[FFmpeg-devel] [PATCH] swscale/aarch64/rgb2rgb: Implemented NEON shuf routines
Krzysztof Pyrkosz
ffmpeg at szaka.eu
Tue Jan 28 20:01:33 EET 2025
The key idea is to pass the pre-generated tables to the TBL instruction
and churn through the data 16 bytes at a time. The remaining 4 elements
are handled with a specialized block located at the end of the routine.
The 3210 variant can be implemented using rev32, but surprisingly it is
slower than the generic TBL on A78, but much faster on A72. I wrapped it
in #if 0 block.
There may be some room for improvement. Possibly instead of handling
last 8 and then 4 bytes separately, we can load these 4 into {v0.s}[2]
and process along with the last 8 bytes.
Speeds measured with checkasm --test=sw_rgb --bench --runs=10 | grep shuf
- A78
shuffle_bytes_0321_c: 75.5 ( 1.00x)
shuffle_bytes_0321_neon: 26.5 ( 2.85x)
shuffle_bytes_1203_c: 136.2 ( 1.00x)
shuffle_bytes_1203_neon: 27.2 ( 5.00x)
shuffle_bytes_1230_c: 135.5 ( 1.00x)
shuffle_bytes_1230_neon: 28.0 ( 4.84x)
shuffle_bytes_2013_c: 138.8 ( 1.00x)
shuffle_bytes_2013_neon: 22.0 ( 6.31x)
shuffle_bytes_2103_c: 76.5 ( 1.00x)
shuffle_bytes_2103_neon: 20.5 ( 3.73x)
shuffle_bytes_2130_c: 137.5 ( 1.00x)
shuffle_bytes_2130_neon: 28.0 ( 4.91x)
shuffle_bytes_3012_c: 138.2 ( 1.00x)
shuffle_bytes_3012_neon: 21.5 ( 6.43x)
shuffle_bytes_3102_c: 138.2 ( 1.00x)
shuffle_bytes_3102_neon: 27.2 ( 5.07x)
shuffle_bytes_3210_c: 138.0 ( 1.00x)
shuffle_bytes_3210_neon: 22.0 ( 6.27x)
shuf3210 using rev32
shuffle_bytes_3210_c: 139.0 ( 1.00x)
shuffle_bytes_3210_neon: 28.5 ( 4.88x)
- A72
shuffle_bytes_0321_c: 120.0 ( 1.00x)
shuffle_bytes_0321_neon: 36.0 ( 3.33x)
shuffle_bytes_1203_c: 188.2 ( 1.00x)
shuffle_bytes_1203_neon: 37.8 ( 4.99x)
shuffle_bytes_1230_c: 195.0 ( 1.00x)
shuffle_bytes_1230_neon: 36.0 ( 5.42x)
shuffle_bytes_2013_c: 195.8 ( 1.00x)
shuffle_bytes_2013_neon: 43.5 ( 4.50x)
shuffle_bytes_2103_c: 117.2 ( 1.00x)
shuffle_bytes_2103_neon: 53.5 ( 2.19x)
shuffle_bytes_2130_c: 203.2 ( 1.00x)
shuffle_bytes_2130_neon: 37.8 ( 5.38x)
shuffle_bytes_3012_c: 183.8 ( 1.00x)
shuffle_bytes_3012_neon: 46.8 ( 3.93x)
shuffle_bytes_3102_c: 180.8 ( 1.00x)
shuffle_bytes_3102_neon: 37.8 ( 4.79x)
shuffle_bytes_3210_c: 195.8 ( 1.00x)
shuffle_bytes_3210_neon: 37.8 ( 5.19x)
shuf3210 using rev32
shuffle_bytes_3210_c: 194.8 ( 1.00x)
shuffle_bytes_3210_neon: 30.8 ( 6.33x)
- x13s:
shuffle_bytes_0321_c: 49.4 ( 1.00x)
shuffle_bytes_0321_neon: 18.1 ( 2.72x)
shuffle_bytes_1203_c: 98.4 ( 1.00x)
shuffle_bytes_1203_neon: 18.4 ( 5.35x)
shuffle_bytes_1230_c: 97.4 ( 1.00x)
shuffle_bytes_1230_neon: 19.1 ( 5.09x)
shuffle_bytes_2013_c: 101.4 ( 1.00x)
shuffle_bytes_2013_neon: 16.9 ( 6.01x)
shuffle_bytes_2103_c: 53.9 ( 1.00x)
shuffle_bytes_2103_neon: 13.9 ( 3.88x)
shuffle_bytes_2130_c: 100.9 ( 1.00x)
shuffle_bytes_2130_neon: 19.1 ( 5.27x)
shuffle_bytes_3012_c: 97.4 ( 1.00x)
shuffle_bytes_3012_neon: 17.1 ( 5.69x)
shuffle_bytes_3102_c: 100.9 ( 1.00x)
shuffle_bytes_3102_neon: 19.1 ( 5.27x)
