Project
Loading...
Searching...
No Matches
GPUChainTrackingDebugAndProfiling.cxx
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#include "GPUChainTracking.h"
18#include "GPUConstantMem.h"
19#include "GPUTPCClusterFilter.h"
20#include <map>
21#include <memory>
22#include <string>
23#include <numeric>
24
25#ifdef GPUCA_TRACKLET_CONSTRUCTOR_DO_PROFILE
26#include "bitmapfile.h"
27#endif
28
29#define PROFILE_MAX_SIZE (100 * 1024 * 1024)
30
31using namespace o2::gpu;
32
33static inline uint32_t RGB(uint8_t r, uint8_t g, uint8_t b) { return (uint32_t)r | ((uint32_t)g << 8) | ((uint32_t)b << 16); }
34
36{
37#ifdef GPUCA_TRACKLET_CONSTRUCTOR_DO_PROFILE
39 processorsShadow()->tpcTrackers[0].mStageAtSync = tmpMem;
40 runKernel<GPUMemClean16>({{BlockCount(), ThreadCount(), -1}}, tmpMem, PROFILE_MAX_SIZE);
41#endif
42 return 0;
43}
44
46{
47#ifdef GPUCA_TRACKLET_CONSTRUCTOR_DO_PROFILE
48 std::unique_ptr<char[]> stageAtSync{new char[PROFILE_MAX_SIZE]};
49 mRec->GPUMemCpy(stageAtSync.get(), processorsShadow()->tpcTrackers[0].mStageAtSync, PROFILE_MAX_SIZE, -1, false);
50
51 FILE* fp = fopen("profile.txt", "w+");
52 FILE* fp2 = fopen("profile.bmp", "w+b");
53
54 const int32_t bmpheight = 8192;
55 BITMAPFILEHEADER bmpFH;
56 BITMAPINFOHEADER bmpIH;
57 memset(&bmpFH, 0, sizeof(bmpFH));
58 memset(&bmpIH, 0, sizeof(bmpIH));
59
60 bmpFH.bfType = 19778; //"BM"
61 bmpFH.bfSize = sizeof(bmpFH) + sizeof(bmpIH) + (ConstructorBlockCount() * ConstructorThreadCount() / 32 * 33 - 1) * bmpheight;
62 bmpFH.bfOffBits = sizeof(bmpFH) + sizeof(bmpIH);
63
64 bmpIH.biSize = sizeof(bmpIH);
65 bmpIH.biWidth = ConstructorBlockCount() * ConstructorThreadCount() / 32 * 33 - 1;
66 bmpIH.biHeight = bmpheight;
67 bmpIH.biPlanes = 1;
68 bmpIH.biBitCount = 32;
69
70 fwrite(&bmpFH, 1, sizeof(bmpFH), fp2);
71 fwrite(&bmpIH, 1, sizeof(bmpIH), fp2);
72
73 [[maybe_unused]] int32_t nEmptySync = 0;
74 for (uint32_t i = 0; i < bmpheight * ConstructorBlockCount() * ConstructorThreadCount(); i += ConstructorBlockCount() * ConstructorThreadCount()) {
75 int32_t fEmpty = 1;
76 for (uint32_t j = 0; j < ConstructorBlockCount() * ConstructorThreadCount(); j++) {
77 fprintf(fp, "%d\t", stageAtSync[i + j]);
78 int32_t color = 0;
79 if (stageAtSync[i + j] == 1) {
80 color = RGB(255, 0, 0);
81 }
82 if (stageAtSync[i + j] == 2) {
83 color = RGB(0, 255, 0);
84 }
85 if (stageAtSync[i + j] == 3) {
86 color = RGB(0, 0, 255);
87 }
88 if (stageAtSync[i + j] == 4) {
89 color = RGB(255, 255, 0);
90 }
91 fwrite(&color, 1, sizeof(int32_t), fp2);
92 if (j > 0 && j % 32 == 0) {
93 color = RGB(255, 255, 255);
94 fwrite(&color, 1, 4, fp2);
95 }
96 if (stageAtSync[i + j]) {
97 fEmpty = 0;
98 }
99 }
100 fprintf(fp, "\n");
101 if (fEmpty) {
102 nEmptySync++;
103 } else {
104 nEmptySync = 0;
105 }
106 // if (nEmptySync == GPUCA_SCHED_ROW_STEP + 2) break;
107 }
108
109 fclose(fp);
110 fclose(fp2);
111#endif
112 return 0;
113}
114
115namespace
116{
117struct GPUChainTrackingMemUsage {
118 void add(size_t n, size_t bound)
119 {
120 nMax = std::max(nMax, n);
121 maxUse = std::max(n / std::max<double>(bound, 1.), maxUse);
122 nSum += n;
123 nBoundSum += bound;
124 count++;
125 }
126 size_t nMax;
127 size_t nSum = 0;
