AI/rocm-systems
2
0
Fork 0
Código Incidencias Pull Requests Acciones Paquetes Proyectos Lanzamientos Wiki Actividad

P4 to Git Change 1182228 by gandryey@gera-dev-w7 on 2015/08/19 12:22:39

Explorar el Código 
EPR #419072 - [OpenCL2.0] Enable 16MB large on device queues
	- Add mask_groups argument into the library for the empty slot spreading

Affected files ...

... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/devenq.h#11 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/enqueue.cl#10 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/eprep.cl#6 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/events.cl#4 edit
... //depot/stg/opencl/drivers/opencl/library/hsa/hsail/src/devenq/schedule.cl#11 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpudevice.cpp#520 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusched.hpp#18 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuvirtual.cpp#373 edit
Este commit está contenido en:
foreman
2015-08-19 12:37:48 -04:00
padre 88e30fccb8
commit 4e5c499ace
Se han modificado 3 ficheros con 4 adiciones y 5 borrados
Descargar archivo de parche Descargar archivo de diferencias Expandir todos los archivos Contraer todos los archivos
rocclr/runtime/device/gpu/gpudevice.cpp
+1 -1
Ver fichero
@@ -489,7 +489,7 @@ void NullDevice::fillDeviceInfo(
info_.queueOnDeviceProperties_ = info_.queueOnDeviceProperties_ =
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_PROFILING_ENABLE; CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_PROFILING_ENABLE;
info_.queueOnDevicePreferredSize_ = 256 * Ki; info_.queueOnDevicePreferredSize_ = 256 * Ki;
info_.queueOnDeviceMaxSize_ = 12 * Mi; info_.queueOnDeviceMaxSize_ = 16 * Mi;
info_.maxOnDeviceQueues_ = 1; info_.maxOnDeviceQueues_ = 1;
info_.maxOnDeviceEvents_ = settings().numDeviceEvents_; info_.maxOnDeviceEvents_ = settings().numDeviceEvents_;
info_.globalVariablePreferredTotalSize_ = static_cast<size_t>(info_.globalMemSize_); info_.globalVariablePreferredTotalSize_ = static_cast<size_t>(info_.globalMemSize_);
rocclr/runtime/device/gpu/gpusched.hpp
+2 -3
Ver fichero
@@ -28,7 +28,7 @@ struct AmdVQueueHeader {
uint32_t command_counter; //!< [LRW] The global counter for the submitted commands into the queue uint32_t command_counter; //!< [LRW] The global counter for the submitted commands into the queue
uint32_t wait_size; //!< [LRO] The wait list size (in clk_event_t) uint32_t wait_size; //!< [LRO] The wait list size (in clk_event_t)
uint32_t arg_size; //!< [LRO] The size of argument buffer (in bytes) uint32_t arg_size; //!< [LRO] The size of argument buffer (in bytes)
uint32_t reserved0; //!< For the future usage uint32_t mask_groups; //!< Processed mask groups by one thread
uint64_t kernel_table; //!< [LRO] Pointer to an array with all kernel objects (ulong for each entry) uint64_t kernel_table; //!< [LRO] Pointer to an array with all kernel objects (ulong for each entry)
uint32_t reserved[2]; //!< For the future usage uint32_t reserved[2]; //!< For the future usage
}; };
@@ -70,8 +70,7 @@ struct SchedulerParam {
uint64_t parentAQL; //!< Host parent AmdAqlWrap packet uint64_t parentAQL; //!< Host parent AmdAqlWrap packet
uint32_t dedicatedQueue; //!< Scheduler uses a dedicated queue uint32_t dedicatedQueue; //!< Scheduler uses a dedicated queue
uint32_t scratchOffset; //!< Scratch buffer offset uint32_t scratchOffset; //!< Scratch buffer offset
uint32_t mask_groups; //!< Processed mask groups by one thread uint32_t reserved[2]; //!< Reserved
uint32_t reserved; //!< Reserved
}; };
} // namespace gpu } // namespace gpu
rocclr/runtime/device/gpu/gpuvirtual.cpp
+1 -1
Ver fichero
@@ -358,6 +358,7 @@ VirtualGPU::createVirtualQueue(uint deviceQueueSize)
header->aql_slot_mask = vaBase + slotMaskOffs; header->aql_slot_mask = vaBase + slotMaskOffs;
header->wait_size = dev().settings().numWaitEvents_; header->wait_size = dev().settings().numWaitEvents_;
header->arg_size = dev().info().maxParameterSize_ + 64; header->arg_size = dev().info().maxParameterSize_ + 64;
header->mask_groups = maskGroups_;
vqHeader_ = new AmdVQueueHeader; vqHeader_ = new AmdVQueueHeader;
if (NULL == vqHeader_) { if (NULL == vqHeader_) {
return false; return false;
@@ -1938,7 +1939,6 @@ VirtualGPU::submitKernelInternalHSA(
param->parentAQL = vmParentWrap; param->parentAQL = vmParentWrap;
param->dedicatedQueue = dev().settings().useDeviceQueue_; param->dedicatedQueue = dev().settings().useDeviceQueue_;
param->useATC = dev().settings().svmFineGrainSystem_; param->useATC = dev().settings().svmFineGrainSystem_;
param->mask_groups = maskGroups_;
// Fill the scratch buffer information // Fill the scratch buffer information
if (hsaKernel.prog().maxScratchRegs() > 0) { if (hsaKernel.prog().maxScratchRegs() > 0) {