TensorRT 10.11.0
NvInferRuntime.h
Go to the documentation of this file.
1/*
2 * SPDX-FileCopyrightText: Copyright (c) 1993-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3 * SPDX-License-Identifier: Apache-2.0
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18#ifndef NV_INFER_RUNTIME_H
19#define NV_INFER_RUNTIME_H
20
26
27#include "NvInferImpl.h"
28#define NV_INFER_INTERNAL_INCLUDE 1
29#include "NvInferPluginBase.h"
30#undef NV_INFER_INTERNAL_INCLUDE
32
33namespace nvinfer1
34{
35
36class IExecutionContext;
37class ICudaEngine;
38class IPluginFactory;
39class IEngineInspector;
40
49
51{
52protected:
53 INoCopy() = default;
54 virtual ~INoCopy() = default;
55 INoCopy(INoCopy const& other) = delete;
56 INoCopy& operator=(INoCopy const& other) = delete;
57 INoCopy(INoCopy&& other) = delete;
58 INoCopy& operator=(INoCopy&& other) = delete;
59};
60
75enum class EngineCapability : int32_t
76{
81 kSTANDARD = 0,
82
89 kSAFETY = 1,
90
97};
98
99namespace impl
100{
102template <>
104{
105 static constexpr int32_t kVALUE = 3;
106};
107} // namespace impl
108
124{
125public:
127 void const* values;
128 int64_t count;
129};
130
141class IHostMemory : public INoCopy
142{
143public:
144 virtual ~IHostMemory() noexcept = default;
145
147 void* data() const noexcept
148 {
149 return mImpl->data();
150 }
151
153 std::size_t size() const noexcept
154 {
155 return mImpl->size();
156 }
157
159 DataType type() const noexcept
160 {
161 return mImpl->type();
162 }
163
164protected:
165 apiv::VHostMemory* mImpl;
166};
167
178enum class DimensionOperation : int32_t
179{
180 kSUM = 0,
181 kPROD = 1,
182 kMAX = 2,
183 kMIN = 3,
184 kSUB = 4,
185 kEQUAL = 5,
186 kLESS = 6,
187 kFLOOR_DIV = 7,
188 kCEIL_DIV = 8
189};
190
192template <>
193constexpr inline int32_t EnumMax<DimensionOperation>() noexcept
194{
195 return 9;
196}
197
203enum class TensorLocation : int32_t
204{
205 kDEVICE = 0,
206 kHOST = 1,
207};
208
209namespace impl
210{
212template <>
214{
215 static constexpr int32_t kVALUE = 2;
216};
217} // namespace impl
218
232{
233public:
237 bool isConstant() const noexcept
238 {
239 return mImpl->isConstant();
240 }
241
248 int64_t getConstantValue() const noexcept
249 {
250 return mImpl->getConstantValue();
251 }
252
253protected:
254 apiv::VDimensionExpr* mImpl;
255 virtual ~IDimensionExpr() noexcept = default;
256
257public:
263 bool isSizeTensor() const noexcept
264 {
265 return mImpl->isSizeTensor();
266 }
267};
268
286class IExprBuilder : public INoCopy
287{
288public:
292 IDimensionExpr const* constant(int64_t value) noexcept
293 {
294 return mImpl->constant(value);
295 }
296
304 DimensionOperation op, IDimensionExpr const& first, IDimensionExpr const& second) noexcept
305 {
306 return mImpl->operation(op, first, second);
307 }
308
309protected:
310 apiv::VExprBuilder* mImpl;
311 virtual ~IExprBuilder() noexcept = default;
312
313public:
338 IDimensionExpr const* declareSizeTensor(int32_t outputIndex, IDimensionExpr const& opt, IDimensionExpr const& upper)
339 {
340 return mImpl->declareSizeTensor(outputIndex, opt, upper);
341 }
342};
343
350{
351public:
352 int32_t nbDims;
354};
355
362{
365
368
371
374};
375
407{
408public:
409 IPluginV2DynamicExt* clone() const noexcept override = 0;
410
435 virtual DimsExprs getOutputDimensions(
436 int32_t outputIndex, DimsExprs const* inputs, int32_t nbInputs, IExprBuilder& exprBuilder) noexcept = 0;
437
441 static constexpr int32_t kFORMAT_COMBINATION_LIMIT = 100;
442
475 virtual bool supportsFormatCombination(
476 int32_t pos, PluginTensorDesc const* inOut, int32_t nbInputs, int32_t nbOutputs) noexcept = 0;
477
515 virtual void configurePlugin(DynamicPluginTensorDesc const* in, int32_t nbInputs,
516 DynamicPluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
517
527 virtual size_t getWorkspaceSize(PluginTensorDesc const* inputs, int32_t nbInputs, PluginTensorDesc const* outputs,
528 int32_t nbOutputs) const noexcept = 0;
529
542 virtual int32_t enqueue(PluginTensorDesc const* inputDesc, PluginTensorDesc const* outputDesc,
543 void const* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept = 0;
544
545protected:
553 int32_t getTensorRTVersion() const noexcept override
554 {
555 return (static_cast<int32_t>(PluginVersion::kV2_DYNAMICEXT) << 24 | (NV_TENSORRT_VERSION & 0xFFFFFF));
556 }
557
558 virtual ~IPluginV2DynamicExt() noexcept {}
559
560private:
561 // Following are obsolete base class methods, and must not be implemented or used.
562
566 void configurePlugin(Dims const*, int32_t, Dims const*, int32_t, DataType const*, DataType const*, bool const*,
567 bool const*, PluginFormat, int32_t) noexcept override final
568 {
569 }
570
574 bool supportsFormat(DataType, PluginFormat) const noexcept override final
575 {
576 return false;
577 }
578
582 Dims getOutputDimensions(int32_t, Dims const*, int32_t) noexcept override final
583 {
584 return Dims{-1, {}};
585 }
586
594 TRT_DEPRECATED bool isOutputBroadcastAcrossBatch(int32_t, bool const*, int32_t) const noexcept override final
595 {
596 return false;
597 }
598
606 TRT_DEPRECATED bool canBroadcastInputAcrossBatch(int32_t) const noexcept override final
607 {
608 return true;
609 }
610
614 size_t getWorkspaceSize(int32_t) const noexcept override final
615 {
616 return 0;
617 }
618
622 int32_t enqueue(int32_t, void const* const*, void* const*, void*, cudaStream_t) noexcept override final
623 {
624 return 1;
625 }
626};
627
628namespace v_1_0
629{
631{
632public:
637 ~IStreamReader() override = default;
638 IStreamReader() = default;
639
643 InterfaceInfo getInterfaceInfo() const noexcept override
644 {
645 return InterfaceInfo{"IStreamReader", 1, 0};
646 }
647
656 virtual int64_t read(void* destination, int64_t nbBytes) = 0;
657
658protected:
659 IStreamReader(IStreamReader const&) = default;
663};
664} // namespace v_1_0
665
675
680enum class SeekPosition : int32_t
681{
683 kSET = 0,
684
686 kCUR = 1,
687
689 kEND = 2,
690};
691
692namespace v_1_0
693{
695{
696public:
701 ~IStreamReaderV2() override = default;
702 IStreamReaderV2() = default;
703
707 InterfaceInfo getInterfaceInfo() const noexcept override
708 {
709 return InterfaceInfo{"IStreamReaderV2", 1, 0};
710 }
711
722 virtual int64_t read(void* destination, int64_t nbBytes, cudaStream_t stream) noexcept = 0;
723
732 virtual bool seek(int64_t offset, SeekPosition where) noexcept = 0;
733
734protected:
739};
740} // namespace v_1_0
741
752
767{
768public:
773 virtual IGpuAllocator* getGpuAllocator() const noexcept = 0;
774
779 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
780 virtual ~IPluginResourceContext() noexcept = default;
781
782protected:
786 IPluginResourceContext& operator=(IPluginResourceContext const&) & = default;
788};
789
790namespace v_1_0
791{
793{
794public:
798 InterfaceInfo getInterfaceInfo() const noexcept override
799 {
800 return InterfaceInfo{"PLUGIN_V3ONE_CORE", 1, 0};
801 }
802
811 virtual AsciiChar const* getPluginName() const noexcept = 0;
812
821 virtual AsciiChar const* getPluginVersion() const noexcept = 0;
822
832 virtual AsciiChar const* getPluginNamespace() const noexcept = 0;
833};
834
836{
837public:
843 static constexpr int32_t kDEFAULT_FORMAT_COMBINATION_LIMIT = 100;
844
848 InterfaceInfo getInterfaceInfo() const noexcept override
849 {
850 return InterfaceInfo{"PLUGIN_V3ONE_BUILD", 1, 0};
851 }
852
872 virtual int32_t configurePlugin(DynamicPluginTensorDesc const* in, int32_t nbInputs,
873 DynamicPluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
874
891 virtual int32_t getOutputDataTypes(
892 DataType* outputTypes, int32_t nbOutputs, const DataType* inputTypes, int32_t nbInputs) const noexcept = 0;
893
915 virtual int32_t getOutputShapes(DimsExprs const* inputs, int32_t nbInputs, DimsExprs const* shapeInputs,
916 int32_t nbShapeInputs, DimsExprs* outputs, int32_t nbOutputs, IExprBuilder& exprBuilder) noexcept = 0;
917
953 int32_t pos, DynamicPluginTensorDesc const* inOut, int32_t nbInputs, int32_t nbOutputs) noexcept = 0;
954
960 virtual int32_t getNbOutputs() const noexcept = 0;
961
971 virtual size_t getWorkspaceSize(DynamicPluginTensorDesc const* inputs, int32_t nbInputs,
972 DynamicPluginTensorDesc const* outputs, int32_t nbOutputs) const noexcept
973 {
974 return 0;
975 }
976
1008 virtual int32_t getValidTactics(int32_t* tactics, int32_t nbTactics) noexcept
1009 {
1010 return 0;
1011 }
1012
1016 virtual int32_t getNbTactics() noexcept
1017 {
1018 return 0;
1019 }
1020
1032 virtual char const* getTimingCacheID() noexcept
1033 {
1034 return nullptr;
1035 }
1036
1040 virtual int32_t getFormatCombinationLimit() noexcept
1041 {
1042 return kDEFAULT_FORMAT_COMBINATION_LIMIT;
1043 }
1044
1051 virtual char const* getMetadataString() noexcept
1052 {
1053 return nullptr;
1054 }
1055};
1056
1058{
1059public:
1063 InterfaceInfo getInterfaceInfo() const noexcept override
1064 {
1065 return InterfaceInfo{"PLUGIN_V3ONE_RUNTIME", 1, 0};
1066 }
1067
1075 virtual int32_t setTactic(int32_t tactic) noexcept
1076 {
1077 return 0;
1078 }
1079
1098 virtual int32_t onShapeChange(
1099 PluginTensorDesc const* in, int32_t nbInputs, PluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
1100
1114 virtual int32_t enqueue(PluginTensorDesc const* inputDesc, PluginTensorDesc const* outputDesc,
1115 void const* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept = 0;
1116
1136 virtual IPluginV3* attachToContext(IPluginResourceContext* context) noexcept = 0;
1137
1143
1147 virtual PluginFieldCollection const* getFieldsToSerialize() noexcept = 0;
1148};
1149} // namespace v_1_0
1150
1151namespace v_2_0
1152{
1153
1155{
1156public:
1157 InterfaceInfo getInterfaceInfo() const noexcept override
1158 {
1159 return InterfaceInfo{"PLUGIN_V3ONE_BUILD", 2, 0};
1160 }
1161
1191 virtual int32_t getAliasedInput(int32_t outputIndex) noexcept
1192 {
1193 return -1;
1194 }
1195};
1196
1197} // namespace v_2_0
1198
1209
1221
1233
1242
1243namespace v_1_0
1244{
1246{
1247public:
1255 virtual void reportLayerTime(char const* layerName, float ms) noexcept = 0;
1256
1257 virtual ~IProfiler() noexcept {}
1258};
1259} // namespace v_1_0
1260
1273
1281enum class WeightsRole : int32_t
1282{
1283 kKERNEL = 0,
1284 kBIAS = 1,
1285 kSHIFT = 2,
1286 kSCALE = 3,
1287 kCONSTANT = 4,
1288 kANY = 5,
1289};
1290
1292template <>
1293constexpr inline int32_t EnumMax<WeightsRole>() noexcept
1294{