128 size_t nBoundSum = 0;
129 double maxUse = 0.;
130 uint32_t count = 0;
131};
132
133void addToMap(std::string name, std::map<std::string, GPUChainTrackingMemUsage>& map, uint64_t n, uint64_t bound)
134{
135 GPUChainTrackingMemUsage& obj = map.insert({name, {}}).first->second;
136 obj.add(n, bound);
137}
138} // namespace
139
141{
142 std::map<std::string, GPUChainTrackingMemUsage> usageMap;
143 for (int32_t i = 0; i < NSECTORS; i++) {
144#ifndef GPUCA_RUN2
145 if (processors()->tpcClusterer[i].mPmemory) {
146 addToMap("TPC Clusterer Sector Peaks", usageMap, processors()->tpcClusterer[i].mPmemory->counters.nPeaks, processors()->tpcClusterer[i].mNMaxPeaks);
147 addToMap("TPC Clusterer Sector Clusters", usageMap, processors()->tpcClusterer[i].mPmemory->counters.nClusters, processors()->tpcClusterer[i].mNMaxClusters);
148 }
149#endif
150 addToMap("TPC Sector Start Hits", usageMap, *processors()->tpcTrackers[i].NStartHits(), processors()->tpcTrackers[i].NMaxStartHits());
151 addToMap("TPC Sector Tracklets", usageMap, *processors()->tpcTrackers[i].NTracklets(), processors()->tpcTrackers[i].NMaxTracklets());
152 addToMap("TPC Sector TrackletHits", usageMap, *processors()->tpcTrackers[i].NRowHits(), processors()->tpcTrackers[i].NMaxRowHits());
153 addToMap("TPC Sector Tracks", usageMap, *processors()->tpcTrackers[i].NTracks(), processors()->tpcTrackers[i].NMaxTracks());
154 addToMap("TPC Sector TrackHits", usageMap, *processors()->tpcTrackers[i].NTrackHits(), processors()->tpcTrackers[i].NMaxTrackHits());
155 }
156 addToMap("TPC Clusterer Clusters", usageMap, mRec->MemoryScalers()->nTPCHits, mRec->MemoryScalers()->NTPCClusters(mRec->MemoryScalers()->nTPCdigits));
157 if (processors()->tpcMerger.Memory()) {
158 addToMap("TPC Tracks", usageMap, processors()->tpcMerger.NMergedTracks(), processors()->tpcMerger.NMaxTracks());
159 addToMap("TPC TrackHits", usageMap, processors()->tpcMerger.NMergedTrackClusters(), processors()->tpcMerger.NMaxMergedTrackClusters());
160 }
161
162 if (mRec->GetProcessingSettings().createO2Output) {
163 addToMap("TPC O2 Tracks", usageMap, processors()->tpcMerger.NOutputTracksTPCO2(), processors()->tpcMerger.NOutputTracksTPCO2());
164 addToMap("TPC O2 ClusRefs", usageMap, processors()->tpcMerger.NOutputClusRefsTPCO2(), processors()->tpcMerger.NOutputClusRefsTPCO2());
165 }
166
167#ifndef GPUCA_RUN2
168 if (processors()->tpcCompressor.mOutput) {
169 addToMap("TPC ComprCache HitsAttached", usageMap, processors()->tpcCompressor.mOutput->nAttachedClusters, processors()->tpcCompressor.mMaxTrackClusters);
170 addToMap("TPC ComprCache HitsUnattached", usageMap, processors()->tpcCompressor.mOutput->nUnattachedClusters, processors()->tpcCompressor.mMaxClustersInCache);
171 addToMap("TPC ComprCache Tracks", usageMap, processors()->tpcCompressor.mOutput->nTracks, processors()->tpcCompressor.mMaxTracks);
172 }
173#endif
174
175 for (auto& elem : usageMap) {
176 printf("Mem Usage %-30s : %'14zu / %'14zu (%3.0f%% / %3.0f%% / count %3u / max %'14zu)\n", elem.first.c_str(), elem.second.nSum, elem.second.nBoundSum, 100. * elem.second.nSum / std::max<size_t>(1, elem.second.nBoundSum), 100. * elem.second.maxUse, elem.second.count, elem.second.nMax);
