[X86] Use correct fp immediate types in _mm_set_ss/sd

Avoids implicit sint_to_fp which wasn't occurring on strict fp codegen Fixes llvm#104848
mgabka · Aug 20, 2024 · 6dcce42 · 6dcce42
1 parent 3b49d27
commit 6dcce42
Show file tree

Hide file tree

Showing 3 changed files with 6 additions and 12 deletions.
diff --git a/clang/lib/Headers/emmintrin.h b/clang/lib/Headers/emmintrin.h
@@ -1774,7 +1774,7 @@ static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_undefined_pd(void) {
 ///    lower 64 bits contain the value of the parameter. The upper 64 bits are
 ///    set to zero.
 static __inline__ __m128d __DEFAULT_FN_ATTRS _mm_set_sd(double __w) {
-  return __extension__(__m128d){__w, 0};
+  return __extension__(__m128d){__w, 0.0};
 }
 
 /// Constructs a 128-bit floating-point vector of [2 x double], with each

diff --git a/clang/lib/Headers/xmmintrin.h b/clang/lib/Headers/xmmintrin.h
@@ -1925,7 +1925,7 @@ _mm_undefined_ps(void)
 static __inline__ __m128 __DEFAULT_FN_ATTRS
 _mm_set_ss(float __w)
 {
-  return __extension__ (__m128){ __w, 0, 0, 0 };
+  return __extension__ (__m128){ __w, 0.0f, 0.0f, 0.0f };
 }
 
 /// Constructs a 128-bit floating-point vector of [4 x float], with each

diff --git a/clang/test/CodeGen/X86/strictfp_patterns.c b/clang/test/CodeGen/X86/strictfp_patterns.c
@@ -3,16 +3,12 @@
 
 #include <immintrin.h>
 
-// TODO: PR104848 - ensure the _mm_set_ss/d headers don't implicity promote any integer/fp values.
+// PR104848 - ensure the _mm_set_ss/d headers don't implicity promote any integer/fp values.
 
 // CHECK-LABEL: @test_mm_set_ss(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <4 x float> poison, float [[NUM:%.*]], i64 0
-// CHECK-NEXT:    [[CONV_I:%.*]] = tail call float @llvm.experimental.constrained.sitofp.f32.i32(i32 0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR2:[0-9]+]]
-// CHECK-NEXT:    [[VECINIT1_I:%.*]] = insertelement <4 x float> [[VECINIT_I]], float [[CONV_I]], i64 1
-// CHECK-NEXT:    [[VECINIT3_I:%.*]] = insertelement <4 x float> [[VECINIT1_I]], float [[CONV_I]], i64 2
-// CHECK-NEXT:    [[VECINIT5_I:%.*]] = insertelement <4 x float> [[VECINIT3_I]], float [[CONV_I]], i64 3
-// CHECK-NEXT:    ret <4 x float> [[VECINIT5_I]]
+// CHECK-NEXT:    [[VECINIT3_I:%.*]] = insertelement <4 x float> <float poison, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>, float [[NUM:%.*]], i64 0
+// CHECK-NEXT:    ret <4 x float> [[VECINIT3_I]]
 //
 __m128 test_mm_set_ss(float num)
 {
@@ -21,9 +17,7 @@ __m128 test_mm_set_ss(float num)
 
 // CHECK-LABEL: @test_mm_set_sd(
 // CHECK-NEXT:  entry:
-// CHECK-NEXT:    [[VECINIT_I:%.*]] = insertelement <2 x double> poison, double [[NUM:%.*]], i64 0
-// CHECK-NEXT:    [[CONV_I:%.*]] = tail call double @llvm.experimental.constrained.sitofp.f64.i32(i32 0, metadata !"round.tonearest", metadata !"fpexcept.maytrap") #[[ATTR2]]
-// CHECK-NEXT:    [[VECINIT1_I:%.*]] = insertelement <2 x double> [[VECINIT_I]], double [[CONV_I]], i64 1
+// CHECK-NEXT:    [[VECINIT1_I:%.*]] = insertelement <2 x double> <double poison, double 0.000000e+00>, double [[NUM:%.*]], i64 0
 // CHECK-NEXT:    ret <2 x double> [[VECINIT1_I]]
 //
 __m128d test_mm_set_sd(double num)