optimizer.cpp 30 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801
  1. // Copyright (c) 2016 Google Inc.
  2. //
  3. // Licensed under the Apache License, Version 2.0 (the "License");
  4. // you may not use this file except in compliance with the License.
  5. // You may obtain a copy of the License at
  6. //
  7. // http://www.apache.org/licenses/LICENSE-2.0
  8. //
  9. // Unless required by applicable law or agreed to in writing, software
  10. // distributed under the License is distributed on an "AS IS" BASIS,
  11. // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12. // See the License for the specific language governing permissions and
  13. // limitations under the License.
  14. #include "spirv-tools/optimizer.hpp"
  15. #include <memory>
  16. #include <string>
  17. #include <unordered_map>
  18. #include <utility>
  19. #include <vector>
  20. #include <source/spirv_optimizer_options.h>
  21. #include "code_sink.h"
  22. #include "source/opt/build_module.h"
  23. #include "source/opt/log.h"
  24. #include "source/opt/pass_manager.h"
  25. #include "source/opt/passes.h"
  26. #include "source/util/make_unique.h"
  27. #include "source/util/string_utils.h"
  28. namespace spvtools {
  29. struct Optimizer::PassToken::Impl {
  30. Impl(std::unique_ptr<opt::Pass> p) : pass(std::move(p)) {}
  31. std::unique_ptr<opt::Pass> pass; // Internal implementation pass.
  32. };
  33. Optimizer::PassToken::PassToken(
  34. std::unique_ptr<Optimizer::PassToken::Impl> impl)
  35. : impl_(std::move(impl)) {}
  36. Optimizer::PassToken::PassToken(std::unique_ptr<opt::Pass>&& pass)
  37. : impl_(MakeUnique<Optimizer::PassToken::Impl>(std::move(pass))) {}
  38. Optimizer::PassToken::PassToken(PassToken&& that)
  39. : impl_(std::move(that.impl_)) {}
  40. Optimizer::PassToken& Optimizer::PassToken::operator=(PassToken&& that) {
  41. impl_ = std::move(that.impl_);
  42. return *this;
  43. }
  44. Optimizer::PassToken::~PassToken() {}
  45. struct Optimizer::Impl {
  46. explicit Impl(spv_target_env env) : target_env(env), pass_manager() {}
  47. spv_target_env target_env; // Target environment.
  48. opt::PassManager pass_manager; // Internal implementation pass manager.
  49. };
  50. Optimizer::Optimizer(spv_target_env env) : impl_(new Impl(env)) {}
  51. Optimizer::~Optimizer() {}
  52. void Optimizer::SetMessageConsumer(MessageConsumer c) {
  53. // All passes' message consumer needs to be updated.
  54. for (uint32_t i = 0; i < impl_->pass_manager.NumPasses(); ++i) {
  55. impl_->pass_manager.GetPass(i)->SetMessageConsumer(c);
  56. }
  57. impl_->pass_manager.SetMessageConsumer(std::move(c));
  58. }
  59. const MessageConsumer& Optimizer::consumer() const {
  60. return impl_->pass_manager.consumer();
  61. }
  62. Optimizer& Optimizer::RegisterPass(PassToken&& p) {
  63. // Change to use the pass manager's consumer.
  64. p.impl_->pass->SetMessageConsumer(consumer());
  65. impl_->pass_manager.AddPass(std::move(p.impl_->pass));
  66. return *this;
  67. }
  68. // The legalization passes take a spir-v shader generated by an HLSL front-end
  69. // and turn it into a valid vulkan spir-v shader. There are two ways in which
  70. // the code will be invalid at the start:
  71. //
  72. // 1) There will be opaque objects, like images, which will be passed around
  73. // in intermediate objects. Valid spir-v will have to replace the use of
  74. // the opaque object with an intermediate object that is the result of the
  75. // load of the global opaque object.
  76. //
  77. // 2) There will be variables that contain pointers to structured or uniform
  78. // buffers. It be legal, the variables must be eliminated, and the
  79. // references to the structured buffers must use the result of OpVariable
  80. // in the Uniform storage class.
