[clang][amdgpu] Add builtin for struct buffer lds load (llvm#148950) (llvm#3252)

Author: vamovsik

clang/include/clang/Basic/BuiltinsAMDGPU.def

@@ -164,6 +164,7 @@ BUILTIN(__builtin_amdgcn_raw_buffer_load_b96, "V3UiQbiiIi", "n")
 BUILTIN(__builtin_amdgcn_raw_buffer_load_b128, "V4UiQbiiIi", "n")
 
 TARGET_BUILTIN(__builtin_amdgcn_raw_ptr_buffer_load_lds, "vQbv*3IUiiiIiIi", "t", "vmem-to-lds-load-insts")
+TARGET_BUILTIN(__builtin_amdgcn_struct_ptr_buffer_load_lds, "vQbv*3IUiiiiIiIi", "t", "vmem-to-lds-load-insts")
 
 //===----------------------------------------------------------------------===//
 // Ballot builtins.

clang/lib/Sema/SemaAMDGPU.cpp

@@ -37,6 +37,7 @@ bool SemaAMDGPU::CheckAMDGCNBuiltinFunctionCall(unsigned BuiltinID,
 
   switch (BuiltinID) {
   case AMDGPU::BI__builtin_amdgcn_raw_ptr_buffer_load_lds:
+  case AMDGPU::BI__builtin_amdgcn_struct_ptr_buffer_load_lds:
   case AMDGPU::BI__builtin_amdgcn_global_load_lds: {
     constexpr const int SizeIdx = 2;
     llvm::APSInt Size;

clang/test/CodeGenOpenCL/builtins-amdgcn-raw-buffer-load-lds.cl

@@ -0,0 +1,21 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// REQUIRES: amdgpu-registered-target
+// RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx90a -emit-llvm -o - %s | FileCheck %s
+
+// CHECK-LABEL: @test_amdgcn_raw_ptr_buffer_load_lds(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    tail call void @llvm.amdgcn.raw.ptr.buffer.load.lds(ptr addrspace(8) [[RSRC:%.*]], ptr addrspace(3) [[LDS:%.*]], i32 1, i32 [[OFFSET:%.*]], i32 [[SOFFSET:%.*]], i32 2, i32 3)
+// CHECK-NEXT:    ret void
+//
+void test_amdgcn_raw_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local void * lds, int offset, int soffset) {
+    __builtin_amdgcn_raw_ptr_buffer_load_lds(rsrc, lds, 1, offset, soffset, 2, 3);
+}
+
+// CHECK-LABEL: @test_amdgcn_struct_ptr_buffer_load_lds(
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    tail call void @llvm.amdgcn.struct.ptr.buffer.load.lds(ptr addrspace(8) [[RSRC:%.*]], ptr addrspace(3) [[LDS:%.*]], i32 4, i32 [[VINDEX:%.*]], i32 [[VOFFSET:%.*]], i32 [[SOFFSET:%.*]], i32 2, i32 3)
+// CHECK-NEXT:    ret void
+//
+void test_amdgcn_struct_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local void * lds, int size, int vindex, int voffset, int soffset) {
+    __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, 4, vindex, voffset, soffset, 2, 3);
+}

clang/test/SemaOpenCL/builtins-amdgcn-raw-ptr-buffer-load-lds-error.cl

@@ -8,3 +8,10 @@ void test_amdgcn_raw_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local vo
   __builtin_amdgcn_raw_ptr_buffer_load_lds(rsrc, lds, 4, offset, soffset, 0, x); //expected-error{{argument to '__builtin_amdgcn_raw_ptr_buffer_load_lds' must be a constant integer}}
   __builtin_amdgcn_raw_ptr_buffer_load_lds(rsrc, lds, 3, offset, soffset, 0, 0); //expected-error{{invalid size value}} gfx950-note{{size must be 1, 2, 4, 12 or 16}} gfx90a-note{{size must be 1, 2, or 4}}
 }
+
+void test_amdgcn_struct_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local void * lds, int size, int vindex, int voffset, int soffset, int x) {
+  __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, x, vindex, voffset, soffset, 0, 0); //expected-error{{argument to '__builtin_amdgcn_struct_ptr_buffer_load_lds' must be a constant integer}}
+  __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, 4, vindex, voffset, soffset, x, 0); //expected-error{{argument to '__builtin_amdgcn_struct_ptr_buffer_load_lds' must be a constant integer}}
+  __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, 4, vindex, voffset, soffset, 0, x); //expected-error{{argument to '__builtin_amdgcn_struct_ptr_buffer_load_lds' must be a constant integer}}
+  __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, 3, vindex, voffset, soffset, 0, 0); //expected-error{{invalid size value}} gfx950-note{{size must be 1, 2, 4, 12 or 16}} gfx90a-note{{size must be 1, 2, or 4}}
+}

clang/test/SemaOpenCL/builtins-amdgcn-raw-ptr-buffer-load-lds-target-error.cl

@@ -1,6 +1,7 @@
 // RUN: %clang_cc1 -triple amdgcn-unknown-unknown -target-cpu gfx1100 -S -verify -o - %s
 // REQUIRES: amdgpu-registered-target
 
-void test_amdgcn_raw_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local void* lds, int offset, int soffset, int x) {
+void test_amdgcn_raw_ptr_buffer_load_lds(__amdgpu_buffer_rsrc_t rsrc, __local void* lds, int vindex, int offset, int soffset) {
   __builtin_amdgcn_raw_ptr_buffer_load_lds(rsrc, lds, 4, offset, soffset, 0, 0); //expected-error{{needs target feature vmem-to-lds-load-insts}}
+  __builtin_amdgcn_struct_ptr_buffer_load_lds(rsrc, lds, 4, vindex, offset, soffset, 0, 0); //expected-error{{needs target feature vmem-to-lds-load-insts}}
 }

llvm/include/llvm/IR/IntrinsicsAMDGPU.td

@@ -1911,7 +1911,9 @@ class AMDGPUStructBufferLoadLDS : Intrinsic <
    ImmArg<ArgIndex<7>>, IntrNoCallback, IntrNoFree], "", [SDNPMemOperand]>, AMDGPURsrcIntrinsic<0>;
 def int_amdgcn_struct_buffer_load_lds : AMDGPUStructBufferLoadLDS;
 
-class AMDGPUStructPtrBufferLoadLDS : Intrinsic <
+class AMDGPUStructPtrBufferLoadLDS :
+  ClangBuiltin<"__builtin_amdgcn_struct_ptr_buffer_load_lds">,
+  Intrinsic <
   [],
   [AMDGPUBufferRsrcTy,        // rsrc(SGPR)
    LLVMQualPointerType<3>,    // LDS base offset