swscale/aarch64: add nv24/nv42 to yuv420p unscaled converter

A55               A76
nv24_yuv420p_128_c:       4956.1            1267.0
nv24_yuv420p_128_neon:    3109.1 ( 1.59x)    640.0 ( 1.98x)
nv24_yuv420p_1920_c:     35728.4           11736.2
nv24_yuv420p_1920_neon:   8011.1 ( 4.46x)   2436.0 ( 4.82x)
nv42_yuv420p_128_c:       4956.4            1270.5
nv42_yuv420p_128_neon:    3074.6 ( 1.61x)    639.5 ( 1.99x)
nv42_yuv420p_1920_c:     35685.9           11732.5
nv42_yuv420p_1920_neon:   7995.1 ( 4.46x)   2437.2 ( 4.81x)

libswscale/aarch64/Makefile

@@ -7,4 +7,5 @@ NEON-OBJS   += aarch64/hscale.o                 \
                aarch64/output.o                 \
                aarch64/range_convert_neon.o     \
                aarch64/rgb2rgb_neon.o           \
+               aarch64/swscale_unscaled_neon.o  \
                aarch64/yuv2rgb_neon.o           \

libswscale/aarch64/swscale_unscaled.c

@@ -139,6 +139,31 @@ static int ifmt##_to_##ofmt##_neon_wrapper(SwsContext *c, const uint8_t *src[],
                                         dst[2] + srcSliceY * dstStride[2], dstStride[2]);   \
 }                                                                                           \
 
+void ff_nv24_to_yuv420p_chroma_neon(uint8_t *dst1, int dstStride1,
+                                    uint8_t *dst2, int dstStride2,
+                                    const uint8_t *src, int srcStride,
+                                    int w, int h);
+
+static int nv24_to_yuv420p_neon_wrapper(SwsContext *c, const uint8_t *src[],
+                                        int srcStride[], int srcSliceY, int srcSliceH,
+                                        uint8_t *dst[], int dstStride[])
+{
+    uint8_t *dst1 = dst[1] + dstStride[1] * srcSliceY / 2;
+    uint8_t *dst2 = dst[2] + dstStride[2] * srcSliceY / 2;
+
+    ff_copyPlane(src[0], srcStride[0], srcSliceY, srcSliceH, c->srcW,
+                 dst[0], dstStride[0]);
+
+    if (c->srcFormat == AV_PIX_FMT_NV24)
+        ff_nv24_to_yuv420p_chroma_neon(dst1, dstStride[1], dst2, dstStride[2],
+                                       src[1], srcStride[1], c->srcW / 2, srcSliceH);
+    else
+        ff_nv24_to_yuv420p_chroma_neon(dst2, dstStride[2], dst1, dstStride[1],
+                                       src[1], srcStride[1], c->srcW / 2, srcSliceH);
+
+    return srcSliceH;
+}
+
 #define DECLARE_FF_NVX_TO_ALL_RGBX_FUNCS(nvx)                                               \
 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, argb)                                                     \
 DECLARE_FF_NVX_TO_RGBX_FUNCS(nvx, rgba)                                                     \
@@ -177,6 +202,11 @@ static void get_unscaled_swscale_neon(SwsContext *c) {
     SET_FF_NVX_TO_ALL_RGBX_FUNC(nv21, NV21, accurate_rnd);
     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv420p, YUV420P, accurate_rnd);
     SET_FF_NVX_TO_ALL_RGBX_FUNC(yuv422p, YUV422P, accurate_rnd);
+
+    if (c->dstFormat == AV_PIX_FMT_YUV420P &&
+        (c->srcFormat == AV_PIX_FMT_NV24 || c->srcFormat == AV_PIX_FMT_NV42) &&
+        !(c->srcH & 1) && !(c->srcW & 15) && !accurate_rnd)
+        c->convert_unscaled = nv24_to_yuv420p_neon_wrapper;
 }
 
 void ff_get_unscaled_swscale_aarch64(SwsContext *c)

libswscale/aarch64/swscale_unscaled_neon.S

@@ -0,0 +1,70 @@
+/*
+ * Copyright (c) 2024 Ramiro Polla
+ *
+ * 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"
+
+function ff_nv24_to_yuv420p_chroma_neon, export=1
+// x0  uint8_t *dst1
+// x1  int dstStride1
+// x2  uint8_t *dst2
+// x3  int dstStride2
+// x4  const uint8_t *src
+// x5  int srcStride
+// w6  int w
+// w7  int h
+
+        add             x9, x4, w5, sxtw            // x9 = src + srcStride
+        lsl             w5, w5, #1                  // srcStride *= 2
+        sub             w5, w5, w6, lsl #2          // srcPadding = (2 * srcStride) - (4 * w)
+        sub             w1, w1, w6                  // dstPadding1 = dstStride1 - w
+        sub             w3, w3, w6                  // dstPadding2 = dstStride2 - w
+
+1:
+        mov             w10, w6                     // w10 = w
+
+2:
+        ld2             { v0.16b, v1.16b }, [x4], #32 // v0 = U1, v1 = V1
+        ld2             { v2.16b, v3.16b }, [x9], #32 // v2 = U2, v3 = V2
+
+        uaddlp          v0.8h, v0.16b               // pairwise add U1 into v0
+        uaddlp          v1.8h, v1.16b               // pairwise add V1 into v1
+        uadalp          v0.8h, v2.16b               // pairwise add U2, accumulate into v0
+        uadalp          v1.8h, v3.16b               // pairwise add V2, accumulate into v1
+
+        shrn            v0.8b, v0.8h, #2            // divide by 4
+        shrn            v1.8b, v1.8h, #2            // divide by 4
+
+        st1             { v0.8b }, [x0], #8         // store U into dst1
+        st1             { v1.8b }, [x2], #8         // store V into dst2
+
+        subs            w10, w10, #8
+        b.gt            2b
+
+        // next row
+        add             x4, x4, w5, sxtw            // src1 += srcPadding
+        add             x9, x9, w5, sxtw            // src2 += srcPadding
+        add             x0, x0, w1, sxtw            // dst1 += dstPadding1
+        add             x2, x2, w3, sxtw            // dst2 += dstPadding2
+
+        subs            w7, w7, #2
+        b.gt            1b
+
+        ret
+endfunc