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

P4 to Git Change 1466505 by wchau@wchau_OCL_boltzmann on 2017/10/04 16:34:02

Browse Source 
SWDEV-134554 - OCLPerfConcurrency is failed on Vega10 Lnx Pro after CL#1465654
	- fixed the issue by limit the number of asyn queues to 8, which matches the number of pipes.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdefs.hpp#12 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#68 edit


[ROCm/clr commit: 051ffe4780]
This commit is contained in:
foreman
2017-10-04 16:45:05 -04:00
parent 3533f385ad
commit ab029df2e0
2 changed files with 3 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/rocclr/runtime/device/rocm/rocdefs.hpp
+2
View File
@@ -58,4 +58,6 @@ static const AMDDeviceInfo DeviceInfo[] = {
/* TARGET VEGA10 */ {HSA_VEGA10_ID, "", "gfx900", "gfx900", 4, 16, 1, 256, 64 * Ki, 32, 0, 0},
/* TARGET VEGA10_HBCC */ {HSA_VEGA10_HBCC_ID, "", "gfx901", "gfx901", 4, 16, 1, 256, 64 * Ki, 32, 0, 0}};
}
const uint kMaxAsyncQueues = 8; // set to match the number of pipes, which is 8
#endif
projects/clr/rocclr/runtime/device/rocm/rocdevice.cpp
+1 -4
View File
@@ -1113,10 +1113,7 @@ bool Device::populateOCLDeviceConstants() {
info_.localMemSizePerCU_ = deviceInfo_.localMemSizePerCU_;
info_.localMemBanks_ = deviceInfo_.localMemBanks_;
info_.gfxipVersion_ = deviceInfo_.gfxipVersion_;
if (HSA_STATUS_SUCCESS !=
hsa_agent_get_info(_bkendDevice, HSA_AGENT_INFO_QUEUES_MAX, &info_.numAsyncQueues_)) {
return false;
}
info_.numAsyncQueues_ = kMaxAsyncQueues;
info_.numRTQueues_ = info_.numAsyncQueues_;
if (HSA_STATUS_SUCCESS !=
hsa_agent_get_info(_bkendDevice, (hsa_agent_info_t)HSA_AMD_AGENT_INFO_COMPUTE_UNIT_COUNT, &info_.numRTCUs_)) {