swscale/x86/rgb2rgb: add AVX512ICL versions of shuffle_bytes

On a AMD 7950x Zen 4

shuffle_bytes_0321_c:                                   56.5 ( 1.00x)
shuffle_bytes_0321_ssse3:                               15.2 ( 3.70x)
shuffle_bytes_0321_avx2:                                10.2 ( 5.51x)
shuffle_bytes_0321_avx512icl:                            9.2 ( 6.11x)
shuffle_bytes_1230_c:                                   84.5 ( 1.00x)
shuffle_bytes_1230_ssse3:                               14.2 ( 5.93x)
shuffle_bytes_1230_avx2:                                15.2 ( 5.54x)
shuffle_bytes_1230_avx512icl:                           11.2 ( 7.51x)
shuffle_bytes_2103_c:                                   48.5 ( 1.00x)
shuffle_bytes_2103_ssse3:                               21.2 ( 2.28x)
shuffle_bytes_2103_avx2:                                13.8 ( 3.53x)
shuffle_bytes_2103_avx512icl:                            9.2 ( 5.24x)
shuffle_bytes_3012_c:                                   84.5 ( 1.00x)
shuffle_bytes_3012_ssse3:                               14.2 ( 5.93x)
shuffle_bytes_3012_avx2:                                16.2 ( 5.20x)
shuffle_bytes_3012_avx512icl:                           10.2 ( 8.24x)
shuffle_bytes_3210_c:                                   89.2 ( 1.00x)
shuffle_bytes_3210_ssse3:                               24.2 ( 3.68x)
shuffle_bytes_3210_avx2:                                16.2 ( 5.49x)
shuffle_bytes_3210_avx512icl:                            9.2 ( 9.65x)

Signed-off-by: Shreesh Adiga <16567adigashreesh@gmail.com>

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

libswscale/x86/rgb_2_rgb.asm

@@ -57,40 +57,53 @@ SECTION .text
 %macro SHUFFLE_BYTES 4
 cglobal shuffle_bytes_%1%2%3%4, 3, 5, 2, src, dst, w, tmp, x
     VBROADCASTI128    m0, [pb_shuffle%1%2%3%4]
-    movsxdifnidn wq, wd
-    mov xq, wq
-
-    add        srcq, wq
-    add        dstq, wq
-    neg          wq
-
-;calc scalar loop
+    movsxdifnidn      wq, wd
+    mov               xq, wq
+
+    add             srcq, wq
+    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, tmpd
+    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
 
-.loop_scalar:
-   mov          tmpb, [srcq + wq + %1]
-   mov [dstq+wq + 0], tmpb
-   mov          tmpb, [srcq + wq + %2]
-   mov [dstq+wq + 1], tmpb
-   mov          tmpb, [srcq + wq + %3]
-   mov [dstq+wq + 2], tmpb
-   mov          tmpb, [srcq + wq + %4]
-   mov [dstq+wq + 3], tmpb
-   add            wq, 4
-   sub            xq, 4
-   jg .loop_scalar
-
-;check if src_size < mmsize
-cmp wq, 0
-jge .end
-
-.loop_simd:
-    movu           m1, [srcq+wq]
-    pshufb         m1, m0
-    movu    [dstq+wq], m1
-    add            wq, mmsize
-    jl .loop_simd
+    .loop_scalar:
+        mov          tmpb, [srcq + wq + %1]
+        mov [dstq+wq + 0], tmpb
+        mov          tmpb, [srcq + wq + %2]
+        mov [dstq+wq + 1], tmpb
+        mov          tmpb, [srcq + wq + %3]
+        mov [dstq+wq + 2], tmpb
+        mov          tmpb, [srcq + wq + %4]
+        mov [dstq+wq + 3], tmpb
+        add            wq, 4
+        sub            xq, 4
+        jg .loop_scalar
+%endif
+
+    ;check if src_size < mmsize
+    cmp wq, 0
+    jge .end
+
+    .loop_simd:
+        movu            m1, [srcq + wq]
+        pshufb          m1, m0
+        movu   [dstq + wq], m1
+        add             wq, mmsize
+        jl .loop_simd
 
 .end:
     RET
@@ -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,