P4 to Git Change 1973183 by gandryey@gera-win10 on 2019/07/24 18:26:45

SWDEV-79445 - OCL generic changes and code clean-up
	- Make sure PAL_DISABLE_SDMA is fully functional. CP DMA is used for buffer transfers currently and kernels for images and buffer rect copies.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.cpp#32 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#150 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#92 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#25 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#142 edit


[ROCm/clr commit: 565ef2dac5]

projects/clr/rocclr/runtime/device/pal/palblit.cpp

@@ -1746,9 +1746,10 @@ bool KernelBlitManager::copyBufferRect(device::Memory& srcMemory, device::Memory
   // Fall into the PAL path for rejected transfers
   if (setup_.disableCopyBufferRect_ || gpuMem(srcMemory).isHostMemDirectAccess() ||
       gpuMem(dstMemory).isHostMemDirectAccess()) {
-    result =
+    if (!dev().settings().disableSdma_) {
+      result =
         DmaBlitManager::copyBufferRect(srcMemory, dstMemory, srcRectIn, dstRectIn, sizeIn, entire);
-
+    }
     if (result) {
       synchronize();
       return result;

projects/clr/rocclr/runtime/device/pal/paldevice.cpp

@@ -934,11 +934,8 @@ bool Device::create(Pal::IDevice* device) {
   palSettings->cmdBufBatchedSubmitChainLimit = 0;
   palSettings->disableResourceProcessingManager = true;
   palSettings->numScratchWavesPerCu = settings().numScratchWavesPerCu_;
-
-  if (PAL_DISABLE_SDMA) {
-    // Make sure CP DMA is used only, since compute path won't work in PAL with OCL
-    palSettings->cpDmaCmdCopyMemoryMaxBytes = 0xFFFFFFFF;
-  }
+  // Make sure CP DMA can be used for all possible transfers
+  palSettings->cpDmaCmdCopyMemoryMaxBytes = 0xFFFFFFFF;
 
   // Commit the new settings for the device
   result = iDev()->CommitSettingsAndInit();

projects/clr/rocclr/runtime/device/pal/palsettings.cpp

@@ -145,6 +145,7 @@ Settings::Settings() {
   lcWavefrontSize64_ = true;
   enableHwP2P_ = false;
   imageBufferWar_ = false;
+  disableSdma_ = PAL_DISABLE_SDMA;
 }
 
 bool Settings::create(const Pal::DeviceProperties& palProp,
@@ -338,6 +339,9 @@ bool Settings::create(const Pal::DeviceProperties& palProp,
     imageBufferWar_ = GPU_IMAGE_BUFFER_WAR;
   }
 
+  // Image DMA must be disabled if SDMA is disabled
+  imageDMA_ &= !disableSdma_;
+
   splitSizeForWin7_ = false;
 
 #if defined(_WIN32)

projects/clr/rocclr/runtime/device/pal/palsettings.hpp

@@ -63,7 +63,8 @@ class Settings : public device::Settings {
       uint splitSizeForWin7_ : 1;       //!< DMA flush split size for Win 7
       uint enableHwP2P_ : 1;            //!< Forces HW P2P path for testing
       uint imageBufferWar_ : 1;         //!< Image buffer workaround for Gfx10
-      uint reserved_ : 9;
+      uint disableSdma_ : 1;            //!< Disable SDMA support
+      uint reserved_ : 8;
     };
     uint value_;
   };

projects/clr/rocclr/runtime/device/pal/palvirtual.cpp

@@ -824,7 +824,7 @@ bool VirtualGPU::create(bool profiling, uint deviceQueueSize, uint rtCUs,
     }
 
     // Check if device has SDMA engines
-    if (dev().numDMAEngines() != 0 && !PAL_DISABLE_SDMA) {
+    if (dev().numDMAEngines() != 0 && !dev().settings().disableSdma_) {
       uint sdma;
       // If only 1 SDMA engine is available then use that one, otherwise it's a round-robin manner
       if ((dev().numDMAEngines() < 2) || ((idx + 1) & 0x1)) {