P4 to Git Change 1092347 by bsumner@bsumner-lin-opencl on 2014/10/30 15:32:01

ECR #304775 - add flag to force CL_FP_DENORM on gpu

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#475 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#216 edit


[ROCm/clr commit: f229d44faf]

projects/clr/rocclr/runtime/device/gpu/gpudevice.cpp

@@ -486,6 +486,10 @@ void Device::fillDeviceInfo(
     info_.singleFPConfig_       = CL_FP_ROUND_TO_NEAREST | CL_FP_ROUND_TO_ZERO
         | CL_FP_ROUND_TO_INF | CL_FP_INF_NAN | CL_FP_FMA;
 
+    if (GPU_FORCE_SINGLE_FP_DENORM) {
+        info_.singleFPConfig_ |= CL_FP_DENORM;
+    }
+
     if (settings().checkExtension(ClKhrFp64)) {
         info_.doubleFPConfig_   = info_.singleFPConfig_ | CL_FP_DENORM;
     }

projects/clr/rocclr/runtime/utils/flags.hpp

@@ -170,6 +170,8 @@ release(bool, GPU_IFH_MODE, false,                                  \
         "1 = Enable GPU IFH (infinitely fast hardware) mode. Any other value keeps setting disabled.") \
 release(bool, HSAIL_IMAGE_HANDLE_ENABLE, false,                               \
         "Pass image/sampler SRD as pointer instead of blob")                  \
+debug(bool, GPU_FORCE_SINGLE_FP_DENORM, false,                                \
+        "Forces reporting CL_FP_DENORM bit for single precision")             \
 
 
 namespace amd {