88 lines · plain
1; RUN: opt -S -passes='dxil-data-scalarization' -mtriple=dxil-pc-shadermodel6.4-library %s | FileCheck %s --check-prefixes=SCHECK,CHECK2; RUN: opt -S -passes='dxil-data-scalarization,function(scalarizer<load-store>),dxil-flatten-arrays' -mtriple=dxil-pc-shadermodel6.4-library %s | FileCheck %s --check-prefixes=FCHECK,CHECK3 4@aTile = hidden addrspace(3) global [10 x [10 x <4 x i32>]] zeroinitializer, align 165@bTile = hidden addrspace(3) global [10 x [10 x i32]] zeroinitializer, align 166@cTile = internal global [2 x [2 x <2 x i32>]] zeroinitializer, align 167@dTile = internal global [2 x [2 x [2 x <2 x i32>]]] zeroinitializer, align 168 9define void @CSMain() {10; CHECK-LABEL: define void @CSMain() {11; CHECK-NEXT: [[ENTRY:.*:]]12; CHECK-NEXT: [[AFRAGPACKED_I_SCALARIZE:%.*]] = alloca [4 x i32], align 1613;14; SCHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds [10 x [10 x [4 x i32]]], ptr addrspace(3) @aTile.scalarized, i32 0, i32 115; SCHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [10 x [4 x i32]], ptr addrspace(3) [[GEP0]], i32 0, i32 216; SCHECK-NEXT: [[LOAD:%.*]] = load <4 x i32>, ptr addrspace(3) [[GEP1]], align 1617; SCHECK-NEXT: store <4 x i32> [[LOAD]], ptr [[AFRAGPACKED_I_SCALARIZE]], align 1618;19; FCHECK-NEXT: [[AFRAGPACKED_I_SCALARIZE_I14:%.*]] = getelementptr [4 x i32], ptr [[AFRAGPACKED_I_SCALARIZE]], i32 0, i32 120; FCHECK-NEXT: [[AFRAGPACKED_I_SCALARIZE_I25:%.*]] = getelementptr [4 x i32], ptr [[AFRAGPACKED_I_SCALARIZE]], i32 0, i32 221; FCHECK-NEXT: [[AFRAGPACKED_I_SCALARIZE_I36:%.*]] = getelementptr [4 x i32], ptr [[AFRAGPACKED_I_SCALARIZE]], i32 0, i32 322; FCHECK-NEXT: [[DOTI07:%.*]] = load i32, ptr addrspace(3) getelementptr inbounds ([400 x i32], ptr addrspace(3) @aTile.scalarized.1dim, i32 0, i32 48), align 1623; FCHECK-NEXT: [[DOTI119:%.*]] = load i32, ptr addrspace(3) getelementptr ([400 x i32], ptr addrspace(3) @aTile.scalarized.1dim, i32 0, i32 49), align 424; FCHECK-NEXT: [[DOTI2211:%.*]] = load i32, ptr addrspace(3) getelementptr ([400 x i32], ptr addrspace(3) @aTile.scalarized.1dim, i32 0, i32 50), align 825; FCHECK-NEXT: [[DOTI3313:%.*]] = load i32, ptr addrspace(3) getelementptr ([400 x i32], ptr addrspace(3) @aTile.scalarized.1dim, i32 0, i32 51), align 426; FCHECK-NEXT: store i32 [[DOTI07]], ptr [[AFRAGPACKED_I_SCALARIZE]], align 1627; FCHECK-NEXT: store i32 [[DOTI119]], ptr [[AFRAGPACKED_I_SCALARIZE_I14]], align 428; FCHECK-NEXT: store i32 [[DOTI2211]], ptr [[AFRAGPACKED_I_SCALARIZE_I25]], align 829; FCHECK-NEXT: store i32 [[DOTI3313]], ptr [[AFRAGPACKED_I_SCALARIZE_I36]], align 430;31; CHECK-NEXT: ret void32entry:33 %aFragPacked.i = alloca <4 x i32>, align 1634 %0 = load <4 x i32>, ptr addrspace(3) getelementptr inbounds ([10 x <4 x i32>], ptr addrspace(3) getelementptr inbounds ([10 x [10 x <4 x i32>]], ptr addrspace(3) @aTile, i32 0, i32 1), i32 0, i32 2), align 1635 store <4 x i32> %0, ptr %aFragPacked.i, align 1636 ret void37}38 39define void @Main() {40; CHECK-LABEL: define void @Main() {41; CHECK-NEXT: [[ENTRY:.