forward fix PR 14245

PR vllm-project#14245 broke compilation for ROCm 6.2.

Signed-off-by: Jeff Daily <jeff.daily@amd.com>

Author: jeffdaily

csrc/quantization/fp8/amd/quant_utils.cuh

@@ -19,12 +19,22 @@ __device__ __forceinline__ fp8_type cvt_c10(float const r) {
   return {};
 }
 
+// __hip_fp8_e4m3 only exists starting in ROCm 6.3. The macro
+// HIP_FP8_TYPE_OCP comes from the hip_fp8.h header and also makes
+// its first appearance in ROCm 6.3. Since VLLM_DISPATCH_FP8_TYPES
+// on ROCm instantiates both OCP and FNUZ kernels, we need to replace
+// the new HW cvt with something reasonable that doesn't rely on the
+// ROCm 6.3 feature. This allows compiling on ROCm 6.2 or newer.
 template <>
 __device__ __forceinline__ c10::Float8_e4m3fn cvt_c10(float const r) {
+    #if HIP_FP8_TYPE_OCP
   return c10::Float8_e4m3fn(
       __hip_cvt_float_to_fp8(r, __hip_fp8_e4m3::__default_saturation,
                              __hip_fp8_e4m3::__default_interpret),
       c10::Float8_e4m3fn::from_bits());
+    #else
+  return static_cast<c10::Float8_e4m3fn>(r);
+    #endif
 }
 
 template <>