intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-19 14:46:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Separate struct EnqueueOperation declaration and implementation

Browse Source 
Change-Id: I537660867a1c98f957280237c14b7a1554fce3db
This commit is contained in:
Zdanowicz, Zbigniew
2018-04-09 16:39:32 +02:00
committed by sys_ocldev
parent 2374b27bf5
commit e51cb6bd0b
15 changed files with 169 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
runtime/command_queue/enqueue_kernel.h
View File

@@ -33,48 +33,6 @@
namespace OCLRT {
template <typename GfxFamily>
struct EnqueueOperation<GfxFamily, CL_COMMAND_NDRANGE_KERNEL> {
static size_t getSizeRequiredCS(bool reserveProfilingCmdsSpace, bool reservePerfCounters, CommandQueue &commandQueue, const Kernel *pKernel) {
size_t size = sizeof(typename GfxFamily::GPGPU_WALKER) + KernelCommandsHelper<GfxFamily>::getSizeRequiredCS() +
sizeof(typename GfxFamily::PIPE_CONTROL) * (KernelCommandsHelper<GfxFamily>::isPipeControlWArequired() ? 2 : 1);
size += PreemptionHelper::getPreemptionWaCsSize<GfxFamily>(commandQueue.getDevice());
if (reserveProfilingCmdsSpace) {
size += 2 * sizeof(typename GfxFamily::PIPE_CONTROL) + 4 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
}
if (reservePerfCounters) {
//start cmds
//P_C: flush CS & TimeStamp BEGIN
size += 2 * sizeof(typename GfxFamily::PIPE_CONTROL);
//SRM NOOPID & Frequency
size += 2 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//gp registers
size += OCLRT::INSTR_GENERAL_PURPOSE_COUNTERS_COUNT * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//report perf count
size += sizeof(typename GfxFamily::MI_REPORT_PERF_COUNT);
//user registers
size += commandQueue.getPerfCountersUserRegistersNumber() * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//end cmds
//P_C: flush CS & TimeStamp END;
size += 2 * sizeof(typename GfxFamily::PIPE_CONTROL);
//OA buffer (status head, tail)
size += 3 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//report perf count
size += sizeof(typename GfxFamily::MI_REPORT_PERF_COUNT);
//gp registers
size += OCLRT::INSTR_GENERAL_PURPOSE_COUNTERS_COUNT * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//SRM NOOPID & Frequency
size += 2 * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
//user registers
size += commandQueue.getPerfCountersUserRegistersNumber() * sizeof(typename GfxFamily::MI_STORE_REGISTER_MEM);
}
size += GpgpuWalkerHelper<GfxFamily>::getSizeForWADisableLSQCROPERFforOCL(pKernel);
return size;
}
};
template <typename GfxFamily>
cl_int CommandQueueHw<GfxFamily>::enqueueKernel(
cl_kernel clKernel,