On x86 use _mm_set_epi32 when _mm_cvtsi64_si128 isn't available

Change-Id: Ibdbd720d4f68892da6164a9849e212e759305005

diff --git a/aom_dsp/simd/v128_intrinsics_x86.h b/aom_dsp/simd/v128_intrinsics_x86.h
index 45049960231e2448925e22016093106a102ff35e..9a10f2055fadfe2a00a5be4b12e5e95917fef1db 100644 (file)
--- a/aom_dsp/simd/v128_intrinsics_x86.h
+++ b/aom_dsp/simd/v128_intrinsics_x86.h
@@ -162,7 +162,11 @@ SIMD_INLINE v128 v128_unziphi_8(v128 a, v128 b) {
 
 SIMD_INLINE v128 v128_unziplo_8(v128 a, v128 b) {
 #if defined(__SSSE3__)
+#ifdef __x86_64__
   v128 order = _mm_cvtsi64_si128(0x0e0c0a0806040200LL);
+#else
+  v128 order = _mm_set_epi32(0, 0, 0x0e0c0a08, 0x06040200);
+#endif
   return _mm_unpacklo_epi64(_mm_shuffle_epi8(b, order),
                             _mm_shuffle_epi8(a, order));
 #else
@@ -176,7 +180,11 @@ SIMD_INLINE v128 v128_unziphi_16(v128 a, v128 b) {
 
 SIMD_INLINE v128 v128_unziplo_16(v128 a, v128 b) {
 #if defined(__SSSE3__)
+#ifdef __x86_64__
   v128 order = _mm_cvtsi64_si128(0x0d0c090805040100LL);
+#else
+  v128 order = _mm_set_epi32(0, 0, 0x0d0c0908, 0x05040100);
+#endif
   return _mm_unpacklo_epi64(_mm_shuffle_epi8(b, order),
                             _mm_shuffle_epi8(a, order));
 #else

diff --git a/aom_dsp/simd/v64_intrinsics_x86.h b/aom_dsp/simd/v64_intrinsics_x86.h
index b951492fdbb6dfe2c44576ad94854d50e95eb85a..09eb160f53709737091dc60619980894f96f9d61 100644 (file)
--- a/aom_dsp/simd/v64_intrinsics_x86.h
+++ b/aom_dsp/simd/v64_intrinsics_x86.h
@@ -47,7 +47,11 @@ SIMD_INLINE v64 v64_from_32(uint32_t x, uint32_t y) {
 }
 
 SIMD_INLINE v64 v64_from_64(uint64_t x) {
+#ifdef __x86_64__
+  return _mm_cvtsi64_si128(x);
+#else
   return _mm_set_epi32(0, 0, x >> 32, (uint32_t)x);
+#endif
 }
 
 SIMD_INLINE uint64_t v64_u64(v64 x) {
@@ -168,7 +172,7 @@ SIMD_INLINE v64 v64_pack_s16_s8(v64 a, v64 b) {
 SIMD_INLINE v64 v64_unziphi_8(v64 a, v64 b) {
 #if defined(__SSSE3__)
   return _mm_shuffle_epi8(_mm_unpacklo_epi64(b, a),
-                          _mm_cvtsi64_si128(0x0f0d0b0907050301LL));
+                          v64_from_64(0x0f0d0b0907050301LL));
 #else
   return _mm_packus_epi16(
       _mm_unpacklo_epi64(_mm_srli_epi16(b, 8), _mm_srli_epi16(a, 8)),
@@ -179,7 +183,7 @@ SIMD_INLINE v64 v64_unziphi_8(v64 a, v64 b) {
 SIMD_INLINE v64 v64_unziplo_8(v64 a, v64 b) {
 #if defined(__SSSE3__)
   return _mm_shuffle_epi8(_mm_unpacklo_epi64(b, a),
-                          _mm_cvtsi64_si128(0x0e0c0a0806040200LL));
+                          v64_from_64(0x0e0c0a0806040200LL));
 #else
   return v64_unziphi_8(_mm_slli_si128(a, 1), _mm_slli_si128(b, 1));
 #endif
@@ -188,7 +192,7 @@ SIMD_INLINE v64 v64_unziplo_8(v64 a, v64 b) {
 SIMD_INLINE v64 v64_unziphi_16(v64 a, v64 b) {
 #if defined(__SSSE3__)
   return _mm_shuffle_epi8(_mm_unpacklo_epi64(b, a),
-                          _mm_cvtsi64_si128(0x0f0e0b0a07060302LL));
+                          v64_from_64(0x0f0e0b0a07060302LL));
 #else
   return _mm_packs_epi32(
       _mm_unpacklo_epi64(_mm_srai_epi32(b, 16), _mm_srai_epi32(a, 16)),
@@ -199,7 +203,7 @@ SIMD_INLINE v64 v64_unziphi_16(v64 a, v64 b) {
 SIMD_INLINE v64 v64_unziplo_16(v64 a, v64 b) {
 #if defined(__SSSE3__)
   return _mm_shuffle_epi8(_mm_unpacklo_epi64(b, a),
-                          _mm_cvtsi64_si128(0x0d0c090805040100LL));
+                          v64_from_64(0x0d0c090805040100LL));
 #else
   return v64_unziphi_16(_mm_slli_si128(a, 2), _mm_slli_si128(b, 2));
 #endif