*:]]42; CHECK-NEXT: [[BFRAGPACKED_I:%.*]] = alloca i32, align 1643;44; SCHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds [10 x [10 x i32]], ptr addrspace(3) @bTile, i32 0, i32 145; SCHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [10 x i32], ptr addrspace(3) [[GEP0]], i32 0, i32 146; SCHECK-NEXT: [[LOAD:%.*]] = load i32, ptr addrspace(3) [[GEP1]], align 1647; SCHECK-NEXT: store i32 [[LOAD]], ptr [[BFRAGPACKED_I]], align 1648;49; FCHECK-NEXT: [[LOAD:%.*]] = load i32, ptr addrspace(3) getelementptr inbounds ([100 x i32], ptr addrspace(3) @bTile.1dim, i32 0, i32 11), align 1650; FCHECK-NEXT: store i32 [[LOAD]], ptr [[BFRAGPACKED_I]], align 1651;52; CHECK-NEXT: ret void53entry:54 %bFragPacked.i = alloca i32, align 1655 %0 = load i32, ptr addrspace(3) getelementptr inbounds ([10 x i32], ptr addrspace(3) getelementptr inbounds ([10 x [10 x i32]], ptr addrspace(3) @bTile, i32 0, i32 1), i32 0, i32 1), align 1656 store i32 %0, ptr %bFragPacked.i, align 1657 ret void58}59 60define void @global_nested_geps_3d() {61; CHECK-LABEL: define void @global_nested_geps_3d() {62; SCHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds [2 x [2 x [2 x i32]]], ptr @cTile.scalarized, i32 0, i32 163; SCHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x [2 x i32]], ptr [[GEP0]], i32 0, i32 164; SCHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x i32], ptr [[GEP1]], i32 0, i32 165; SCHECK-NEXT: load i32, ptr [[GEP2]], align 466;67; FCHECK-NEXT: load i32, ptr getelementptr inbounds ([8 x i32], ptr @cTile.scalarized.1dim, i32 0, i32 7), align 468;69; CHECK-NEXT: ret void70 %1 = load i32, i32* getelementptr inbounds (<2 x i32>, <2 x i32>* getelementptr inbounds ([2 x <2 x i32>], [2 x <2 x i32>]* getelementptr inbounds ([2 x [2 x <2 x i32>]], [2 x [2 x <2 x i32>]]* @cTile, i32 0, i32 1), i32 0, i32 1), i32 0, i32 1), align 471 ret void72}73 74define void @global_nested_geps_4d() {75; CHECK-LABEL: define void @global_nested_geps_4d() {76; SCHECK-NEXT: [[GEP0:%.*]] = getelementptr inbounds [2 x [2 x [2 x [2 x i32]]]], ptr @dTile.scalarized, i32 0, i32 177; SCHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds [2 x [2 x [2 x i32]]], ptr [[GEP0]], i32 0, i32 178; SCHECK-NEXT: [[GEP2:%.*]] = getelementptr inbounds [2 x [2 x i32]], ptr [[GEP1]], i32 0, i32 179; SCHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds [2 x i32], ptr [[GEP2]], i32 0, i32 180; SCHECK-NEXT: load i32, ptr [[GEP3]], align 481;82; FCHECK-NEXT: load i32, ptr getelementptr inbounds ([16 x i32], ptr @dTile.scalarized.1dim, i32 0, i32 15), align 483;84; CHECK-NEXT: ret void85 %1 = load i32, i32* getelementptr inbounds (<2 x i32>, <2 x i32>* getelementptr inbounds ([2 x <2 x i32>], [2 x <2 x i32>]* getelementptr inbounds ([2 x [2 x <2 x i32>]], [2 x [2 x <2 x i32>]]* getelementptr inbounds ([2 x [2 x [2 x <2 x i32>]]], [2 x [2 x [2 x <2 x i32>]]]* @dTile, i32 0, i32 1), i32 0, i32 1), i32 0, i32 1), i32 0, i32 1), align 486 ret void87}88