[ESIMD] Fix errors caused by move of bfloat16 from experimental ns (#… (#7981)

…9143)

The error in LIT test esimd/intel_fp16_converts.cpp is caused by
#6524 which
- moved 'bfloat16' out of 'experimental' namespace
- created a wrapper __devicelib_ConvertBF16ToFINTEL() which simply calls
__spirv_ConvertBF16ToFINTEL()

The fix
- creates a test for bfloat16 conversions.
- allows __devicelib_ConvertBF16ToFINTEL() and
__devicelib_ConvertFToBF16INTEL() for ESIMD context.

Signed-off-by: Vyacheslav N Klochkov <vyacheslav.n.klochkov@intel.com>

Author: v-klochkov

clang/lib/Sema/SemaExpr.cpp

@@ -261,7 +261,9 @@ bool Sema::DiagnoseUseOfDecl(NamedDecl *D, ArrayRef<SourceLocation> Locs,
       const IdentifierInfo *Id = FDecl->getIdentifier();
       if ((getEmissionReason(FDecl) == Sema::DeviceDiagnosticReason::Sycl) &&
           Id && !Id->getName().startswith("__spirv_") &&
-          !Id->getName().startswith("__sycl_")) {
+          !Id->getName().startswith("__sycl_") &&
+          !Id->getName().startswith("__devicelib_ConvertBF16ToFINTEL") &&
+          !Id->getName().startswith("__devicelib_ConvertFToBF16INTEL")) {
         SYCLDiagIfDeviceCode(
             *Locs.begin(), diag::err_sycl_device_function_is_called_from_esimd,
             Sema::DeviceDiagnosticReason::Esimd);

llvm/lib/SYCLLowerIR/ESIMD/ESIMDVerifier.cpp

@@ -54,7 +54,7 @@ static const char *LegalSYCLFunctions[] = {
     "^sycl::_V1::ext::oneapi::sub_group::.+",
     "^sycl::_V1::ext::oneapi::experimental::spec_constant<.+>::.+",
     "^sycl::_V1::ext::oneapi::experimental::this_sub_group",
-    "^sycl::_V1::ext::oneapi::experimental::bfloat16::.+",
+    "^sycl::_V1::ext::oneapi::bfloat16::.+",
     "^sycl::_V1::ext::oneapi::experimental::if_architecture_is"};
 
 static const char *LegalSYCLFunctionsInStatelessMode[] = {

sycl/test/esimd/fp16_converts.cpp

@@ -0,0 +1,50 @@
+// The test verifies support of bfloat16 <-> float conversions
+
+// Checks host+device compilation
+// RUN: %clangxx -fsycl -fsyntax-only %s
+
+// Checks that lowerESIMD pass builds proper vc-intrinsics
+// RUN: %clangxx -O2 -fsycl -c -fsycl-device-only -Xclang -emit-llvm %s -o %t
+// RUN: sycl-post-link -split-esimd -lower-esimd -O0 -S %t -o %t.table
+// RUN: FileCheck %s -input-file=%t_esimd_0.ll
+
+#include <sycl/ext/intel/esimd.hpp>
+#include <sycl/sycl.hpp>
+
+using namespace sycl::ext::intel::esimd;
+
+SYCL_ESIMD_FUNCTION SYCL_EXTERNAL void bf16_vector();
+SYCL_ESIMD_FUNCTION SYCL_EXTERNAL void bf16_scalar();
+
+using bfloat16 = sycl::ext::oneapi::bfloat16;
+
+class EsimdFunctor {
+public:
+  void operator()() __attribute__((sycl_explicit_simd)) {
+    bf16_vector();
+    bf16_scalar();
+  }
+};
+
+template <typename name, typename Func>
+__attribute__((sycl_kernel)) void kernel(Func kernelFunc) {
+  kernelFunc();
+}
+
+SYCL_ESIMD_FUNCTION SYCL_EXTERNAL void bf16_vector() {
+  simd<float, 8> F32 = 0;
+  simd<bfloat16, 8> BF16 = F32;
+  // CHECK: call <8 x half> @llvm.genx.bf.cvt.v8f16.v8f32(<8 x float> {{[^)]+}})
+  simd<float, 8> F32_conv = BF16;
+  // CHECK: call <8 x float> @llvm.genx.bf.cvt.v8f32.v8f16(<8 x half> {{[^)]+}})
+}
+
+SYCL_ESIMD_FUNCTION SYCL_EXTERNAL void bf16_scalar() {
+  // Note that this is the compilation test only. It checks that IR is correct.
+  // The actual support in GPU RT is on the way though.
+  float F32_scalar = 1;
+  bfloat16 BF16_scalar = F32_scalar;
+  // CHECK: call spir_func zeroext i16 @__devicelib_ConvertFToBF16INTEL(float {{[^)]+}})
+  float F32_scalar_conv = BF16_scalar;
+  // CHECK: call spir_func float @__devicelib_ConvertBF16ToFINTEL(i16 {{[^)]+}})
+}