1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211
|
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=infer-address-spaces %s | FileCheck %s
; Ports of most of test/CodeGen/NVPTX/access-non-generic.ll
@scalar = internal addrspace(3) global float 0.0, align 4
@array = internal addrspace(3) global [10 x float] zeroinitializer, align 4
define amdgpu_kernel void @load_store_lds_f32(i32 %i, float %v) #0 {
; CHECK-LABEL: define amdgpu_kernel void @load_store_lds_f32(
; CHECK-SAME: i32 [[I:%.*]], float [[V:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT: [[BB:.*:]]
; CHECK-NEXT: [[TMP:%.*]] = load float, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT: call void @use(float [[TMP]])
; CHECK-NEXT: store float [[V]], ptr addrspace(3) @scalar, align 4
; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT: [[TMP2:%.*]] = load float, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT: call void @use(float [[TMP2]])
; CHECK-NEXT: store float [[V]], ptr addrspace(3) @scalar, align 4
; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT: [[TMP3:%.*]] = load float, ptr addrspace(3) getelementptr inbounds ([10 x float], ptr addrspace(3) @array, i32 0, i32 5), align 4
; CHECK-NEXT: call void @use(float [[TMP3]])
; CHECK-NEXT: store float [[V]], ptr addrspace(3) getelementptr inbounds ([10 x float], ptr addrspace(3) @array, i32 0, i32 5), align 4
; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [10 x float], ptr addrspace(3) @array, i32 0, i32 5
; CHECK-NEXT: [[TMP5:%.*]] = load float, ptr addrspace(3) [[TMP4]], align 4
; CHECK-NEXT: call void @use(float [[TMP5]])
; CHECK-NEXT: store float [[V]], ptr addrspace(3) [[TMP4]], align 4
; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [10 x float], ptr addrspace(3) @array, i32 0, i32 [[I]]
; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr addrspace(3) [[TMP7]], align 4
; CHECK-NEXT: call void @use(float [[TMP8]])
; CHECK-NEXT: store float [[V]], ptr addrspace(3) [[TMP7]], align 4
; CHECK-NEXT: call void @llvm.amdgcn.s.barrier()
; CHECK-NEXT: ret void
;
bb:
%tmp = load float, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
call void @use(float %tmp)
store float %v, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
call void @llvm.amdgcn.s.barrier()
%tmp1 = addrspacecast ptr addrspace(3) @scalar to ptr
%tmp2 = load float, ptr %tmp1, align 4
call void @use(float %tmp2)
store float %v, ptr %tmp1, align 4
call void @llvm.amdgcn.s.barrier()
%tmp3 = load float, ptr getelementptr inbounds ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5), align 4
call void @use(float %tmp3)
store float %v, ptr getelementptr inbounds ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5), align 4
call void @llvm.amdgcn.s.barrier()
%tmp4 = getelementptr inbounds [10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i32 0, i32 5
%tmp5 = load float, ptr %tmp4, align 4
call void @use(float %tmp5)
store float %v, ptr %tmp4, align 4
call void @llvm.amdgcn.s.barrier()
%tmp6 = addrspacecast ptr addrspace(3) @array to ptr
%tmp7 = getelementptr inbounds [10 x float], ptr %tmp6, i32 0, i32 %i
%tmp8 = load float, ptr %tmp7, align 4
call void @use(float %tmp8)
store float %v, ptr %tmp7, align 4
call void @llvm.amdgcn.s.barrier()
ret void
}
define i32 @constexpr_load_int_from_float_lds() #0 {
; CHECK-LABEL: define i32 @constexpr_load_int_from_float_lds(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[BB:.*:]]
; CHECK-NEXT: [[TMP:%.*]] = load i32, ptr addrspace(3) @scalar, align 4
; CHECK-NEXT: ret i32 [[TMP]]
;
bb:
%tmp = load i32, ptr addrspacecast (ptr addrspace(3) @scalar to ptr), align 4
ret i32 %tmp
}
define i32 @load_int_from_global_float(ptr addrspace(1) %input, i32 %i, i32 %j) #0 {
; CHECK-LABEL: define i32 @load_int_from_global_float(
; CHECK-SAME: ptr addrspace(1) [[INPUT:%.*]], i32 [[I:%.*]], i32 [[J:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[BB:.*:]]
; CHECK-NEXT: [[TMP1:%.*]] = getelementptr float, ptr addrspace(1) [[INPUT]], i32 [[I]]
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr float, ptr addrspace(1) [[TMP1]], i32 [[J]]
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(1) [[TMP2]], align 4
; CHECK-NEXT: ret i32 [[TMP4]]
;
bb:
%tmp = addrspacecast ptr addrspace(1) %input to ptr
%tmp1 = getelementptr float, ptr %tmp, i32 %i
%tmp2 = getelementptr float, ptr %tmp1, i32 %j
