1; RUN: llc < %s -mtriple=x86_64-apple-darwin | FileCheck %s 2 3define <4 x float> @foo(<4 x float> %val, <4 x float> %test) nounwind { 4; CHECK-LABEL: LCPI0_0: 5; CHECK-NEXT: .long 1065353216 ## 0x3f800000 6; CHECK-NEXT: .long 1065353216 ## 0x3f800000 7; CHECK-NEXT: .long 1065353216 ## 0x3f800000 8; CHECK-NEXT: .long 1065353216 ## 0x3f800000 9; CHECK-LABEL: foo: 10; CHECK: cmpeqps %xmm1, %xmm0 11; CHECK-NEXT: andps LCPI0_0(%rip), %xmm0 12; CHECK-NEXT: retq 13 14 %cmp = fcmp oeq <4 x float> %val, %test 15 %ext = zext <4 x i1> %cmp to <4 x i32> 16 %result = sitofp <4 x i32> %ext to <4 x float> 17 ret <4 x float> %result 18} 19 20; Make sure the operation doesn't try to get folded when the sizes don't match, 21; as that ends up crashing later when trying to form a bitcast operation for 22; the folded nodes. 23define void @foo1(<4 x float> %val, <4 x float> %test, <4 x double>* %p) nounwind { 24; CHECK-LABEL: LCPI1_0: 25; CHECK-NEXT: .long 1 ## 0x1 26; CHECK-NEXT: .long 1 ## 0x1 27; CHECK-NEXT: .long 1 ## 0x1 28; CHECK-NEXT: .long 1 ## 0x1 29; CHECK-LABEL: foo1: 30; CHECK: cvtdq2pd 31; CHECK: cvtdq2pd 32 %cmp = fcmp oeq <4 x float> %val, %test 33 %ext = zext <4 x i1> %cmp to <4 x i32> 34 %result = sitofp <4 x i32> %ext to <4 x double> 35 store <4 x double> %result, <4 x double>* %p 36 ret void 37} 38 39; Also test the general purpose constant folding of int->fp. 40define void @foo2(<4 x float>* noalias %result) nounwind { 41; CHECK-LABEL: LCPI2_0: 42; CHECK-NEXT: .long 1082130432 ## float 4 43; CHECK-NEXT: .long 1084227584 ## float 5 44; CHECK-NEXT: .long 1086324736 ## float 6 45; CHECK-NEXT: .long 1088421888 ## float 7 46; CHECK-LABEL: foo2: 47; CHECK: movaps LCPI2_0(%rip), %xmm0 48 49 %val = uitofp <4 x i32> <i32 4, i32 5, i32 6, i32 7> to <4 x float> 50 store <4 x float> %val, <4 x float>* %result 51 ret void 52} 53 54; Fold explicit AND operations when the constant isn't a splat of a single 55; scalar value like what the zext creates. 56define <4 x float> @foo3(<4 x float> %val, <4 x float> %test) nounwind { 57; CHECK-LABEL: LCPI3_0: 58; CHECK-NEXT: .long 1065353216 ## 0x3f800000 59; CHECK-NEXT: .long 0 ## 0x0 60; CHECK-NEXT: .long 1065353216 ## 0x3f800000 61; CHECK-NEXT: .long 0 ## 0x0 62; CHECK-LABEL: foo3: 63; CHECK: cmpeqps %xmm1, %xmm0 64; CHECK-NEXT: andps LCPI3_0(%rip), %xmm0 65 %cmp = fcmp oeq <4 x float> %val, %test 66 %ext = zext <4 x i1> %cmp to <4 x i32> 67 %and = and <4 x i32> %ext, <i32 255, i32 256, i32 257, i32 258> 68 %result = sitofp <4 x i32> %and to <4 x float> 69 ret <4 x float> %result 70} 71 72; Test the general purpose constant folding of uint->fp. 73define void @foo4(<4 x float>* noalias %result) nounwind { 74; CHECK-LABEL: LCPI4_0: 75; CHECK-NEXT: .long 1065353216 ## float 1 76; CHECK-NEXT: .long 1123942400 ## float 127 77; CHECK-NEXT: .long 1124073472 ## float 128 78; CHECK-NEXT: .long 1132396544 ## float 255 79; CHECK-LABEL: foo4: 80; CHECK: movaps LCPI4_0(%rip), %xmm0 81 82 %val = uitofp <4 x i8> <i8 1, i8 127, i8 -128, i8 -1> to <4 x float> 83 store <4 x float> %val, <4 x float>* %result 84 ret void 85} 86