[clang] ee815ff - [PowerPC] Fix the implicit casting for the emulated intrinsics

[Maryam Moghadas via cfe-commits]

Author: Maryam Moghadas
Date: 2023-02-21T15:44:57-06:00
New Revision: ee815ff2ce8d0782cf04f5cad1d4038207f661cf

URL: https://github.com/llvm/llvm-project/commit/ee815ff2ce8d0782cf04f5cad1d4038207f661cf
DIFF: https://github.com/llvm/llvm-project/commit/ee815ff2ce8d0782cf04f5cad1d4038207f661cf.diff

LOG: [PowerPC] Fix the implicit casting for the emulated intrinsics

This patch is to fix some implicit castings for emulated intrinsics
so that there are no lax-vector-conversions errors and warnings.

Reviewed By: nemanjai, #powerpc

Differential Revision: https://reviews.llvm.org/D144293

Added: 
    

Modified: 
    clang/lib/Headers/ppc_wrappers/emmintrin.h
    clang/lib/Headers/ppc_wrappers/smmintrin.h
    clang/test/CodeGen/PowerPC/ppc-smmintrin.c

Removed: 
    


################################################################################
diff  --git a/clang/lib/Headers/ppc_wrappers/emmintrin.h b/clang/lib/Headers/ppc_wrappers/emmintrin.h
index 0814ea5593bad..fc18ab9d43b15 100644
--- a/clang/lib/Headers/ppc_wrappers/emmintrin.h
+++ b/clang/lib/Headers/ppc_wrappers/emmintrin.h
@@ -46,6 +46,7 @@
 
 /* SSE2 */
 typedef __vector double __v2df;
+typedef __vector float __v4f;
 typedef __vector long long __v2di;
 typedef __vector unsigned long long __v2du;
 typedef __vector int __v4si;
@@ -951,7 +952,7 @@ extern __inline __m128d
     _mm_cvtpi32_pd(__m64 __A) {
   __v4si __temp;
   __v2di __tmp2;
-  __v2df __result;
+  __v4f __result;
 
   __temp = (__v4si)vec_splats(__A);
   __tmp2 = (__v2di)vec_unpackl(__temp);

diff  --git a/clang/lib/Headers/ppc_wrappers/smmintrin.h b/clang/lib/Headers/ppc_wrappers/smmintrin.h
index 6fe6d2a157a59..349b395c4f00b 100644
--- a/clang/lib/Headers/ppc_wrappers/smmintrin.h
+++ b/clang/lib/Headers/ppc_wrappers/smmintrin.h
@@ -305,9 +305,9 @@ extern __inline int
 extern __inline __m128i
     __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     _mm_blend_epi16(__m128i __A, __m128i __B, const int __imm8) {
-  __v16qi __charmask = vec_splats((signed char)__imm8);
+  __v16qu __charmask = vec_splats((unsigned char)__imm8);
   __charmask = vec_gb(__charmask);
-  __v8hu __shortmask = (__v8hu)vec_unpackh(__charmask);
+  __v8hu __shortmask = (__v8hu)vec_unpackh((__v16qi)__charmask);
 #ifdef __BIG_ENDIAN__
   __shortmask = vec_reve(__shortmask);
 #endif

diff  --git a/clang/test/CodeGen/PowerPC/ppc-smmintrin.c b/clang/test/CodeGen/PowerPC/ppc-smmintrin.c
index 5033526fd3369..220b65c1ce164 100644
--- a/clang/test/CodeGen/PowerPC/ppc-smmintrin.c
+++ b/clang/test/CodeGen/PowerPC/ppc-smmintrin.c
@@ -73,7 +73,7 @@ test_blend() {
 
 // CHECK-LABEL: define available_externally <2 x i64> @_mm_blend_epi16(<2 x i64> noundef %{{[0-9a-zA-Z_.]+}}, <2 x i64> noundef %{{[0-9a-zA-Z_.]+}}, i32 noundef signext %{{[0-9a-zA-Z_.]+}})
 // CHECK: %[[TRUNC:[0-9a-zA-Z_.]+]] = trunc i32 %{{[0-9a-zA-Z_.]+}} to i8
-// CHECK: call <16 x i8> @vec_splats(signed char)(i8 noundef signext %[[TRUNC]])
+// CHECK: call <16 x i8> @vec_splats(unsigned char)(i8 noundef zeroext %[[TRUNC]])
 // CHECK: call <16 x i8> @llvm.ppc.altivec.vgbbd(<16 x i8> %{{[0-9a-zA-Z_.]+}})
 // CHECK: %[[PACK:[0-9a-zA-Z_.]+]] = call <8 x i16> @vec_unpackh(signed char vector[16])
 // CHECK: store <8 x i16> %[[PACK]], ptr %{{[0-9a-zA-Z_.]+}}, align 16
@@ -232,8 +232,8 @@ void __attribute__((noinline))
 test_round() {
   _mm_round_ps(mn1, 0);
   _mm_round_ss(mn1, mn2, 0);
-  _mm_round_pd(mn1, 0);
-  _mm_round_sd(mn1, mn2, 0);
+  _mm_round_pd(md1, 0);
+  _mm_round_sd(md1, md2, 0);
 }
 
 // CHECK-LABEL: @test_round