177 }
178}
179
181{
182 for (int32_t i = 0; i < NSECTORS; i++) {
183 GPUInfo("MEMREL StartHits NCl %d NTrkl %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NStartHits());
184 GPUInfo("MEMREL Tracklets NCl %d NTrkl %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTracklets());
185 GPUInfo("MEMREL Tracklets NCl %d NTrkl %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NRowHits());
186 GPUInfo("MEMREL SectorTracks NCl %d NTrk %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTracks());
187 GPUInfo("MEMREL SectorTrackHits NCl %d NTrkH %d", processors()->tpcTrackers[i].NHitsTotal(), *processors()->tpcTrackers[i].NTrackHits());
188 }
189 if (processors()->tpcMerger.Memory()) {
190 GPUInfo("MEMREL Tracks NCl %d NTrk %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NMergedTracks());
191 GPUInfo("MEMREL TrackHitss NCl %d NTrkH %d", processors()->tpcMerger.NMaxClusters(), processors()->tpcMerger.NMergedTrackClusters());
192 }
193}
194
196{
197#ifdef GPUCA_KERNEL_DEBUGGER_OUTPUT
198 const auto& threadContext = GetThreadContext();
199 if (mRec->IsGPU()) {
200 SetupGPUProcessor(&processors()->debugOutput, false);
201 WriteToConstantMemory(RecoStep::NoRecoStep, (char*)&processors()->debugOutput - (char*)processors(), &processorsShadow()->debugOutput, sizeof(processors()->debugOutput), -1);
202 memset(processors()->debugOutput.memory(), 0, processors()->debugOutput.memorySize() * sizeof(processors()->debugOutput.memory()[0]));
203 }
204 runKernel<GPUMemClean16>({{BlockCount(), ThreadCount(), 0, RecoStep::TPCSectorTracking}}, (mRec->IsGPU() ? processorsShadow() : processors())->debugOutput.memory(), processorsShadow()->debugOutput.memorySize() * sizeof(processors()->debugOutput.memory()[0]));
205#endif
206}
207
209{
210#ifdef GPUCA_KERNEL_DEBUGGER_OUTPUT
211 const auto& threadContext = GetThreadContext();
212 TransferMemoryResourcesToHost(RecoStep::NoRecoStep, &processors()->debugOutput, -1);
213 processors()->debugOutput.Print();
214#endif
215}
216
218{
219 int32_t nTracks = 0, nAttachedClusters = 0, nAttachedClustersFitted = 0, nAdjacentClusters = 0;
220 uint32_t nCls = GetProcessingSettings().doublePipeline ? mIOPtrs.clustersNative->nClustersTotal : processors()->tpcMerger.NMaxClusters();
221 if (GetProcessingSettings().createO2Output > 1) {
222 nTracks = mIOPtrs.nOutputTracksTPCO2;
223 nAttachedClusters = mIOPtrs.nMergedTrackHits;
224 } else {
225 for (uint32_t k = 0; k < mIOPtrs.nMergedTracks; k++) {
226 if (mIOPtrs.mergedTracks[k].OK()) {
227 if (!mIOPtrs.mergedTracks[k].MergedLooper()) {
228 nTracks++;
229 }
230 nAttachedClusters += mIOPtrs.mergedTracks[k].NClusters();
231 nAttachedClustersFitted += mIOPtrs.mergedTracks[k].NClustersFitted();
232 }
233 }
234 for (uint32_t k = 0; k < nCls; k++) {
237 nAdjacentClusters++;
238 }
239 }
240 }
241
242 char trdText[1024] = "";
244 int32_t nTRDTracks = 0;
245 int32_t nTRDTracklets = 0;
246 for (uint32_t k = 0; k < mIOPtrs.nTRDTracks; k++) {
247 if (mIOPtrs.trdTracksO2) {
248 auto& trk = mIOPtrs.trdTracksO2[k];
249 nTRDTracklets += trk.getNtracklets();
250 nTRDTracks += trk.getNtracklets() != 0;
251 } else {
252 auto& trk = mIOPtrs.trdTracks[k];
