[HLSL] Tests for local resource arrays

clang/test/CodeGenHLSL/resources/res-array-local-multi-dim.hlsl

@@ -0,0 +1,49 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -finclude-default-header \
+// RUN:   -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies handling of multi-dimensional local arrays of resources
+// when used as a function argument and local variable.
+
+// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<float> A : register(u10);
+RWBuffer<float> B : register(u20);
+RWStructuredBuffer<float> Out;
+
+// NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for RWBuffer<float> and
+//       _ZN4hlsl18RWStructuredBufferIfEixEj is the subscript operator for RWStructuredBuffer<float>
+
+// CHECK: define {{.*}} float @_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef byval([2 x [2 x %"class.hlsl::RWBuffer"]]) align 4 %Arr)
+// CHECK-NEXT: entry:
+float foo(RWBuffer<float> Arr[2][2]) {
+// CHECK-NEXT: %[[Arr_1_Ptr:.*]] = getelementptr inbounds [2 x [2 x %"class.hlsl::RWBuffer"]], ptr %Arr, i32 0, i32 1
+// CHECK-NEXT: %[[Arr_1_1_Ptr:.*]] = getelementptr inbounds [2 x %"class.hlsl::RWBuffer"], ptr %[[Arr_1_Ptr]], i32 0, i32 1
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr {{.*}} %[[Arr_1_1_Ptr]], i32 noundef 0)
+// CHECK-NEXT: %[[Value:.*]] = load float, ptr %[[BufPtr]], align 4
+// CHECK-NEXT: ret float %[[Value]]
+  return Arr[1][1][0];
+}
+
+// CHECK: define internal void @_Z4mainv()
+// CHECK-NEXT: entry:
+[numthreads(4,1,1)]
+void main() {
+// CHECK-NEXT: %L = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align 4
+// CHECK-NEXT: %[[Tmp:.*]] = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align 4
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %L, ptr align 4 @_ZL1A, i32 4, i1 false)
+// CHECK-NEXT: %[[Ptr1:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %L, i32 1
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr1]], ptr align 4 @_ZL1B, i32 4, i1 false)
+// CHECK-NEXT: %[[Ptr2:.*]] = getelementptr inbounds [2 x %"class.hlsl::RWBuffer"], ptr %L, i32 1
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr2]], ptr align 4 @_ZL1A, i32 4, i1 false)
+// CHECK-NEXT: %[[Ptr3:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %[[Ptr2]], i32 1
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr3]], ptr align 4 @_ZL1B, i32 4, i1 false)
+  RWBuffer<float> L[2][2] = { { A, B }, { A, B } };
+
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr align 4 %L, i32 16, i1 false)
+// CHECK-NEXT: %[[ReturnedValue:.*]] = call {{.*}}float @_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef byval([2 x [2 x %"class.hlsl::RWBuffer"]]) align 4 %[[Tmp]])
+// CHECK-NEXT: %[[OutBufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl18RWStructuredBufferIfEixEj(ptr {{.*}} @_ZL3Out, i32 noundef 0)
+// CHECK-NEXT: store float %[[ReturnedValue]], ptr %[[OutBufPtr]], align 4
+// CHECK-NEXT: ret void
+  Out[0] = foo(L);
+}

