1 ; RUN: opt -S -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -amdgpu-promote-alloca < %s | FileCheck %s 2 3 ; This normally would be fixed by instcombine to be compare to the GEP 4 ; indices 5 6 ; CHECK-LABEL: @lds_promoted_alloca_icmp_same_derived_pointer( 7 ; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_same_derived_pointer.alloca, i32 0, i32 %{{[0-9]+}} 8 ; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a 9 ; CHECK: %ptr1 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %b 10 ; CHECK: %cmp = icmp eq i32 addrspace(3)* %ptr0, %ptr1 11 define void @lds_promoted_alloca_icmp_same_derived_pointer(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 { 12 %alloca = alloca [16 x i32], align 4 13 %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 14 %ptr1 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %b 15 %cmp = icmp eq i32* %ptr0, %ptr1 16 %zext = zext i1 %cmp to i32 17 store volatile i32 %zext, i32 addrspace(1)* %out 18 ret void 19 } 20 21 ; CHECK-LABEL: @lds_promoted_alloca_icmp_null_rhs( 22 ; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_rhs.alloca, i32 0, i32 %{{[0-9]+}} 23 ; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a 24 ; CHECK: %cmp = icmp eq i32 addrspace(3)* %ptr0, null 25 define void @lds_promoted_alloca_icmp_null_rhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 { 26 %alloca = alloca [16 x i32], align 4 27 %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 28 %cmp = icmp eq i32* %ptr0, null 29 %zext = zext i1 %cmp to i32 30 store volatile i32 %zext, i32 addrspace(1)* %out 31 ret void 32 } 33 34 ; CHECK-LABEL: @lds_promoted_alloca_icmp_null_lhs( 35 ; CHECK: [[ARRAYGEP:%[0-9]+]] = getelementptr inbounds [256 x [16 x i32]], [256 x [16 x i32]] addrspace(3)* @lds_promoted_alloca_icmp_null_lhs.alloca, i32 0, i32 %{{[0-9]+}} 36 ; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32] addrspace(3)* [[ARRAYGEP]], i32 0, i32 %a 37 ; CHECK: %cmp = icmp eq i32 addrspace(3)* null, %ptr0 38 define void @lds_promoted_alloca_icmp_null_lhs(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 { 39 %alloca = alloca [16 x i32], align 4 40 %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 41 %cmp = icmp eq i32* null, %ptr0 42 %zext = zext i1 %cmp to i32 43 store volatile i32 %zext, i32 addrspace(1)* %out 44 ret void 45 } 46 47 ; CHECK-LABEL: @lds_promoted_alloca_icmp_unknown_ptr( 48 ; CHECK: %alloca = alloca [16 x i32], align 4 49 ; CHECK: %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 50 ; CHECK: %ptr1 = call i32* @get_unknown_pointer() 51 ; CHECK: %cmp = icmp eq i32* %ptr0, %ptr1 52 define void @lds_promoted_alloca_icmp_unknown_ptr(i32 addrspace(1)* %out, i32 %a, i32 %b) #0 { 53 %alloca = alloca [16 x i32], align 4 54 %ptr0 = getelementptr inbounds [16 x i32], [16 x i32]* %alloca, i32 0, i32 %a 55 %ptr1 = call i32* @get_unknown_pointer() 56 %cmp = icmp eq i32* %ptr0, %ptr1 57 %zext = zext i1 %cmp to i32 58 store volatile i32 %zext, i32 addrspace(1)* %out 59 ret void 60 } 61 62 declare i32* @get_unknown_pointer() #0 63 64 attributes #0 = { nounwind "amdgpu-max-waves-per-eu"="1" } 65
