SWDEV-361730 - Fix to match the kernel arg allocation sizes between gpu visible mem and host/runtime mem

Change-Id: I6495aa57e97b46373c1d90cc230435780fdcf3ef


[ROCm/clr commit: b2665ccc4b]

projects/clr/rocclr/device/pal/palkernel.cpp

@@ -409,7 +409,7 @@ hsa_kernel_dispatch_packet_t* HSAILKernel::loadArguments(VirtualGPU& gpu, const
         }
         break;
       case amd::KernelParameterDescriptor::HiddenRemainderX:
-        WriteAqlArgAt(hidden_arguments, static_cast<uint16_t>(global[0] % local[0]), 
+        WriteAqlArgAt(hidden_arguments, static_cast<uint16_t>(global[0] % local[0]),
                       it.size_, it.offset_);
         break;
       case amd::KernelParameterDescriptor::HiddenRemainderY:
@@ -447,7 +447,8 @@ hsa_kernel_dispatch_packet_t* HSAILKernel::loadArguments(VirtualGPU& gpu, const
 
   // Load all kernel arguments
   if (signature.version() == kernel.signature().version()) {
-    memcpy(aqlArgBuf, parameters, argsBufferSize());
+    memcpy(aqlArgBuf, parameters, std::min(static_cast<uint32_t>(argsBufferSize()),
+                                           signature.paramsSize()));
   }
 
   // hsa_kernel_dispatch_packet_t disp;

projects/clr/rocclr/device/rocm/rocvirtual.cpp

@@ -2964,7 +2964,8 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes,
       argBuffer = reinterpret_cast<address>(allocKernArg(gpuKernel.KernargSegmentByteSize(),
                                             gpuKernel.KernargSegmentAlignment()));
       // Load all kernel arguments
-      memcpy(argBuffer, parameters, gpuKernel.KernargSegmentByteSize());
+      memcpy(argBuffer, parameters, std::min(gpuKernel.KernargSegmentByteSize(),
+                                             signature.paramsSize()));
     }
 
     // Check for group memory overflow