253 nTRDTracklets += trk.getNtracklets();
254 nTRDTracks += trk.getNtracklets() != 0;
255 }
256 }
257 snprintf(trdText, 1024, " - TRD Tracker reconstructed %d tracks (%d tracklets)", nTRDTracks, nTRDTracklets);
258 }
259 GPUInfo("Output Tracks: %d (%d / %d / %d / %d clusters (fitted / attached / adjacent / total) - %s format)%s", nTracks, nAttachedClustersFitted, nAttachedClusters, nAdjacentClusters, nCls, GetProcessingSettings().createO2Output > 1 ? "O2" : "GPU", trdText);
260}
261
262void GPUChainTracking::OutputSanityCheck()
263{
264 size_t nErrors = 0;
265
266 for (uint32_t i = 0; i < mIOPtrs.nOutputTracksTPCO2; i++) {
267 const auto& trk = mIOPtrs.outputTracksTPCO2[i];
268 const auto& ref = trk.getClusterRef();
269 if (ref.getFirstEntry() > mIOPtrs.nOutputClusRefsTPCO2) {
270 if (nErrors++ < 1000) {
271 GPUError("Invalid getFirst() entry in cluster reference: %u > %u", ref.getFirstEntry(), mIOPtrs.nOutputClusRefsTPCO2);
272 continue;
273 }
274 }
275 if (ref.getFirstEntry() + (ref.getEntries() * 3 + 1) / 2 > mIOPtrs.nOutputClusRefsTPCO2) {
276 if (nErrors++ < 1000) {
277 GPUError("Invalid getEntries() entry in cluster reference: %u > %u", ref.getFirstEntry() + (ref.getEntries() * 3 + 1) / 2, mIOPtrs.nOutputClusRefsTPCO2);
278 continue;
279 }
280 }
281 for (int32_t j = 0; j < trk.getNClusters(); j++) {
283 uint32_t cl;
284 trk.getClusterReference(mIOPtrs.outputClusRefsTPCO2, j, sector, row, cl);
285 if (sector >= GPUTPCGeometry::NSECTORS || row >= GPUTPCGeometry::NROWS) {
286 if (nErrors++ < 1000) {
287 GPUError("Invalid sector / row %d / %d", (int32_t)sector, (int32_t)row);
288 continue;
289 }
290 }
292 if (nErrors++ < 1000) {
293 GPUError("Invalid cluster index %d >= %d", cl, mIOPtrs.clustersNative->nClusters[sector][row]);
294 }
295 }
296 }
297 }
298
299 if (nErrors == 0) {
300 GPUInfo("Sanity check passed");
301 } else {
302 GPUError("Sanity check found %lu errors", nErrors);
303 }
304}
305
306void GPUChainTracking::RunTPCClusterFilter(o2::tpc::ClusterNativeAccess* clusters, std::function<o2::tpc::ClusterNative*(size_t)> allocator, bool applyClusterCuts)
307{
308 const uint8_t filterType = GetProcessingSettings().tpcApplyClusterFilterOnCPU;
309 GPUTPCClusterFilter clusterFilter(*clusters, filterType);
310 o2::tpc::ClusterNative* outputBuffer = nullptr;
311 for (int32_t iPhase = 0; iPhase < 2; iPhase++) {
312 uint32_t countTotal = 0;
313 for (uint32_t iSector = 0; iSector < GPUTPCGeometry::NSECTORS; iSector++) {
314 for (uint32_t iRow = 0; iRow < GPUTPCGeometry::NROWS; iRow++) {
315 uint32_t count = 0;
316 for (uint32_t k = 0; k < clusters->nClusters[iSector][iRow]; k++) {
317 o2::tpc::ClusterNative cl = clusters->clusters[iSector][iRow][k];
318 bool keep = true;
319 if (applyClusterCuts) {
320 keep = keep && cl.getQtot() > param().rec.tpc.cfQTotCutoff && cl.qMax > param().rec.tpc.cfQMaxCutoff;
321 keep = keep && (!(cl.getFlags() & o2::tpc::ClusterNative::flagSingle) || ((cl.sigmaPadPacked || cl.qMax > param().rec.tpc.cfQMaxCutoffSinglePad) && (cl.sigmaTimePacked || cl.qMax > param().rec.tpc.cfQMaxCutoffSingleTime)));
322 }
323 if (param().tpcCutTimeBin > 0) {
324 keep = keep && cl.getTime() < param().tpcCutTimeBin;
