Project
Loading...
Searching...
No Matches
GPUReconstructionOCL.h
Go to the documentation of this file.
1// Copyright 2019-2020 CERN and copyright holders of ALICE O2.
2// See https://alice-o2.web.cern.ch/copyright for details of the copyright holders.
3// All rights not expressly granted are reserved.
4//
5// This software is distributed under the terms of the GNU General Public
6// License v3 (GPL Version 3), copied verbatim in the file "COPYING".
7//
8// In applying this license CERN does not waive the privileges and immunities
9// granted to it by virtue of its status as an Intergovernmental Organization
10// or submit itself to any jurisdiction.
11
14
15#ifndef GPURECONSTRUCTIONOCL_H
16#define GPURECONSTRUCTIONOCL_H
17
19
20#ifdef _WIN32
22#else
24#endif
25
26namespace o2::gpu
27{
28struct GPUReconstructionOCLInternals;
29
30class GPUReconstructionOCL : public GPUReconstructionProcessing::KernelInterface<GPUReconstructionOCL, GPUReconstructionDeviceBase>
31{
32 public:
34 ~GPUReconstructionOCL() override;
35
36 template <class T, int32_t I = 0, typename... Args>
37 void runKernelBackend(const krnlSetupTime& _xyz, const Args&... args);
38
39 protected:
40 int32_t InitDevice_Runtime() override;
41 int32_t ExitDevice_Runtime() override;
42
43 virtual int32_t GPUChkErrInternal(const int64_t retval, const char* file, int32_t line) const override;
44
45 void SynchronizeGPU() override;
46 int32_t GPUDebug(const char* state = "UNKNOWN", int32_t stream = -1, bool force = false) override;
47 void SynchronizeStream(int32_t stream) override;
48 void SynchronizeEvents(deviceEvent* evList, int32_t nEvents = 1) override;
49 void StreamWaitForEvents(int32_t stream, deviceEvent* evList, int32_t nEvents = 1) override;
50 bool IsEventDone(deviceEvent* evList, int32_t nEvents = 1) override;
51
52 size_t WriteToConstantMemory(size_t offset, const void* src, size_t size, int32_t stream = -1, deviceEvent* ev = nullptr) override;
53 size_t GPUMemCpy(void* dst, const void* src, size_t size, int32_t stream, int32_t toGPU, deviceEvent* ev = nullptr, deviceEvent* evList = nullptr, int32_t nEvents = 1) override;
54 void ReleaseEvent(deviceEvent ev) override;
55 void RecordMarker(deviceEvent* ev, int32_t stream) override;
56
57 template <class T, int32_t I = 0>
59
62
63 template <class S, class T, int32_t I>
65
67
68 private:
69 static const char* convertErrorToString(int32_t errorCode);
70 template <typename T, typename... Args>
71 static inline int64_t OCLsetKernelParameters_helper(cl_kernel& kernel, int32_t i, const T& firstParameter, const Args&... restOfParameters);
72 template <typename... Args>
73 static int64_t OCLsetKernelParameters(cl_kernel& kernel, const Args&... args);
74 static int64_t clExecuteKernelA(cl_command_queue queue, cl_kernel krnl, size_t local_size, size_t global_size, cl_event* pEvent = nullptr, cl_event* wait = nullptr, cl_int nWaitEvents = 1);
75 int32_t AddKernels();
76};
77
78} // namespace o2::gpu
79
80#endif
int32_t i
o2::gpu::GPUReconstruction * GPUReconstruction_Create_OCL(const o2::gpu::GPUSettingsDeviceBackend &cfg)
configStandalone Light OpenGL display display_light Apply smoothing to points Enable Z buffer Highlight links Highlight seeds Highlight sector tracks Highlight final tracks Show individual Show related Highlight grid Show only individual collision Color clusters belonging to track objects Set line Enable drawing TPC data Enable drawing TOF data Enable drawing magnetic field Set field line step count Invert colors configStandalone Heavy OpenGL display display_heavy Show tracks with ITS contribution Show tracks with TOF contribution Enable propagation of loopers Show clusters at nominal x of pad row for early transformed data Split CE tracks when they cross the central Mark clusters Mark adjacent clusters Hide rejected tracks Apply filter on tracks to be displayed Separate increase X h configStandalone Camera window renderer OpenGL display display_camera Orient the camera such that the y axis is always upwards Full Screen MultiSample Anti Downsampling Anti Enable Vertical Sync Target frame or if the backend cannot limit to the refresh Use OpenGL indirect draws to reduce number of draw calls Display h configStandalone Vulkan display display_vulkan Have uniform buffers in host accessible device memory h configStandalone Internal display display_internal Number of updates the Qt GUI may send ahead of the last one the display has before it waits for the display Opacity of the console background Number of console lines the history is shortened to this when it reaches twice the size Time in seconds an info message fades out after setInfoTime h OpenGL display display Font size Do not use Freetype for font Renderer for event Frontend to use for event ROOT macros used as track filter Console script file
