[ROCm][Quantization][Kernel] Use FP8 FNUZ when OCP flag is 0 or undef…

…ined (vllm-project#13851) Signed-off-by: Hollow Man <hollowman@opensuse.org>
martihns24 · Feb 27, 2025 · a31614e · a31614e
1 parent f959039
commit a31614e
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/csrc/quantization/fp8/amd/quant_utils.cuh b/csrc/quantization/fp8/amd/quant_utils.cuh
@@ -24,12 +24,12 @@ __inline__ __device__ Tout scaled_vec_conversion(const Tin& x,
   return x;
 }
 
-    #if HIP_FP8_TYPE_FNUZ
-using fp8_type = __hip_fp8_e4m3_fnuz;
-using fp8x2_type = __hip_fp8x2_e4m3_fnuz;
-    #elif HIP_FP8_TYPE_OCP
+    #if HIP_FP8_TYPE_OCP
 using fp8_type = __hip_fp8_e4m3;
 using fp8x2_type = __hip_fp8x2_e4m3;
+    #else
+using fp8_type = __hip_fp8_e4m3_fnuz;
+using fp8x2_type = __hip_fp8x2_e4m3_fnuz;
     #endif
 
 // fp8 -> half