x9 // dst_u = ru * r + const_offset smaddl x8, w4, w11, x8 // dst_u += gu * g @@ -170,12 +203,28 @@ function ff_rgb24ToUV_half_neon, export=1 smaddl x8, w7, w15, x8 // dst_v += bv * b asr x8, x8, #10 // dst_v >>= 10 sub w5, w5, #1 - add x3, x3, #6 // src += 6 + ldrb w4, [x3, #1] // w4: g1 + add x3, x3, #(2*\element) strh w8, [x1], #2 // store dst_v cbnz w5, 2b 3: ret endfunc +.endm + +rgbToUV_half_neon fmt=rgb24, element=3 + +function ff_bgra32ToUV_half_neon, export=1 + cmp w5, #0 // check width > 0 + b.le 3f + + ldp w12, w11, [x6, #12] + ldp w10, w15, [x6, #20] + ldp w14, w13, [x6, #28] + b 4f +endfunc + +rgbToUV_half_neon fmt=rgba32, element=4 function ff_bgr24ToUV_neon, export=1 cmp w5, #0 // check width > 0 @@ -187,7 +236,8 @@ function ff_bgr24ToUV_neon, export=1 b 4f endfunc -function ff_rgb24ToUV_neon, export=1 +.macro rgbToUV_neon fmt, element +function ff_\fmt\()ToUV_neon, export=1 cmp w5, #0 // check width > 0 b.le 3f @@ -199,13 +249,13 @@ function ff_rgb24ToUV_neon, export=1 rgb_set_uv_coeff half=0 b.lt 2f 1: - rgb_to_yuv_load_rgb x3 + rgb_to_yuv_load_rgb x3, \element 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 rgb_to_yuv_product v22, v23, v24, v27, v28, v19, v3, v4, v5, #9 sub w5, w5, #16 - add x3, x3, #48 // src += 48 + add x3, x3, #(16*\element) cmp w5, #16 stp q16, q17, [x0], #32 // store to dst_u stp q18, q19, [x1], #32 // store to dst_v @@ -227,9 +277,24 @@ function ff_rgb24ToUV_neon, export=1 smaddl x8, w4, w15, x8 // x8 += bv * b asr w8, w8, #9 // x8 >>= 9 sub w5, w5, #1 // width-- - add x3, x3, #3 // src += 3 + add x3, x3, #\element strh w8, [x1], #2 // store to dst_v cbnz w5, 2b 3: ret endfunc +.endm + +rgbToUV_neon fmt=rgb24, element=3 + +function ff_bgra32ToUV_neon, export=1 + cmp w5, #0 // check width > 0 + b.le 3f + + ldp w12, w11, [x6, #12] + ldp w10, w15, [x6, #20] + ldp w14, w13, [x6, #28] + b 4f +endfunc + +rgbToUV_neon fmt=rgba32, element=4 diff --git a/libswscale/aarch64/swscale.c b/libswscale/aarch64/swscale.c index c6594944c3..92af662014 100644 --- a/libswscale/aarch64/swscale.c +++ b/libswscale/aarch64/swscale.c @@ -212,7 +212,9 @@ void ff_##name##ToUV_half_neon(uint8_t *, uint8_t *, const uint8_t *, \ uint32_t *coeffs, void *) NEON_INPUT(bgr24); +NEON_INPUT(bgra32); NEON_INPUT(rgb24); +NEON_INPUT(rgba32); void ff_lumRangeFromJpeg_neon(int16_t *dst, int width); void ff_chrRangeFromJpeg_neon(int16_t *dstU, int16_t *dstV, int width); @@ -253,6 +255,13 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c) else c->chrToYV12 = ff_bgr24ToUV_neon; break; + case AV_PIX_FMT_BGRA: + c->lumToYV12 = ff_bgra32ToY_neon; + if (c->chrSrcHSubSample) + c->chrToYV12 = ff_bgra32ToUV_half_neon; + else + c->chrToYV12 = ff_bgra32ToUV_neon; + break; case AV_PIX_FMT_RGB24: c->lumToYV12 = ff_rgb24ToY_neon; if (c->chrSrcHSubSample) @@ -260,6 +269,13 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c) else c->chrToYV12 = ff_rgb24ToUV_neon; break; + case AV_PIX_FMT_RGBA: + c->lumToYV12 = ff_rgba32ToY_neon; + if (c->chrSrcHSubSample) + c->chrToYV12 = ff_rgba32ToUV_half_neon; + else + c->chrToYV12 = ff_rgba32ToUV_neon; + break; default: break; }
From: Zhao Zhili <zhilizhao@tencent.com> Test on Apple M1 with kperf : -O3 : -O3 -fno-vectorize bgra_to_uv_8_c : 13.4 : 27.5 bgra_to_uv_8_neon : 37.4 : 41.7 bgra_to_uv_128_c : 155.9 : 550.2 bgra_to_uv_128_neon : 91.7 : 92.7 bgra_to_uv_1080_c : 1173.2 : 4558.2 bgra_to_uv_1080_neon : 822.7 : 809.5 bgra_to_uv_1920_c : 2078.2 : 8115.2 bgra_to_uv_1920_neon : 1437.7 : 1438.7 bgra_to_uv_half_8_c : 17.9 : 14.2 bgra_to_uv_half_8_neon : 37.4 : 10.5 bgra_to_uv_half_128_c : 103.9 : 326.0 bgra_to_uv_half_128_neon : 73.9 : 68.7 bgra_to_uv_half_1080_c : 850.2 : 3732.0 bgra_to_uv_half_1080_neon : 484.2 : 490.0 bgra_to_uv_half_1920_c : 1479.2 : 4942.7 bgra_to_uv_half_1920_neon : 824.2 : 824.7 bgra_to_y_8_c : 8.2 : 29.5 bgra_to_y_8_neon : 18.2 : 32.7 bgra_to_y_128_c : 101.4 : 361.5 bgra_to_y_128_neon : 74.9 : 73.7 bgra_to_y_1080_c : 739.4 : 3018.0 bgra_to_y_1080_neon : 613.4 : 544.2 bgra_to_y_1920_c : 1298.7 : 5326.0 bgra_to_y_1920_neon : 918.7 : 934.2 --- libswscale/aarch64/input.S | 91 ++++++++++++++++++++++++++++++------ libswscale/aarch64/swscale.c | 16 +++++++ 2 files changed, 94 insertions(+), 13 deletions(-)