[FFmpeg-devel] [WIP PATCH 2/2] swscale/aarch64: Add rgb24 to yuv implementation

Zhao Zhili quinkblack at foxmail.com
Mon Jun 3 16:02:33 EEST 2024


From: Zhao Zhili <zhilizhao at tencent.com>

Test on Apple M1:

rgb24_to_uv_1080_c: 7.0
rgb24_to_uv_1080_neon: 5.5
rgb24_to_uv_1280_c: 8.2
rgb24_to_uv_1280_neon: 6.0
rgb24_to_uv_4096_c: 26.2
rgb24_to_uv_4096_neon: 20.7

rgb24_to_uv_half_540_c: 6.5
rgb24_to_uv_half_540_neon: 3.0
rgb24_to_uv_half_640_c: 8.0
rgb24_to_uv_half_640_neon: 3.2
rgb24_to_uv_half_2048_c: 24.0
rgb24_to_uv_half_2048_neon: 11.2

rgb24_to_y_1080_c: 4.5
rgb24_to_y_1080_neon: 3.2
rgb24_to_y_1280_c: 5.5
rgb24_to_y_1280_neon: 3.7
rgb24_to_y_4096_c: 17.7
rgb24_to_y_4096_neon: 12.2

Signed-off-by: Zhao Zhili <zhilizhao at tencent.com>
---
 libswscale/aarch64/Makefile  |   1 +
 libswscale/aarch64/input.S   | 229 +++++++++++++++++++++++++++++++++++
 libswscale/aarch64/swscale.c |  21 ++++
 3 files changed, 251 insertions(+)
 create mode 100644 libswscale/aarch64/input.S

diff --git a/libswscale/aarch64/Makefile b/libswscale/aarch64/Makefile
index da1d909561..adfd90a1b6 100644
--- a/libswscale/aarch64/Makefile
+++ b/libswscale/aarch64/Makefile
@@ -3,6 +3,7 @@ OBJS        += aarch64/rgb2rgb.o                \
                aarch64/swscale_unscaled.o       \
 
