P4 to Git Change 1531829 by gandryey@gera-w8 on 2018/03/23 18:34:36

SWDEV-79445 - OCL generic changes and code clean-up
	- Implement ConstantBuffer management through a single managed buffer. It reduces the amount of allocated buffers on GPU

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palconstbuf.cpp#6 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palconstbuf.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#48 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#79 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#44 edit

rocclr/runtime/device/pal/palvirtual.cpp

@@ -702,6 +702,7 @@ VirtualGPU::VirtualGPU(Device& device)
       printfDbgHSA_(nullptr),
       tsCache_(nullptr),
       dmaFlushMgmt_(device),
+      writeBuffer_(nullptr),
       hwRing_(0),
       readjustTimeGPU_(0),
       lastTS_(nullptr),
@@ -812,6 +813,13 @@ bool VirtualGPU::create(bool profiling, uint deviceQueueSize, uint rtCUs,
     Unimplemented();
   }
 
+  writeBuffer_ = new ManagedBuffer(*this, dev().settings().stagedXferSize_);
+  if ((writeBuffer_ == nullptr) || !writeBuffer_->create(Resource::RemoteUSWC)) {
+    // We failed to create a constant buffer
+    delete writeBuffer_;
+    return false;
+  }
+
   // Diable double copy optimization,
   // since UAV read from nonlocal is fast enough
   blitSetup.disableCopyBufferToImageOpt_ = true;
@@ -934,6 +942,8 @@ VirtualGPU::~VirtualGPU() {
     delete constBufs_[i];
   }
 
+  delete writeBuffer_;
+
   //! @todo Temporarily keep the buffer mapped for debug purpose
   if (nullptr != schedParams_) {
     schedParams_->unmap(this);
@@ -2803,15 +2813,16 @@ void VirtualGPU::waitEventLock(CommandBatch* cb) {
 }
 
 bool VirtualGPU::allocConstantBuffers() {
-  // Allocate constant buffers, GCN doesn't really have a limit
-  static constexpr uint32_t MinCbSize = 256 * Ki;
+  // Allocate constant buffers. 
+  // Use double size, reported to the app to account for internal arguments
+  const uint32_t MinCbSize = 2 * dev().info().maxParameterSize_;
   uint i;
 
   // Create/reallocate constant buffer resources
   for (i = 0; i < MaxConstBuffersArguments; ++i) {
-    ManagedBuffer* constBuf = new ManagedBuffer(*this, MinCbSize);
+    ConstantBuffer* constBuf = new ConstantBuffer(*writeBuffer_, MinCbSize);
 
-    if ((constBuf != nullptr) && constBuf->create(Resource::RemoteUSWC, true)) {
+    if ((constBuf != nullptr) && constBuf->Create()) {
       addConstBuffer(constBuf);
     } else {
       // We failed to create a constant buffer