1 ; RUN: llc < %s -mtriple=x86_64-apple-darwin | FileCheck %s 2 3 define <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. 23 define 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 ; FIXME: The operation gets scalarized. If/when the compiler learns to better 31 ; use [V]CVTDQ2PD, this will need updated. 32 ; CHECK: cvtsi2sdq 33 ; CHECK: cvtsi2sdq 34 ; CHECK: cvtsi2sdq 35 ; CHECK: cvtsi2sdq 36 %cmp = fcmp oeq <4 x float> %val, %test 37 %ext = zext <4 x i1> %cmp to <4 x i32> 38 %result = sitofp <4 x i32> %ext to <4 x double> 39 store <4 x double> %result, <4 x double>* %p 40 ret void 41 } 42 43 ; Also test the general purpose constant folding of int->fp. 44 define void @foo2(<4 x float>* noalias %result) nounwind { 45 ; CHECK-LABEL: LCPI2_0: 46 ; CHECK-NEXT: .long 1082130432 ## float 4.000000e+00 47 ; CHECK-NEXT: .long 1084227584 ## float 5.000000e+00 48 ; CHECK-NEXT: .long 1086324736 ## float 6.000000e+00 49 ; CHECK-NEXT: .long 1088421888 ## float 7.000000e+00 50 ; CHECK-LABEL: foo2: 51 ; CHECK: movaps LCPI2_0(%rip), %xmm0 52 53 %val = uitofp <4 x i32> <i32 4, i32 5, i32 6, i32 7> to <4 x float> 54 store <4 x float> %val, <4 x float>* %result 55 ret void 56 } 57 58 ; Fold explicit AND operations when the constant isn't a splat of a single 59 ; scalar value like what the zext creates. 60 define <4 x float> @foo3(<4 x float> %val, <4 x float> %test) nounwind { 61 ; CHECK-LABEL: LCPI3_0: 62 ; CHECK-NEXT: .long 1065353216 ## 0x3f800000 63 ; CHECK-NEXT: .long 0 ## 0x0 64 ; CHECK-NEXT: .long 1065353216 ## 0x3f800000 65 ; CHECK-NEXT: .long 0 ## 0x0 66 ; CHECK-LABEL: foo3: 67 ; CHECK: cmpeqps %xmm1, %xmm0 68 ; CHECK-NEXT: andps LCPI3_0(%rip), %xmm0 69 %cmp = fcmp oeq <4 x float> %val, %test 70 %ext = zext <4 x i1> %cmp to <4 x i32> 71 %and = and <4 x i32> %ext, <i32 255, i32 256, i32 257, i32 258> 72 %result = sitofp <4 x i32> %and to <4 x float> 73 ret <4 x float> %result 74 } 75 76 ; Test the general purpose constant folding of uint->fp. 77 define void @foo4(<4 x float>* noalias %result) nounwind { 78 ; CHECK-LABEL: LCPI4_0: 79 ; CHECK-NEXT: .long 1065353216 ## float 1.000000e+00 80 ; CHECK-NEXT: .long 1123942400 ## float 1.270000e+02 81 ; CHECK-NEXT: .long 1124073472 ## float 1.280000e+02 82 ; CHECK-NEXT: .long 1132396544 ## float 2.550000e+02 83 ; CHECK-LABEL: foo4: 84 ; CHECK: movaps LCPI4_0(%rip), %xmm0 85 86 %val = uitofp <4 x i8> <i8 1, i8 127, i8 -128, i8 -1> to <4 x float> 87 store <4 x float> %val, <4 x float>* %result 88 ret void 89 } 90
