1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -amdgpu-late-codegenprepare %s | FileCheck %s -check-prefix=GFX9 3; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 -amdgpu-late-codegenprepare %s | FileCheck %s -check-prefix=GFX12 4; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -passes=amdgpu-late-codegenprepare %s | FileCheck %s -check-prefix=GFX9 5 6; Make sure we don't crash when trying to create a bitcast between 7; address spaces 8define amdgpu_kernel void @constant_from_offset_cast_generic_null() { 9; GFX9-LABEL: @constant_from_offset_cast_generic_null( 10; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 4), align 4 11; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16 12; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8 13; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1 14; GFX9-NEXT: ret void 15; 16; GFX12-LABEL: @constant_from_offset_cast_generic_null( 17; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 6), align 1 18; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 19; GFX12-NEXT: ret void 20; 21 %load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr null to ptr addrspace(4)), i64 6), align 1 22 store i8 %load, ptr addrspace(1) undef 23 ret void 24} 25 26define amdgpu_kernel void @constant_from_offset_cast_global_null() { 27; GFX9-LABEL: @constant_from_offset_cast_global_null( 28; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 4), align 4 29; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16 30; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8 31; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1 32; GFX9-NEXT: ret void 33; 34; GFX12-LABEL: @constant_from_offset_cast_global_null( 35; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 6), align 1 36; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 37; GFX12-NEXT: ret void 38; 39 %load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) null to ptr addrspace(4)), i64 6), align 1 40 store i8 %load, ptr addrspace(1) undef 41 ret void 42} 43 44@gv = unnamed_addr addrspace(1) global [64 x i8] undef, align 4 45 46define amdgpu_kernel void @constant_from_offset_cast_global_gv() { 47; GFX9-LABEL: @constant_from_offset_cast_global_gv( 48; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 4), align 4 49; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16 50; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8 51; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1 52; GFX9-NEXT: ret void 53; 54; GFX12-LABEL: @constant_from_offset_cast_global_gv( 55; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 6), align 1 56; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 57; GFX12-NEXT: ret void 58; 59 %load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr addrspace(1) @gv to ptr addrspace(4)), i64 6), align 1 60 store i8 %load, ptr addrspace(1) undef 61 ret void 62} 63 64define amdgpu_kernel void @constant_from_offset_cast_generic_inttoptr() { 65; GFX9-LABEL: @constant_from_offset_cast_generic_inttoptr( 66; GFX9-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(4) getelementptr (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 4), align 4 67; GFX9-NEXT: [[TMP2:%.*]] = lshr i32 [[TMP1]], 16 68; GFX9-NEXT: [[TMP3:%.*]] = trunc i32 [[TMP2]] to i8 69; GFX9-NEXT: store i8 [[TMP3]], ptr addrspace(1) undef, align 1 70; GFX9-NEXT: ret void 71; 72; GFX12-LABEL: @constant_from_offset_cast_generic_inttoptr( 73; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 6), align 1 74; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 75; GFX12-NEXT: ret void 76; 77 %load = load i8, ptr addrspace(4) getelementptr inbounds (i8, ptr addrspace(4) addrspacecast (ptr inttoptr (i64 128 to ptr) to ptr addrspace(4)), i64 6), align 1 78 store i8 %load, ptr addrspace(1) undef 79 ret void 80} 81 82define amdgpu_kernel void @constant_from_inttoptr() { 83; GFX9-LABEL: @constant_from_inttoptr( 84; GFX9-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 4 85; GFX9-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 86; GFX9-NEXT: ret void 87; 88; GFX12-LABEL: @constant_from_inttoptr( 89; GFX12-NEXT: [[LOAD:%.*]] = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 1 90; GFX12-NEXT: store i8 [[LOAD]], ptr addrspace(1) undef, align 1 91; GFX12-NEXT: ret void 92; 93 %load = load i8, ptr addrspace(4) inttoptr (i64 128 to ptr addrspace(4)), align 1 94 store i8 %load, ptr addrspace(1) undef 95 ret void 96} 97 98define void @broken_phi() { 99; GFX9-LABEL: @broken_phi( 100; GFX9-NEXT: bb: 101; GFX9-NEXT: br label [[BB1:%.*]] 102; GFX9: bb1: 103; GFX9-NEXT: [[I:%.*]] = phi <4 x i8> [ splat (i8 1), [[BB:%.*]] ], [ [[I8:%.*]], [[BB7:%.*]] ] 104; GFX9-NEXT: br i1 false, label [[BB3:%.*]], label [[BB2:%.*]] 105; GFX9: bb2: 106; GFX9-NEXT: br label [[BB3]] 107; GFX9: bb3: 108; GFX9-NEXT: [[I4:%.*]] = phi <4 x i8> [ zeroinitializer, [[BB2]] ], [ [[I]], [[BB1]] ] 109; GFX9-NEXT: br i1 false, label [[BB7]], label [[BB5:%.*]] 110; GFX9: bb5: 111; GFX9-NEXT: [[I6:%.*]] = call <4 x i8> @llvm.smax.v4i8(<4 x i8> [[I4]], <4 x i8> zeroinitializer) 112; GFX9-NEXT: br label [[BB7]] 113; GFX9: bb7: 114; GFX9-NEXT: [[I8]] = phi <4 x i8> [ zeroinitializer, [[BB5]] ], [ zeroinitializer, [[BB3]] ] 115; GFX9-NEXT: br label [[BB1]] 116; 117; GFX12-LABEL: @broken_phi( 118; GFX12-NEXT: bb: 119; GFX12-NEXT: br label [[BB1:%.*]] 120; GFX12: bb1: 121; GFX12-NEXT: [[I:%.*]] = phi <4 x i8> [ splat (i8 1), [[BB:%.*]] ], [ [[I8:%.*]], [[BB7:%.*]] ] 122; GFX12-NEXT: br i1 false, label [[BB3:%.*]], label [[BB2:%.*]] 123; GFX12: bb2: 124; GFX12-NEXT: br label [[BB3]] 125; GFX12: bb3: 126; GFX12-NEXT: [[I4:%.*]] = phi <4 x i8> [ zeroinitializer, [[BB2]] ], [ [[I]], [[BB1]] ] 127; GFX12-NEXT: br i1 false, label [[BB7]], label [[BB5:%.*]] 128; GFX12: bb5: 129; GFX12-NEXT: [[I6:%.*]] = call <4 x i8> @llvm.smax.v4i8(<4 x i8> [[I4]], <4 x i8> zeroinitializer) 130; GFX12-NEXT: br label [[BB7]] 131; GFX12: bb7: 132; GFX12-NEXT: [[I8]] = phi <4 x i8> [ zeroinitializer, [[BB5]] ], [ zeroinitializer, [[BB3]] ] 133; GFX12-NEXT: br label [[BB1]] 134; 135bb: 136 br label %bb1 137bb1: 138 %i = phi <4 x i8> [ <i8 1, i8 1, i8 1, i8 1>, %bb ], [ %i8, %bb7 ] 139 br i1 false, label %bb3, label %bb2 140bb2: 141 br label %bb3 142bb3: 143 %i4 = phi <4 x i8> [ zeroinitializer, %bb2 ], [ %i, %bb1 ] 144 br i1 false, label %bb7, label %bb5 145bb5: 146 %i6 = call <4 x i8> @llvm.smax.v4i8(<4 x i8> %i4, <4 x i8> zeroinitializer) 147 br label %bb7 148bb7: 149 %i8 = phi <4 x i8> [ zeroinitializer, %bb5 ], [ zeroinitializer, %bb3 ] 150 br label %bb1 151} 152