  81. //
  82. // Optimization in this list must accept shaders with these relaxation of the
  83. // rules. There is not guarantee that this list of optimizations is able to
  84. // legalize all inputs, but it is on a best effort basis.
  85. //
  86. // The legalization problem is essentially a very general copy propagation
  87. // problem. The optimization we use are all used to either do copy propagation
  88. // or enable more copy propagation.
  89. Optimizer& Optimizer::RegisterLegalizationPasses() {
  90. return
  91. // Remove unreachable block so that merge return works.
  92. RegisterPass(CreateDeadBranchElimPass())
  93. // Merge the returns so we can inline.
  94. .RegisterPass(CreateMergeReturnPass())
  95. // Make sure uses and definitions are in the same function.
  96. .RegisterPass(CreateInlineExhaustivePass())
  97. // Make private variable function scope
  98. .RegisterPass(CreateEliminateDeadFunctionsPass())
  99. .RegisterPass(CreatePrivateToLocalPass())
  100. // Propagate the value stored to the loads in very simple cases.
  101. .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
  102. .RegisterPass(CreateLocalSingleStoreElimPass())
  103. .RegisterPass(CreateAggressiveDCEPass())
  104. // Split up aggregates so they are easier to deal with.
  105. .RegisterPass(CreateScalarReplacementPass(0))
  106. // Remove loads and stores so everything is in intermediate values.
  107. // Takes care of copy propagation of non-members.
  108. .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
  109. .RegisterPass(CreateLocalSingleStoreElimPass())
  110. .RegisterPass(CreateAggressiveDCEPass())
  111. .RegisterPass(CreateLocalMultiStoreElimPass())
  112. .RegisterPass(CreateAggressiveDCEPass())
  113. // Propagate constants to get as many constant conditions on branches
  114. // as possible.
  115. .RegisterPass(CreateCCPPass())
  116. .RegisterPass(CreateLoopUnrollPass(true))
  117. .RegisterPass(CreateDeadBranchElimPass())
  118. // Copy propagate members. Cleans up code sequences generated by
  119. // scalar replacement. Also important for removing OpPhi nodes.
  120. .RegisterPass(CreateSimplificationPass())
  121. .RegisterPass(CreateAggressiveDCEPass())
  122. .RegisterPass(CreateCopyPropagateArraysPass())
  123. // May need loop unrolling here see
  124. // https://github.com/Microsoft/DirectXShaderCompiler/pull/930
  125. // Get rid of unused code that contain traces of illegal code
  126. // or unused references to unbound external objects
  127. .RegisterPass(CreateVectorDCEPass())
  128. .RegisterPass(CreateDeadInsertElimPass())
  129. .RegisterPass(CreateReduceLoadSizePass())
  130. .RegisterPass(CreateAggressiveDCEPass());
  131. }
  132. Optimizer& Optimizer::RegisterPerformancePasses() {
  133. return RegisterPass(CreateDeadBranchElimPass())
  134. .RegisterPass(CreateMergeReturnPass())
  135. .RegisterPass(CreateInlineExhaustivePass())
  136. .RegisterPass(CreateAggressiveDCEPass())
  137. .RegisterPass(CreatePrivateToLocalPass())
  138. .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
  139. .RegisterPass(CreateLocalSingleStoreElimPass())
  140. .RegisterPass(CreateAggressiveDCEPass())
  141. .RegisterPass(CreateScalarReplacementPass())
  142. .RegisterPass(CreateLocalAccessChainConvertPass())
  143. .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
  144. .RegisterPass(CreateLocalSingleStoreElimPass())
  145. .RegisterPass(CreateAggressiveDCEPass())
  146. .RegisterPass(CreateLocalMultiStoreElimPass())
  147. .RegisterPass(CreateAggressiveDCEPass())