325 }
326 keep = keep && (!filterType || clusterFilter.filter(iSector, iRow, cl));
327 if (iPhase && keep) {
328 outputBuffer[countTotal] = cl;
329 }
330 count += keep;
331 countTotal += keep;
332 }
333 if (iPhase) {
334 clusters->nClusters[iSector][iRow] = count;
335 }
336 }
337 }
338 if (iPhase) {
339 clusters->clustersLinear = outputBuffer;
340 clusters->setOffsetPtrs();
341 } else {
342 outputBuffer = allocator(countTotal);
343 }
344 }
345}
346
348{
349 out << "\nTPC Clusters:\n";
350 for (uint32_t iSec = 0; iSec < GPUTPCGeometry::NSECTORS; iSec++) {
351 out << "TPCClusters - Sector " << iSec << "\n";
352 for (uint32_t i = 0; i < GPUTPCGeometry::NROWS; i++) {
353 out << " Row: " << i << ": " << clusters->nClusters[iSec][i] << " clusters:\n";
354 for (uint32_t j = 0; j < clusters->nClusters[iSec][i]; j++) {
355 const auto& cl = clusters->clusters[iSec][i][j];
356 out << " " << std::hex << cl.timeFlagsPacked << std::dec << " " << cl.padPacked << " " << int32_t{cl.sigmaTimePacked} << " " << int32_t{cl.sigmaPadPacked} << " " << cl.qMax << " " << cl.qTotPacked << "\n";
357 }
358 }
359 }
360}
361
363{
365 std::vector<uint32_t> sorted(c.nTracks), offsets(c.nTracks);
366 std::iota(sorted.begin(), sorted.end(), 0);
367 auto sorter = [&c](const auto a, const auto b) {
368 return std::tie(c.sliceA[a], c.rowA[a], c.timeA[a], c.padA[a], c.qPtA[a]) <
369 std::tie(c.sliceA[b], c.rowA[b], c.timeA[b], c.padA[b], c.qPtA[b]);
370 };
371 std::sort(sorted.begin(), sorted.end(), sorter);
372 uint32_t offset = 0;
373 for (uint32_t i = 0; i < c.nTracks; i++) {
374 offsets[i] = offset;
375 offset += c.nTrackClusters[i];
376 }
377
378 auto sortArray = [&c, &sorted, &offsets](auto* src, size_t totalSize, auto getOffset, auto getSize) {
379 auto buf = std::make_unique<std::remove_reference_t<decltype(src[0])>[]>(totalSize);
380 memcpy(buf.get(), src, totalSize * sizeof(*src));
381 uint32_t targetOffset = 0;
382 for (uint32_t i = 0; i < c.nTracks; i++) {
383 const uint32_t j = sorted[i];
384 memcpy(src + targetOffset, buf.get() + getOffset(offsets[j], j), getSize(j) * sizeof(*src));
385 targetOffset += getSize(j);
386 }
387 };
388 auto sortMultiple = [&sortArray](size_t totalSize, auto getOffset, auto getSize, auto&&... arrays) {
389 (..., sortArray(std::forward<decltype(arrays)>(arrays), totalSize, getOffset, getSize));
390 };
391 auto getFullOffset = [](uint32_t off, uint32_t ind) { return off; };
392 auto getReducedOffset = [](uint32_t off, uint32_t ind) { return off - ind; };
393 auto getIndex = [](uint32_t off, uint32_t ind) { return ind; };
394 auto getN = [&c](uint32_t j) { return c.nTrackClusters[j]; };
395 auto getN1 = [&c](uint32_t j) { return c.nTrackClusters[j] - 1; };
396 auto get1 = [](uint32_t j) { return 1; };
397
398 sortMultiple(c.nAttachedClusters, getFullOffset, getN, c.qTotA, c.qMaxA, c.flagsA, c.sigmaPadA, c.sigmaTimeA);
399 sortMultiple(c.nAttachedClustersReduced, getReducedOffset, getN1, c.rowDiffA, c.sliceLegDiffA, c.padResA, c.timeResA);
400 sortMultiple(c.nTracks, getIndex, get1, c.qPtA, c.rowA, c.sliceA, c.timeA, c.padA, c.nTrackClusters); // NOTE: This must be last, since nTrackClusters is used for handling the arrays above!