clang/test/CodeGenHLSL/resources/res-array-local1.hlsl

@@ -0,0 +1,64 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -finclude-default-header \
+// RUN:   -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies local arrays of resources in HLSL.
+
+// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1C = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<float> A : register(u1);
+RWBuffer<float> B : register(u2);
+RWBuffer<float> C : register(u3);
+RWStructuredBuffer<float> Out : register(u0);
+
+// CHECK: define internal void @_Z4mainv()
+// CHECK-NEXT: entry:
+[numthreads(4,1,1)]
+void main() {
+// CHECK-NEXT:  %First = alloca [3 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT:  %Second = alloca [4 x %"class.hlsl::RWBuffer"], align 4
+  RWBuffer<float> First[3] = { A, B, C };
+  RWBuffer<float> Second[4];
+
+// Verify initialization of First array from an initialization list
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %First, ptr align 4 @_ZL1A, i32 4, i1 false)
+// CHECK-NEXT: %[[Ptr1:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %First, i32 1
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr1]], ptr align 4 @_ZL1B, i32 4, i1 false)
+// CHECK-NEXT: %[[Ptr2:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %First, i32 2
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr2]], ptr align 4 @_ZL1C, i32 4, i1 false)
+
+// Verify default initialization of Second array, which means there is a loop iterating
+// over the array elements and calling the default constructor for each
+// CHECK-NEXT: %[[ArrayBeginPtr:.*]] = getelementptr inbounds [4 x %"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 0
+// CHECK-NEXT: %[[ArrayEndPtr:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %[[ArrayBeginPtr]], i32 4
+// CHECK-NEXT: br label %[[ArrayInitLoop:.*]]
+// CHECK: [[ArrayInitLoop]]:
+// CHECK-NEXT: %[[ArrayCurPtr:.*]] = phi ptr [ %[[ArrayBeginPtr]], %entry ], [ %[[ArrayNextPtr:.*]], %[[ArrayInitLoop]] ]
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1Ev(ptr {{.*}} %[[ArrayCurPtr]])
+// CHECK-NEXT: %[[ArrayNextPtr]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %[[ArrayCurPtr]], i32 1
+// CHECK-NEXT: %[[ArrayInitDone:.*]] = icmp eq ptr %[[ArrayNextPtr]], %[[ArrayEndPtr]]
+// CHECK-NEXT: br i1 %[[ArrayInitDone]], label %[[AfterArrayInit:.*]], label %[[ArrayInitLoop]]
+// CHECK: [[AfterArrayInit]]:
+
+// Initialize First[2] with C
+// CHECK: %[[Ptr3:.*]] = getelementptr inbounds [4 x %"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 2
+// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr3]], ptr align 4 @_ZL1C, i32 4, i1 false)
+  Second[2] = C;
+
+  // NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for RWBuffer<float>
+
+// get First[1][0] value
+// CHECK: %[[First_1_Ptr:.*]] = getelementptr inbounds [3 x %"class.hlsl::RWBuffer"], ptr %First, i32 0, i32 1
+// CHECK: %[[BufPtr1:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr {{.*}} %[[First_1_Ptr]], i32 noundef 0)
+// CHECK: %[[Value1:.*]] = load float, ptr %[[BufPtr1]], align 4
+
+// get Second[2][0] value
+// CHECK: %[[Second_2_Ptr:.*]] = getelementptr inbounds [4 x %"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 2
+// CHECK: %[[BufPtr2:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr {{.*}} %[[Second_2_Ptr]], i32 noundef 0)
+// CHECK: %[[Value2:.*]] = load float, ptr %[[BufPtr2]], align 4
+
+// add them
+// CHECK: %{{.*}} = fadd {{.*}} float %[[Value1]], %[[Value2]]
+  Out[0] = First[1][0] + Second[2][0];
+}

clang/test/CodeGenHLSL/resources/res-array-local2.hlsl

@@ -0,0 +1,37 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -finclude-default-header \
+// RUN:   -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies handling of local arrays of resources when used as a function argument.
+
+// CHECK: @_ZL1A = internal global [3 x %"class.hlsl::RWBuffer"] poison, align 4
+
+RWBuffer<float> A[3] : register(u0);
+RWStructuredBuffer<float> Out : register(u0);
+
+// NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for RWBuffer<float> and
+//       _ZN4hlsl18RWStructuredBufferIfEixEj is the subscript operator for RWStructuredBuffer<float>
+
+// CHECK: define {{.*}} float @_Z3fooA3_N4hlsl8RWBufferIfEE(ptr noundef byval([3 x %"class.hlsl::RWBuffer"]) align 4 %LocalA)
+// CHECK-NEXT: entry:
+float foo(RWBuffer<float> LocalA[3]) {
+// CHECK-NEXT: %[[LocalA_2_Ptr:.*]] = getelementptr inbounds [3 x %"class.hlsl::RWBuffer"], ptr %LocalA, i32 0, i32 2
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr {{.*}} %[[LocalA_2_Ptr]], i32 noundef 0)
+// CHECK-NEXT: %[[Value:.*]] = load float, ptr %[[BufPtr]], align 4
+// CHECK-NEXT: ret float %[[Value]]
+  return LocalA[2][0];
+}
+
+// CHECK: define internal void @_Z4mainv()
+// CHECK-NEXT: entry:
+[numthreads(4,1,1)]
+void main() {
+// Check that the `main` function calls `foo` with a local copy of the array
+// CHECK-NEXT: %[[Tmp:.*]] = alloca [3 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr align 4 @_ZL1A, i32 12, i1 false)
+
+// CHECK-NEXT: %[[ReturnedValue:.*]] = call {{.*}} float @_Z3fooA3_N4hlsl8RWBufferIfEE(ptr noundef byval([3 x %"class.hlsl::RWBuffer"]) align 4 %[[Tmp]])
+// CHECK-NEXT: %[[OutBufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl18RWStructuredBufferIfEixEj(ptr {{.*}} @_ZL3Out, i32 noundef 0)
+// CHECK-NEXT: store float %[[ReturnedValue]], ptr %[[OutBufPtr]], align 4
+// CHECK-NEXT: ret void
+  Out[0] = foo(A);
+}