shuffle_bytes_3210_c: 100.6 ( 1.00x)
shuffle_bytes_3210_neon: 16.9 ( 5.96x)
shuf3210 using rev32
shuffle_bytes_3210_c: 100.6 ( 1.00x)
shuffle_bytes_3210_neon: 18.6 ( 5.40x)
Krzysztof
---
libswscale/aarch64/rgb2rgb.c | 18 ++++
libswscale/aarch64/rgb2rgb_neon.S | 159 ++++++++++++++++++++++++++++++
tests/checkasm/sw_rgb.c | 12 +++
3 files changed, 189 insertions(+)
diff --git a/libswscale/aarch64/rgb2rgb.c b/libswscale/aarch64/rgb2rgb.c
index 401c7d09c1..7e1dba572d 100644
--- a/libswscale/aarch64/rgb2rgb.c
+++ b/libswscale/aarch64/rgb2rgb.c
@@ -57,6 +57,15 @@ void ff_interleave_bytes_neon(const uint8_t *src1, const uint8_t *src2,
void ff_deinterleave_bytes_neon(const uint8_t *src, uint8_t *dst1, uint8_t *dst2,
int width, int height, int srcStride,
int dst1Stride, int dst2Stride);
+void ff_shuffle_bytes_0321_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_2103_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_1230_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3012_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3210_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3102_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_2013_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_2130_neon(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_1203_neon(const uint8_t *src, uint8_t *dst, int src_size);
av_cold void rgb2rgb_init_aarch64(void)
{
@@ -66,5 +75,14 @@ av_cold void rgb2rgb_init_aarch64(void)
ff_rgb24toyv12 = rgb24toyv12;
interleaveBytes = ff_interleave_bytes_neon;
deinterleaveBytes = ff_deinterleave_bytes_neon;
+ shuffle_bytes_0321 = ff_shuffle_bytes_0321_neon;
+ shuffle_bytes_1230 = ff_shuffle_bytes_1230_neon;
+ shuffle_bytes_2103 = ff_shuffle_bytes_2103_neon;
+ shuffle_bytes_3012 = ff_shuffle_bytes_3012_neon;
+ shuffle_bytes_3210 = ff_shuffle_bytes_3210_neon;
+ shuffle_bytes_3102 = ff_shuffle_bytes_3102_neon;
+ shuffle_bytes_2013 = ff_shuffle_bytes_2013_neon;
+ shuffle_bytes_2130 = ff_shuffle_bytes_2130_neon;
+ shuffle_bytes_1203 = ff_shuffle_bytes_1203_neon;
}
}
diff --git a/libswscale/aarch64/rgb2rgb_neon.S b/libswscale/aarch64/rgb2rgb_neon.S
index 1382e00261..a69a211ad4 100644
--- a/libswscale/aarch64/rgb2rgb_neon.S
+++ b/libswscale/aarch64/rgb2rgb_neon.S
@@ -34,6 +34,69 @@
#define Y_OFFSET v22
#define UV_OFFSET v23
+const shuf_0321_tbl, align=4
+ .byte 0, 3, 2, 1
+ .byte 4, 7, 6, 5
+ .byte 8, 11, 10, 9
+ .byte 12, 15, 14, 13
+endconst
+
+const shuf_1230_tbl, align=4
+ .byte 1, 2, 3, 0
+ .byte 5, 6, 7, 4
+ .byte 9, 10, 11, 8
+ .byte 13, 14, 15, 12
+endconst
+
+const shuf_2103_tbl, align=4
+ .byte 2, 1, 0, 3
+ .byte 6, 5, 4, 7
+ .byte 10, 9, 8, 11
+ .byte 14, 13, 12, 15
+endconst
+
+const shuf_3012_tbl, align=4
+ .byte 3, 0, 1, 2
+ .byte 7, 4, 5, 6
+ .byte 11, 8, 9, 10
+ .byte 15, 12, 13, 14
+endconst
+
+const shuf_3210_tbl, align=4
+ .byte 3, 2, 1, 0
+ .byte 7, 6, 5, 4
+ .byte 11, 10, 9, 8
+ .byte 15, 14, 13, 12
+endconst
+
+const shuf_3102_tbl, align=4
+ .byte 3, 1, 0, 2
+ .byte 7, 5, 4, 6
