[AMDGPU] Add builtin functions image_bvh_intersect_ray

[arsenm]

clang/include/clang/Basic/BuiltinsAMDGPU.def

221–224

The intrinsic signature suggests the 1st and 4th/5th arguments are overloadable. How does this handle the various supported types?

[yaxunl]

clang/include/clang/Basic/BuiltinsAMDGPU.def

221–224

By convention, we do not define overloaded target specific clang builtins. In stead we use postfixes to indicate the types. In this case, we use the following naming convention:

h - 4/5-th args are half
d - 4/5-th args are double
l - first arg is i64

by default - 1st arg is i32, 4/5-th args are float.

Currently we are missing functions for 4/5-th args are double. I will add them.

[yaxunl]

clang/include/clang/Basic/BuiltinsAMDGPU.def

221–224

It seems we do not support 4/50th args to be double. So the current definitions are all that we support.

[yaxunl]

Add check for ISA. Add comments about argument types.

clang/include/clang/Basic/BuiltinsAMDGPU.def

 //===----------------------------------------------------------------------===//
 // GFX10+ only builtins.
 //===----------------------------------------------------------------------===//
 TARGET_BUILTIN(__builtin_amdgcn_permlane16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts")
 TARGET_BUILTIN(__builtin_amdgcn_permlanex16, "UiUiUiUiUiIbIb", "nc", "gfx10-insts")
 TARGET_BUILTIN(__builtin_amdgcn_mov_dpp8, "UiUiIUi", "nc", "gfx10-insts")
 
 //===----------------------------------------------------------------------===//
+// Raytracing builtins.
+// By default the 1st argument is i32 and the 4/5-th arguments are float4.
+// Postfix l indicates the 1st argument is i64.
+// Postfix h indicates the 4/5-th arguments are half4.
+//===----------------------------------------------------------------------===//
+TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray, "V4UiUifV4fV4fV4fV4Ui", "nc", "gfx10-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_h, "V4UiUifV4fV4hV4hV4Ui", "nc", "gfx10-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_l, "V4UiWUifV4fV4fV4fV4Ui", "nc", "gfx10-insts")
+TARGET_BUILTIN(__builtin_amdgcn_image_bvh_intersect_ray_lh, "V4UiWUifV4fV4hV4hV4Ui", "nc", "gfx10-insts")
+
+//===----------------------------------------------------------------------===//
 // Special builtins.
 //===----------------------------------------------------------------------===//
 BUILTIN(__builtin_amdgcn_read_exec, "WUi", "nc")
 BUILTIN(__builtin_amdgcn_read_exec_lo, "Ui", "nc")
 BUILTIN(__builtin_amdgcn_read_exec_hi, "Ui", "nc")
 
 BUILTIN(__builtin_amdgcn_endpgm, "v", "nr")
 

clang/lib/CodeGen/CGBuiltin.cpp

   case AMDGPU::BI__builtin_amdgcn_read_exec_hi: {
     StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ?
       "exec_lo" : "exec_hi";
     CallInst *CI = cast<CallInst>(
       EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, NormalRead, RegName));
     CI->setConvergent();
     return CI;
   }
+  case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray:
+  case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_h:
+  case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_l:
+  case AMDGPU::BI__builtin_amdgcn_image_bvh_intersect_ray_lh: {
+    llvm::Value *NodePtr = EmitScalarExpr(E->getArg(0));
+    llvm::Value *RayExtent = EmitScalarExpr(E->getArg(1));
+    llvm::Value *RayOrigin = EmitScalarExpr(E->getArg(2));
+    llvm::Value *RayDir = EmitScalarExpr(E->getArg(3));
+    llvm::Value *RayInverseDir = EmitScalarExpr(E->getArg(4));
+    llvm::Value *TextureDescr = EmitScalarExpr(E->getArg(5));
+
+    Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_image_bvh_intersect_ray,
+                                   {NodePtr->getType(), RayDir->getType()});
+    return Builder.CreateCall(F, {NodePtr, RayExtent, RayOrigin, RayDir,
+                                  RayInverseDir, TextureDescr});
+  }
+
   // amdgcn workitem
   case AMDGPU::BI__builtin_amdgcn_workitem_id_x:
     return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024);
   case AMDGPU::BI__builtin_amdgcn_workitem_id_y:
     return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024);
   case AMDGPU::BI__builtin_amdgcn_workitem_id_z:
     return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024);
 

clang/test/CodeGenOpenCL/builtins-amdgcn-raytracing.cl

+// REQUIRES: amdgpu-registered-target
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1030 -S \
+// RUN:   -emit-llvm -cl-std=CL2.0 -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1030 -S \
+// RUN:   -cl-std=CL2.0 -o - %s | FileCheck -check-prefix=ISA %s
+
+// Test llvm.amdgcn.image.bvh.intersect.ray intrinsic.
+
+// The clang builtin functions __builtin_amdgcn_image_bvh_intersect_ray* use
+// postfixes to indicate the types of the 1st, 4th, and 5th arguments.
+// By default, the 1st argument is i32, the 4/5-th arguments are float4.
+// Postfix l indicates the 1st argument is i64 and postfix h indicates
+// the 4/5-th arguments are half4.
+
+typedef unsigned int uint;
+typedef unsigned long ulong;
+typedef float float4 __attribute__((ext_vector_type(4)));
+typedef double double4 __attribute__((ext_vector_type(4)));
+typedef half half4 __attribute__((ext_vector_type(4)));
+typedef uint uint4 __attribute__((ext_vector_type(4)));
+
+// CHECK: call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f32
+// ISA: image_bvh_intersect_ray
+void test_image_bvh_intersect_ray(global uint4* out, uint node_ptr,
+  float ray_extent, float4 ray_origin, float4 ray_dir, float4 ray_inv_dir,
+  uint4 texture_descr)
+{
+  *out = __builtin_amdgcn_image_bvh_intersect_ray(node_ptr, ray_extent,
+           ray_origin, ray_dir, ray_inv_dir, texture_descr);
+}
+
+// CHECK: call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i32.v4f16
+// ISA: image_bvh_intersect_ray
+void test_image_bvh_intersect_ray_h(global uint4* out, uint node_ptr,
+  float ray_extent, float4 ray_origin, half4 ray_dir, half4 ray_inv_dir,
+  uint4 texture_descr)
+{
+  *out = __builtin_amdgcn_image_bvh_intersect_ray_h(node_ptr, ray_extent,
+           ray_origin, ray_dir, ray_inv_dir, texture_descr);
+}
+
+// CHECK: call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f32
+// ISA: image_bvh_intersect_ray
+void test_image_bvh_intersect_ray_l(global uint4* out, ulong node_ptr,
+  float ray_extent, float4 ray_origin, float4 ray_dir, float4 ray_inv_dir,
+  uint4 texture_descr)
+{
+  *out = __builtin_amdgcn_image_bvh_intersect_ray_l(node_ptr, ray_extent,
+           ray_origin, ray_dir, ray_inv_dir, texture_descr);
+}
+
+// CHECK: call <4 x i32> @llvm.amdgcn.image.bvh.intersect.ray.i64.v4f16
+// ISA: image_bvh_intersect_ray
+void test_image_bvh_intersect_ray_lh(global uint4* out, ulong node_ptr,
+  float ray_extent, float4 ray_origin, half4 ray_dir, half4 ray_inv_dir,
+  uint4 texture_descr)
+{
+  *out = __builtin_amdgcn_image_bvh_intersect_ray_lh(node_ptr, ray_extent,
+           ray_origin, ray_dir, ray_inv_dir, texture_descr);
+}
+