1295 return 6;
1296}
1297
1303enum class DeviceType : int32_t
1304{
1305 kGPU = 0,
1306 kDLA = 1,
1307};
1308
1310template <>
1311constexpr inline int32_t EnumMax<DeviceType>() noexcept
1312{
1313 return 2;
1314}
1315
1326enum class TempfileControlFlag : int32_t
1327{
1330
1335};
1336
1338template <>
1339constexpr inline int32_t EnumMax<TempfileControlFlag>() noexcept
1340{
1341 return 2;
1342}
1343
1350using TempfileControlFlags = uint32_t;
1351
1382enum class TensorFormat : int32_t
1383{
1390 kLINEAR = 0,
1391
1396 kCHW2 = 1,
1397
1401 kHWC8 = 2,
1402
1416 kCHW4 = 3,
1417
1425 kCHW16 = 4,
1426
1434 kCHW32 = 5,
1435
1440 kDHWC8 = 6,
1441
1446 kCDHW32 = 7,
1447
1451 kHWC = 8,
1452
1461 kDLA_LINEAR = 9,
1462
1476 kDLA_HWC4 = 10,
1477
1482 kHWC16 = 11,
1483
1488 kDHWC = 12
1489};
1490
1491namespace impl
1492{
1494template <>
1496{
1498 static constexpr int32_t kVALUE = 13;
1499};
1500} // namespace impl
1501
1507enum class AllocatorFlag : int32_t
1508{
1510 kRESIZABLE = 0,
1511};
1512
1513namespace impl
1514{
1516template <>
1518{
1520 static constexpr int32_t kVALUE = 1;
1521};
1522} // namespace impl
1523
1524using AllocatorFlags = uint32_t;
1525
1528
1542{
1543public:
1549 enum class Severity : int32_t
1550 {
1552 kINTERNAL_ERROR = 0,
1554 kERROR = 1,
1556 kWARNING = 2,
1558 kINFO = 3,
1560 kVERBOSE = 4,
1561 };
1562
1581 virtual void log(Severity severity, AsciiChar const* msg) noexcept = 0;
1582
1583 ILogger() = default;
1584 virtual ~ILogger() = default;
1585
1586protected:
1587 // @cond SuppressDoxyWarnings
1588 ILogger(ILogger const&) = default;
1589 ILogger(ILogger&&) = default;
1590 ILogger& operator=(ILogger const&) & = default;
1591 ILogger& operator=(ILogger&&) & = default;
1592 // @endcond
1593};
1594
1595namespace impl
1596{
1598template <>
1599struct EnumMaxImpl<ILogger::Severity>
1600{
1602 static constexpr int32_t kVALUE = 5;
1603};
1604} // namespace impl
1605
1606namespace v_1_0
1607{
1608
1610{
1611public:
1637 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept = 0;
1638
1639 ~IGpuAllocator() override = default;
1640 IGpuAllocator() = default;
1641
1679 virtual void* reallocate(void* const /*baseAddr*/, uint64_t /*alignment*/, uint64_t /*newSize*/) noexcept
1680 {
1681 return nullptr;
1682 }
1683
1702 TRT_DEPRECATED virtual bool deallocate(void* const memory) noexcept = 0;
1703
1732 virtual void* allocateAsync(
1733 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags, cudaStream_t /*stream*/) noexcept
1734 {
1735 return allocate(size, alignment, flags);
1736 }
1765 virtual bool deallocateAsync(void* const memory, cudaStream_t /*stream*/) noexcept
1766 {
1767 return deallocate(memory);
1768 }
1769
1773 InterfaceInfo getInterfaceInfo() const noexcept override
1774 {
1775 return {"IGpuAllocator", 1, 0};
1776 }
1777
1778protected:
1779 // @cond SuppressDoxyWarnings
1780 IGpuAllocator(IGpuAllocator const&) = default;
1781 IGpuAllocator(IGpuAllocator&&) = default;
1782 IGpuAllocator& operator=(IGpuAllocator const&) & = default;
1783 IGpuAllocator& operator=(IGpuAllocator&&) & = default;
1784 // @endcond
1785};
1786
1787} // namespace v_1_0
1788
1810
1818class IRuntime : public INoCopy
1819{
1820public:
1821 virtual ~IRuntime() noexcept = default;
1822
1834 void setDLACore(int32_t dlaCore) noexcept
1835 {
1836 mImpl->setDLACore(dlaCore);
1837 }
1838
1844 int32_t getDLACore() const noexcept
1845 {
1846 return mImpl->getDLACore();
1847 }
1848
1852 int32_t getNbDLACores() const noexcept
1853 {
1854 return mImpl->getNbDLACores();
1855 }
1856
1868 void setGpuAllocator(IGpuAllocator* allocator) noexcept
1869 {
1870 mImpl->setGpuAllocator(allocator);
1871 }
1872
1884 //
1887 void setErrorRecorder(IErrorRecorder* recorder) noexcept
1888 {
1889 mImpl->setErrorRecorder(recorder);
1890 }
1891
1903 {
1904 return mImpl->getErrorRecorder();
1905 }
1906
1917 ICudaEngine* deserializeCudaEngine(void const* blob, std::size_t size) noexcept
1918 {
1919 return mImpl->deserializeCudaEngine(blob, size);
1920 }
1921
1939 {
1940 return mImpl->deserializeCudaEngine(streamReader);
1941 }
1942
1959 {
1960 return mImpl->deserializeCudaEngineV2(streamReader);
1961 }
1962
1968 ILogger* getLogger() const noexcept
1969 {
1970 return mImpl->getLogger();
1971 }
1972
1983 bool setMaxThreads(int32_t maxThreads) noexcept
1984 {
1985 return mImpl->setMaxThreads(maxThreads);
1986 }
1987
1997 int32_t getMaxThreads() const noexcept
1998 {
1999 return mImpl->getMaxThreads();
2000 }
2001
2032 void setTemporaryDirectory(char const* path) noexcept
2033 {
2034 return mImpl->setTemporaryDirectory(path);
2035 }
2036
2043 char const* getTemporaryDirectory() const noexcept
2044 {
2045 return mImpl->getTemporaryDirectory();
2046 }
2047
2060 {
2061 return mImpl->setTempfileControlFlags(flags);
2062 }
2063
2072 {
2073 return mImpl->getTempfileControlFlags();
2074 }
2075
2082 {
2083 return mImpl->getPluginRegistry();
2084 }
2085
2099 IRuntime* loadRuntime(char const* path) noexcept
2100 {
2101 return mImpl->loadRuntime(path);
2102 }
2103
2111 void setEngineHostCodeAllowed(bool allowed) noexcept
2112 {
2113 return mImpl->setEngineHostCodeAllowed(allowed);
2114 }
2115
2121 bool getEngineHostCodeAllowed() const noexcept
2122 {
2123 return mImpl->getEngineHostCodeAllowed();
2124 }
2125
2126protected:
2127 apiv::VRuntime* mImpl;
2128};
2129
2137class IRefitter : public INoCopy
2138{
2139public:
2140 virtual ~IRefitter() noexcept = default;
2141
2157 bool setWeights(char const* layerName, WeightsRole role, Weights weights) noexcept
2158 {
2159 return mImpl->setWeights(layerName, role, weights);
2160 }
2161
2174 bool refitCudaEngine() noexcept
2175 {
2176 return mImpl->refitCudaEngine();
2177 }
2178
2195 int32_t getMissing(int32_t size, char const** layerNames, WeightsRole* roles) noexcept
2196 {
2197 return mImpl->getMissing(size, layerNames, roles);
2198 }
2199
2212 int32_t getAll(int32_t size, char const** layerNames, WeightsRole* roles) noexcept
2213 {
2214 return mImpl->getAll(size, layerNames, roles);
2215 }
2216
2234 TRT_DEPRECATED bool setDynamicRange(char const* tensorName, float min, float max) noexcept
2235 {
2236 return mImpl->setDynamicRange(tensorName, min, max);
2237 }
2238
2250 TRT_DEPRECATED float getDynamicRangeMin(char const* tensorName) const noexcept
2251 {
2252 return mImpl->getDynamicRangeMin(tensorName);
2253 }
2254
2266 TRT_DEPRECATED float getDynamicRangeMax(char const* tensorName) const noexcept
2267 {
2268 return mImpl->getDynamicRangeMax(tensorName);
2269 }
2270
2284 TRT_DEPRECATED int32_t getTensorsWithDynamicRange(int32_t size, char const** tensorNames) const noexcept
2285 {
2286 return mImpl->getTensorsWithDynamicRange(size, tensorNames);
2287 }
2288
2300 //
2303 void setErrorRecorder(IErrorRecorder* recorder) noexcept
2304 {
2305 mImpl->setErrorRecorder(recorder);
2306 }
2307
2319 {
2320 return mImpl->getErrorRecorder();
2321 }
2322
2343 bool setNamedWeights(char const* name, Weights weights) noexcept
2344 {
2345 return mImpl->setNamedWeights(name, weights);
2346 }
2347
2363 int32_t getMissingWeights(int32_t size, char const** weightsNames) noexcept
2364 {
2365 return mImpl->getMissingWeights(size, weightsNames);
2366 }
2367
2379 int32_t getAllWeights(int32_t size, char const** weightsNames) noexcept
2380 {
2381 return mImpl->getAllWeights(size, weightsNames);
2382 }
2383
2389 ILogger* getLogger() const noexcept
2390 {
2391 return mImpl->getLogger();
2392 }
2393
2405 bool setMaxThreads(int32_t maxThreads) noexcept
2406 {
2407 return mImpl->setMaxThreads(maxThreads);
2408 }
2409
2419 int32_t getMaxThreads() const noexcept
2420 {
2421 return mImpl->getMaxThreads();
2422 }
2423
2446 bool setNamedWeights(char const* name, Weights weights, TensorLocation location) noexcept
2447 {
2448 return mImpl->setNamedWeightsWithLocation(name, weights, location);
2449 }
2450
2462 Weights getNamedWeights(char const* weightsName) const noexcept
2463 {
2464 return mImpl->getNamedWeights(weightsName);
2465 }
2466
2478 TensorLocation getWeightsLocation(char const* weightsName) const noexcept
2479 {
2480 return mImpl->getWeightsLocation(weightsName);
2481 }
2482
2494 bool unsetNamedWeights(char const* weightsName) noexcept
2495 {
2496 return mImpl->unsetNamedWeights(weightsName);
2497 }
2498
2510 void setWeightsValidation(bool weightsValidation) noexcept
2511 {
2512 return mImpl->setWeightsValidation(weightsValidation);
2513 }
2514
2518 bool getWeightsValidation() const noexcept
2519 {
2520 return mImpl->getWeightsValidation();
2521 }
2522
2540 bool refitCudaEngineAsync(cudaStream_t stream) noexcept
2541 {
2542 return mImpl->refitCudaEngineAsync(stream);
2543 }
2544
2558 Weights getWeightsPrototype(char const* weightsName) const noexcept
2559 {
2560 return mImpl->getWeightsPrototype(weightsName);
2561 }
2562
2563protected:
2564 apiv::VRefitter* mImpl;
2565};
2566
2577enum class OptProfileSelector : int32_t
2578{
2579 kMIN = 0,
2580 kOPT = 1,
2581 kMAX = 2
2582};
2583
2589template <>
2590constexpr inline int32_t EnumMax<OptProfileSelector>() noexcept
2591{
2592 return 3;
2593}
2594
2618{
2619public:
2647 bool setDimensions(char const* inputName, OptProfileSelector select, Dims const& dims) noexcept
2648 {
2649 return mImpl->setDimensions(inputName, select, dims);
2650 }
2651
2659 Dims getDimensions(char const* inputName, OptProfileSelector select) const noexcept
2660 {
2661 return mImpl->getDimensions(inputName, select);
2662 }
2663
2712 char const* inputName, OptProfileSelector select, int32_t const* values, int32_t nbValues) noexcept
2713 {
2714 return mImpl->setShapeValues(inputName, select, values, nbValues);
2715 }
2716
2725 int32_t getNbShapeValues(char const* inputName) const noexcept
2726 {
2727 return mImpl->getNbShapeValues(inputName);
2728 }
2729
2739 TRT_DEPRECATED int32_t const* getShapeValues(char const* inputName, OptProfileSelector select) const noexcept
2740 {
2741 return mImpl->getShapeValues(inputName, select);
2742 }
2743
2757 bool setExtraMemoryTarget(float target) noexcept
2758 {
2759 return mImpl->setExtraMemoryTarget(target);
2760 }
2761
2769 float getExtraMemoryTarget() const noexcept
2770 {
2771 return mImpl->getExtraMemoryTarget();
2772 }
2773
2786 bool isValid() const noexcept
2787 {
2788 return mImpl->isValid();
2789 }
2790
2837 char const* inputName, OptProfileSelector select, int64_t const* values, int32_t nbValues) noexcept
2838 {