  148. .RegisterPass(CreateCCPPass())
  149. .RegisterPass(CreateAggressiveDCEPass())
  150. .RegisterPass(CreateRedundancyEliminationPass())
  151. .RegisterPass(CreateCombineAccessChainsPass())
  152. .RegisterPass(CreateSimplificationPass())
  153. .RegisterPass(CreateVectorDCEPass())
  154. .RegisterPass(CreateDeadInsertElimPass())
  155. .RegisterPass(CreateDeadBranchElimPass())
  156. .RegisterPass(CreateSimplificationPass())
  157. .RegisterPass(CreateIfConversionPass())
  158. .RegisterPass(CreateCopyPropagateArraysPass())
  159. .RegisterPass(CreateReduceLoadSizePass())
  160. .RegisterPass(CreateAggressiveDCEPass())
  161. .RegisterPass(CreateBlockMergePass())
  162. .RegisterPass(CreateRedundancyEliminationPass())
  163. .RegisterPass(CreateDeadBranchElimPass())
  164. .RegisterPass(CreateBlockMergePass())
  165. .RegisterPass(CreateSimplificationPass());
  166. // Currently exposing driver bugs resulting in crashes (#946)
  167. // .RegisterPass(CreateCommonUniformElimPass())
  168. }
  169. Optimizer& Optimizer::RegisterSizePasses() {
  170. return RegisterPass(CreateDeadBranchElimPass())
  171. .RegisterPass(CreateMergeReturnPass())
  172. .RegisterPass(CreateInlineExhaustivePass())
  173. .RegisterPass(CreateAggressiveDCEPass())
  174. .RegisterPass(CreatePrivateToLocalPass())
  175. .RegisterPass(CreateScalarReplacementPass())
  176. .RegisterPass(CreateLocalAccessChainConvertPass())
  177. .RegisterPass(CreateLocalSingleBlockLoadStoreElimPass())
  178. .RegisterPass(CreateLocalSingleStoreElimPass())
  179. .RegisterPass(CreateAggressiveDCEPass())
  180. .RegisterPass(CreateSimplificationPass())
  181. .RegisterPass(CreateDeadInsertElimPass())
  182. .RegisterPass(CreateLocalMultiStoreElimPass())
  183. .RegisterPass(CreateAggressiveDCEPass())
  184. .RegisterPass(CreateCCPPass())
  185. .RegisterPass(CreateAggressiveDCEPass())
  186. .RegisterPass(CreateDeadBranchElimPass())
  187. .RegisterPass(CreateIfConversionPass())
  188. .RegisterPass(CreateAggressiveDCEPass())
  189. .RegisterPass(CreateBlockMergePass())
  190. .RegisterPass(CreateSimplificationPass())
  191. .RegisterPass(CreateDeadInsertElimPass())
  192. .RegisterPass(CreateRedundancyEliminationPass())
  193. .RegisterPass(CreateCFGCleanupPass())
  194. // Currently exposing driver bugs resulting in crashes (#946)
  195. // .RegisterPass(CreateCommonUniformElimPass())
  196. .RegisterPass(CreateAggressiveDCEPass());
  197. }
  198. Optimizer& Optimizer::RegisterWebGPUPasses() {
  199. return RegisterPass(CreateAggressiveDCEPass())
  200. .RegisterPass(CreateDeadBranchElimPass());
  201. }
  202. bool Optimizer::RegisterPassesFromFlags(const std::vector<std::string>& flags) {
  203. for (const auto& flag : flags) {
  204. if (!RegisterPassFromFlag(flag)) {
  205. return false;
  206. }
  207. }
  208. return true;
  209. }
  210. bool Optimizer::FlagHasValidForm(const std::string& flag) const {
  211. if (flag == "-O" || flag == "-Os") {
  212. return true;
  213. } else if (flag.size() > 2 && flag.substr(0, 2) == "--") {
  214. return true;
  215. }
  216. Errorf(consumer(), nullptr, {},
  217. "%s is not a valid flag. Flag passes should have the form "
  218. "'--pass_name[=pass_args]'. Special flag names also accepted: -O "
  219. "and -Os.",
  220. flag.c_str());
  221. return false;
  222. }
  223. bool Optimizer::RegisterPassFromFlag(const std::string& flag) {
  224. if (!FlagHasValidForm(flag)) {
  225. return false;
  226. }
  227. // Split flags of the form --pass_name=pass_args.