+ .byte 11, 9, 8, 10
+ .byte 15, 13, 12, 14
+endconst
+
+const shuf_2013_tbl, align=4
+ .byte 2, 0, 1, 3
+ .byte 6, 4, 5, 7
+ .byte 10, 8, 9, 11
+ .byte 14, 12, 13, 15
+endconst
+
+const shuf_1203_tbl, align=4
+ .byte 1, 2, 0, 3
+ .byte 5, 6, 4, 7
+ .byte 9, 10, 8, 11
+ .byte 13, 14, 12, 15
+endconst
+
+const shuf_2130_tbl, align=4
+ .byte 2, 1, 3, 0
+ .byte 6, 5, 7, 4
+ .byte 10, 9, 11, 8
+ .byte 14, 13, 15, 12
+endconst
+
// convert rgb to 16-bit y, u, or v
// uses v3 and v4
.macro rgbconv16 dst, b, g, r, bc, gc, rc
@@ -296,3 +359,99 @@ function ff_deinterleave_bytes_neon, export=1
0:
ret
endfunc
+
+.macro neon_shuf shuf
+function ff_shuffle_bytes_\shuf\()_neon, export=1
+ movrel x9, shuf_\shuf\()_tbl
+ ld1 {v1.16b}, [x9]
+ and w5, w2, #~15
+ and w3, w2, #8
+ and w4, w2, #4
+ cbz w5, 2f
+1:
+ subs w5, w5, #16
+ ld1 {v0.16b}, [x0], #16
+ tbl v0.16b, {v0.16b}, v1.16b
+ st1 {v0.16b}, [x1], #16
+ b.gt 1b
+2:
+ cbz w3, 3f
+ ld1 {v0.8b}, [x0], #8
+ tbl v0.8b, {v0.16b}, v1.8b
+ st1 {v0.8b}, [x1], #8
+3:
+ cbz w4, 4f
+.if \shuf == 0321
+ ldr w5, [x0]
+ rev w5, w5
+ ror w5, w5, #24
+ str w5, [x1]
+.endif
+.if \shuf == 1230
+ ldr w5, [x0]
+ ror w5, w5, #8
+ str w5, [x1]
+.endif
+.if \shuf == 2103
+ ldr w5, [x0]
+ rev w5, w5
+ ror w5, w5, #8
+ str w5, [x1]
+.endif
+.if \shuf == 3012
+ ldr w5, [x0]
+ ror w5, w5, #24
+ str w5, [x1]
+.endif
+.if \shuf == 3210
+ ldr w5, [x0]
+ rev w5, w5
+ str w5, [x1]
+.endif
+.if \shuf == 3102 || \shuf == 2013 || \shuf == 1203 || \shuf == 2130
+ ld1 {v0.s}[0], [x0]
+ tbl v0.8b, {v0.16b}, v1.8b
+ st1 {v0.s}[0], [x1]
+.endif
+4:
+ ret
+endfunc
+.endm
+
+#
+#if 0
+function ff_shuffle_bytes_3210_neon, export=1
+ and w5, w2, #~(15)
+ and w3, w2, #8
+ and w4, w2, #4
+ cbz w5, 2f
+1:
+ subs w5, w5, #16
+ ld1 {v0.16b}, [x0], #16
+ rev32 v0.16b, v0.16b
+ st1 {v0.16b}, [x1], #16
+ b.gt 1b
+2:
+ cbz w3, 3f
+ ld1 {v0.8b}, [x0], #8
+ rev32 v0.8b, v0.8b
+ st1 {v0.8b}, [x1], #8
+3:
+ cbz w4, 4f
+ ldr w5, [x0]
+ rev w5, w5
+ str w5, [x1]
+4:
+ ret
+endfunc
+#endif
+
+neon_shuf 0321
+neon_shuf 1230
+neon_shuf 2103
+neon_shuf 3012
+neon_shuf 3102
+neon_shuf 2013
+neon_shuf 1203
+neon_shuf 2130
+neon_shuf 3210
diff --git a/tests/checkasm/sw_rgb.c b/tests/checkasm/sw_rgb.c
index 5714bfaa1e..6e8c63cffe 100644
--- a/tests/checkasm/sw_rgb.c
+++ b/tests/checkasm/sw_rgb.c
@@ -822,6 +822,18 @@ void checkasm_check_sw_rgb(void)
check_shuffle_bytes(shuffle_bytes_3210, "shuffle_bytes_3210");
report("shuffle_bytes_3210");
+ check_shuffle_bytes(shuffle_bytes_3102, "shuffle_bytes_3102");
+ report("shuffle_bytes_3102");
+
+ check_shuffle_bytes(shuffle_bytes_2013, "shuffle_bytes_2013");
+ report("shuffle_bytes_2013");
+
+ check_shuffle_bytes(shuffle_bytes_1203, "shuffle_bytes_1203");
+ report("shuffle_bytes_1203");
+
+ check_shuffle_bytes(shuffle_bytes_2130, "shuffle_bytes_2130");
+ report("shuffle_bytes_2130");
+
check_uyvy_to_422p();
report("uyvytoyuv422");
--
2.45.2
More information about the ffmpeg-devel
mailing list