[llvm-branch-commits] [clang] [HLSL] Constant buffers codegen (PR #124886)

[Damyan Pepper via llvm-branch-commits]

================
@@ -0,0 +1,199 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.3-compute \
+// RUN:   -fnative-half-type -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// CHECK: %struct.__cblayout_CBScalars = type <{ float, double, half, i64, i32, i16, i32, i64 }>
+// CHECK: %struct.__cblayout_CBVectors = type <{ <3 x float>, <3 x double>, <2 x half>, <3 x i64>, <4 x i32>, <3 x i16>, <3 x i64> }>
+// CHECK: %struct.__cblayout_CBArrays = type <{ [3 x float], [2 x <3 x double>], [2 x [2 x half]], [3 x i64], [2 x [3 x [4 x <4 x i32>]]], [1 x i16], [2 x i64], [4 x i32] }>
+// CHECK: %struct.__cblayout_CBStructs = type { %struct.A, %struct.B, %struct.C, [5 x %struct.A], %struct.__cblayout_D, half, %struct.B, <3 x i16> }
+// CHECK: %struct.A = type { <2 x float> }
+// CHECK: %struct.C = type { i32, %struct.A }
+// CHECK: %struct.__cblayout_D = type { [2 x [3 x %struct.B]] }
+// CHECK: %struct.B = type { %struct.A, <3 x i16> }
+
+cbuffer CBScalars : register(b1, space5) {
+  float a1;
+  double a2;
+  float16_t a3;
+  uint64_t a4;
+  int a5;
+  uint16_t a6;
+  bool a7;
+  int64_t a8;
+}
+
+// CHECK: @CBScalars.cb = external constant target("dx.CBuffer", %struct.__cblayout_CBScalars)
+// CHECK: @a1 = external addrspace(2) global float, align 4
+// CHECK: @a2 = external addrspace(2) global double, align 8
+// CHECK: @a3 = external addrspace(2) global half, align 2
+// CHECK: @a4 = external addrspace(2) global i64, align 8
+// CHECK: @a5 = external addrspace(2) global i32, align 4
+// CHECK: @a6 = external addrspace(2) global i16, align 2
+// CHECK: @a7 = external addrspace(2) global i32, align 4
+// CHECK: @a8 = external addrspace(2) global i64, align 8
+
+cbuffer CBVectors {
+  float3 b1;
+  double3 b2;
+  float16_t2 b3;
+  uint64_t3 b4;
+  int4 b5;
+  uint16_t3 b6;
+  int64_t3 b7;
+  // FIXME: add s bool vectors after llvm-project/llvm#91639 is added
+}
+
+// CHECK: @CBVectors.cb = external constant target("dx.CBuffer", %struct.__cblayout_CBVectors)
+// CHECK: @b1 = external addrspace(2) global <3 x float>, align 16
+// CHECK: @b2 = external addrspace(2) global <3 x double>, align 32
+// CHECK: @b3 = external addrspace(2) global <2 x half>, align 4
+// CHECK: @b4 = external addrspace(2) global <3 x i64>, align 32
+// CHECK: @b5 = external addrspace(2) global <4 x i32>, align 16
+// CHECK: @b6 = external addrspace(2) global <3 x i16>, align 8
+// CHECK: @b7 = external addrspace(2) global <3 x i64>, align 32
+
+cbuffer CBArrays : register(b2) {
+  float c1[3];
+  double3 c2[2];
+  float16_t c3[2][2];
+  uint64_t c4[3];
+  int4 c5[2][3][4];
+  uint16_t c6[1];
+  int64_t c7[2];
+  bool c8[4];
+}
+
+// CHECK: @CBArrays.cb = external constant target("dx.CBuffer", %struct.__cblayout_CBArrays)
+// CHECK: @c1 = external addrspace(2) global [3 x float], align 4
+// CHECK: @c2 = external addrspace(2) global [2 x <3 x double>], align 32
+// CHECK: @c3 = external addrspace(2) global [2 x [2 x half]], align 2
+// CHECK: @c4 = external addrspace(2) global [3 x i64], align 8
