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

P4 to Git Change 1089823 by gandryey@gera-dev-w7 on 2014/10/21 17:52:16

Browse Source 
ECR #304775 - Move OCL runtime to the latest HSA1.0 spec

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/brig_loader.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/gpu/build/Makefile.gpu#24 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/build/Makefile.gpu#57 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudefs.hpp#115 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#471 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#267 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.hpp#102 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuprogram.cpp#181 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusched.hpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#335 edit
This commit is contained in:
foreman
2014-10-21 18:07:39 -04:00
parent 1a5e2fabe0
commit 692a1b9f9a
7 changed files with 76 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/gpu/gpuvirtual.cpp
+12 -11
View File
@@ -14,7 +14,7 @@
#include "device/gpu/gputhreadtrace.hpp"
#include "device/gpu/gputimestamp.hpp"
#include "device/gpu/gpublit.hpp"
#include "newcore.h"
#include "hsa.h"
#include "sc-hsa/Interface/SCHSAInterface.h"
#include <fstream>
#include <sstream>
@@ -342,7 +342,7 @@ VirtualGPU::createVirtualQueue(uint deviceQueueSize)
AmdAqlWrap* slots = reinterpret_cast<AmdAqlWrap*>(&header[1]);
for (uint i = 0; i < numSlots; ++i) {
uint64_t argStart = vaBase + argOffs + i * singleArgSize;
slots[i].aql.kernel_arg_address = argStart;
slots[i].aql.kernarg_address = argStart;
slots[i].wait_list = argStart + dev().info().maxParameterSize_ + 64;
}
// Upload data back to local memory
@@ -576,7 +576,8 @@ VirtualGPU::create(
bool
VirtualGPU::allocHsaQueueMem()
{
amd_queue_t queue = {0};
amd_queue_t queue;
memset(&queue, 0, sizeof(queue));
hsaQueueMem_ = new gpu::Memory(dev(), sizeof(queue));
if (hsaQueueMem_ == NULL) {
return false;
@@ -1726,7 +1727,7 @@ VirtualGPU::submitKernelInternalHSA(
}
// Program the kernel arguments for the GPU execution
HsaAqlDispatchPacket* aqlPkt =
hsa_dispatch_packet_t* aqlPkt =
hsaKernel.loadArguments(*this, kernel, sizes, parameters, nativeMem,
vmDefQueue, &vmParentWrap, memList);
if (NULL == aqlPkt) {
@@ -1793,12 +1794,12 @@ VirtualGPU::submitKernelInternalHSA(
print << "\tState: " << eventD->state <<
"; Counter: " << eventD->counter << "\n";
}
print << "WorkGroupSize[ " << wraps[i].aql.workgroup_size[0] << ", ";
print << wraps[i].aql.workgroup_size[1] << ", ";
print << wraps[i].aql.workgroup_size[2] << "]\n";
print << "GridSize[ " << wraps[i].aql.grid_size[0] << ", ";
print << wraps[i].aql.grid_size[1] << ", ";
print << wraps[i].aql.grid_size[2] << "]\n";
print << "WorkGroupSize[ " << wraps[i].aql.workgroup_size_x << ", ";
print << wraps[i].aql.workgroup_size_y << ", ";
print << wraps[i].aql.workgroup_size_z << "]\n";
print << "GridSize[ " << wraps[i].aql.grid_size_x << ", ";
print << wraps[i].aql.grid_size_y << ", ";
print << wraps[i].aql.grid_size_z << "]\n";
uint64_t* kernels = (uint64_t*)(
const_cast<Memory*>(hsaKernel.prog().kernelTable())->map(this));
@@ -1819,7 +1820,7 @@ VirtualGPU::submitKernelInternalHSA(
printf("Error: couldn't find child kernel!\n");
continue;
}
uint offsArg = wraps[i].aql.kernel_arg_address -
uint offsArg = wraps[i].aql.kernarg_address -
gpuDefQueue->virtualQueue_->vmAddress();
address argum = gpuDefQueue->virtualQueue_->data() + offsArg;
print << "Kernel: " << child->name() << "\n";