  228. auto p = utils::SplitFlagArgs(flag);
  229. std::string pass_name = p.first;
  230. std::string pass_args = p.second;
  231. // FIXME(dnovillo): This should be re-factored so that pass names can be
  232. // automatically checked against Pass::name() and PassToken instances created
  233. // via a template function. Additionally, class Pass should have a desc()
  234. // method that describes the pass (so it can be used in --help).
  235. //
  236. // Both Pass::name() and Pass::desc() should be static class members so they
  237. // can be invoked without creating a pass instance.
  238. if (pass_name == "strip-debug") {
  239. RegisterPass(CreateStripDebugInfoPass());
  240. } else if (pass_name == "strip-reflect") {
  241. RegisterPass(CreateStripReflectInfoPass());
  242. } else if (pass_name == "set-spec-const-default-value") {
  243. if (pass_args.size() > 0) {
  244. auto spec_ids_vals =
  245. opt::SetSpecConstantDefaultValuePass::ParseDefaultValuesString(
  246. pass_args.c_str());
  247. if (!spec_ids_vals) {
  248. Errorf(consumer(), nullptr, {},
  249. "Invalid argument for --set-spec-const-default-value: %s",
  250. pass_args.c_str());
  251. return false;
  252. }
  253. RegisterPass(
  254. CreateSetSpecConstantDefaultValuePass(std::move(*spec_ids_vals)));
  255. } else {
  256. Errorf(consumer(), nullptr, {},
  257. "Invalid spec constant value string '%s'. Expected a string of "
  258. "<spec id>:<default value> pairs.",
  259. pass_args.c_str());
  260. return false;
  261. }
  262. } else if (pass_name == "if-conversion") {
  263. RegisterPass(CreateIfConversionPass());
  264. } else if (pass_name == "freeze-spec-const") {
  265. RegisterPass(CreateFreezeSpecConstantValuePass());
  266. } else if (pass_name == "inline-entry-points-exhaustive") {
  267. RegisterPass(CreateInlineExhaustivePass());
  268. } else if (pass_name == "inline-entry-points-opaque") {
  269. RegisterPass(CreateInlineOpaquePass());
  270. } else if (pass_name == "combine-access-chains") {
  271. RegisterPass(CreateCombineAccessChainsPass());
  272. } else if (pass_name == "convert-local-access-chains") {
  273. RegisterPass(CreateLocalAccessChainConvertPass());
  274. } else if (pass_name == "eliminate-dead-code-aggressive") {
  275. RegisterPass(CreateAggressiveDCEPass());
  276. } else if (pass_name == "propagate-line-info") {
  277. RegisterPass(CreatePropagateLineInfoPass());
  278. } else if (pass_name == "eliminate-redundant-line-info") {
  279. RegisterPass(CreateRedundantLineInfoElimPass());
  280. } else if (pass_name == "eliminate-insert-extract") {
  281. RegisterPass(CreateInsertExtractElimPass());
  282. } else if (pass_name == "eliminate-local-single-block") {
  283. RegisterPass(CreateLocalSingleBlockLoadStoreElimPass());
  284. } else if (pass_name == "eliminate-local-single-store") {
  285. RegisterPass(CreateLocalSingleStoreElimPass());
  286. } else if (pass_name == "merge-blocks") {
  287. RegisterPass(CreateBlockMergePass());
  288. } else if (pass_name == "merge-return") {
  289. RegisterPass(CreateMergeReturnPass());
  290. } else if (pass_name == "eliminate-dead-branches") {
  291. RegisterPass(CreateDeadBranchElimPass());
  292. } else if (pass_name == "eliminate-dead-functions") {
  293. RegisterPass(CreateEliminateDeadFunctionsPass());
  294. } else if (pass_name == "eliminate-local-multi-store") {
  295. RegisterPass(CreateLocalMultiStoreElimPass());
  296. } else if (pass_name == "eliminate-common-uniform") {
