arm64: mc: NEON implementation of warp8x8{,t}

Relative speedup vs C code:
                 Cortex A53    A72    A73
warp_8x8_8bpc_neon:    3.19   2.60   3.66
warp_8x8t_8bpc_neon:   3.09   2.50   3.58

1 files changed, 29 insertions, 0 deletions

diff --git a/src/arm/64/util.S b/src/arm/64/util.S
index 1181a6e..0dd4617 100644
--- a/src/arm/64/util.S
+++ b/src/arm/64/util.S
@@ -59,4 +59,33 @@
 #endif
 .endm
 
+.macro transpose_8x8b r0, r1, r2, r3, r4, r5, r6, r7, r8, r9
+    trn1        \r8\().8b,  \r0\().8b,  \r1\().8b
+    trn2        \r9\().8b,  \r0\().8b,  \r1\().8b
+    trn1        \r1\().8b,  \r2\().8b,  \r3\().8b
+    trn2        \r3\().8b,  \r2\().8b,  \r3\().8b
+    trn1        \r0\().8b,  \r4\().8b,  \r5\().8b
+    trn2        \r5\().8b,  \r4\().8b,  \r5\().8b
+    trn1        \r2\().8b,  \r6\().8b,  \r7\().8b
+    trn2        \r7\().8b,  \r6\().8b,  \r7\().8b
+
+    trn1        \r4\().4h,  \r0\().4h,  \r2\().4h
+    trn2        \r2\().4h,  \r0\().4h,  \r2\().4h
+    trn1        \r6\().4h,  \r5\().4h,  \r7\().4h
+    trn2        \r7\().4h,  \r5\().4h,  \r7\().4h
+    trn1        \r5\().4h,  \r9\().4h,  \r3\().4h
+    trn2        \r9\().4h,  \r9\().4h,  \r3\().4h
+    trn1        \r3\().4h,  \r8\().4h,  \r1\().4h
+    trn2        \r8\().4h,  \r8\().4h,  \r1\().4h
+
+    trn1        \r0\().2s,  \r3\().2s,  \r4\().2s
+    trn2        \r4\().2s,  \r3\().2s,  \r4\().2s
+    trn1        \r1\().2s,  \r5\().2s,  \r6\().2s
+    trn2        \r5\().2s,  \r5\().2s,  \r6\().2s
+    trn2        \r6\().2s,  \r8\().2s,  \r2\().2s
+    trn1        \r2\().2s,  \r8\().2s,  \r2\().2s
+    trn1        \r3\().2s,  \r9\().2s,  \r7\().2s
+    trn2        \r7\().2s,  \r9\().2s,  \r7\().2s
+.endm
+
 #endif /* DAVID_SRC_ARM_64_UTIL_S */