[x265] [PATCH v2 1/7] AArch64: Add Neon implementation of 4x4 DST

[chen]

Thank for the patch, my comments inline

At 2024-12-04 23:37:53, "Micro Daryl Robles" <microdaryl.robles at arm.com> wrote:
>Also optimize transpose_4x4_s16 implementation.
>
>diff --git a/source/common/aarch64/dct-prim.cpp b/source/common/aarch64/dct-prim.cpp
>index 8b523ceb0..fcc86f6d6 100644
>--- a/source/common/aarch64/dct-prim.cpp
>+++ b/source/common/aarch64/dct-prim.cpp
>@@ -21,21 +21,23 @@ namespace
> {
> using namespace X265_NS;
> 
>-static void transpose_4x4x16(int16x4_t &x0, int16x4_t &x1, int16x4_t &x2, int16x4_t &x3)
>+static inline void transpose_4x4_s16(int16x4_t &s0, int16x4_t &s1, int16x4_t &s2, int16x4_t &s3)
> {
>-    int32x2_t s0, s1, s2, s3;
>+    int16x8_t s0q = vcombine_s16(s0, vdup_n_s16(0));
>+    int16x8_t s1q = vcombine_s16(s1, vdup_n_s16(0));
>+    int16x8_t s2q = vcombine_s16(s2, vdup_n_s16(0));

>+    int16x8_t s3q = vcombine_s16(s3, vdup_n_s16(0));


Why clear high 64-bits? it will overwrite by ZIP1 below



>+    int16x8x2_t s0123 = vzipq_s16(s02, s13);





>+void dst4_neon(const int16_t *src, int16_t *dst, intptr_t srcStride)
>+{
>+    const int shift_pass1 = 1 + X265_DEPTH - 8;
>+    const int shift_pass2 = 8;
>+
>+    ALIGN_VAR_32(int16_t, coef[4 * 4]);
>+    ALIGN_VAR_32(int16_t, block[4 * 4]);
>+
>+    for (int i = 0; i < 4; i++)
>+    {
>+        memcpy(&block[i * 4], &src[i * srcStride], 4 * sizeof(int16_t));
>+    }

We need not this loop to copy data from input buffer

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20241206/a782e07a/attachment.htm>