+// CHECK: @c5 = external addrspace(2) global [2 x [3 x [4 x <4 x i32>]]], align 16
+// CHECK: @c6 = external addrspace(2) global [1 x i16], align 2
+// CHECK: @c7 = external addrspace(2) global [2 x i64], align 8
+// CHECK: @c8 = external addrspace(2) global [4 x i32], align 4
+
+struct Empty {};
+
+struct A {
+  float2 f1;
+};
+
+struct B : A {
+  uint16_t3 f2;
+};
+
+struct C {
+  int i;
+  A f3;
+};
+
+struct D {
+  B array_of_B[2][3];
+  Empty es;
+};
+
+// CHECK: @CBStructs.cb = external constant target("dx.CBuffer", %struct.__cblayout_CBStructs)
+// CHECK: @a = external addrspace(2) global %struct.A, align 8
+// CHECK: @b = external addrspace(2) global %struct.B, align 8
+// CHECK: @c = external addrspace(2) global %struct.C, align 8
+// CHECK: @array_of_A = external addrspace(2) global [5 x %struct.A], align 8
+// CHECK: @d = external addrspace(2) global %struct.__cblayout_D, align 8
+// CHECK: @e = external addrspace(2) global half, align 2
+
+cbuffer CBStructs {
+  A a;
+  B b;
+  C c;
+  A array_of_A[5];
+  D d;
+  half e;
+  B f;
+  uint16_t3 g;
+};
+
+struct Test {
+    float a, b;
+};
+
+// CHECK: @CBMix.cb = external constant target("dx.CBuffer", %struct.__cblayout_CBMix)
+// CHECK: @test = external addrspace(2) global [2 x %struct.Test], align 4
+// CHECK: @f1 = external addrspace(2) global float, align 4
+// CHECK: @f2 = external addrspace(2) global [3 x [2 x <2 x float>]], align 8
+// CHECK: @f3 = external addrspace(2) global float, align 4
+// CHECK: @s = external addrspace(2) global %struct.anon, align 4
+// CHECK: @dd = external addrspace(2) global double, align 8
+// CHECK: @f4 = external addrspace(2) global float, align 4
+// CHECK: @dv = external addrspace(2) global <1 x double>, align 8
+// CHECK: @uv = external addrspace(2) global i16, align 2
+
+cbuffer CBMix {
+    Test test[2];
+    float f1;
+    float2 f2[3][2];
+    float f3;
+    struct { float c; } s;
+    double dd;
+    float f4;
+    vector<double,1> dv;
+    uint16_t uv;
+};  
+
+// CHECK: efine internal void @_init_resource_CBScalars.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[HANDLE1:.*]] = call target("dx.CBuffer", %struct.__cblayout_CBScalars)
+// CHECK-SAME: @llvm.dx.resource.handlefrombinding.tdx.CBuffer_s_struct.__cblayout_CBScalarsst(i32 5, i32 1, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", %struct.__cblayout_CBScalars) %[[HANDLE1]], ptr @CBScalars.cb, align 4
+
+// CHECK: define internal void @_init_resource_CBArrays.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[HANDLE2:.*]] = call target("dx.CBuffer", %struct.__cblayout_CBArrays)
+// CHECK-SAME: @llvm.dx.resource.handlefrombinding.tdx.CBuffer_s_struct.__cblayout_CBArraysst(i32 0, i32 2, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", %struct.__cblayout_CBArrays) %[[HANDLE2]], ptr @CBArrays.cb, align 4
+
+RWBuffer<float> Buf;
+
+[numthreads(4,1,1)]
+void main() {
+  //Buf[0] = a1 + b1.z + c1[2] + a.f1.y;
----------------
damyanp wrote:

```
  //Buf[0] = a1 + b1.z + c1[2] + a.f1.y;
```

Is there a reason to keep this commented out line around?

https://github.com/llvm/llvm-project/pull/124886