P4 to Git Change 1174182 by jatang@jatang-opencl-hsa-stg1 on 2015/07/27 10:26:56

ECR #333753 - denorms-are-zero runtime change.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#514 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#196 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#315 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.hpp#92 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#236 edit


[ROCm/clr commit: b96feac3b6]

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

@@ -146,6 +146,9 @@ Settings::Settings()
 
     // Use host queue for device enqueuing by default
     useDeviceQueue_ = GPU_USE_DEVICE_QUEUE;
+
+    // Don't support Denormals for single precision by default
+    singleFpDenorm_ = false;
 }
 
 bool
@@ -251,6 +254,7 @@ Settings::create(
     case CAL_TARGET_BAFFIN:
         // Disable tiling aperture on VI+
         linearPersistentImage_ = true;
+        singleFpDenorm_ = true;
         viPlus_ = true;
         // Fall through to CI ...
     case CAL_TARGET_KALINDI:
@@ -547,6 +551,19 @@ Settings::override()
     if (!flagIsDefault(GPU_RESOURCE_CACHE_SIZE)) {
         resourceCacheSize_ = GPU_RESOURCE_CACHE_SIZE * Mi;
     }
+
+    if (!flagIsDefault(AMD_GPU_FORCE_SINGLE_FP_DENORM)) {
+        switch (AMD_GPU_FORCE_SINGLE_FP_DENORM) {
+        case 0:
+            singleFpDenorm_ = false;
+            break;
+        case 1:
+            singleFpDenorm_ = true;
+            break;
+        default:
+            break;
+        }
+    }
 }
 
 } // namespace gpu