2839 return mImpl->setShapeValuesV2(inputName, select, values, nbValues);
2840 }
2841
2849 int64_t const* getShapeValuesV2(char const* inputName, OptProfileSelector select) const noexcept
2850 {
2851 return mImpl->getShapeValuesV2(inputName, select);
2852 }
2853
2854protected:
2855 apiv::VOptimizationProfile* mImpl;
2856 virtual ~IOptimizationProfile() noexcept = default;
2857};
2858
2866enum class TacticSource : int32_t
2867{
2872
2876
2881
2886
2890};
2891
2892template <>
2893constexpr inline int32_t EnumMax<TacticSource>() noexcept
2894{
2895 return 5;
2896}
2897
2904using TacticSources = uint32_t;
2905
2915enum class ProfilingVerbosity : int32_t
2916{
2917 kLAYER_NAMES_ONLY = 0,
2918 kNONE = 1,
2919 kDETAILED = 2,
2920};
2921
2923template <>
2924constexpr inline int32_t EnumMax<ProfilingVerbosity>() noexcept
2925{
2926 return 3;
2927}
2928
2935using SerializationFlags = uint32_t;
2936
2944enum class SerializationFlag : int32_t
2945{
2946 kEXCLUDE_WEIGHTS = 0,
2948};
2949
2951template <>
2952constexpr inline int32_t EnumMax<SerializationFlag>() noexcept
2953{
2954 return 2;
2955}
2956
2965{
2966public:
2967 virtual ~ISerializationConfig() noexcept = default;
2968
2980 bool setFlags(SerializationFlags serializationFlags) noexcept
2981 {
2982 return mImpl->setFlags(serializationFlags);
2983 }
2984
2993 {
2994 return mImpl->getFlags();
2995 }
2996
3004 bool clearFlag(SerializationFlag serializationFlag) noexcept
3005 {
3006 return mImpl->clearFlag(serializationFlag);
3007 }
3008
3016 bool setFlag(SerializationFlag serializationFlag) noexcept
3017 {
3018 return mImpl->setFlag(serializationFlag);
3019 }
3020
3028 bool getFlag(SerializationFlag serializationFlag) const noexcept
3029 {
3030 return mImpl->getFlag(serializationFlag);
3031 }
3032
3033protected:
3034 apiv::VSerializationConfig* mImpl;
3035};
3036
3049{
3050 kSTATIC = 0,
3051 kON_PROFILE_CHANGE = 1,
3052 kUSER_MANAGED = 2,
3053};
3054
3060template <>
3061constexpr inline int32_t EnumMax<ExecutionContextAllocationStrategy>() noexcept
3062{
3063 return 3;
3064}
3065
3066
3074{
3075public:
3076 virtual ~IRuntimeConfig() noexcept = default;
3077
3083 void setExecutionContextAllocationStrategy(ExecutionContextAllocationStrategy strategy) noexcept
3084 {
3085 return mImpl->setExecutionContextAllocationStrategy(strategy);
3086 }
3087
3094 {
3095 return mImpl->getExecutionContextAllocationStrategy();
3096 }
3097
3098
3099protected:
3100 apiv::VRuntimeConfig* mImpl;
3101}; // class IRuntimeConfig
3102
3110class ICudaEngine : public INoCopy
3111{
3112public:
3113 virtual ~ICudaEngine() noexcept = default;
3114
3125 Dims getTensorShape(char const* tensorName) const noexcept
3126 {
3127 return mImpl->getTensorShape(tensorName);
3128 }
3129
3140 DataType getTensorDataType(char const* tensorName) const noexcept
3141 {
3142 return mImpl->getTensorDataType(tensorName);
3143 }
3144
3154 int32_t getNbLayers() const noexcept
3155 {
3156 return mImpl->getNbLayers();
3157 }
3158
3168 IHostMemory* serialize() const noexcept
3169 {
3170 return mImpl->serialize();
3171 }
3172
3187 {
3188 return mImpl->createExecutionContext(strategy);
3189 }
3190
3203 TensorLocation getTensorLocation(char const* tensorName) const noexcept
3204 {
3205 return mImpl->getTensorLocation(tensorName);
3206 }
3207
3223 bool isShapeInferenceIO(char const* tensorName) const noexcept
3224 {
3225 return mImpl->isShapeInferenceIO(tensorName);
3226 }
3227
3237 TensorIOMode getTensorIOMode(char const* tensorName) const noexcept
3238 {
3239 return mImpl->getTensorIOMode(tensorName);
3240 }
3241
3250 {
3251 return mImpl->createExecutionContextWithoutDeviceMemory();
3252 }
3253
3262 {
3263 return mImpl->createExecutionContextWithRuntimeConfig(runtimeConfig);
3264 }
3265
3275 {
3276 return mImpl->createRuntimeConfig();
3277 }
3278
3286 TRT_DEPRECATED size_t getDeviceMemorySize() const noexcept
3287 {
3288 return mImpl->getDeviceMemorySize();
3289 }
3290
3298 TRT_DEPRECATED size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept
3299 {
3300 return mImpl->getDeviceMemorySizeForProfile(profileIndex);
3301 }
3302
3314 int64_t getDeviceMemorySizeV2() const noexcept
3315 {
3316 return mImpl->getDeviceMemorySizeV2();
3317 }
3318
3330 int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept
3331 {
3332 return mImpl->getDeviceMemorySizeForProfileV2(profileIndex);
3333 }
3334
3340 bool isRefittable() const noexcept
3341 {
3342 return mImpl->isRefittable();
3343 }
3344
3361 int32_t getTensorBytesPerComponent(char const* tensorName) const noexcept
3362 {
3363 return mImpl->getTensorBytesPerComponent(tensorName);
3364 }
3365
3379 int32_t getTensorBytesPerComponent(char const* tensorName, int32_t profileIndex) const noexcept
3380 {
3381 return mImpl->getTensorBytesPerComponentV2(tensorName, profileIndex);
3382 }
3383
3400 int32_t getTensorComponentsPerElement(char const* tensorName) const noexcept
3401 {
3402 return mImpl->getTensorComponentsPerElement(tensorName);
3403 }
3404
3418 int32_t getTensorComponentsPerElement(char const* tensorName, int32_t profileIndex) const noexcept
3419 {
3420 return mImpl->getTensorComponentsPerElementV2(tensorName, profileIndex);
3421 }
3422
3433 TensorFormat getTensorFormat(char const* tensorName) const noexcept
3434 {
3435 return mImpl->getTensorFormat(tensorName);
3436 }
3437
3447 TensorFormat getTensorFormat(char const* tensorName, int32_t profileIndex) const noexcept
3448 {
3449 return mImpl->getTensorFormatV2(tensorName, profileIndex);
3450 }
3451
3471 char const* getTensorFormatDesc(char const* tensorName) const noexcept
3472 {
3473 return mImpl->getTensorFormatDesc(tensorName);
3474 }
3475
3494 char const* getTensorFormatDesc(char const* tensorName, int32_t profileIndex) const noexcept
3495 {
3496 return mImpl->getTensorFormatDescV2(tensorName, profileIndex);
3497 }
3498
3511 int32_t getTensorVectorizedDim(char const* tensorName) const noexcept
3512 {
3513 return mImpl->getTensorVectorizedDim(tensorName);
3514 }
3515
3527 int32_t getTensorVectorizedDim(char const* tensorName, int32_t profileIndex) const noexcept
3528 {
3529 return mImpl->getTensorVectorizedDimV2(tensorName, profileIndex);
3530 }
3531
3542 char const* getName() const noexcept
3543 {
3544 return mImpl->getName();
3545 }
3546
3553 int32_t getNbOptimizationProfiles() const noexcept
3554 {
3555 return mImpl->getNbOptimizationProfiles();
3556 }
3557
3573 Dims getProfileShape(char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
3574 {
3575 return mImpl->getProfileShape(tensorName, profileIndex, select);
3576 }
3577
3599 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
3600 {
3601 return mImpl->getProfileTensorValues(tensorName, profileIndex, select);
3602 }
3603
3615 {
3616 return mImpl->getEngineCapability();
3617 }
3618
3633 void setErrorRecorder(IErrorRecorder* recorder) noexcept
3634 {
3635 return mImpl->setErrorRecorder(recorder);
3636 }
3637
3649 {
3650 return mImpl->getErrorRecorder();
3651 }
3652
3663 {
3664 return mImpl->hasImplicitBatchDimension();
3665 }
3666
3679 {
3680 return mImpl->getTacticSources();
3681 }
3682
3691 {
3692 return mImpl->getProfilingVerbosity();
3693 }
3694
3701 {
3702 return mImpl->createEngineInspector();
3703 }
3704
3713 int32_t getNbIOTensors() const noexcept
3714 {
3715 return mImpl->getNbIOTensors();
3716 }
3717
3725 char const* getIOTensorName(int32_t index) const noexcept
3726 {
3727 return mImpl->getIOTensorName(index);
3728 }
3729
3737 {
3738 return mImpl->getHardwareCompatibilityLevel();
3739 }
3740
3751 int32_t getNbAuxStreams() const noexcept
3752 {
3753 return mImpl->getNbAuxStreams();
3754 }
3755
3762 {
3763 return mImpl->createSerializationConfig();
3764 }
3765
3778 {
3779 return mImpl->serializeWithConfig(config);
3780 }
3781
3822 TRT_DEPRECATED bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept
3823 {
3824 return mImpl->setWeightStreamingBudget(gpuMemoryBudget);
3825 }
3826
3843 {
3844 return mImpl->getWeightStreamingBudget();
3845 }
3846
3866 {
3867 return mImpl->getMinimumWeightStreamingBudget();
3868 }
3869
3881 int64_t getStreamableWeightsSize() const noexcept
3882 {
3883 return mImpl->getStreamableWeightsSize();
3884 }
3885
3923 bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept
3924 {
3925 return mImpl->setWeightStreamingBudgetV2(gpuMemoryBudget);
3926 }
3927
3941 int64_t getWeightStreamingBudgetV2() const noexcept
3942 {
3943 return mImpl->getWeightStreamingBudgetV2();
3944 }
3945
3966 int64_t getWeightStreamingAutomaticBudget() const noexcept
3967 {
3968 return mImpl->getWeightStreamingAutomaticBudget();
3969 }
3970
3995 {
3996 return mImpl->getWeightStreamingScratchMemorySize();
3997 }
3998
4008 bool isDebugTensor(char const* name) const noexcept
4009 {
4010 return mImpl->isDebugTensor(name);
4011 }
4012
4033 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
4034 {
4035 return mImpl->getProfileTensorValuesV2(tensorName, profileIndex, select);
4036 }
4037
4038protected:
4039 apiv::VCudaEngine* mImpl;
4040};
4041
4042namespace v_1_0
4043{
4045{
4046public:
4050 InterfaceInfo getInterfaceInfo() const noexcept override
4051 {
4052 return {"IOutputAllocator", 1, 0};
4053 }
4054
4075 char const* tensorName, void* currentMemory, uint64_t size, uint64_t alignment) noexcept
4076 {
4077 return nullptr;
4078 }
4079
4103 char const* tensorName, void* currentMemory, uint64_t size, uint64_t alignment, cudaStream_t /*stream*/)
4104 {
4105 return reallocateOutput(tensorName, currentMemory, size, alignment);
4106 }
4107
4116 virtual void notifyShape(char const* tensorName, Dims const& dims) noexcept = 0;
4117};
4118} // namespace v_1_0
4119
4128
4129namespace v_1_0
4130{
4132{
4133public:
4137 InterfaceInfo getInterfaceInfo() const noexcept override
4138 {
4139 return {"IDebugListener", 1, 0};
4140 }
4141
4155 virtual bool processDebugTensor(void const* addr, TensorLocation location, DataType type, Dims const& shape,
4156 char const* name, cudaStream_t stream)
4157 = 0;
4158
4159 ~IDebugListener() override = default;
4160};
4161} // namespace v_1_0
4162
4169
4181{
4182public:
4183 virtual ~IExecutionContext() noexcept = default;
4184
4193 void setDebugSync(bool sync) noexcept
4194 {
4195 mImpl->setDebugSync(sync);
4196 }
4197
4203 bool getDebugSync() const noexcept
4204 {
4205 return mImpl->getDebugSync();
4206 }
4207
4213 void setProfiler(IProfiler* profiler) noexcept
4214 {
4215 mImpl->setProfiler(profiler);
