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

SWDEV-276120 - Remove support for barrier sync

Browse Source 
ROC_BARRIER_SYNC will not work with direct dispatch.
Remove and cleanup.

Change-Id: I81368b2e65039477bd0343bb92708dab48867db6
This commit is contained in:
Saleel Kudchadker
2021-04-05 15:23:07 -07:00
parent 48e470a14c
commit aa38af8c96
8 changed files with 30 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/device/rocm/rocdevice.cpp
+3 -9
View File
@@ -1606,11 +1606,6 @@ device::VirtualDevice* Device::createVirtualDevice(amd::CommandQueue* queue) {
cooperative = amd::IS_HIP && settings().enableCoopGroups_;
profiling = amd::IS_HIP;
}
// If barrier is disabled, then profiling should be enabled to make sure HSA signal is
// attached for every dispatch
else if (!settings().barrier_sync_) {
queue->properties().set(CL_QUEUE_PROFILING_ENABLE);
}
// Initialization of heap and other resources occur during the command
// queue creation time.
const std::vector<uint32_t> defaultCuMask = {};
@@ -1872,7 +1867,7 @@ void* Device::hostAlloc(size_t size, size_t alignment, MemorySegment mem_seg) co
}
case kNoAtomics :
// If runtime disables barrier, then all host allocations must have L2 disabled
if ((settings().barrier_sync_) && (system_coarse_segment_.handle != 0)) {
if (system_coarse_segment_.handle != 0) {
segment = system_coarse_segment_;
break;
}
@@ -1908,10 +1903,9 @@ void* Device::hostAgentAlloc(size_t size, const AgentInfo& agentInfo, bool atomi
void* ptr = nullptr;
const hsa_amd_memory_pool_t segment =
// If runtime disables barrier, then all host allocations must have L2 disabled
(!atomics && settings().barrier_sync_) ?
(agentInfo.coarse_grain_pool.handle != 0) ?
!atomics ? (agentInfo.coarse_grain_pool.handle != 0) ?
agentInfo.coarse_grain_pool : agentInfo.fine_grain_pool
: agentInfo.fine_grain_pool;
: agentInfo.fine_grain_pool;
assert(segment.handle != 0);
hsa_status_t stat = hsa_amd_memory_pool_allocate(segment, size, 0, &ptr);
ClPrint(amd::LOG_DEBUG, amd::LOG_MEM, "Allocate hsa host memory %p, size 0x%zx", ptr, size);