neon,load_unaligned_*: use dup for lane 0

this produces better assembly with gcc (11.3.0-3); no change in assembly
using clang from the r24 android sdk (Android (8075178, based on
r437112b) clang version 14.0.1
(https://android.googlesource.com/toolchain/llvm-project
8671348b81b95fc603505dfc881b45103bee1731)

Change-Id: Ifec252d4f499f23be1cd94aa8516caf6b3fbbc11

diff --git a/vpx_dsp/arm/mem_neon.h b/vpx_dsp/arm/mem_neon.h
index 50aaa94fe0919d34b50696b9336ff783c1ff6544..84aae161b3cadb18c21138a3cafd6ca4adc2b54e 100644 (file)
--- a/vpx_dsp/arm/mem_neon.h
+++ b/vpx_dsp/arm/mem_neon.h
@@ -116,11 +116,11 @@ static INLINE void uint32_to_mem(uint8_t *buf, uint32_t a) {
 static INLINE uint8x8_t load_unaligned_u8(const uint8_t *buf,
                                           ptrdiff_t stride) {
   uint32_t a;
-  uint32x2_t a_u32 = vdup_n_u32(0);
+  uint32x2_t a_u32;
   if (stride == 4) return vld1_u8(buf);
   memcpy(&a, buf, 4);
   buf += stride;
-  a_u32 = vset_lane_u32(a, a_u32, 0);
+  a_u32 = vdup_n_u32(a);
   memcpy(&a, buf, 4);
   a_u32 = vset_lane_u32(a, a_u32, 1);
   return vreinterpret_u8_u32(a_u32);
@@ -143,11 +143,11 @@ static INLINE void store_unaligned_u8(uint8_t *buf, ptrdiff_t stride,
 static INLINE uint8x16_t load_unaligned_u8q(const uint8_t *buf,
                                             ptrdiff_t stride) {
   uint32_t a;
-  uint32x4_t a_u32 = vdupq_n_u32(0);
+  uint32x4_t a_u32;
   if (stride == 4) return vld1q_u8(buf);
   memcpy(&a, buf, 4);
   buf += stride;
-  a_u32 = vsetq_lane_u32(a, a_u32, 0);
+  a_u32 = vdupq_n_u32(a);
   memcpy(&a, buf, 4);
   buf += stride;
   a_u32 = vsetq_lane_u32(a, a_u32, 1);

diff --git a/vpx_dsp/arm/sad4d_neon.c b/vpx_dsp/arm/sad4d_neon.c
index 03f716c3d58e62db1e3a9c827cc64bc4b9f79ad4..53866296ce8d064575218da6e5f210ca6dd3266b 100644 (file)
--- a/vpx_dsp/arm/sad4d_neon.c
+++ b/vpx_dsp/arm/sad4d_neon.c
@@ -20,9 +20,9 @@
 static INLINE uint8x8_t load_unaligned_2_buffers(const void *const buf0,
                                                  const void *const buf1) {
   uint32_t a;
-  uint32x2_t aa = vdup_n_u32(0);
+  uint32x2_t aa;
   memcpy(&a, buf0, 4);
-  aa = vset_lane_u32(a, aa, 0);
+  aa = vdup_n_u32(a);
   memcpy(&a, buf1, 4);
   aa = vset_lane_u32(a, aa, 1);
   return vreinterpret_u8_u32(aa);