P4 to Git Change 1567428 by gandryey@gera-w8 on 2018/06/12 18:39:23

SWDEV-79445 - OCL generic changes and code clean-up
	- Optimize setup of kernel arguments. Stage 2.
	- Add HW ABI support in the abstraction layer
	- Remove arguments parsing loop from the kernel launch. Memory processing will be responsible for dependency tracking and  patching of arguments.

	http://ocltc.amd.com/reviews/r/15122/

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/device.cpp#221 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#307 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#325 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palblit.cpp#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#53 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#17 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.hpp#9 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#107 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#53 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#36 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.cpp#30 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/kernel.hpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/program.cpp#95 edit


[ROCm/clr commit: 1be400ff01]

projects/clr/rocclr/runtime/platform/kernel.cpp

@@ -243,13 +243,17 @@ void KernelParameters::release(address mem, const amd::Device& device) const {
 }
 
 KernelSignature::KernelSignature(const std::vector<KernelParameterDescriptor>& params,
-  const std::string& attrib)
+  const std::string& attrib,
+    const std::vector<KernelParameterDescriptor>& hiddenParams,
+    uint32_t version)
   : params_(params)
+  , hiddenParams_(hiddenParams)
   , attributes_(attrib)
   , paramsSize_(0)
   , numMemories_(0)
   , numSamplers_(0)
-  , numQueues_(0) {
+  , numQueues_(0)
+  , version_(version) {
   size_t maxOffset = 0;
   size_t last = 0;
   // Find the last entry
@@ -283,7 +287,15 @@ KernelSignature::KernelSignature(const std::vector<KernelParameterDescriptor>& p
     if (lastSize == 0 /* local mem */) {
       lastSize = sizeof(cl_mem);
     }
-    paramsSize_ = params[last].offset_ + alignUp(lastSize, sizeof(intptr_t));
+    // Note: It's a special case. HW ABI expects 64 bit for SRD, regardless of the binary.
+    // Force the size to 64 bit for those cases.
+    if ((params[last].info_.oclObject_ == amd::KernelParameterDescriptor::ImageObject) ||
+        (params[last].info_.oclObject_ == amd::KernelParameterDescriptor::SamplerObject) ||
+        (params[last].info_.oclObject_ == amd::KernelParameterDescriptor::QueueObject)) {
+      lastSize = alignUp(lastSize, sizeof(uint64_t));
+    }
+    paramsSize_ = params[last].offset_ + lastSize;
+    paramsSize_ = alignUp(paramsSize_, sizeof(intptr_t));
   }
 }
 }  // namespace amd

projects/clr/rocclr/runtime/platform/kernel.hpp

@@ -36,18 +36,30 @@ class Program;
 class KernelSignature : public HeapObject {
  private:
   std::vector<KernelParameterDescriptor> params_;
+  std::vector<KernelParameterDescriptor> hiddenParams_;
   std::string attributes_;  //!< The kernel attributes
   uint32_t  paramsSize_;
   uint32_t  numMemories_;
   uint32_t  numSamplers_;
   uint32_t  numQueues_;
+  uint32_t  version_;
 
  public:
+  enum {
+    ABIVersion_0 = 0,   //! ABI constructed based on the OCL semantics
+    ABIVersion_1 = 1    //! ABI constructed based on the HW ABI returned from the compiler
+  };
+
   //! Default constructor
-  KernelSignature() : paramsSize_(0), numMemories_(0), numSamplers_(0), numQueues_(0) {}
+  KernelSignature():
+    paramsSize_(0), numMemories_(0), numSamplers_(0),
+    numQueues_(0), version_(ABIVersion_0) {}
 
   //! Construct a new signature.
-  KernelSignature(const std::vector<KernelParameterDescriptor>& params, const std::string& attrib);
+  KernelSignature(const std::vector<KernelParameterDescriptor>& params,
+    const std::string& attrib,
+    const std::vector<KernelParameterDescriptor>& hiddenParams,
+    uint32_t version);
 
   //! Return the number of parameters
   size_t numParameters() const { return params_.size(); }
@@ -72,8 +84,17 @@ class KernelSignature : public HeapObject {
   //! Returns the number of queue objects.
   uint32_t numQueues() const { return numQueues_; }
 
+  //! Returns the signature version
+  uint32_t version() const { return version_; }
+
   //! Return the kernel attributes
   const std::string& attributes() const { return attributes_; }
+
+  const std::vector<KernelParameterDescriptor>& hiddenParameters() const
+    { return hiddenParams_; }
+
+  const std::vector<KernelParameterDescriptor>& parameters() const
+    { return params_; }
 };
 
 // @todo: look into a copy-on-write model instead of copy-on-read.

projects/clr/rocclr/runtime/platform/program.cpp

@@ -604,8 +604,8 @@ bool Program::ParseAllOptions(const std::string& options, option::Options& parse
 }
 
 bool Symbol::setDeviceKernel(const Device& device, const device::Kernel* func) {
-  // FIXME_lmoriche: check that the signatures are compatible
-  if (deviceKernels_.size() == 0) {
+  if (deviceKernels_.size() == 0 ||
+      (func->signature().version() > KernelSignature::ABIVersion_0)) {
     signature_ = func->signature();
   }
   deviceKernels_[&device] = func;