401}
402
404{
405 std::string dirName = mRec->getDebugFolder("tpc_raw");
406 if (dirName == "") {
407 return;
408 }
410 if (mIOPtrs.tpcZS) {
411 ioPtrs.tpcZS = mIOPtrs.tpcZS;
412 } else if (mIOPtrs.tpcPackedDigits) {
414 } else if (mIOPtrs.clustersNative) {
416 }
417
418 GPUInfo("Doing debug raw dump");
419 mRec->DumpSettings((dirName + "/").c_str());
420 DumpData((dirName + "/event.0.dump").c_str(), &ioPtrs);
421}
unsigned int uint32_t
int32_t i
uint32_t iSector
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 sector
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
uint32_t j
Definition RawData.h:0
uint32_t c
Definition RawData.h:2
static void DebugSortCompressedClusters(o2::tpc::CompressedClustersFlat *cls)
static void DumpClusters(std::ostream &out, const o2::tpc::ClusterNativeAccess *clusters)
void DumpData(const char *filename, const GPUTrackingInOutPointers *ioPtrs=nullptr)
GPUTrackingInOutPointers & mIOPtrs
GPUReconstruction::RecoStepField GetRecoSteps() const
Definition GPUChain.h:71
void WriteToConstantMemory(RecoStep step, size_t offset, const void *src, size_t size, int32_t stream=-1, deviceEvent *ev=nullptr)
Definition GPUChain.h:128
uint32_t ThreadCount() const
Definition GPUChain.h:226
virtual std::unique_ptr< GPUReconstructionProcessing::threadContext > GetThreadContext()
Definition GPUChain.h:109
GPUConstantMem * processors()
Definition GPUChain.h:84
GPUParam & param()
Definition GPUChain.h:87
void SetupGPUProcessor(T *proc, bool allocate)
Definition GPUChain.h:231
const GPUSettingsProcessing & GetProcessingSettings() const
Definition GPUChain.h:76
GPUReconstructionCPU * mRec
Definition GPUChain.h:79
GPUConstantMem * processorsShadow()
Definition GPUChain.h:85
static constexpr int32_t NSECTORS
Definition GPUChain.h:58
void TransferMemoryResourcesToHost(RecoStep step, GPUProcessor *proc, int32_t stream=-1, bool all=false)
Definition GPUChain.h:123
uint32_t BlockCount() const
Definition GPUChain.h:224
GPUReconstruction * rec()
Definition GPUChain.h:66
virtual 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)
std::string getDebugFolder(const std::string &prefix="")
GPUMemorySizeScalers * MemoryScalers()
const GPUSettingsProcessing & GetProcessingSettings() const
void DumpSettings(const char *dir="")
void * AllocateDirectMemory(size_t size, int32_t type)
GLboolean GLboolean g
Definition glcorearb.h:1233
GLdouble n
Definition glcorearb.h:1982
GLenum src
Definition glcorearb.h:1767
GLint GLsizei count
Definition glcorearb.h:399
GLuint GLsizei const GLuint const GLintptr * offsets
Definition glcorearb.h:2595
GLuint const GLchar * name
Definition glcorearb.h:781
GLint first
Definition glcorearb.h:399
GLboolean GLboolean GLboolean b
Definition glcorearb.h:1233
GLintptr offset
Definition glcorearb.h:660
const GLuint * arrays
Definition glcorearb.h:1314
GLenum GLfloat param
Definition glcorearb.h:271
GLboolean GLboolean GLboolean GLboolean a
Definition glcorearb.h:1233
GLint ref
Definition glcorearb.h:291
GLenum GLuint GLenum GLsizei const GLchar * buf
Definition glcorearb.h:2514
GPUdoubleBinary64 b
uint32_t bfSize
Definition bitmapfile.h:17
uint32_t bfOffBits
Definition bitmapfile.h:19
uint16_t bfType
Definition bitmapfile.h:16
uint32_t biWidth
Definition bitmapfile.h:24
uint32_t biSize
Definition bitmapfile.h:23
uint16_t biBitCount
Definition bitmapfile.h:27
uint16_t biPlanes
Definition bitmapfile.h:26
uint32_t biHeight
Definition bitmapfile.h:25
GPUTPCTracker tpcTrackers[GPUTPCGeometry::NSECTORS]
size_t NTPCClusters(size_t tpcDigits, bool perSector=false)
const o2::tpc::ClusterNativeAccess * clustersNative
const GPUTRDTrackGPU * trdTracks
const GPUTrackingInOutZS * tpcZS
const o2::tpc::TrackTPC * outputTracksTPCO2
const GPUTrackingInOutDigits * tpcPackedDigits
const GPUTPCGMMergedTrack * mergedTracks
unsigned int nClusters[constants::MAXSECTOR][constants::MAXGLOBALPADROW]
std::vector< Cluster > clusters
std::vector< int > row
auto getOffset(const map_T &resultsMap) -> typename map_T::key_type