1 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck %s 2 3 ; CHECK: 'extractelement_v2i32' 4 ; CHECK: estimated cost of 0 for {{.*}} extractelement <2 x i32> 5 define void @extractelement_v2i32(i32 addrspace(1)* %out, <2 x i32> addrspace(1)* %vaddr) { 6 %vec = load <2 x i32>, <2 x i32> addrspace(1)* %vaddr 7 %elt = extractelement <2 x i32> %vec, i32 1 8 store i32 %elt, i32 addrspace(1)* %out 9 ret void 10 } 11 12 ; CHECK: 'extractelement_v2f32' 13 ; CHECK: estimated cost of 0 for {{.*}} extractelement <2 x float> 14 define void @extractelement_v2f32(float addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr) { 15 %vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr 16 %elt = extractelement <2 x float> %vec, i32 1 17 store float %elt, float addrspace(1)* %out 18 ret void 19 } 20 21 ; CHECK: 'extractelement_v3i32' 22 ; CHECK: estimated cost of 0 for {{.*}} extractelement <3 x i32> 23 define void @extractelement_v3i32(i32 addrspace(1)* %out, <3 x i32> addrspace(1)* %vaddr) { 24 %vec = load <3 x i32>, <3 x i32> addrspace(1)* %vaddr 25 %elt = extractelement <3 x i32> %vec, i32 1 26 store i32 %elt, i32 addrspace(1)* %out 27 ret void 28 } 29 30 ; CHECK: 'extractelement_v4i32' 31 ; CHECK: estimated cost of 0 for {{.*}} extractelement <4 x i32> 32 define void @extractelement_v4i32(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %vaddr) { 33 %vec = load <4 x i32>, <4 x i32> addrspace(1)* %vaddr 34 %elt = extractelement <4 x i32> %vec, i32 1 35 store i32 %elt, i32 addrspace(1)* %out 36 ret void 37 } 38 39 ; CHECK: 'extractelement_v8i32' 40 ; CHECK: estimated cost of 0 for {{.*}} extractelement <8 x i32> 41 define void @extractelement_v8i32(i32 addrspace(1)* %out, <8 x i32> addrspace(1)* %vaddr) { 42 %vec = load <8 x i32>, <8 x i32> addrspace(1)* %vaddr 43 %elt = extractelement <8 x i32> %vec, i32 1 44 store i32 %elt, i32 addrspace(1)* %out 45 ret void 46 } 47 48 ; FIXME: Should be non-0 49 ; CHECK: 'extractelement_v8i32_dynindex' 50 ; CHECK: estimated cost of 2 for {{.*}} extractelement <8 x i32> 51 define void @extractelement_v8i32_dynindex(i32 addrspace(1)* %out, <8 x i32> addrspace(1)* %vaddr, i32 %idx) { 52 %vec = load <8 x i32>, <8 x i32> addrspace(1)* %vaddr 53 %elt = extractelement <8 x i32> %vec, i32 %idx 54 store i32 %elt, i32 addrspace(1)* %out 55 ret void 56 } 57 58 ; CHECK: 'extractelement_v2i64' 59 ; CHECK: estimated cost of 0 for {{.*}} extractelement <2 x i64> 60 define void @extractelement_v2i64(i64 addrspace(1)* %out, <2 x i64> addrspace(1)* %vaddr) { 61 %vec = load <2 x i64>, <2 x i64> addrspace(1)* %vaddr 62 %elt = extractelement <2 x i64> %vec, i64 1 63 store i64 %elt, i64 addrspace(1)* %out 64 ret void 65 } 66 67 ; CHECK: 'extractelement_v3i64' 68 ; CHECK: estimated cost of 0 for {{.*}} extractelement <3 x i64> 69 define void @extractelement_v3i64(i64 addrspace(1)* %out, <3 x i64> addrspace(1)* %vaddr) { 70 %vec = load <3 x i64>, <3 x i64> addrspace(1)* %vaddr 71 %elt = extractelement <3 x i64> %vec, i64 1 72 store i64 %elt, i64 addrspace(1)* %out 73 ret void 74 } 75 76 ; CHECK: 'extractelement_v4i64' 77 ; CHECK: estimated cost of 0 for {{.*}} extractelement <4 x i64> 78 define void @extractelement_v4i64(i64 addrspace(1)* %out, <4 x i64> addrspace(1)* %vaddr) { 79 %vec = load <4 x i64>, <4 x i64> addrspace(1)* %vaddr 80 %elt = extractelement <4 x i64> %vec, i64 1 81 store i64 %elt, i64 addrspace(1)* %out 82 ret void 83 } 84 85 ; CHECK: 'extractelement_v8i64' 86 ; CHECK: estimated cost of 0 for {{.*}} extractelement <8 x i64> 87 define void @extractelement_v8i64(i64 addrspace(1)* %out, <8 x i64> addrspace(1)* %vaddr) { 88 %vec = load <8 x i64>, <8 x i64> addrspace(1)* %vaddr 89 %elt = extractelement <8 x i64> %vec, i64 1 90 store i64 %elt, i64 addrspace(1)* %out 91 ret void 92 } 93 94 ; CHECK: 'extractelement_v4i8' 95 ; CHECK: estimated cost of 0 for {{.*}} extractelement <4 x i8> 96 define void @extractelement_v4i8(i8 addrspace(1)* %out, <4 x i8> addrspace(1)* %vaddr) { 97 %vec = load <4 x i8>, <4 x i8> addrspace(1)* %vaddr 98 %elt = extractelement <4 x i8> %vec, i8 1 99 store i8 %elt, i8 addrspace(1)* %out 100 ret void 101 } 102 103 ; CHECK: 'extractelement_v2i16' 104 ; CHECK: estimated cost of 0 for {{.*}} extractelement <2 x i16> 105 define void @extractelement_v2i16(i16 addrspace(1)* %out, <2 x i16> addrspace(1)* %vaddr) { 106 %vec = load <2 x i16>, <2 x i16> addrspace(1)* %vaddr 107 %elt = extractelement <2 x i16> %vec, i16 1 108 store i16 %elt, i16 addrspace(1)* %out 109 ret void 110 } 111