clang/test/CodeGenHLSL/resources/res-array-local3.hlsl

@@ -0,0 +1,62 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -finclude-default-header \
+// RUN:   -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies handling of local arrays of resources when used
+// as a function argument that is modified inside the function.
+
+// CHECK: @_ZL1X = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1Y = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<int> X : register(u0);
+RWBuffer<int> Y : register(u1);
+
+// CHECK: define {{.*}} @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(
+// CHECK-SAME: ptr noundef byval([2 x %"class.hlsl::RWBuffer"]) align 4 %B, i32 noundef %Idx, i32 noundef %Val0)
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[Idx_addr:.*]] = alloca i32, align 4
+// CHECK-NEXT: %[[Val0_addr:.*]] = alloca i32, align 4
+// CHECK-NEXT: store i32 %Idx, ptr %[[Idx_addr]], align 4
+// CHECK-NEXT: store i32 %Val0, ptr %[[Val0_addr]], align 4
+void SomeFn(RWBuffer<int> B[2], uint Idx, int Val0) {
+
+// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x %"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[B_0_Ptr]], ptr align 4 @_ZL1Y, i32 4, i1 false)
+  B[0] = Y;
+
+// NOTE: _ZN4hlsl8RWBufferIiEixEj is the subscript operator for RWBuffer<int>
+
+// CHECK-NEXT: %[[Val0:.*]] = load i32, ptr %[[Val0_addr]], align 4
+// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x %"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
+// CHECK-NEXT: %[[Idx:.*]] = load i32, ptr %[[Idx_addr]], align 4
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr {{.*}} %[[B_0_Ptr]], i32 noundef %[[Idx]])
+// CHECK-NEXT: store i32 %[[Val0]], ptr %[[BufPtr]], align 4
+  B[0][Idx] = Val0;
+}
+
+// CHECK: define {{.*}} void @_Z4mainj(i32 noundef %GI)
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[GI_addr:.*]] = alloca i32, align 4
+[numthreads(4,1,1)]
+void main(uint GI : SV_GroupIndex) {
+// CHECK-NEXT: %A = alloca [2 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: %[[Tmp:.*]] = alloca [2 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: store i32 %GI, ptr %GI.addr, align 4
+
+// Initialization of array A with resources X and Y
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %A, ptr align 4 @_ZL1X, i32 4, i1 false)
+// CHECK-NEXT: %[[A_1_Ptr:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %A, i32 1
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[A_1_Ptr]], ptr align 4 @_ZL1Y, i32 4, i1 false)
+  RWBuffer<int> A[2] = {X, Y};
+
+// Verify that SomeFn is called with a local copy of the array A
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr align 4 %A, i32 8, i1 false)
+// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
+// CHECK-NEXT: call void @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(ptr noundef byval([2 x %"class.hlsl::RWBuffer"]) align 4 %[[Tmp]], i32 noundef %[[GI]], i32 noundef 1)
+  SomeFn(A, GI, 1);
+
+// CHECK-NEXT: %[[A_0_Ptr:.*]] = getelementptr inbounds [2 x %"class.hlsl::RWBuffer"], ptr %A, i32 0, i32 0
+// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr {{.*}} %[[A_0_Ptr]], i32 noundef %[[GI]])
+// CHECK-NEXT: store i32 2, ptr %[[BufPtr]], align 4
+  A[0][GI] = 2;
+}