4216 }
4217
4223 IProfiler* getProfiler() const noexcept
4224 {
4225 return mImpl->getProfiler();
4226 }
4227
4233 ICudaEngine const& getEngine() const noexcept
4234 {
4235 return mImpl->getEngine();
4236 }
4237
4247 void setName(char const* name) noexcept
4248 {
4249 mImpl->setName(name);
4250 }
4251
4257 char const* getName() const noexcept
4258 {
4259 return mImpl->getName();
4260 }
4261
4284 void setDeviceMemory(void* memory) noexcept
4285 {
4286 mImpl->setDeviceMemory(memory);
4287 }
4288
4306 void setDeviceMemoryV2(void* memory, int64_t size) noexcept
4307 {
4308 return mImpl->setDeviceMemoryV2(memory, size);
4309 }
4310
4327 Dims getTensorStrides(char const* tensorName) const noexcept
4328 {
4329 return mImpl->getTensorStrides(tensorName);
4330 }
4331
4332public:
4342 int32_t getOptimizationProfile() const noexcept
4343 {
4344 return mImpl->getOptimizationProfile();
4345 }
4346
4360 bool setInputShape(char const* tensorName, Dims const& dims) noexcept
4361 {
4362 return mImpl->setInputShape(tensorName, dims);
4363 }
4364
4397 Dims getTensorShape(char const* tensorName) const noexcept
4398 {
4399 return mImpl->getTensorShape(tensorName);
4400 }
4401
4413 bool allInputDimensionsSpecified() const noexcept
4414 {
4415 return mImpl->allInputDimensionsSpecified();
4416 }
4417
4431 {
4432 return mImpl->allInputShapesSpecified();
4433 }
4434
4449 void setErrorRecorder(IErrorRecorder* recorder) noexcept
4450 {
4451 mImpl->setErrorRecorder(recorder);
4452 }
4453
4465 {
4466 return mImpl->getErrorRecorder();
4467 }
4468
4481 bool executeV2(void* const* bindings) noexcept
4482 {
4483 return mImpl->executeV2(bindings);
4484 }
4485
4525 bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept
4526 {
4527 return mImpl->setOptimizationProfileAsync(profileIndex, stream);
4528 }
4529
4541 void setEnqueueEmitsProfile(bool enqueueEmitsProfile) noexcept
4542 {
4543 mImpl->setEnqueueEmitsProfile(enqueueEmitsProfile);
4544 }
4545
4553 bool getEnqueueEmitsProfile() const noexcept
4554 {
4555 return mImpl->getEnqueueEmitsProfile();
4556 }
4557
4583 bool reportToProfiler() const noexcept
4584 {
4585 return mImpl->reportToProfiler();
4586 }
4587
4627 bool setTensorAddress(char const* tensorName, void* data) noexcept
4628 {
4629 return mImpl->setTensorAddress(tensorName, data);
4630 }
4631
4644 void const* getTensorAddress(char const* tensorName) const noexcept
4645 {
4646 return mImpl->getTensorAddress(tensorName);
4647 }
4648
4667 bool setOutputTensorAddress(char const* tensorName, void* data) noexcept
4668 {
4669 return mImpl->setOutputTensorAddress(tensorName, data);
4670 }
4671
4689 bool setInputTensorAddress(char const* tensorName, void const* data) noexcept
4690 {
4691 return mImpl->setInputTensorAddress(tensorName, data);
4692 }
4693
4708 void* getOutputTensorAddress(char const* tensorName) const noexcept
4709 {
4710 return mImpl->getOutputTensorAddress(tensorName);
4711 }
4712
4741 int32_t inferShapes(int32_t nbMaxNames, char const** tensorNames) noexcept
4742 {
4743 return mImpl->inferShapes(nbMaxNames, tensorNames);
4744 }
4745
4759 {
4760 return mImpl->updateDeviceMemorySizeForShapes();
4761 }
4762
4774 bool setInputConsumedEvent(cudaEvent_t event) noexcept
4775 {
4776 return mImpl->setInputConsumedEvent(event);
4777 }
4778
4784 cudaEvent_t getInputConsumedEvent() const noexcept
4785 {
4786 return mImpl->getInputConsumedEvent();
4787 }
4788
4803 bool setOutputAllocator(char const* tensorName, IOutputAllocator* outputAllocator) noexcept
4804 {
4805 return mImpl->setOutputAllocator(tensorName, outputAllocator);
4806 }
4807
4816 IOutputAllocator* getOutputAllocator(char const* tensorName) const noexcept
4817 {
4818 return mImpl->getOutputAllocator(tensorName);
4819 }
4820
4834 int64_t getMaxOutputSize(char const* tensorName) const noexcept
4835 {
4836 return mImpl->getMaxOutputSize(tensorName);
4837 }
4838
4855 {
4856 return mImpl->setTemporaryStorageAllocator(allocator);
4857 }
4858
4865 {
4866 return mImpl->getTemporaryStorageAllocator();
4867 }
4868
4888 bool enqueueV3(cudaStream_t stream) noexcept
4889 {
4890 return mImpl->enqueueV3(stream);
4891 }
4892
4904 void setPersistentCacheLimit(size_t size) noexcept
4905 {
4906 mImpl->setPersistentCacheLimit(size);
4907 }
4908
4915 size_t getPersistentCacheLimit() const noexcept
4916 {
4917 return mImpl->getPersistentCacheLimit();
4918 }
4919
4939 bool setNvtxVerbosity(ProfilingVerbosity verbosity) noexcept
4940 {
4941 return mImpl->setNvtxVerbosity(verbosity);
4942 }
4943
4952 {
4953 return mImpl->getNvtxVerbosity();
4954 }
4955
4982 void setAuxStreams(cudaStream_t* auxStreams, int32_t nbStreams) noexcept
4983 {
4984 mImpl->setAuxStreams(auxStreams, nbStreams);
4985 }
4986
4994 bool setDebugListener(IDebugListener* listener) noexcept
4995 {
4996 return mImpl->setDebugListener(listener);
4997 }
4998
5005 {
5006 return mImpl->getDebugListener();
5007 }
5008
5023 bool setTensorDebugState(char const* name, bool flag) noexcept
5024 {
5025 return mImpl->setTensorDebugState(name, flag);
5026 }
5027
5036 bool setAllTensorsDebugState(bool flag) noexcept
5037 {
5038 return mImpl->setAllTensorsDebugState(flag);
5039 }
5040
5046 bool getDebugState(char const* name) const noexcept
5047 {
5048 return mImpl->getDebugState(name);
5049 }
5050
5057 {
5058 return mImpl->getRuntimeConfig();
5059 }
5060
5061protected:
5062 apiv::VExecutionContext* mImpl;
5063}; // class IExecutionContext
5064
5072enum class LayerInformationFormat : int32_t
5073{
5074 kONELINE = 0,
5075 kJSON = 1,
5076};
5077
5080template <>
5081constexpr inline int32_t EnumMax<LayerInformationFormat>() noexcept
5082{
5083 return 2;
5084}
5085
5102{
5103public:
5104 virtual ~IEngineInspector() noexcept = default;
5105
5118 bool setExecutionContext(IExecutionContext const* context) noexcept
5119 {
5120 return mImpl->setExecutionContext(context);
5121 }
5122
5131 {
5132 return mImpl->getExecutionContext();
5133 }
5134
5155 char const* getLayerInformation(int32_t layerIndex, LayerInformationFormat format) const noexcept
5156 {
5157 return mImpl->getLayerInformation(layerIndex, format);
5158 }
5159
5178 char const* getEngineInformation(LayerInformationFormat format) const noexcept
5179 {
5180 return mImpl->getEngineInformation(format);
5181 }
5182
5197 void setErrorRecorder(IErrorRecorder* recorder) noexcept
5198 {
5199 mImpl->setErrorRecorder(recorder);
5200 }
5201
5213 {
5214 return mImpl->getErrorRecorder();
5215 }
5216
5217protected:
5218 apiv::VEngineInspector* mImpl;
5219}; // class IEngineInspector
5220
5221} // namespace nvinfer1
5222
5227extern "C" TENSORRTAPI void* createInferRuntime_INTERNAL(void* logger, int32_t version) noexcept;
5228
5233extern "C" TENSORRTAPI void* createInferRefitter_INTERNAL(void* engine, void* logger, int32_t version) noexcept;
5234
5239
5245extern "C" TENSORRTAPI nvinfer1::ILogger* getLogger() noexcept;
5246
5247namespace nvinfer1
5248{
5249namespace // unnamed namespace avoids linkage surprises when linking objects built with different versions of this
5250 // header.
5251{
5257inline IRuntime* createInferRuntime(ILogger& logger) noexcept
5258{
5259 return static_cast<IRuntime*>(createInferRuntime_INTERNAL(&logger, NV_TENSORRT_VERSION));
5260}
5261
5268inline IRefitter* createInferRefitter(ICudaEngine& engine, ILogger& logger) noexcept
5269{
5270 return static_cast<IRefitter*>(createInferRefitter_INTERNAL(&engine, &logger, NV_TENSORRT_VERSION));
5271}
5272
5273} // namespace
5274
5286template <typename T>
5288{
5289public:
5291 {
5292 getPluginRegistry()->registerCreator(instance, "");
5293 }
5294
5295private:
5297 T instance{};
5298};
5299
5300} // namespace nvinfer1
5301
5302#define REGISTER_TENSORRT_PLUGIN(name) \
5303 static nvinfer1::PluginRegistrar<name> pluginRegistrar##name {}
5304
5305namespace nvinfer1
5306{
5316{
5317public:
5325 virtual ILogger* findLogger() = 0;
5326
5327protected:
5328 virtual ~ILoggerFinder() = default;
5329};
5330
5333namespace v_1_0
5334{
5335
5337{
5338public:
5340 ~IGpuAsyncAllocator() override = default;
5341
5371 void* allocateAsync(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags,
5372 cudaStream_t /*stream*/) noexcept override = 0;
5373
5399 bool deallocateAsync(void* const memory, cudaStream_t /*stream*/) noexcept override = 0;
5400
5424 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept override
5425 {
5426 return allocateAsync(size, alignment, flags, nullptr);
5427 }
5428
5447 TRT_DEPRECATED bool deallocate(void* const memory) noexcept override
5448 {
5449 return deallocateAsync(memory, nullptr);
5450 }
5451
5455 InterfaceInfo getInterfaceInfo() const noexcept override
5456 {
5457 return {"IGpuAllocator", 1, 0};
5458 }
5459};
5460
5462{
5463public:
5467 InterfaceInfo getInterfaceInfo() const noexcept override
5468 {
5469 return InterfaceInfo{"PLUGIN CREATOR_V3ONE", 1, 0};
5470 }
5471
5489 AsciiChar const* name, PluginFieldCollection const* fc, TensorRTPhase phase) noexcept = 0;
5490
5497 virtual PluginFieldCollection const* getFieldNames() noexcept = 0;
5498
5505 virtual AsciiChar const* getPluginName() const noexcept = 0;
5506
5513 virtual AsciiChar const* getPluginVersion() const noexcept = 0;
5514
5521 virtual AsciiChar const* getPluginNamespace() const noexcept = 0;
5522
5524 virtual ~IPluginCreatorV3One() = default;
5525
5526protected:
5529 IPluginCreatorV3One& operator=(IPluginCreatorV3One const&) & = default;
5530 IPluginCreatorV3One& operator=(IPluginCreatorV3One&&) & = default;
5531};
5532
5533} // namespace v_1_0
5534
5549
5559
5560} // namespace nvinfer1
5561
5565extern "C" TENSORRTAPI int32_t getInferLibMajorVersion() noexcept;
5569extern "C" TENSORRTAPI int32_t getInferLibMinorVersion() noexcept;
5573extern "C" TENSORRTAPI int32_t getInferLibPatchVersion() noexcept;
5577extern "C" TENSORRTAPI int32_t getInferLibBuildVersion() noexcept;
5578
5579#endif // NV_INFER_RUNTIME_H
nvinfer1::ILogger * getLogger() noexcept
Return the logger object.
int32_t getInferLibMajorVersion() noexcept
Return the library major version number.
nvinfer1::IPluginRegistry * getPluginRegistry() noexcept
Return the plugin registry.
int32_t getInferLibPatchVersion() noexcept
Return the library patch version number.
int32_t getInferLibMinorVersion() noexcept
Return the library minor version number.
int32_t getInferLibBuildVersion() noexcept
Return the library build version number.
