77 lines · plain
1// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -finclude-default-header \
2// RUN: -emit-llvm -disable-llvm-passes -o - %s | llvm-cxxfilt | FileCheck %s -check-prefixes=CHECK,DXIL
3// RUN: %clang_cc1 -finclude-default-header -triple spirv-unknown-vulkan-compute \
4// RUN: -emit-llvm -disable-llvm-passes -o - %s | llvm-cxxfilt | FileCheck %s -check-prefixes=CHECK,SPV
5
6// CHECK: @[[BufA:.*]] = private unnamed_addr constant [2 x i8] c"A\00", align 1
7// CHECK: @[[BufB:.*]] = private unnamed_addr constant [2 x i8] c"B\00", align 1
8
9RWBuffer<float> A[] : register(u10, space1);
10RWBuffer<int> B[][5][4];
11
12RWStructuredBuffer<float> Out;
13
14float foo(RWBuffer<int> Arr[4], uint Index) {
15 return (float)Arr[Index][0];
16}
17
18[numthreads(4,1,1)]
19void main(uint GI : SV_GroupIndex) {
20 // CHECK: define internal{{.*}} void @main(unsigned int)(i32 noundef %GI)
21 // CHECK: %[[GI_alloca:.*]] = alloca i32, align 4
22 // CHECK-NEXT: %a = alloca float, align 4
23 // CHECK-NEXT: %[[Tmp0:.*]] = alloca %"class.hlsl::RWBuffer
24 // CHECK-NEXT: %b = alloca float, align 4
25 // CHECK-NEXT: %[[Tmp1:.*]] = alloca [4 x %"class.hlsl::RWBuffer"]
26 // CHECK-NEXT: %[[Tmp2:.*]] = alloca [4 x %"class.hlsl::RWBuffer"]
27 // CHECK-NEXT: store i32 %GI, ptr %[[GI_alloca]], align 4
28
29 // Make sure A[100] is translated to a RWBuffer<float> constructor call with range -1 and index 100
30 // and explicit binding (u10, space1)
31 // CHECK: @hlsl::RWBuffer<float>::__createFromBinding(unsigned int, unsigned int, int, unsigned int, char const*)
32 // CHECK-SAME: (ptr {{.*}} sret(%"class.hlsl::RWBuffer.0") align {{(4|8)}} %[[Tmp0]],
33 // CHECK-SAME: i32 noundef 10, i32 noundef 1, i32 noundef -1, i32 noundef 100, ptr noundef @A.str)
34 // CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr{{.*}} @hlsl::RWBuffer<float>::operator[](unsigned int)(ptr {{.*}} %[[Tmp0]], i32 noundef 0)
35 // CHECK-NEXT: %[[Value1:.*]] = load float, ptr{{.*}} %[[BufPtr]], align 4
36 // CHECK-NEXT: store float %[[Value1]], ptr %a, align 4
37 float a = A[100][0];
38
39 // Make sure B[2][3] is translated to a local RWBuffer<int>[4] array where each array element
40 // is initialized by a constructor call with range -1 and index 52-55 and implicit binding
41 // (space 0, order_id 0)
42 // The first index is calculated from the array dimensions (unbounded x 5 x 4) and indices (2, 3)
43 // as 2 * 5 * 4 + 3 * 4 = 52 and the following indices are sequential.
44
45 // CHECK-NEXT: %[[Ptr_Tmp2_0:.*]] = getelementptr [4 x %"class.hlsl::RWBuffer"], ptr %[[Tmp2]], i32 0, i32 0
46 // CHECK-NEXT: call void @hlsl::RWBuffer<int>::__createFromImplicitBinding(unsigned int, unsigned int, int, unsigned int, char const*)
47 // CHECK-SAME: (ptr {{.*}} sret(%"class.hlsl::RWBuffer") align {{(4|8)}} %[[Ptr_Tmp2_0]],
48 // CHECK-SAME: i32 noundef 0, i32 noundef 0, i32 noundef -1, i32 noundef 52, ptr noundef @[[BufB]])
49
50 // CHECK-NEXT: %[[Ptr_Tmp2_1:.*]] = getelementptr [4 x %"class.hlsl::RWBuffer"], ptr %[[Tmp2]], i32 0, i32 1
51 // CHECK-NEXT: call void @hlsl::RWBuffer<int>::__createFromImplicitBinding(unsigned int, unsigned int, int, unsigned int, char const*)
52 // CHECK-SAME: (ptr {{.*}} sret(%"class.hlsl::RWBuffer") align {{(4|8)}} %[[Ptr_Tmp2_1]],
53 // CHECK-SAME: i32 noundef 0, i32 noundef 0, i32 noundef -1, i32 noundef 53, ptr noundef @[[BufB]])
54
55 // CHECK-NEXT: %[[Ptr_Tmp2_2:.*]] = getelementptr [4 x %"class.hlsl::RWBuffer"], ptr %[[Tmp2]], i32 0, i32 2
56 // CHECK-NEXT: call void @hlsl::RWBuffer<int>::__createFromImplicitBinding(unsigned int, unsigned int, int, unsigned int, char const*)
57 // CHECK-SAME: (ptr {{.*}} sret(%"class.hlsl::RWBuffer") align {{(4|8)}} %[[Ptr_Tmp2_2]],
58 // CHECK-SAME: i32 noundef 0, i32 noundef 0, i32 noundef -1, i32 noundef 54, ptr noundef @[[BufB]])
59
60 // CHECK-NEXT: %[[Ptr_Tmp2_3:.*]] = getelementptr [4 x %"class.hlsl::RWBuffer"], ptr %[[Tmp2]], i32 0, i32 3
61 // CHECK-NEXT: call void @hlsl::RWBuffer<int>::__createFromImplicitBinding(unsigned int, unsigned int, int, unsigned int, char const*)
62 // CHECK-SAME: (ptr {{.*}} sret(%"class.hlsl::RWBuffer") align {{(4|8)}} %[[Ptr_Tmp2_3]],
63 // CHECK-SAME: i32 noundef 0, i32 noundef 0, i32 noundef -1, i32 noundef 55, ptr noundef @[[BufB]])
64
65 // DXIL-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp1]], ptr align 4 %[[Tmp2]], i32 16, i1 false)
66 // SPV-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 %[[Tmp1]], ptr align 8 %[[Tmp2]], i64 32, i1 false)
67
68 // CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_alloca]], align 4
69 // CHECK-NEXT: %[[Value2:.*]] = call {{.*}} float @foo(hlsl::RWBuffer<int> [4], unsigned int)
70 // CHECK-SAME: (ptr noundef byval([4 x %"class.hlsl::RWBuffer"]) align {{(4|8)}} %[[Tmp1]], i32 noundef %[[GI]])
71
72 // CHECK-NEXT: store float %[[Value2]], ptr %b, align 4
73 float b = foo(B[2][3], GI);
74
75 Out[0] = a + b;
76}
77