configStandalone Reconstruction rec_tpc multiplier for the combined cluster track error during track following area in cm for the search of for z only used if decrease removal by when above this local occupancy rowx Shift of TPC revert cluster flag edge bit distance to edge exceeds this ZOffset for secondary multiple of radius max P of electron K BB bands crossing max P of electron p BB bands crossing max P of electron d BB bands crossing max P of electron t BB bands crossing min accepted dEdxMaxTPC of the track Additive extra cluster error for Y2 if edge flag set Additive extra cluster error for Y2 if single set Additive extra cluster error for Y2 if or single set Additive extra cluster error for Z2 if or single set Extra errors to add to track when crossing depending on addErrorsCECrossing Inner portion of y range in sector that is not used in searching for extrapolated track candidates factor2YS for track merging factor2K for track merging Margin in cm of Y position when rejecting edge clusters based on uncorrected track Y Max fraction of shared clusters for tracklet Except pad from cluster finding if total number of charges in a fragment is above this Except pad from cluster finding if number of consecutive charges in a fragment is above this Threshold where a timebin is considered disabling the noisy pad check for that pad Treat signal above this minimum as saturated Smoothing factor for the exponential Highly Ionising Particle tail filter Number of timebins to use for the averaging Add additional custom track errors when crossing relative as fraction of Do N fit passes in final fit of< 0:reject at max -n hits") AddOptionRTC(rejectIFCLowRadiusCluster, uint8_t, 1, "", 0, "Reject clusters that get the IFC mask error during refit") AddOptionRTC(dEdxTruncLow, uint8_t, 2, "", 0, "Low truncation threshold, fraction of 128") AddOptionRTC(dEdxTruncHigh, uint8_t, 77, "", 0, "High truncation threshold, fraction of 128") AddOptionRTC(extrapolationTracking, int8_t, 1, "", 0, "Enable Extrapolation Tracking(prolong tracks to adjacent sectors to find short segments)") AddOptionRTC(disableRefitAttachment, uint8_t, 0, "", 0, "Bitmask to disable certain attachment steps during refit(1:attachment, 2:propagation, 4:loop following, 8:mirroring)") AddOptionRTC(rejectionStrategy, uint8_t, o2::gpu::GPUSettings::RejectionStrategyA, "", 0, "Enable rejection of TPC clusters for compression(0=no, 1=strategy A, 2=strategy B)") AddOptionRTC(mergeLoopersAfterburner, uint8_t, 1, "", 0, "Run afterburner for additional looper merging") AddOptionRTC(compressionTypeMask, uint8_t, o2::gpu::GPUSettings::CompressionFull, "", 0, "TPC Compression mode bits(1=truncate charge/width LSB, 2=differences, 4=track-model)") AddOptionRTC(compressionSortOrder, uint8_t, o2::gpu::GPUSettings::SortTime, "", 0, "Sort order of TPC compression(0=time, 1=pad, 2=Z-time-pad, 3=Z-pad-time, 4=no sorting(use incoming order))") AddOptionRTC(sigBitsCharge, uint8_t, 4, "", 0, "Number of significant bits for TPC cluster charge in compression mode 1") AddOptionRTC(sigBitsWidth, uint8_t, 3, "", 0, "Number of significant bits for TPC cluster width in compression mode 1") AddOptionRTC(dropLoopers, uint8_t, 0, "", 0, "Drop looping tracks starting from second loop") AddOptionRTC(mergerInterpolateErrors, uint8_t, 1, "", 0, "Use interpolation instead of extrapolation for chi2 based cluster rejection") AddOptionRTC(mergerInterpolateRejectAlsoOnCurrentPosition, uint8_t, 1, "", 0, "When using mergerInterpolateErrors, reject based on chi2 twice computed with interpolated and current track position starting from NDF > mergerNonInterpolateRejectMinNDF Merge tracks accross the central electrode Enable PID response Add sub threshold clusters in TPC dEdx computation OR mask of TPC flags that will reject the cluster in alternative dEdx Reject edge clusters based on uncorrected track Y during track fit Increase decrease removal by when on this number of pads from the edge h configStandalone Reconstruction rec_trd Max chi2 for TRD tracklets to be matched to a track Chi2 cut for strict matching mode Number of sigmas for ITS TPC track time error estimate Addition to search road around track prolongation along Y in cm Optimal tracklet rphi resolution in maximum chi2 per attached tracklet for TRD tracks take into account the full likelihood of the slope including exponential tails Tracks with less attached tracklets are discarded after the tracking Post trigger Pile up integration time in BCs fits all iterations but the last one in projections Use material lookup table for TPC refit Apply workarounds to avoid cyclic merge should not be needed tpc dyn configStandalone Processing proc_rtc Replace constant variables static by constexpr expressions Compile RTC in deterministic with NO_FAST_MATH flags and GPUCA_DETERMINISTIC_MODE define Use RTC to optimize GPU code h configStandalone Processing proc_rtctech Use a file lock to serialize access to the cache folder int32_t
SurfaceTrackState state
void ReleaseEvent(deviceEvent ev) override
void StreamWaitForEvents(int32_t stream, deviceEvent *evList, int32_t nEvents=1) override
void RecordMarker(deviceEvent *ev, int32_t stream) override
void SynchronizeEvents(deviceEvent *evList, int32_t nEvents=1) override
size_t GPUMemCpy(void *dst, const void *src, size_t size, int32_t stream, int32_t toGPU, deviceEvent *ev=nullptr, deviceEvent *evList=nullptr, int32_t nEvents=1) override
size_t WriteToConstantMemory(size_t offset, const void *src, size_t size, int32_t stream=-1, deviceEvent *ev=nullptr) override
int32_t GPUDebug(const char *state="UNKNOWN", int32_t stream=-1, bool force=false) override
void SynchronizeStream(int32_t stream) override
void runKernelBackend(const krnlSetupTime &_xyz, const Args &... args)
GPUReconstructionOCLInternals * mInternals
bool IsEventDone(deviceEvent *evList, int32_t nEvents=1) override
virtual int32_t GPUChkErrInternal(const int64_t retval, const char *file, int32_t line) const override
GLsizeiptr size
Definition glcorearb.h:659
GLenum GLenum dst
Definition glcorearb.h:1767
GLintptr offset
Definition glcorearb.h:660
GLuint GLuint stream
Definition glcorearb.h:1806
const int nEvents
Definition test_Fifo.cxx:27