#define TENSORRTAPI
Definition: NvInferRuntimeBase.h:69
#define NV_TENSORRT_VERSION
Definition: NvInferRuntimeBase.h:101
#define TRT_DEPRECATED
Definition: NvInferRuntimeBase.h:42
#define TRT_DEPRECATED_ENUM
Definition: NvInferRuntimeBase.h:43
Structure to define the dimensions of a tensor.
Definition: NvInferRuntimeBase.h:214
static constexpr int32_t MAX_DIMS
The maximum rank (number of dimensions) supported for a tensor.
Definition: NvInferRuntimeBase.h:217
Analog of class Dims with expressions instead of constants for the dimensions.
Definition: NvInferRuntime.h:350
IDimensionExpr const * d[Dims::MAX_DIMS]
The extent of each dimension.
Definition: NvInferRuntime.h:353
int32_t nbDims
The number of dimensions.
Definition: NvInferRuntime.h:352
An engine for executing inference on a built network, with functionally unsafe features.
Definition: NvInferRuntime.h:3111
int32_t getTensorBytesPerComponent(char const *tensorName) const noexcept
Return the number of bytes per component of an element, or -1 if the tensor is not vectorized or prov...
Definition: NvInferRuntime.h:3361
ISerializationConfig * createSerializationConfig() noexcept
Create a serialization configuration object.
Definition: NvInferRuntime.h:3761
TRT_DEPRECATED int64_t getWeightStreamingBudget() const noexcept
Returns the current weight streaming device memory budget in bytes.
Definition: NvInferRuntime.h:3842
char const * getIOTensorName(int32_t index) const noexcept
Return name of an IO tensor.
Definition: NvInferRuntime.h:3725
int64_t getWeightStreamingBudgetV2() const noexcept
Returns the current weight streaming device memory budget in bytes.
Definition: NvInferRuntime.h:3941
EngineCapability getEngineCapability() const noexcept
Determine what execution capability this engine has.
Definition: NvInferRuntime.h:3614
IErrorRecorder * getErrorRecorder() const noexcept
Get the ErrorRecorder assigned to this interface.
Definition: NvInferRuntime.h:3648
TensorFormat getTensorFormat(char const *tensorName, int32_t profileIndex) const noexcept
Return the tensor format of given profile, or TensorFormat::kLINEAR if the provided name does not map...
Definition: NvInferRuntime.h:3447
int64_t const * getProfileTensorValuesV2(char const *tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum values (not dimensions) for an input tensor given its name under ...
Definition: NvInferRuntime.h:4032
TRT_DEPRECATED bool hasImplicitBatchDimension() const noexcept
Query whether the engine was built with an implicit batch dimension.
Definition: NvInferRuntime.h:3662
apiv::VCudaEngine * mImpl
Definition: NvInferRuntime.h:4039
TRT_DEPRECATED size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept
Return the maximum device memory required by the context for a profile.
Definition: NvInferRuntime.h:3298
IExecutionContext * createExecutionContext(ExecutionContextAllocationStrategy strategy=ExecutionContextAllocationStrategy::kSTATIC) noexcept
Create an execution context and specify the strategy for allocating internal activation memory.
Definition: NvInferRuntime.h:3185
char const * getTensorFormatDesc(char const *tensorName) const noexcept
Return the human readable description of the tensor format, or empty string if the provided name does...
Definition: NvInferRuntime.h:3471
Dims getProfileShape(char const *tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum dimensions for an input tensor given its name under an optimizati...
Definition: NvInferRuntime.h:3573
bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept
Limit the maximum amount of GPU memory usable for network weights in bytes.
Definition: NvInferRuntime.h:3923
IExecutionContext * createExecutionContext(IRuntimeConfig *runtimeConfig) noexcept
Create an execution context with TensorRT JIT runtime config.
Definition: NvInferRuntime.h:3261
int32_t getNbAuxStreams() const noexcept
Return the number of auxiliary streams used by this engine.
Definition: NvInferRuntime.h:3751
int64_t getStreamableWeightsSize() const noexcept
Get the total size in bytes of all streamable weights.
Definition: NvInferRuntime.h:3881
DataType getTensorDataType(char const *tensorName) const noexcept
Determine the required data type for a buffer from its tensor name.
Definition: NvInferRuntime.h:3140
void setErrorRecorder(IErrorRecorder *recorder) noexcept
Set the ErrorRecorder for this interface.
Definition: NvInferRuntime.h:3633
TacticSources getTacticSources() const noexcept
return the tactic sources required by this engine.
Definition: NvInferRuntime.h:3678
IHostMemory * serializeWithConfig(ISerializationConfig &config) const noexcept
Serialize the network to a stream with the provided SerializationConfig.
Definition: NvInferRuntime.h:3777
virtual ~ICudaEngine() noexcept=default
int64_t getWeightStreamingAutomaticBudget() const noexcept
TensorRT automatically determines a device memory budget for the model to run. The budget is close to...
Definition: NvInferRuntime.h:3966
bool isDebugTensor(char const *name) const noexcept
Check if a tensor is marked as a debug tensor.
Definition: NvInferRuntime.h:4008
int32_t getTensorVectorizedDim(char const *tensorName, int32_t profileIndex) const noexcept
Return the dimension index that the buffer is vectorized of given profile, or -1 if the provided name...
Definition: NvInferRuntime.h:3527
char const * getName() const noexcept
Returns the name of the network associated with the engine.
Definition: NvInferRuntime.h:3542
ProfilingVerbosity getProfilingVerbosity() const noexcept
Return the ProfilingVerbosity the builder config was set to when the engine was built.
Definition: NvInferRuntime.h:3690
bool isShapeInferenceIO(char const *tensorName) const noexcept
True if tensor is required as input for shape calculations or is output from shape calculations.
Definition: NvInferRuntime.h:3223
TRT_DEPRECATED int32_t const * getProfileTensorValues(char const *tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum values (not dimensions) for an input tensor given its name under ...
Definition: NvInferRuntime.h:3598
int64_t getWeightStreamingScratchMemorySize() const noexcept
Returns the size of the scratch memory required by the current weight streaming budget.
Definition: NvInferRuntime.h:3994
TRT_DEPRECATED bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept
Limit the maximum amount of GPU memory usable for network weights in bytes.
Definition: NvInferRuntime.h:3822
int64_t getDeviceMemorySizeV2() const noexcept
Return the maximum device memory required by the context over all profiles.
Definition: NvInferRuntime.h:3314
int32_t getTensorVectorizedDim(char const *tensorName) const noexcept
Return the dimension index that the buffer is vectorized, or -1 if the provided name does not map to ...
Definition: NvInferRuntime.h:3511
TRT_DEPRECATED size_t getDeviceMemorySize() const noexcept
Return the maximum device memory required by the context over all profiles.
Definition: NvInferRuntime.h:3286
int32_t getTensorComponentsPerElement(char const *tensorName, int32_t profileIndex) const noexcept
Return the number of components included in one element of given profile, or -1 if tensor is not vect...
Definition: NvInferRuntime.h:3418
int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept
Return the maximum device memory required by the context for a profile.
Definition: NvInferRuntime.h:3330
IRuntimeConfig * createRuntimeConfig() noexcept
Create a runtime config for TensorRT JIT. The caller is responsible for ownership of the returned IRu...
Definition: NvInferRuntime.h:3274
TensorFormat getTensorFormat(char const *tensorName) const noexcept
Return the tensor format, or TensorFormat::kLINEAR if the provided name does not map to an input or o...
Definition: NvInferRuntime.h:3433
IHostMemory * serialize() const noexcept
Serialize the network to a stream.
Definition: NvInferRuntime.h:3168
TensorLocation getTensorLocation(char const *tensorName) const noexcept
Get whether an input or output tensor must be on GPU or CPU.
Definition: NvInferRuntime.h:3203
IEngineInspector * createEngineInspector() const noexcept
Create a new engine inspector which prints the layer information in an engine or an execution context...
Definition: NvInferRuntime.h:3700
int32_t getTensorBytesPerComponent(char const *tensorName, int32_t profileIndex) const noexcept
Return the number of bytes per component of an element given of given profile, or -1 if the tensor is...
Definition: NvInferRuntime.h:3379
HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept
Return the hardware compatibility level of this engine.
Definition: NvInferRuntime.h:3736
int32_t getNbOptimizationProfiles() const noexcept
Get the number of optimization profiles defined for this engine.
Definition: NvInferRuntime.h:3553
TRT_DEPRECATED IExecutionContext * createExecutionContextWithoutDeviceMemory() noexcept
create an execution context without any device memory allocated
Definition: NvInferRuntime.h:3249
char const * getTensorFormatDesc(char const *tensorName, int32_t profileIndex) const noexcept
Return the human readable description of the tensor format of given profile, or empty string if the p...
Definition: NvInferRuntime.h:3494
TRT_DEPRECATED int64_t getMinimumWeightStreamingBudget() const noexcept
The minimum number of bytes of GPU memory required by network weights for successful weight streaming...
Definition: NvInferRuntime.h:3865
TensorIOMode getTensorIOMode(char const *tensorName) const noexcept
Determine whether a tensor is an input or output tensor.
Definition: NvInferRuntime.h:3237
int32_t getNbLayers() const noexcept
Get the number of layers in the network.
Definition: NvInferRuntime.h:3154
int32_t getNbIOTensors() const noexcept
Return number of IO tensors.
Definition: NvInferRuntime.h:3713
int32_t getTensorComponentsPerElement(char const *tensorName) const noexcept
Return the number of components included in one element, or -1 if tensor is not vectorized or if the ...
Definition: NvInferRuntime.h:3400
bool isRefittable() const noexcept
Return true if an engine can be refit.
Definition: NvInferRuntime.h:3340
An IDimensionExpr represents an integer expression constructed from constants, input dimensions,...
Definition: NvInferRuntime.h:232
bool isConstant() const noexcept
Return true if expression is a build-time constant.
Definition: NvInferRuntime.h:237
bool isSizeTensor() const noexcept
Return true if this denotes the value of a size tensor.
Definition: NvInferRuntime.h:263
virtual ~IDimensionExpr() noexcept=default
apiv::VDimensionExpr * mImpl
Definition: NvInferRuntime.h:254
int64_t getConstantValue() const noexcept
Get the value of the constant.
Definition: NvInferRuntime.h:248
An engine inspector which prints out the layer information of an engine or an execution context.
Definition: NvInferRuntime.h:5102
char const * getLayerInformation(int32_t layerIndex, LayerInformationFormat format) const noexcept
Get a string describing the information about a specific layer in the current engine or the execution...
Definition: NvInferRuntime.h:5155
IErrorRecorder * getErrorRecorder() const noexcept
Get the ErrorRecorder assigned to this interface.
Definition: NvInferRuntime.h:5212
void setErrorRecorder(IErrorRecorder *recorder) noexcept
Set the ErrorRecorder for this interface.
Definition: NvInferRuntime.h:5197
IExecutionContext const * getExecutionContext() const noexcept
Get the context currently being inspected.
Definition: NvInferRuntime.h:5130
apiv::VEngineInspector * mImpl
Definition: NvInferRuntime.h:5218
virtual ~IEngineInspector() noexcept=default
char const * getEngineInformation(LayerInformationFormat format) const noexcept
Get a string describing the information about all the layers in the current engine or the execution c...
Definition: NvInferRuntime.h:5178
Context for executing inference using an engine, with functionally unsafe features.
Definition: NvInferRuntime.h:4181
IOutputAllocator * getOutputAllocator(char const *tensorName) const noexcept
Get output allocator associated with output tensor of given name, or nullptr if the provided name doe...
Definition: NvInferRuntime.h:4816
IErrorRecorder * getErrorRecorder() const noexcept
Get the ErrorRecorder assigned to this interface.
Definition: NvInferRuntime.h:4464
bool reportToProfiler() const noexcept
Calculate layer timing info for the current optimization profile in IExecutionContext and update the ...
Definition: NvInferRuntime.h:4583
virtual ~IExecutionContext() noexcept=default
void setDeviceMemory(void *memory) noexcept
Set the device memory for use by this execution context.
Definition: NvInferRuntime.h:4284
TRT_DEPRECATED bool allInputShapesSpecified() const noexcept
Whether all input shape bindings have been specified.
Definition: NvInferRuntime.h:4430
bool setTensorDebugState(char const *name, bool flag) noexcept
Set debug state of tensor given the tensor name.
