P4 to Git Change 1519129 by skudchad@skudchad_test2_win_opencl on 2018/02/23 22:04:04

SWDEV-145570 -  Add new info stats for VDI layer which are reported by HIP

	ReviewBoardURL = http://ocltc.amd.com/reviews/r/14290/diff/

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_device.cpp#70 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/cpu/cpudevice.cpp#283 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#296 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#585 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#413 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#74 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#75 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#83 edit

rocclr/runtime/device/pal/palvirtual.cpp

@@ -479,7 +479,7 @@ void VirtualGPU::MemoryDependency::clear(bool all) {
 
 VirtualGPU::DmaFlushMgmt::DmaFlushMgmt(const Device& dev) : cbWorkload_(0), dispatchSplitSize_(0) {
   aluCnt_ = dev.info().simdPerCU_ * dev.info().simdWidth_ * dev.info().maxComputeUnits_;
-  maxDispatchWorkload_ = static_cast<uint64_t>(dev.info().maxClockFrequency_) *
+  maxDispatchWorkload_ = static_cast<uint64_t>(dev.info().maxEngineClockFrequency_) *
       // find time in us
       dev.settings().maxWorkloadTime_ * aluCnt_;
   resetCbWorkload(dev);
@@ -487,7 +487,7 @@ VirtualGPU::DmaFlushMgmt::DmaFlushMgmt(const Device& dev) : cbWorkload_(0), disp
 
 void VirtualGPU::DmaFlushMgmt::resetCbWorkload(const Device& dev) {
   cbWorkload_ = 0;
-  maxCbWorkload_ = static_cast<uint64_t>(dev.info().maxClockFrequency_) *
+  maxCbWorkload_ = static_cast<uint64_t>(dev.info().maxEngineClockFrequency_) *
       // find time in us
       dev.settings().minWorkloadTime_ * aluCnt_;
 }
@@ -2010,7 +2010,7 @@ void VirtualGPU::PostDeviceEnqueue(
     gpuDefQueue->schedParams_->data())[gpuDefQueue->schedParamIdx_];
   param->signal = 1;
   // Scale clock to 1024 to avoid 64 bit div in the scheduler
-  param->eng_clk = (1000 * 1024) / dev().info().maxClockFrequency_;
+  param->eng_clk = (1000 * 1024) / dev().info().maxEngineClockFrequency_;
   param->hw_queue = patchStart + sizeof(uint32_t) /* Rewind packet*/;
   param->hsa_queue = gpuDefQueue->hsaQueueMem()->vmAddress();
   param->releaseHostCP = 0;