P4 to Git Change 1529710 by wchau@wchau_OCL_boltzmann on 2018/03/20 17:05:04

SWDEV-145760 - OCL on ROCm Support for respecting target's xnack setting (subtask of SWDEV-134107)

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcompiler.cpp#21 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldefs.hpp#34 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#77 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#58 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/roccompiler.cpp#41 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#16 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#84 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#80 edit

rocclr/runtime/device/rocm/rocdevice.cpp

@@ -592,6 +592,9 @@ bool Device::create() {
   // with dash as delimiter to be compatible with Windows directory name
   std::ostringstream cacheTarget;
   cacheTarget << "AMD-AMDGPU-" << gfxipMajor << "-" << gfxipMinor << "-" << gfxipStepping;
+  if (deviceInfo_.xnackEnabled_) {
+    cacheTarget << "-xnack";
+  }
 
   amd::CacheCompilation* compObj = new amd::CacheCompilation(
       cacheTarget.str(), "_rocm", OCL_CODE_CACHE_ENABLE, OCL_CODE_CACHE_RESET);
@@ -772,6 +775,10 @@ bool Device::populateOCLDeviceConstants() {
 
   std::ostringstream oss;
   oss << "gfx" << gfxipMajor << gfxipMinor << gfxipStepping;
+  if (IS_LIGHTNING && deviceInfo_.xnackEnabled_) {
+    oss << "-xnack";
+  }
+
   ::strcpy(info_.name_, oss.str().c_str());
 
   char device_name[64] = {0};