Definition: NvInferRuntime.h:5023
char const * getName() const noexcept
Return the name of the execution context.
Definition: NvInferRuntime.h:4257
IGpuAllocator * getTemporaryStorageAllocator() const noexcept
Get allocator set by setTemporaryStorageAllocator.
Definition: NvInferRuntime.h:4864
void setEnqueueEmitsProfile(bool enqueueEmitsProfile) noexcept
Set whether enqueue emits layer timing to the profiler.
Definition: NvInferRuntime.h:4541
Dims getTensorShape(char const *tensorName) const noexcept
Return the shape of the given input or output.
Definition: NvInferRuntime.h:4397
bool getDebugState(char const *name) const noexcept
Definition: NvInferRuntime.h:5046
bool setInputShape(char const *tensorName, Dims const &dims) noexcept
Set shape of given input.
Definition: NvInferRuntime.h:4360
bool executeV2(void *const *bindings) noexcept
Synchronously execute a network.
Definition: NvInferRuntime.h:4481
bool getEnqueueEmitsProfile() const noexcept
Get the enqueueEmitsProfile state.
Definition: NvInferRuntime.h:4553
void const * getTensorAddress(char const *tensorName) const noexcept
Get memory address bound to given input or output tensor, or nullptr if the provided name does not ma...
Definition: NvInferRuntime.h:4644
bool setOutputAllocator(char const *tensorName, IOutputAllocator *outputAllocator) noexcept
Set output allocator to use for output tensor of given name. Pass nullptr to outputAllocator to unset...
Definition: NvInferRuntime.h:4803
bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept
Select an optimization profile for the current context with async semantics.
Definition: NvInferRuntime.h:4525
apiv::VExecutionContext * mImpl
Definition: NvInferRuntime.h:5062
bool setOutputTensorAddress(char const *tensorName, void *data) noexcept
Set the memory address for a given output tensor.
Definition: NvInferRuntime.h:4667
void setPersistentCacheLimit(size_t size) noexcept
Set the maximum size for persistent cache usage.
Definition: NvInferRuntime.h:4904
size_t getPersistentCacheLimit() const noexcept
Get the maximum size for persistent cache usage.
Definition: NvInferRuntime.h:4915
bool setAllTensorsDebugState(bool flag) noexcept
Definition: NvInferRuntime.h:5036
ICudaEngine const & getEngine() const noexcept
Get the associated engine.
Definition: NvInferRuntime.h:4233
ProfilingVerbosity getNvtxVerbosity() const noexcept
Get the NVTX verbosity of the execution context.
Definition: NvInferRuntime.h:4951
size_t updateDeviceMemorySizeForShapes() noexcept
Recompute the internal activation buffer sizes based on the current input shapes, and return the tota...
Definition: NvInferRuntime.h:4758
void setAuxStreams(cudaStream_t *auxStreams, int32_t nbStreams) noexcept
Set the auxiliary streams that TensorRT should launch kernels on in the next enqueueV3() call.
Definition: NvInferRuntime.h:4982
int64_t getMaxOutputSize(char const *tensorName) const noexcept
Get upper bound on an output tensor's size, in bytes, based on the current optimization profile and i...
Definition: NvInferRuntime.h:4834
int32_t inferShapes(int32_t nbMaxNames, char const **tensorNames) noexcept
Run shape calculations.
Definition: NvInferRuntime.h:4741
bool setDebugListener(IDebugListener *listener) noexcept
Set DebugListener for this execution context.
Definition: NvInferRuntime.h:4994
bool setTensorAddress(char const *tensorName, void *data) noexcept
Set memory address for given input or output tensor.
Definition: NvInferRuntime.h:4627
bool setTemporaryStorageAllocator(IGpuAllocator *allocator) noexcept
Specify allocator to use for internal temporary storage.
Definition: NvInferRuntime.h:4854
void * getOutputTensorAddress(char const *tensorName) const noexcept
Get memory address for given output.
Definition: NvInferRuntime.h:4708
bool enqueueV3(cudaStream_t stream) noexcept
Enqueue inference on a stream.
Definition: NvInferRuntime.h:4888
IDebugListener * getDebugListener() noexcept
Get the DebugListener of this execution context.
Definition: NvInferRuntime.h:5004
int32_t getOptimizationProfile() const noexcept
Get the index of the currently selected optimization profile.
Definition: NvInferRuntime.h:4342
bool setInputTensorAddress(char const *tensorName, void const *data) noexcept
Set memory address for given input.
Definition: NvInferRuntime.h:4689
bool getDebugSync() const noexcept
Get the debug sync flag.
Definition: NvInferRuntime.h:4203
bool setInputConsumedEvent(cudaEvent_t event) noexcept
Mark input as consumed.
Definition: NvInferRuntime.h:4774
Dims getTensorStrides(char const *tensorName) const noexcept
Return the strides of the buffer for the given tensor name.
Definition: NvInferRuntime.h:4327
bool setNvtxVerbosity(ProfilingVerbosity verbosity) noexcept
Set the verbosity of the NVTX markers in the execution context.
Definition: NvInferRuntime.h:4939
IProfiler * getProfiler() const noexcept
Get the profiler.
Definition: NvInferRuntime.h:4223
void setErrorRecorder(IErrorRecorder *recorder) noexcept
Set the ErrorRecorder for this interface.
Definition: NvInferRuntime.h:4449
void setDeviceMemoryV2(void *memory, int64_t size) noexcept
Set the device memory and its corresponding size for use by this execution context.
Definition: NvInferRuntime.h:4306
bool allInputDimensionsSpecified() const noexcept
Whether all dynamic dimensions of input tensors have been specified.
Definition: NvInferRuntime.h:4413
void setProfiler(IProfiler *profiler) noexcept
Set the profiler.
Definition: NvInferRuntime.h:4213
void setName(char const *name) noexcept
Set the name of the execution context.
Definition: NvInferRuntime.h:4247
cudaEvent_t getInputConsumedEvent() const noexcept
The event associated with consuming the input.
Definition: NvInferRuntime.h:4784
IRuntimeConfig * getRuntimeConfig() const noexcept
Get the runtime config object used during execution context creation.
Definition: NvInferRuntime.h:5056
Object for constructing IDimensionExpr.
Definition: NvInferRuntime.h:287
IDimensionExpr const * operation(DimensionOperation op, IDimensionExpr const &first, IDimensionExpr const &second) noexcept
Get the operation.
Definition: NvInferRuntime.h:303
virtual ~IExprBuilder() noexcept=default
IDimensionExpr const * constant(int64_t value) noexcept
Return pointer to IDimensionExpr for given value.
Definition: NvInferRuntime.h:292
apiv::VExprBuilder * mImpl
Definition: NvInferRuntime.h:310
IDimensionExpr const * declareSizeTensor(int32_t outputIndex, IDimensionExpr const &opt, IDimensionExpr const &upper)
Declare a size tensor at the given output index, with the specified auto-tuning formula and upper bou...
Definition: NvInferRuntime.h:338
Class to handle library allocated memory that is accessible to the user.
Definition: NvInferRuntime.h:142
void * data() const noexcept
A pointer to the raw data that is owned by the library.
Definition: NvInferRuntime.h:147
DataType type() const noexcept
The type of the memory that was allocated.
Definition: NvInferRuntime.h:159
std::size_t size() const noexcept
The size in bytes of the data that was allocated.
Definition: NvInferRuntime.h:153
apiv::VHostMemory * mImpl
Definition: NvInferRuntime.h:165
virtual ~IHostMemory() noexcept=default
A virtual base class to find a logger. Allows a plugin to find an instance of a logger if it needs to...
Definition: NvInferRuntime.h:5316
virtual ILogger * findLogger()=0
Get the logger used by the engine or execution context which called the plugin method.
virtual ~ILoggerFinder()=default
Application-implemented logging interface for the builder, refitter and runtime.
Definition: NvInferRuntime.h:1542
virtual ~ILogger()=default
Severity
The severity corresponding to a log message.
Definition: NvInferRuntime.h:1550
virtual void log(Severity severity, AsciiChar const *msg) noexcept=0
A callback implemented by the application to handle logging messages;.
Forward declaration of IEngineInspector for use by other interfaces.
Definition: NvInferRuntime.h:51
INoCopy & operator=(INoCopy &&other)=delete
INoCopy(INoCopy const &other)=delete
INoCopy(INoCopy &&other)=delete
virtual ~INoCopy()=default
INoCopy & operator=(INoCopy const &other)=delete
Optimization profile for dynamic input dimensions and shape tensors.
Definition: NvInferRuntime.h:2618
TRT_DEPRECATED int32_t const * getShapeValues(char const *inputName, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum values for an input shape tensor.
Definition: NvInferRuntime.h:2739
apiv::VOptimizationProfile * mImpl
Definition: NvInferRuntime.h:2855
Dims getDimensions(char const *inputName, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum dimensions for a dynamic input tensor.
Definition: NvInferRuntime.h:2659
TRT_DEPRECATED bool setShapeValues(char const *inputName, OptProfileSelector select, int32_t const *values, int32_t nbValues) noexcept
Set the minimum / optimum / maximum values for an input shape tensor.
Definition: NvInferRuntime.h:2711
virtual ~IOptimizationProfile() noexcept=default
float getExtraMemoryTarget() const noexcept
Get the extra memory target that has been defined for this profile.
Definition: NvInferRuntime.h:2769
bool setExtraMemoryTarget(float target) noexcept
Set a target for extra GPU memory that may be used by this profile.
Definition: NvInferRuntime.h:2757
bool setDimensions(char const *inputName, OptProfileSelector select, Dims const &dims) noexcept
Set the minimum / optimum / maximum dimensions for a dynamic input tensor.
Definition: NvInferRuntime.h:2647
bool isValid() const noexcept
Check whether the optimization profile can be passed to an IBuilderConfig object.
Definition: NvInferRuntime.h:2786
int64_t const * getShapeValuesV2(char const *inputName, OptProfileSelector select) const noexcept
Get the minimum / optimum / maximum values for an input shape tensor.
Definition: NvInferRuntime.h:2849
bool setShapeValuesV2(char const *inputName, OptProfileSelector select, int64_t const *values, int32_t nbValues) noexcept
Set the minimum / optimum / maximum values for an input shape tensor.
Definition: NvInferRuntime.h:2836
int32_t getNbShapeValues(char const *inputName) const noexcept
Get the number of values for an input shape tensor.
Definition: NvInferRuntime.h:2725
Single registration point for all plugins in an application. It is used to find plugin implementation...
Definition: NvInferRuntimeCommon.h:56
virtual TRT_DEPRECATED bool registerCreator(IPluginCreator &creator, AsciiChar const *const pluginNamespace) noexcept=0
Register a plugin creator implementing IPluginCreator. Returns false if any plugin creator with the s...
Interface for plugins to access per context resources provided by TensorRT.
Definition: NvInferRuntime.h:767
virtual IErrorRecorder * getErrorRecorder() const noexcept=0
Get the error recorder associated with the resource context.
IPluginResourceContext & operator=(IPluginResourceContext const &) &=default
virtual IGpuAllocator * getGpuAllocator() const noexcept=0
Get the GPU allocator associated with the resource context.
Similar to IPluginV2Ext, but with support for dynamic shapes.
Definition: NvInferRuntime.h:407
IPluginV2DynamicExt * clone() const noexcept override=0
Clone the plugin object. This copies over internal plugin parameters as well and returns a new plugin...
virtual ~IPluginV2DynamicExt() noexcept
Definition: NvInferRuntime.h:558
Plugin class for user-implemented layers.
Definition: NvInferRuntimePlugin.h:474
Updates weights in an engine.
Definition: NvInferRuntime.h:2138
bool refitCudaEngineAsync(cudaStream_t stream) noexcept
Enqueue weights refitting of the associated engine on the given stream.
Definition: NvInferRuntime.h:2540
int32_t getMaxThreads() const noexcept
get the maximum number of threads that can be used by the refitter.
Definition: NvInferRuntime.h:2419
TensorLocation getWeightsLocation(char const *weightsName) const noexcept
Get location for the weights associated with the given name.
Definition: NvInferRuntime.h:2478
bool setNamedWeights(char const *name, Weights weights) noexcept
Specify new weights of given name.
Definition: NvInferRuntime.h:2343
int32_t getAllWeights(int32_t size, char const **weightsNames) noexcept
Get names of all weights that could be refit.
