[SYCL] Fixed AMD compilation for bfloat16 to use generic impl. (#10009)

We don't have AMD native bfloat16 support but this fix at least allows
us to use a generic implementation.

---------

Signed-off-by: JackAKirk <[email protected]>

Author: JackAKirk

sycl/include/sycl/ext/oneapi/bfloat16.hpp

@@ -115,16 +115,11 @@ class bfloat16 {
 
   // Unary minus operator overloading
   friend bfloat16 operator-(bfloat16 &lhs) {
-#if defined(__SYCL_DEVICE_ONLY__)
-#if defined(__NVPTX__)
-#if (__SYCL_CUDA_ARCH__ >= 800)
+#if defined(__SYCL_DEVICE_ONLY__) && defined(__NVPTX__) &&                     \
+    (__SYCL_CUDA_ARCH__ >= 800)
     return detail::bitsToBfloat16(__nvvm_neg_bf16(lhs.value));
-#else
-    return -to_float(lhs.value);
-#endif
-#else
+#elif defined(__SYCL_DEVICE_ONLY__) && defined(__SPIR__)
     return bfloat16{-__devicelib_ConvertBF16ToFINTEL(lhs.value)};
-#endif
 #else
     return -to_float(lhs.value);
 #endif

sycl/test-e2e/BFloat16/bfloat16_type.cpp

@@ -6,8 +6,6 @@
 // TODO currently the feature isn't supported on FPGA.
 // UNSUPPORTED: accelerator
 //
-// Not currently supported on HIP.
-// UNSUPPORTED: hip
 
 //==----------- bfloat16_type.cpp - SYCL bfloat16 type test ----------------==//
 //