%tmp4 = load i32, ptr %tmp2
ret i32 %tmp4
}
define amdgpu_kernel void @nested_const_expr() #0 {
; CHECK-LABEL: define amdgpu_kernel void @nested_const_expr(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: store i32 1, ptr addrspace(3) getelementptr ([10 x float], ptr addrspace(3) @array, i64 0, i64 1), align 4
; CHECK-NEXT: ret void
;
store i32 1, ptr bitcast (ptr getelementptr ([10 x float], ptr addrspacecast (ptr addrspace(3) @array to ptr), i64 0, i64 1) to ptr), align 4
ret void
}
define amdgpu_kernel void @rauw(ptr addrspace(1) %input) #0 {
; CHECK-LABEL: define amdgpu_kernel void @rauw(
; CHECK-SAME: ptr addrspace(1) [[INPUT:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[BB:.*:]]
; CHECK-NEXT: [[ADDR:%.*]] = getelementptr float, ptr addrspace(1) [[INPUT]], i64 10
; CHECK-NEXT: [[V:%.*]] = load float, ptr addrspace(1) [[ADDR]], align 4
; CHECK-NEXT: store float [[V]], ptr addrspace(1) [[ADDR]], align 4
; CHECK-NEXT: ret void
;
bb:
%generic_input = addrspacecast ptr addrspace(1) %input to ptr
%addr = getelementptr float, ptr %generic_input, i64 10
%v = load float, ptr %addr
store float %v, ptr %addr
ret void
}
; FIXME: Should be able to eliminate the cast inside the loop
define amdgpu_kernel void @loop() #0 {
; CHECK-LABEL: define amdgpu_kernel void @loop(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[END:%.*]] = getelementptr float, ptr addrspace(3) @array, i64 10
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[I:%.*]] = phi ptr addrspace(3) [ @array, %[[ENTRY]] ], [ [[I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT: [[V:%.*]] = load float, ptr addrspace(3) [[I]], align 4
; CHECK-NEXT: call void @use(float [[V]])
; CHECK-NEXT: [[I2]] = getelementptr float, ptr addrspace(3) [[I]], i64 1
; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp eq ptr addrspace(3) [[I2]], [[END]]
; CHECK-NEXT: br i1 [[EXIT_COND]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
entry:
%p = addrspacecast ptr addrspace(3) @array to ptr
%end = getelementptr float, ptr %p, i64 10
br label %loop
loop: ; preds = %loop, %entry
%i = phi ptr [ %p, %entry ], [ %i2, %loop ]
%v = load float, ptr %i
call void @use(float %v)
%i2 = getelementptr float, ptr %i, i64 1
%exit_cond = icmp eq ptr %i2, %end
br i1 %exit_cond, label %exit, label %loop
exit: ; preds = %loop
ret void
}
@generic_end = external addrspace(1) global ptr
define amdgpu_kernel void @loop_with_generic_bound() #0 {
; CHECK-LABEL: define amdgpu_kernel void @loop_with_generic_bound(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[END:%.*]] = load ptr, ptr addrspace(1) @generic_end, align 8
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[I:%.*]] = phi ptr addrspace(3) [ @array, %[[ENTRY]] ], [ [[I2:%.*]], %[[LOOP]] ]
; CHECK-NEXT: [[V:%.*]] = load float, ptr addrspace(3) [[I]], align 4
; CHECK-NEXT: call void @use(float [[V]])
; CHECK-NEXT: [[I2]] = getelementptr float, ptr addrspace(3) [[I]], i64 1
; CHECK-NEXT: [[TMP0:%.*]] = addrspacecast ptr addrspace(3) [[I2]] to ptr
; CHECK-NEXT: [[EXIT_COND:%.*]] = icmp eq ptr [[TMP0]], [[END]]
; CHECK-NEXT: br i1 [[EXIT_COND]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
entry:
%p = addrspacecast ptr addrspace(3) @array to ptr
%end = load ptr, ptr addrspace(1) @generic_end
br label %loop
loop: ; preds = %loop, %entry
%i = phi ptr [ %p, %entry ], [ %i2, %loop ]
%v = load float, ptr %i
call void @use(float %v)
%i2 = getelementptr float, ptr %i, i64 1
%exit_cond = icmp eq ptr %i2, %end
br i1 %exit_cond, label %exit, label %loop
exit: ; preds = %loop
ret void
}
define void @select_bug() #0 {
; CHECK-LABEL: define void @select_bug(
; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[CMP:%.*]] = icmp ne ptr inttoptr (i64 4873 to ptr), null
; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], i64 73, i64 93
; CHECK-NEXT: [[ADD_PTR157:%.*]] = getelementptr inbounds i64, ptr undef, i64 [[SEL]]
; CHECK-NEXT: [[CMP169:%.*]] = icmp uge ptr undef, [[ADD_PTR157]]
; CHECK-NEXT: unreachable
;
%cmp = icmp ne ptr inttoptr (i64 4873 to ptr), null
%sel = select i1 %cmp, i64 73, i64 93
%add.ptr157 = getelementptr inbounds i64, ptr undef, i64 %sel
%cmp169 = icmp uge ptr undef, %add.ptr157
unreachable
}
declare void @llvm.amdgcn.s.barrier() #1
declare void @use(float) #0
attributes #0 = { nounwind }
attributes #1 = { convergent nounwind }
|