Definition: NvInferRuntime.h:2379
ILogger * getLogger() const noexcept
get the logger with which the refitter was created
Definition: NvInferRuntime.h:2389
bool refitCudaEngine() noexcept
Refits associated engine.
Definition: NvInferRuntime.h:2174
int32_t getMissingWeights(int32_t size, char const **weightsNames) noexcept
Get names of missing weights.
Definition: NvInferRuntime.h:2363
TRT_DEPRECATED float getDynamicRangeMax(char const *tensorName) const noexcept
Get maximum of dynamic range.
Definition: NvInferRuntime.h:2266
int32_t getMissing(int32_t size, char const **layerNames, WeightsRole *roles) noexcept
Get description of missing weights.
Definition: NvInferRuntime.h:2195
Weights getNamedWeights(char const *weightsName) const noexcept
Get weights associated with the given name.
Definition: NvInferRuntime.h:2462
bool unsetNamedWeights(char const *weightsName) noexcept
Unset weights associated with the given name.
Definition: NvInferRuntime.h:2494
Weights getWeightsPrototype(char const *weightsName) const noexcept
Get the Weights prototype associated with the given name.
Definition: NvInferRuntime.h:2558
bool setMaxThreads(int32_t maxThreads) noexcept
Set the maximum number of threads.
Definition: NvInferRuntime.h:2405
TRT_DEPRECATED float getDynamicRangeMin(char const *tensorName) const noexcept
Get minimum of dynamic range.
Definition: NvInferRuntime.h:2250
TRT_DEPRECATED int32_t getTensorsWithDynamicRange(int32_t size, char const **tensorNames) const noexcept
Get names of all tensors that have refittable dynamic ranges.
Definition: NvInferRuntime.h:2284
bool setNamedWeights(char const *name, Weights weights, TensorLocation location) noexcept
Specify new weights on a specified device of given name.
Definition: NvInferRuntime.h:2446
void setWeightsValidation(bool weightsValidation) noexcept
Set whether to validate weights during refitting.
Definition: NvInferRuntime.h:2510
TRT_DEPRECATED bool setDynamicRange(char const *tensorName, float min, float max) noexcept
Definition: NvInferRuntime.h:2234
apiv::VRefitter * mImpl
Definition: NvInferRuntime.h:2564
int32_t getAll(int32_t size, char const **layerNames, WeightsRole *roles) noexcept
Get description of all weights that could be refit.
Definition: NvInferRuntime.h:2212
virtual ~IRefitter() noexcept=default
bool getWeightsValidation() const noexcept
Get whether to validate weights values during refitting.
Definition: NvInferRuntime.h:2518
void setErrorRecorder(IErrorRecorder *recorder) noexcept
Set the ErrorRecorder for this interface.
Definition: NvInferRuntime.h:2303
IErrorRecorder * getErrorRecorder() const noexcept
Get the ErrorRecorder assigned to this interface.
Definition: NvInferRuntime.h:2318
A class for runtime configuration. This class is used during execution context creation.
Definition: NvInferRuntime.h:3074
virtual ~IRuntimeConfig() noexcept=default
apiv::VRuntimeConfig * mImpl
Definition: NvInferRuntime.h:3100
ExecutionContextAllocationStrategy getExecutionContextAllocationStrategy() const noexcept
Get the execution context allocation strategy.
Definition: NvInferRuntime.h:3093
Allows a serialized functionally unsafe engine to be deserialized.
Definition: NvInferRuntime.h:1819
bool setMaxThreads(int32_t maxThreads) noexcept
Set the maximum number of threads.
Definition: NvInferRuntime.h:1983
IRuntime * loadRuntime(char const *path) noexcept
Load IRuntime from the file.
Definition: NvInferRuntime.h:2099
bool getEngineHostCodeAllowed() const noexcept
Get whether the runtime is allowed to deserialize engines with host executable code.
Definition: NvInferRuntime.h:2121
TempfileControlFlags getTempfileControlFlags() const noexcept
Get the tempfile control flags for this runtime.
Definition: NvInferRuntime.h:2071
void setEngineHostCodeAllowed(bool allowed) noexcept
Set whether the runtime is allowed to deserialize engines with host executable code.
Definition: NvInferRuntime.h:2111
virtual ~IRuntime() noexcept=default
void setTemporaryDirectory(char const *path) noexcept
Set the directory that will be used by this runtime for temporary files.
Definition: NvInferRuntime.h:2032
IPluginRegistry & getPluginRegistry() noexcept
Get the local plugin registry that can be used by the runtime.
Definition: NvInferRuntime.h:2081
apiv::VRuntime * mImpl
Definition: NvInferRuntime.h:2127
int32_t getNbDLACores() const noexcept
Returns number of DLA hardware cores accessible or 0 if DLA is unavailable.
Definition: NvInferRuntime.h:1852
ICudaEngine * deserializeCudaEngine(void const *blob, std::size_t size) noexcept
Deserialize an engine from host memory.
Definition: NvInferRuntime.h:1917
void setTempfileControlFlags(TempfileControlFlags flags) noexcept
Set the tempfile control flags for this runtime.
Definition: NvInferRuntime.h:2059
int32_t getDLACore() const noexcept
Get the DLA core that the engine executes on.
Definition: NvInferRuntime.h:1844
void setGpuAllocator(IGpuAllocator *allocator) noexcept
Set the GPU allocator.
Definition: NvInferRuntime.h:1868
IErrorRecorder * getErrorRecorder() const noexcept
get the ErrorRecorder assigned to this interface.
Definition: NvInferRuntime.h:1902
ICudaEngine * deserializeCudaEngine(IStreamReaderV2 &streamReader)
Deserialize an engine from a stream. IStreamReaderV2 is expected to support reading to both host and ...
Definition: NvInferRuntime.h:1958
ILogger * getLogger() const noexcept
get the logger with which the runtime was created
Definition: NvInferRuntime.h:1968
int32_t getMaxThreads() const noexcept
Get the maximum number of threads that can be used by the runtime.
Definition: NvInferRuntime.h:1997
char const * getTemporaryDirectory() const noexcept
Get the directory that will be used by this runtime for temporary files.
Definition: NvInferRuntime.h:2043
TRT_DEPRECATED ICudaEngine * deserializeCudaEngine(IStreamReader &streamReader)
Deserialize an engine from a stream.
Definition: NvInferRuntime.h:1938
void setErrorRecorder(IErrorRecorder *recorder) noexcept
Set the ErrorRecorder for this interface.
Definition: NvInferRuntime.h:1887
Holds properties for configuring an engine to serialize the binary.
Definition: NvInferRuntime.h:2965
virtual ~ISerializationConfig() noexcept=default
bool clearFlag(SerializationFlag serializationFlag) noexcept
clear a serialization flag.
Definition: NvInferRuntime.h:3004
bool setFlag(SerializationFlag serializationFlag) noexcept
Set a serialization flag.
Definition: NvInferRuntime.h:3016
SerializationFlags getFlags() const noexcept
Get the serialization flags for this config.
Definition: NvInferRuntime.h:2992
bool getFlag(SerializationFlag serializationFlag) const noexcept
Returns true if the serialization flag is set.
Definition: NvInferRuntime.h:3028
apiv::VSerializationConfig * mImpl
Definition: NvInferRuntime.h:3034
An Interface class for version control.
Definition: NvInferRuntimeBase.h:274
Version information associated with a TRT interface.
Definition: NvInferRuntimeBase.h:239
Register the plugin creator to the registry The static registry object will be instantiated when the ...
Definition: NvInferRuntime.h:5288
PluginRegistrar()
Definition: NvInferRuntime.h:5290
An array of weights used as a layer parameter.
Definition: NvInferRuntime.h:124
DataType type
The type of the weights.
Definition: NvInferRuntime.h:126
int64_t count
The number of weights in the array.
Definition: NvInferRuntime.h:128
void const * values
The weight values, in a contiguous array.
Definition: NvInferRuntime.h:127
Definition: NvInferRuntime.h:4132
virtual bool processDebugTensor(void const *addr, TensorLocation location, DataType type, Dims const &shape, char const *name, cudaStream_t stream)=0
Callback function that is called when a debug tensor’s value is updated and the debug state of the te...
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:4137
~IDebugListener() override=default
Definition: NvInferRuntimeBase.h:411
Definition: NvInferRuntime.h:1610
virtual void * allocateAsync(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags, cudaStream_t) noexcept
A thread-safe callback implemented by the application to handle stream-ordered acquisition of GPU mem...
Definition: NvInferRuntime.h:1732
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:1773
virtual TRT_DEPRECATED bool deallocate(void *const memory) noexcept=0
A thread-safe callback implemented by the application to handle release of GPU memory.
~IGpuAllocator() override=default
virtual void * reallocate(void *const, uint64_t, uint64_t) noexcept
A thread-safe callback implemented by the application to resize an existing allocation.
Definition: NvInferRuntime.h:1679
virtual TRT_DEPRECATED void * allocate(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept=0
A thread-safe callback implemented by the application to handle acquisition of GPU memory.
virtual bool deallocateAsync(void *const memory, cudaStream_t) noexcept
A thread-safe callback implemented by the application to handle stream-ordered release of GPU memory.
Definition: NvInferRuntime.h:1765
Definition: NvInferRuntime.h:5337
bool deallocateAsync(void *const memory, cudaStream_t) noexcept override=0
A thread-safe callback implemented by the application to handle stream-ordered asynchronous release o...
void * allocateAsync(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags, cudaStream_t) noexcept override=0
A thread-safe callback implemented by the application to handle stream-ordered asynchronous acquisiti...
TRT_DEPRECATED void * allocate(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept override
A thread-safe callback implemented by the application to handle acquisition of GPU memory.
Definition: NvInferRuntime.h:5423
TRT_DEPRECATED bool deallocate(void *const memory) noexcept override
A thread-safe callback implemented by the application to handle release of GPU memory.
Definition: NvInferRuntime.h:5447
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:5455
~IGpuAsyncAllocator() override=default
Definition: NvInferRuntime.h:4045
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:4050
virtual void * reallocateOutputAsync(char const *tensorName, void *currentMemory, uint64_t size, uint64_t alignment, cudaStream_t)
Return a pointer to memory for an output tensor, or nullptr if memory cannot be allocated....
Definition: NvInferRuntime.h:4102
virtual void notifyShape(char const *tensorName, Dims const &dims) noexcept=0
Called by TensorRT when the shape of the output tensor is known.
virtual TRT_DEPRECATED void * reallocateOutput(char const *tensorName, void *currentMemory, uint64_t size, uint64_t alignment) noexcept
Return a pointer to memory for an output tensor, or nullptr if memory cannot be allocated....
Definition: NvInferRuntime.h:4074
Definition: NvInferPluginBase.h:141
Definition: NvInferPluginBase.h:193
Definition: NvInferRuntime.h:5462
virtual PluginFieldCollection const * getFieldNames() noexcept=0
Return a list of fields that need to be passed to createPlugin() when creating a plugin for use in th...
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:5467
virtual IPluginV3 * createPlugin(AsciiChar const *name, PluginFieldCollection const *fc, TensorRTPhase phase) noexcept=0
Return a plugin object. Return nullptr in case of error.
Definition: NvInferPluginBase.h:206
Definition: NvInferRuntime.h:836
virtual int32_t getFormatCombinationLimit() noexcept
Return the maximum number of format combinations that will be timed by TensorRT during the build phas...
Definition: NvInferRuntime.h:1040
virtual int32_t getNbOutputs() const noexcept=0
Get the number of outputs from the plugin.
virtual int32_t configurePlugin(DynamicPluginTensorDesc const *in, int32_t nbInputs, DynamicPluginTensorDesc const *out, int32_t nbOutputs) noexcept=0
Configure the plugin.
virtual int32_t getNbTactics() noexcept
Query for the number of custom tactics the plugin intends to use.
Definition: NvInferRuntime.h:1016
virtual char const * getMetadataString() noexcept
Query for a string representing the configuration of the plugin. May be called anytime after plugin c...
Definition: NvInferRuntime.h:1051
virtual char const * getTimingCacheID() noexcept
Called to query the suffix to use for the timing cache ID. May be called anytime after plugin creatio...
Definition: NvInferRuntime.h:1032
virtual bool supportsFormatCombination(int32_t pos, DynamicPluginTensorDesc const *inOut, int32_t nbInputs, int32_t nbOutputs) noexcept=0
Return true if plugin supports the format and datatype for the input/output indexed by pos.
virtual int32_t getOutputDataTypes(DataType *outputTypes, int32_t nbOutputs, const DataType *inputTypes, int32_t nbInputs) const noexcept=0
Provide the data types of the plugin outputs if the input tensors have the data types provided.