 NEON-OBJS   += aarch64/hscale.o                 \
+               aarch64/input.o                  \
                aarch64/output.o                 \
                aarch64/rgb2rgb_neon.o           \
                aarch64/yuv2rgb_neon.o           \
diff --git a/libswscale/aarch64/input.S b/libswscale/aarch64/input.S
new file mode 100644
index 0000000000..ee0d223c6e
--- /dev/null
+++ b/libswscale/aarch64/input.S
@@ -0,0 +1,229 @@
+/*
+ * Copyright (c) 2024 Zhao Zhili <quinkblack at foxmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/aarch64/asm.S"
+
+.macro rgb24_to_yuv_load_rgb, src
+        ld3             { v16.16b, v17.16b, v18.16b }, [\src]
+        ushll           v19.8h, v16.8b, #0         // v19: r
+        ushll           v20.8h, v17.8b, #0         // v20: g
+        ushll           v21.8h, v18.8b, #0         // v21: b
+        ushll2          v22.8h, v16.16b, #0        // v22: r
+        ushll2          v23.8h, v17.16b, #0        // v23: g
+        ushll2          v24.8h, v18.16b, #0        // v24: b
+.endm
+
+.macro rgb24_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
+        smlal           \dst1\().4s, \coef0\().4h, \r\().4h     // dst1 += rx * r
+        smlal2          \dst2\().4s, \coef0\().8h, \r\().8h     // dst2 += rx * r
+        smlal           \dst1\().4s, \coef1\().4h, \g\().4h     // dst1 += gx * g
+        smlal2          \dst2\().4s, \coef1\().8h, \g\().8h     // dst2 += gx * g
+        smlal           \dst1\().4s, \coef2\().4h, \b\().4h     // dst1 += bx * b
+        smlal2          \dst2\().4s, \coef2\().8h, \b\().8h     // dst2 += bx * b
+        sqshrn          \dst\().4h, \dst1\().4s, \right_shift   // dst_lower_half = dst1 >> right_shift
+        sqshrn2         \dst\().8h, \dst2\().4s, \right_shift   // dst_higher_half = dst2 >> right_shift
+.endm
+
+function ff_rgb24ToY_neon, export=1
+        cmp             w4, #0                  // check width > 0
+        b.le            4f
+
+        ldp             w10, w11, [x5], #8       // w10: ry, w11: gy
+        dup             v0.8h, w10
+        dup             v1.8h, w11
+        ldr             w12, [x5]               // w12: by
+        dup             v2.8h, w12
+
+        mov             w9, #256                // w9 = 1 << (RGB2YUV_SHIFT - 7)
+        movk            w9, #8, lsl #16         // w9 += 32 << (RGB2YUV_SHIFT - 1)
+        dup             v6.4s, w9               // w9: const_offset
+
+        mov             x2, #0                  // w2: i
+        and             w3, w4, #0xFFFFFFF0     // w3 = width / 16 * 16
+        cbz             w3, 3f
+1:
+        rgb24_to_yuv_load_rgb x1
+        rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v0, v1, v2, #9
+        rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v0, v1, v2, #9
+        stp             q16, q17, [x0], #32     // store to dst
+
+        add             w2, w2, #16             // i += 16
+        add             x1, x1, #48             // src += 48
+        cmp             w2, w3                  // i < (width / 16 * 16)
+        b.lt            1b
+        b               3f
+2:
+        ldrb            w13, [x1]               // w13: r
+        ldrb            w14, [x1, #1]           // w14: g
+        ldrb            w15, [x1, #2]           // w15: b
+
+        smaddl          x13, w13, w10, x9       // x13 = ry * r + const_offset
+        smaddl          x13, w14, w11, x13      // x13 += gy * g
+        smaddl          x13, w15, w12, x13      // x13 += by * b
+        asr             w13, w13, #9            // x13 >>= 9
+        strh            w13, [x0], #2           // store to dst
+
+        add             w2, w2, #1              // i++
+        add             x1, x1, #3              // src += 3
+3:
+        cmp             w2, w4                  // i < width
+        b.lt            2b
+4:
+        ret
+endfunc
+
+.macro rgb24_load_uv_coeff half
+        add             x6, x6, #12
+
+        ldp             w10, w11, [x6], #8      // w10: ru, w11: gu
+        dup             v0.8h, w10
+        dup             v1.8h, w11
+
+        ldp             w12, w13, [x6], #8      // w12: bu, w13: rv
+        dup             v2.8h, w12
+        dup             v3.8h, w13
+
+        ldp             w14, w15, [x6], #8      // w14: gv, w15: bv
+        dup             v4.8h, w14
+        dup             v5.8h, w15
+
+    .if \half
+        mov             w9, #512
+        movk            w9, #128, lsl #16       // w9: const_offset
+    .else
+        mov             w9, #256
+        movk            w9, #64, lsl #16        // w9: const_offset
+    .endif
+        dup             v6.4s, w9
+.endm
+
+function ff_rgb24ToUV_half_neon, export=1
+        cmp             w5, #0          // check width > 0
+        b.le            4f
+
+        rgb24_load_uv_coeff half=1
+
+        mov             x9, #0                  // x9: i
+        and             w7, w5, #0xFFFFFFF8     // w7 = width / 8 * 8
+        cbz             w7, 3f
+1:
+        ld3             { v16.16b, v17.16b, v18.16b }, [x3]
+        uaddlp          v19.8h, v16.16b         // v19: r
+        uaddlp          v20.8h, v17.16b         // v20: g
+        uaddlp          v21.8h, v18.16b         // v21: b
+
+        rgb24_to_yuv_product v19, v20, v21, v22, v23, v16, v0, v1, v2, #10
+        str             q16, [x0], #16          // store dst_u
+        rgb24_to_yuv_product v19, v20, v21, v24, v25, v17, v3, v4, v5, #10
+        str             q17, [x1], #16          // store dst_v
+
+        add             w9, w9, #8              // i += 8
+        add             x3, x3, #48             // src += 48
+        cmp             w9, w7                  // i < (width * 8 / 8)
+        b.lt            1b
+        b               3f
+2:
+        ldrb            w2, [x3]                // w2: r1
+        ldrb            w4, [x3, #3]            // w4: r2
+        add             w2, w2, w4              // w2 = r1 + r2
+
+        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
+
+        umov            w8, v6.s[0]             // dst_u = const_offset
+        smaddl          x8, w2, w10, x8         // dst_u += ru * r
+        smaddl          x8, w4, w11, x8         // dst_u += gu * g
+        smaddl          x8, w7, w12, x8         // dst_u += bu * b
+        asr             x8, x8, #10             // dst_u >>= 10
+        strh            w8, [x0], #2            // store dst_u
+
+        umov            w8, v6.s[0]             // dst_v = const_offset
+        smaddl          x8, w2, w13, x8         // dst_v += rv * r
+        smaddl          x8, w4, w14, x8         // dst_v += gv * g
+        smaddl          x8, w7, w15, x8         // dst_v += bv * b
+        asr             x8, x8, #10             // dst_v >>= 10
+        strh            w8, [x1], #2            // store dst_v
+
+        add             w9, w9, #1              // i++
+        add             x3, x3, #6              // src += 6
+3:
+        cmp             w9, w5
+        b.lt            2b
+4:
+        ret
+endfunc
+
+function ff_rgb24ToUV_neon, export=1
+        cmp             w5, #0                  // check width > 0
+        b.le            4f
+
+        rgb24_load_uv_coeff half=0
+
+        mov             x2, #0                  // w2: i
+        and             w4, w5, #0xFFFFFFF0     // w4: width / 16 * 16
+        cbz             w4, 3f
+1:
+        rgb24_to_yuv_load_rgb x3
+        rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v0, v1, v2, #9
+        rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v0, v1, v2, #9
+        stp             q16, q17, [x0], #32      // store to dst_u
+        rgb24_to_yuv_product v19, v20, v21, v25, v26, v16, v3, v4, v5, #9
+        rgb24_to_yuv_product v22, v23, v24, v27, v28, v17, v3, v4, v5, #9
+        stp             q16, q17, [x1], #32      // store to dst_v
+
+        add             w2, w2, #16             // i += 16
+        add             x3, x3, #48             // src += 48
+        cmp             w2, w4                  // i < (width / 16 * 16)
+        b.lt            1b
+        b               3f
+2:
+        ldrb            w16, [x3]               // w16: r
+        ldrb            w17, [x3, #1]           // w17: g
+        ldrb            w4, [x3, #2]            // w4: b
+
+        umov            w7, v6.s[0]            // w7 = const_offset
+
+        smaddl          x8, w16, w10, x7        // x8 = ru * r + const_offset
+        smaddl          x8, w17, w11, x8        // x8 += gu * g
+        smaddl          x8, w4, w12, x8         // x8 += bu * b
+        asr             w8, w8, #9              // x8 >>= 9
+        strh            w8, [x0], #2            // store to dst_u
+
+        smaddl          x8, w16, w13, x7        // x8 = rv * r + const_offset
+        smaddl          x8, w17, w14, x8        // x8 += gv * g
+        smaddl          x8, w4, w15, x8         // x8 += bv * b
+        asr             w8, w8, #9              // x8 >>= 9
+        strh            w8, [x1], #2            // store to dst_v
+
+        add             w2, w2, #1              // i++
+        add             x3, x3, #3              // src += 3
+3:
+        cmp             w2, w5                  // i < width
+        b.lt            2b
+4:
+        ret
+endfunc
diff --git a/libswscale/aarch64/swscale.c b/libswscale/aarch64/swscale.c
index bbd9719a44..5dd700a16b 100644
--- a/libswscale/aarch64/swscale.c
+++ b/libswscale/aarch64/swscale.c
@@ -201,6 +201,20 @@ void ff_yuv2plane1_8_neon(
     default: break;                                                     \
     }
 
+void ff_rgb24ToY_neon(uint8_t *_dst, const uint8_t *src, const uint8_t *unused1,
+                      const uint8_t *unused2, int width,
+                      uint32_t *rgb2yuv, void *opq);
+
+void ff_rgb24ToUV_neon(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused0,
+                       const uint8_t *src1,
+                       const uint8_t *src2, int width, uint32_t *rgb2yuv,
+                       void *opq);
+
+void ff_rgb24ToUV_half_neon(uint8_t *_dstU, uint8_t *_dstV, const uint8_t *unused0,
+                       const uint8_t *src1,
+                       const uint8_t *src2, int width, uint32_t *rgb2yuv,
+                       void *opq);
+
 av_cold void ff_sws_init_swscale_aarch64(SwsContext *c)
 {
     int cpu_flags = av_get_cpu_flags();
@@ -212,5 +226,12 @@ av_cold void ff_sws_init_swscale_aarch64(SwsContext *c)
         if (c->dstBpc == 8) {
             c->yuv2planeX = ff_yuv2planeX_8_neon;
         }
+        if (c->srcFormat == AV_PIX_FMT_RGB24) {
+            c->lumToYV12 = ff_rgb24ToY_neon;
+            if (c->chrSrcHSubSample)
+                c->chrToYV12 = ff_rgb24ToUV_half_neon;
+            else
+                c->chrToYV12 = ff_rgb24ToUV_neon;
+        }
     }
 }
-- 
2.42.0



More information about the ffmpeg-devel mailing list