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

P4 to Git Change 1594574 by vsytchen@vsytchen-win10 on 2018/08/16 13:43:33

Browse Source 
SWDEV-159881 - [OCL][ROCm] Add SVM coarse-grain buffer support with device memory (Part 2)

	1. Implement clEnqueueSvmMap/Unmap using a staging buffer
	2. Enable device memory coarse grain SVM for OCL only with single device contexts.

	ReviewBoardURL = http://ocltc.amd.com/reviews/r/15616/diff/

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#94 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#30 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#38 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#63 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#296 edit


[ROCm/clr commit: 8dcc948d37]
This commit is contained in:
foreman
2018-08-16 13:59:59 -04:00
parent f5c414dfba
commit 9c2c23a3a3
6 changed files with 136 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/rocm/rocmemory.cpp
+19 -8
View File
@@ -105,11 +105,6 @@ void* Memory::allocMapTarget(const amd::Coord3D& origin, const amd::Coord3D& reg
if (IsPersistentDirectMap()) {
return (static_cast<char*>(persistent_host_ptr_) + origin[0]);
}
// Otherwise, check for host memory.
void* hostMem = owner()->getHostMem();
if (hostMem != nullptr) {
return (static_cast<char*>(hostMem) + origin[0]);
}
// Allocate one if needed.
if (indirectMapCount_ == 1) {
@@ -124,7 +119,17 @@ void* Memory::allocMapTarget(const amd::Coord3D& origin, const amd::Coord3D& reg
return nullptr;
}
}
return reinterpret_cast<address>(mapMemory_->getHostMem()) + origin[0];
void* mappedMemory = nullptr;
if (owner()->getSvmPtr() != nullptr) {
owner()->commitSvmMemory();
mappedMemory = owner()->getSvmPtr();
} else {
mappedMemory = reinterpret_cast<address>(mapMemory_->getHostMem()) + origin[0];
}
return mappedMemory;
}
void Memory::decIndMapCount() {
@@ -584,7 +589,7 @@ void Buffer::destroy() {
cl_mem_flags memFlags = owner()->getMemFlags();
if (owner()->getSvmPtr() != nullptr) {
if (!dev().settings().enableCoarseGrainSVM_) {
if (dev().forceFineGrain(owner())) {
memFlags |= CL_MEM_SVM_FINE_GRAIN_BUFFER;
}
const bool isFineGrain = memFlags & CL_MEM_SVM_FINE_GRAIN_BUFFER;
@@ -652,7 +657,7 @@ bool Buffer::create() {
cl_mem_flags memFlags = owner()->getMemFlags();
if (owner()->getSvmPtr() != nullptr) {
if (!dev().settings().enableCoarseGrainSVM_) {
if (dev().forceFineGrain(owner())) {
memFlags |= CL_MEM_SVM_FINE_GRAIN_BUFFER;
flags_ |= HostMemoryDirectAccess;
}
@@ -669,6 +674,12 @@ bool Buffer::create() {
deviceMemory_ = owner()->getSvmPtr();
}
if (!isFineGrain &&
(owner()->parent() != nullptr) &&
(owner()->parent()->getSvmPtr() != nullptr)) {
owner()->parent()->commitSvmMemory();
}
if (dev().settings().apuSystem_ || !isFineGrain) {
const_cast<Device&>(dev()).updateFreeMemory(size(), false);
}