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:848
virtual int32_t getOutputShapes(DimsExprs const *inputs, int32_t nbInputs, DimsExprs const *shapeInputs, int32_t nbShapeInputs, DimsExprs *outputs, int32_t nbOutputs, IExprBuilder &exprBuilder) noexcept=0
Provide expressions for computing dimensions of the output tensors from dimensions of the input tenso...
virtual int32_t getValidTactics(int32_t *tactics, int32_t nbTactics) noexcept
Query for any custom tactics that the plugin intends to use.
Definition: NvInferRuntime.h:1008
Definition: NvInferRuntime.h:793
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:798
virtual AsciiChar const * getPluginName() const noexcept=0
Return the plugin name. Should match the plugin name returned by the corresponding plugin creator.
Definition: NvInferRuntime.h:1058
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:1063
virtual int32_t onShapeChange(PluginTensorDesc const *in, int32_t nbInputs, PluginTensorDesc const *out, int32_t nbOutputs) noexcept=0
Called when a plugin is being prepared for execution for specific dimensions. This could happen multi...
virtual PluginFieldCollection const * getFieldsToSerialize() noexcept=0
Get the plugin fields which should be serialized.
virtual int32_t setTactic(int32_t tactic) noexcept
Set the tactic to be used in the subsequent call to enqueue(). If no custom tactics were advertised,...
Definition: NvInferRuntime.h:1075
virtual int32_t enqueue(PluginTensorDesc const *inputDesc, PluginTensorDesc const *outputDesc, void const *const *inputs, void *const *outputs, void *workspace, cudaStream_t stream) noexcept=0
Execute the layer.
virtual IPluginV3 * attachToContext(IPluginResourceContext *context) noexcept=0
Clone the plugin, attach the cloned plugin object to a execution context and grant the cloned plugin ...
Definition: NvInferRuntime.h:1246
virtual ~IProfiler() noexcept
Definition: NvInferRuntime.h:1257
virtual void reportLayerTime(char const *layerName, float ms) noexcept=0
Layer time reporting callback.
Definition: NvInferRuntime.h:631
~IStreamReader() override=default
IStreamReader & operator=(IStreamReader const &) &=default
IStreamReader & operator=(IStreamReader &&) &=default
virtual int64_t read(void *destination, int64_t nbBytes)=0
Read the next number of bytes in the stream.
IStreamReader(IStreamReader &&)=default
IStreamReader(IStreamReader const &)=default
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:643
Definition: NvInferRuntime.h:695
IStreamReaderV2 & operator=(IStreamReaderV2 const &) &=default
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:707
IStreamReaderV2(IStreamReaderV2 &&)=default
~IStreamReaderV2() override=default
virtual int64_t read(void *destination, int64_t nbBytes, cudaStream_t stream) noexcept=0
Read the next number of bytes in the stream asynchronously.
IStreamReaderV2(IStreamReaderV2 const &)=default
virtual bool seek(int64_t offset, SeekPosition where) noexcept=0
Sets the position of the stream to the given offset.
IStreamReaderV2 & operator=(IStreamReaderV2 &&) &=default
Definition: NvInferRuntime.h:1155
virtual int32_t getAliasedInput(int32_t outputIndex) noexcept
Communicates to TensorRT that the output at the specified output index is aliased to the input at the...
Definition: NvInferRuntime.h:1191
InterfaceInfo getInterfaceInfo() const noexcept override
Return version information associated with this interface. Applications must not override this method...
Definition: NvInferRuntime.h:1157
IRefitter * createInferRefitter(ICudaEngine &engine, ILogger &logger) noexcept
Create an instance of an IRefitter class.
Definition: NvInferRuntime.h:5268
IRuntime * createInferRuntime(ILogger &logger) noexcept
Create an instance of an IRuntime class.
Definition: NvInferRuntime.h:5257
The TensorRT API version 1 namespace.
uint32_t TacticSources
Represents a collection of one or more TacticSource values combine using bitwise-OR operations.
Definition: NvInferRuntime.h:2904
v_1_0::IOutputAllocator IOutputAllocator
Definition: NvInferRuntime.h:4127
EngineCapability
List of supported engine capability flows.
Definition: NvInferRuntime.h:76
DimensionOperation
An operation on two IDimensionExpr, which represent integer expressions used in dimension computation...
Definition: NvInferRuntime.h:179
@ kSUB
Substract the second element from the first.
@ kSUM
Sum of the two operands.
@ kPROD
Product of the two operands.
@ kFLOOR_DIV
Floor division of the first element by the second.
@ kEQUAL
1 if operands are equal, 0 otherwise.
@ kMIN
Minimum of the two operands.
@ kLESS
1 if first operand is less than second operand, 0 otherwise.
@ kMAX
Maximum of the two operands.
@ kCEIL_DIV
Division rounding up.
v_1_0::IPluginV3OneCore IPluginV3OneCore
Definition: NvInferRuntime.h:1208
TensorIOMode
Definition of tensor IO Mode.
Definition: NvInferRuntimeBase.h:654
HardwareCompatibilityLevel
Describes requirements of compatibility with GPU architectures other than that of the GPU on which th...
Definition: NvInfer.h:9254
SerializationFlag
List of valid flags that the engine can enable when serializing the bytes.
Definition: NvInferRuntime.h:2945
@ kEXCLUDE_WEIGHTS
Exclude the weights that can be refitted.
v_1_0::IProfiler IProfiler
Definition: NvInferRuntime.h:1272
SeekPosition
Controls the seek mode of IStreamReaderV2.
Definition: NvInferRuntime.h:681
@ kSET
From the beginning of the file.
@ kCUR
From the current position of the file.
@ kEND
From the tail of the file.
v_1_0::IStreamReaderV2 IStreamReaderV2
Definition: NvInferRuntime.h:751
uint32_t TempfileControlFlags
Represents a collection of one or more TempfileControlFlag values combined using bitwise-OR operation...
Definition: NvInferRuntime.h:1350
v_1_0::IGpuAllocator IGpuAllocator
Definition: NvInferRuntime.h:1809
constexpr int32_t EnumMax< WeightsRole >() noexcept
Maximum number of elements in WeightsRole enum.
Definition: NvInferRuntime.h:1293
char_t AsciiChar
Definition: NvInferRuntimeBase.h:115
TensorRTPhase
Indicates a phase of operation of TensorRT.
Definition: NvInferPluginBase.h:116
@ kV2_DYNAMICEXT
IPluginV2DynamicExt.
constexpr int32_t EnumMax< LayerInformationFormat >() noexcept
Definition: NvInferRuntime.h:5081
DataType
The type of weights and tensors.
Definition: NvInferRuntimeBase.h:143
DeviceType
The device that this layer/network will execute on.
Definition: NvInferRuntime.h:1304
@ kSCALE
Scale layer.
@ kCONSTANT
Constant layer.
v_1_0::IDebugListener IDebugListener
Definition: NvInferRuntime.h:4168
TempfileControlFlag
Flags used to control TensorRT's behavior when creating executable temporary files.
Definition: NvInferRuntime.h:1327
@ kALLOW_IN_MEMORY_FILES
Allow creating and loading files in-memory (or unnamed files).
constexpr int32_t EnumMax< OptProfileSelector >() noexcept
Number of different values of OptProfileSelector enum.
Definition: NvInferRuntime.h:2590
WeightsRole
How a layer uses particular Weights.
Definition: NvInferRuntime.h:1282
@ kSHIFT
shift part of IScaleLayer
@ kANY
Any other weights role.
@ kBIAS
bias for IConvolutionLayer or IDeconvolutionLayer
@ kKERNEL
kernel for IConvolutionLayer or IDeconvolutionLayer
constexpr int32_t EnumMax< ProfilingVerbosity >() noexcept
Maximum number of profile verbosity levels in ProfilingVerbosity enum.
Definition: NvInferRuntime.h:2924
ProfilingVerbosity
List of verbosity levels of layer information exposed in NVTX annotations and in IEngineInspector.
Definition: NvInferRuntime.h:2916
@ kLAYER_NAMES_ONLY
Print only the layer names. This is the default setting.
@ kDETAILED
Print detailed layer information including layer names and layer parameters.
TacticSource
List of tactic sources for TensorRT.
Definition: NvInferRuntime.h:2867
TensorFormat PluginFormat
PluginFormat is reserved for backward compatibility.
Definition: NvInferRuntimePlugin.h:54
v_1_0::IPluginV3OneRuntime IPluginV3OneRuntime
Definition: NvInferRuntime.h:1232
@ kMIN
Minimum of the two elements.
constexpr int32_t EnumMax< TempfileControlFlag >() noexcept
Maximum number of elements in TempfileControlFlag enum.
Definition: NvInferRuntime.h:1339
uint32_t SerializationFlags
Represents one or more SerializationFlag values using binary OR operations, e.g., 1U << Serialization...
Definition: NvInferRuntime.h:2935
@ kLINEAR
Supports linear (1D), bilinear (2D), and trilinear (3D) interpolation.
v_1_0::IPluginV3OneBuild IPluginV3OneBuild
Definition: NvInferRuntime.h:1220
TensorFormat
Format of the input/output tensors.
Definition: NvInferRuntime.h:1383
ExecutionContextAllocationStrategy
Different memory allocation behaviors for IExecutionContext.
Definition: NvInferRuntime.h:3049
@ kSTATIC
Default static allocation with the maximum size across all profiles.
@ kUSER_MANAGED
The user supplies custom allocation to the execution context.
@ kON_PROFILE_CHANGE
Reallocate for a profile when it's selected.
constexpr int32_t EnumMax< TacticSource >() noexcept
Maximum number of tactic sources in TacticSource enum.
Definition: NvInferRuntime.h:2893
LayerInformationFormat
The format in which the IEngineInspector prints the layer information.
Definition: NvInferRuntime.h:5073
@ kJSON
Print layer information in JSON format.
@ kONELINE
Print layer information in one line per layer.
v_1_0::IStreamReader IStreamReader
Definition: NvInferRuntime.h:674
AllocatorFlag
Allowed type of memory allocation.
Definition: NvInferRuntime.h:1508
@ kRESIZABLE
TensorRT may call realloc() on this allocation.
@ kMAX
Maximum over elements.
constexpr int32_t EnumMax< DeviceType >() noexcept
Maximum number of elements in DeviceType enum.
Definition: NvInferRuntime.h:1311
constexpr int32_t EnumMax< DimensionOperation >() noexcept
Maximum number of elements in DimensionOperation enum.
Definition: NvInferRuntime.h:193
constexpr int32_t EnumMax< ExecutionContextAllocationStrategy >() noexcept
Maximum number of memory allocation strategies in ExecutionContextAllocationStrategy enum.
Definition: NvInferRuntime.h:3061
constexpr int32_t EnumMax< SerializationFlag >() noexcept
Maximum number of serialization flags in SerializationFlag enum.
Definition: NvInferRuntime.h:2952
TensorLocation
The location for tensor data storage, device or host.
Definition: NvInferRuntime.h:204
@ kHOST
Data stored on host.
@ kDEVICE
Data stored on device.
OptProfileSelector
When setting or querying optimization profile parameters (such as shape tensor inputs or dynamic dime...
Definition: NvInferRuntime.h:2578
@ kOPT
This is used to set or get the value that is used in the optimization (kernel selection).
uint32_t AllocatorFlags
Definition: NvInferRuntime.h:1524
Summarizes tensors that a plugin might see for an input or output.
Definition: NvInferRuntime.h:362
Dims min
Lower bounds on tensor’s dimensions.
Definition: NvInferRuntime.h:367
Dims max
Upper bounds on tensor’s dimensions.
Definition: NvInferRuntime.h:370
Dims opt
Optimum value of tensor’s dimensions specified for auto-tuning.
Definition: NvInferRuntime.h:373
PluginTensorDesc desc
Information required to interpret a pointer to tensor data, except that desc.dims has -1 in place of ...
Definition: NvInferRuntime.h:364
Plugin field collection struct.
Definition: NvInferPluginBase.h:103
Fields that a plugin might see for an input or output.
Definition: NvInferRuntimePlugin.h:73
Declaration of EnumMaxImpl struct to store maximum number of elements in an enumeration type.
Definition: NvInferRuntimeBase.h:128

  Copyright © 2024 NVIDIA Corporation
  Privacy Policy | Manage My Privacy | Do Not Sell or Share My Data | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact