P4 to Git Change 1767752 by wchau@wchau_OCL_Linux on 2019/04/09 22:58:03

SWDEV-165259 - Update OpenCL runtime to support MsgPack metadata
	- Add support for the V3 code objects

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/devkernel.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devkernel.hpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devprogram.cpp#39 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/devprogram.hpp#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#336 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#134 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/inc/core/palCmdBuffer.h#63 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/src/core/hw/gfxip/gfx6/gfx6ComputeCmdBuffer.cpp#63 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palbe/src/core/hw/gfxip/gfx9/gfx9ComputeCmdBuffer.cpp#69 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#77 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#90 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.cpp#76 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#21 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#130 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.cpp#52 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rockernel.hpp#27 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.cpp#103 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocprogram.hpp#47 edit

rocclr/runtime/device/pal/palprogram.cpp

@@ -747,6 +747,25 @@ bool LightningProgram::setKernels(amd::option::Options* options, void* binary, s
     return false;
   }
 
+#if defined(USE_COMGR_LIBRARY)
+  for (const auto &kernelMeta : kernelMetadataMap_) {
+    auto kernelName = kernelMeta.first;
+    auto kernel = new LightningKernel(kernelName, this,
+                                      options->origOptionStr + ProcessOptions(options));
+    kernels()[kernelName] = kernel;
+
+    if (!kernel->init()) {
+      return false;
+    }
+
+    kernel->setUniformWorkGroupSize(options->oVariables->UniformWorkGroupSize);
+
+    // Find max scratch regs used in the program. It's used for scratch buffer preallocation
+    // with dynamic parallelism, since runtime doesn't know which child kernel will be called
+    maxScratchRegs_ =
+        std::max(static_cast<uint>(kernel->workGroupInfo()->scratchRegs_), maxScratchRegs_);
+  }
+#else
   // Get the list of kernels
   std::vector<std::string> kernelNameList;
   status = executable_->IterateSymbols(GetKernelNamesCallback, &kernelNameList);
@@ -781,7 +800,7 @@ bool LightningProgram::setKernels(amd::option::Options* options, void* binary, s
     maxScratchRegs_ =
         std::max(static_cast<uint>(kernel->workGroupInfo()->scratchRegs_), maxScratchRegs_);
   }
-
+#endif // defined(USE_COMGR_LIBRARY)
   DestroySegmentCpuAccess();
 #endif // defined(WITH_LIGHTNING_COMPILER) || defined(USE_COMGR_LIBRARY)
   return true;