AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-307185 - Create heap for device memory allocator

Browse Source 
Pass the allocated heap with the kernel arguments

Change-Id: Icdec09b7f937845c39e21cbca7071dc3ba791af9


[ROCm/clr commit: 7b114a2b8b]
This commit is contained in:
German Andryeyev
2022-03-02 19:19:29 -05:00
parent 8fd4a43f3e
commit 7d5ed33e8f
11 changed files with 103 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/device/pal/palkernel.cpp
+8
View File
@@ -358,6 +358,14 @@ hsa_kernel_dispatch_packet_t* HSAILKernel::loadArguments(VirtualGPU& gpu, const
break;
case amd::KernelParameterDescriptor::HiddenMultiGridSync:
break;
case amd::KernelParameterDescriptor::HiddenHeap:
if (gpu.dev().HeapBuffer() != nullptr) {
// Add heap pointer to the code
size_t heap_ptr = static_cast<size_t>(gpu.dev().HeapBuffer()->virtualAddress());
gpu.addVmMemory(reinterpret_cast<Memory*>(gpu.dev().HeapBuffer()));
WriteAqlArgAt(hidden_arguments, heap_ptr, it.size_, it.offset_);
}
break;
case amd::KernelParameterDescriptor::HiddenBlockCountX:
WriteAqlArgAt(hidden_arguments, static_cast<uint32_t>(global[0] / local[0]),
it.size_, it.offset_);