72 lines · plain
1; RUN: opt -S -scalarizer -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library %s | FileCheck %s2 3; This test exercises the lowering of the intrinsic @llvm.uadd.with.overflow.i32 to the UAddc DXIL op4 5; CHECK-DAG: [[DX_TYPES_I32C:%dx\.types\.i32c]] = type { i32, i1 }6 7; NOTE: The uint2 overload of AddUint64 HLSL uses @llvm.uadd.with.overflow.i32, resulting in one UAddc op8define noundef i32 @test_UAddc(i32 noundef %a, i32 noundef %b) {9; CHECK-LABEL: define noundef i32 @test_UAddc(10; CHECK-SAME: i32 noundef [[A:%.*]], i32 noundef [[B:%.*]]) {11; CHECK-NEXT: [[UADDC:%.*]] = call [[DX_TYPES_I32C]] @dx.op.binaryWithCarryOrBorrow.i32(i32 44, i32 [[A]], i32 [[B]]) #[[ATTR0:[0-9]+]]12; CHECK-NEXT: [[CARRY:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC]], 113; CHECK-NEXT: [[SUM:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC]], 014; CHECK-NEXT: [[CARRY_ZEXT:%.*]] = zext i1 [[CARRY]] to i3215; CHECK-NEXT: [[RESULT:%.*]] = add i32 [[SUM]], [[CARRY_ZEXT]]16; CHECK-NEXT: ret i32 [[RESULT]]17;18 %uaddc = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)19 %carry = extractvalue { i32, i1 } %uaddc, 120 %sum = extractvalue { i32, i1 } %uaddc, 021 %carry_zext = zext i1 %carry to i3222 %result = add i32 %sum, %carry_zext23 ret i32 %result24}25 26; NOTE: The uint4 overload of AddUint64 HLSL uses @llvm.uadd.with.overflow.v2i32, resulting in two UAddc ops after scalarization27define noundef <2 x i32> @test_UAddc_vec2(<2 x i32> noundef %a, <2 x i32> noundef %b) {28; CHECK-LABEL: define noundef <2 x i32> @test_UAddc_vec2(29; CHECK-SAME: <2 x i32> noundef [[A:%.*]], <2 x i32> noundef [[B:%.*]]) {30; CHECK-NEXT: [[A_I0:%.*]] = extractelement <2 x i32> [[A]], i64 031; CHECK-NEXT: [[B_I0:%.*]] = extractelement <2 x i32> [[B]], i64 032; CHECK-NEXT: [[UADDC_I0:%.*]] = call [[DX_TYPES_I32C]] @dx.op.binaryWithCarryOrBorrow.i32(i32 44, i32 [[A_I0]], i32 [[B_I0]]) #[[ATTR0]]33; CHECK-NEXT: [[A_I1:%.*]] = extractelement <2 x i32> [[A]], i64 134; CHECK-NEXT: [[B_I1:%.*]] = extractelement <2 x i32> [[B]], i64 135; CHECK-NEXT: [[UADDC_I1:%.*]] = call [[DX_TYPES_I32C]] @dx.op.binaryWithCarryOrBorrow.i32(i32 44, i32 [[A_I1]], i32 [[B_I1]]) #[[ATTR0]]36; CHECK-NEXT: [[CARRY_ELEM0:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC_I0]], 137; CHECK-NEXT: [[CARRY_ELEM1:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC_I1]], 138; CHECK-NEXT: [[SUM_ELEM0:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC_I0]], 039; CHECK-NEXT: [[SUM_ELEM1:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC_I1]], 040; CHECK-NEXT: [[CARRY_ZEXT_I0:%.*]] = zext i1 [[CARRY_ELEM0]] to i3241; CHECK-NEXT: [[CARRY_ZEXT_I1:%.*]] = zext i1 [[CARRY_ELEM1]] to i3242; CHECK-NEXT: [[RESULT_I0:%.*]] = add i32 [[SUM_ELEM0]], [[CARRY_ZEXT_I0]]43; CHECK-NEXT: [[RESULT_I1:%.*]] = add i32 [[SUM_ELEM1]], [[CARRY_ZEXT_I1]]44; CHECK-NEXT: [[RESULT_UPTO0:%.*]] = insertelement <2 x i32> poison, i32 [[RESULT_I0]], i64 045; CHECK-NEXT: [[RESULT:%.*]] = insertelement <2 x i32> [[RESULT_UPTO0]], i32 [[RESULT_I1]], i64 146; CHECK-NEXT: ret <2 x i32> [[RESULT]]47;48 %uaddc = call { <2 x i32>, <2 x i1> } @llvm.uadd.with.overflow.v2i32(<2 x i32> %a, <2 x i32> %b)49 %carry = extractvalue { <2 x i32>, <2 x i1> } %uaddc, 150 %sum = extractvalue { <2 x i32>, <2 x i1> } %uaddc, 051 %carry_zext = zext <2 x i1> %carry to <2 x i32>52 %result = add <2 x i32> %sum, %carry_zext53 ret <2 x i32> %result54}55 56define noundef i32 @test_UAddc_insert(i32 noundef %a, i32 noundef %b) {57; CHECK-LABEL: define noundef i32 @test_UAddc_insert(58; CHECK-SAME: i32 noundef [[A:%.*]], i32 noundef [[B:%.*]]) {59; CHECK-NEXT: [[UADDC:%.*]] = call [[DX_TYPES_I32C]] @dx.op.binaryWithCarryOrBorrow.i32(i32 44, i32 [[A]], i32 [[B]]) #[[ATTR0]]60; CHECK-NEXT: [[UNUSED:%.*]] = insertvalue [[DX_TYPES_I32C]] [[UADDC]], i32 [[A]], 061; CHECK-NEXT: [[RESULT:%.*]] = extractvalue [[DX_TYPES_I32C]] [[UADDC]], 062; CHECK-NEXT: ret i32 [[RESULT]]63;64 %uaddc = call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)65 insertvalue { i32, i1 } %uaddc, i32 %a, 066 %result = extractvalue { i32, i1 } %uaddc, 067 ret i32 %result68}69 70declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32)71 72