[AMDGPU][Clang] Support bfloat16 arithmetic. (#147541)

Co-authored-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

Author: kosarev

clang/lib/Basic/Targets/AMDGPU.h

@@ -439,6 +439,7 @@ class LLVM_LIBRARY_VISIBILITY AMDGPUTargetInfo final : public TargetInfo {
   // pre-defined macros.
   bool handleTargetFeatures(std::vector<std::string> &Features,
                             DiagnosticsEngine &Diags) override {
+    HasFullBFloat16 = true;
     auto TargetIDFeatures =
         getAllPossibleTargetIDFeatures(getTriple(), getArchNameAMDGCN(GPUKind));
     for (const auto &F : Features) {

clang/test/CodeGen/AMDGPU/full-bf16.c

@@ -0,0 +1,28 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu tahiti -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu fiji -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu gfx900 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu gfx950 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu gfx1010 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu gfx1100 -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa-gnu -target-cpu gfx1250 -emit-llvm -o - %s | FileCheck %s
+
+// CHECK-LABEL: define dso_local bfloat @div(
+// CHECK-SAME: bfloat noundef [[A:%.*]], bfloat noundef [[B:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[RETVAL:%.*]] = alloca bfloat, align 2, addrspace(5)
+// CHECK-NEXT:    [[A_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
+// CHECK-NEXT:    [[B_ADDR:%.*]] = alloca bfloat, align 2, addrspace(5)
+// CHECK-NEXT:    [[RETVAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[RETVAL]] to ptr
+// CHECK-NEXT:    [[A_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[A_ADDR]] to ptr
+// CHECK-NEXT:    [[B_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[B_ADDR]] to ptr
+// CHECK-NEXT:    store bfloat [[A]], ptr [[A_ADDR_ASCAST]], align 2
+// CHECK-NEXT:    store bfloat [[B]], ptr [[B_ADDR_ASCAST]], align 2
+// CHECK-NEXT:    [[TMP0:%.*]] = load bfloat, ptr [[A_ADDR_ASCAST]], align 2
+// CHECK-NEXT:    [[TMP1:%.*]] = load bfloat, ptr [[B_ADDR_ASCAST]], align 2
+// CHECK-NEXT:    [[DIV:%.*]] = fdiv bfloat [[TMP0]], [[TMP1]]
+// CHECK-NEXT:    ret bfloat [[DIV]]
+//
+__bf16 div(__bf16 a, __bf16 b) {
+  return a / b;
+}