[FFmpeg-devel] [PATCH] swscale/x86/rgb2rgb: add AVX512ICL versions of shuffle_bytes

[Shreesh Adiga]

Signed-off-by: Shreesh Adiga <16567adigashreesh at gmail.com>
---
 libswscale/x86/rgb2rgb.c     | 21 +++++++++++++++++++++
 libswscale/x86/rgb_2_rgb.asm | 28 ++++++++++++++++++++++++++++
 2 files changed, 49 insertions(+)

diff --git a/libswscale/x86/rgb2rgb.c b/libswscale/x86/rgb2rgb.c
index 6790551a38..4cbed54b35 100644
--- a/libswscale/x86/rgb2rgb.c
+++ b/libswscale/x86/rgb2rgb.c
@@ -2364,6 +2364,16 @@ void ff_shuffle_bytes_2013_avx2(const uint8_t *src, uint8_t *dst, int src_size);
 void ff_shuffle_bytes_2130_avx2(const uint8_t *src, uint8_t *dst, int src_size);
 void ff_shuffle_bytes_1203_avx2(const uint8_t *src, uint8_t *dst, int src_size);
 
+void ff_shuffle_bytes_2103_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_0321_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_1230_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3012_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3210_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_3102_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_2013_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_2130_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+void ff_shuffle_bytes_1203_avx512icl(const uint8_t *src, uint8_t *dst, int src_size);
+
 void ff_uyvytoyuv422_sse2(uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
                           const uint8_t *src, int width, int height,
                           int lumStride, int chromStride, int srcStride);
@@ -2454,6 +2464,17 @@ av_cold void rgb2rgb_init_x86(void)
         shuffle_bytes_2130 = ff_shuffle_bytes_2130_avx2;
         shuffle_bytes_1203 = ff_shuffle_bytes_1203_avx2;
     }
+    if (EXTERNAL_AVX512ICL(cpu_flags)) {
+        shuffle_bytes_0321 = ff_shuffle_bytes_0321_avx512icl;
+        shuffle_bytes_2103 = ff_shuffle_bytes_2103_avx512icl;
+        shuffle_bytes_1230 = ff_shuffle_bytes_1230_avx512icl;
+        shuffle_bytes_3012 = ff_shuffle_bytes_3012_avx512icl;
+        shuffle_bytes_3210 = ff_shuffle_bytes_3210_avx512icl;
+        shuffle_bytes_3102 = ff_shuffle_bytes_3102_avx512icl;
+        shuffle_bytes_2013 = ff_shuffle_bytes_2013_avx512icl;
+        shuffle_bytes_2130 = ff_shuffle_bytes_2130_avx512icl;
+        shuffle_bytes_1203 = ff_shuffle_bytes_1203_avx512icl;
+    }
     if (EXTERNAL_AVX2_FAST(cpu_flags)) {
         uyvytoyuv422 = ff_uyvytoyuv422_avx2;
 #endif
diff --git a/libswscale/x86/rgb_2_rgb.asm b/libswscale/x86/rgb_2_rgb.asm
index b468beb12d..64b0988c4a 100644
--- a/libswscale/x86/rgb_2_rgb.asm
+++ b/libswscale/x86/rgb_2_rgb.asm
@@ -64,6 +64,18 @@ cglobal shuffle_bytes_%1%2%3%4, 3, 5, 2, src, dst, w, tmp, x
     add        dstq, wq
     neg          wq
 
+%if mmsize == 64
+    and xq, mmsize-4
+    shr xq, 2
+    mov tmpd, -1
+    shlx tmpd, tmpd, xd
+    not tmpd
+    kmovw k7, tmpw
+    vmovdqu32 m1{k7}{z}, [srcq + wq]
+    pshufb         m1, m0
+    vmovdqu32 [dstq + wq]{k7}, m1
+    lea wq, [wq + 4 * xq]
+%else
 ;calc scalar loop
     and xq, mmsize-4
     je .loop_simd
@@ -80,6 +92,7 @@ cglobal shuffle_bytes_%1%2%3%4, 3, 5, 2, src, dst, w, tmp, x
    add            wq, 4
    sub            xq, 4
    jg .loop_scalar
+%endif
 
 ;check if src_size < mmsize
 cmp wq, 0
@@ -122,6 +135,21 @@ SHUFFLE_BYTES 1, 2, 0, 3
 %endif
 %endif
 
+%if ARCH_X86_64
+%if HAVE_AVX512ICL_EXTERNAL
+INIT_ZMM avx512icl
+SHUFFLE_BYTES 2, 1, 0, 3
+SHUFFLE_BYTES 0, 3, 2, 1
+SHUFFLE_BYTES 1, 2, 3, 0
+SHUFFLE_BYTES 3, 0, 1, 2
+SHUFFLE_BYTES 3, 2, 1, 0
+SHUFFLE_BYTES 3, 1, 0, 2
+SHUFFLE_BYTES 2, 0, 1, 3
+SHUFFLE_BYTES 2, 1, 3, 0
+SHUFFLE_BYTES 1, 2, 0, 3
+%endif
+%endif
+
 ;-----------------------------------------------------------------------------------------------
 ; uyvytoyuv422(uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
 ;              const uint8_t *src, int width, int height,
-- 
2.45.3