  297. RegisterPass(CreateCommonUniformElimPass());
  298. } else if (pass_name == "eliminate-dead-const") {
  299. RegisterPass(CreateEliminateDeadConstantPass());
  300. } else if (pass_name == "eliminate-dead-inserts") {
  301. RegisterPass(CreateDeadInsertElimPass());
  302. } else if (pass_name == "eliminate-dead-variables") {
  303. RegisterPass(CreateDeadVariableEliminationPass());
  304. } else if (pass_name == "fold-spec-const-op-composite") {
  305. RegisterPass(CreateFoldSpecConstantOpAndCompositePass());
  306. } else if (pass_name == "loop-unswitch") {
  307. RegisterPass(CreateLoopUnswitchPass());
  308. } else if (pass_name == "scalar-replacement") {
  309. if (pass_args.size() == 0) {
  310. RegisterPass(CreateScalarReplacementPass());
  311. } else {
  312. int limit = -1;
  313. if (pass_args.find_first_not_of("0123456789") == std::string::npos) {
  314. limit = atoi(pass_args.c_str());
  315. }
  316. if (limit >= 0) {
  317. RegisterPass(CreateScalarReplacementPass(limit));
  318. } else {
  319. Error(consumer(), nullptr, {},
  320. "--scalar-replacement must have no arguments or a non-negative "
  321. "integer argument");
  322. return false;
  323. }
  324. }
  325. } else if (pass_name == "strength-reduction") {
  326. RegisterPass(CreateStrengthReductionPass());
  327. } else if (pass_name == "unify-const") {
  328. RegisterPass(CreateUnifyConstantPass());
  329. } else if (pass_name == "flatten-decorations") {
  330. RegisterPass(CreateFlattenDecorationPass());
  331. } else if (pass_name == "compact-ids") {
  332. RegisterPass(CreateCompactIdsPass());
  333. } else if (pass_name == "cfg-cleanup") {
  334. RegisterPass(CreateCFGCleanupPass());
  335. } else if (pass_name == "local-redundancy-elimination") {
  336. RegisterPass(CreateLocalRedundancyEliminationPass());
  337. } else if (pass_name == "loop-invariant-code-motion") {
  338. RegisterPass(CreateLoopInvariantCodeMotionPass());
  339. } else if (pass_name == "reduce-load-size") {
  340. RegisterPass(CreateReduceLoadSizePass());
  341. } else if (pass_name == "redundancy-elimination") {
  342. RegisterPass(CreateRedundancyEliminationPass());
  343. } else if (pass_name == "private-to-local") {
  344. RegisterPass(CreatePrivateToLocalPass());
  345. } else if (pass_name == "remove-duplicates") {
  346. RegisterPass(CreateRemoveDuplicatesPass());
  347. } else if (pass_name == "workaround-1209") {
  348. RegisterPass(CreateWorkaround1209Pass());
  349. } else if (pass_name == "replace-invalid-opcode") {
  350. RegisterPass(CreateReplaceInvalidOpcodePass());
  351. } else if (pass_name == "inst-bindless-check") {
  352. RegisterPass(CreateInstBindlessCheckPass(7, 23));
  353. RegisterPass(CreateSimplificationPass());
  354. RegisterPass(CreateDeadBranchElimPass());
  355. RegisterPass(CreateBlockMergePass());
  356. RegisterPass(CreateAggressiveDCEPass());
  357. } else if (pass_name == "simplify-instructions") {
  358. RegisterPass(CreateSimplificationPass());
  359. } else if (pass_name == "ssa-rewrite") {
  360. RegisterPass(CreateSSARewritePass());
  361. } else if (pass_name == "copy-propagate-arrays") {
  362. RegisterPass(CreateCopyPropagateArraysPass());
  363. } else if (pass_name == "loop-fission") {
  364. int register_threshold_to_split =
  365. (pass_args.size() > 0) ? atoi(pass_args.c_str()) : -1;
  366. if (register_threshold_to_split > 0) {
  367. RegisterPass(CreateLoopFissionPass(
  368. static_cast<size_t>(register_threshold_to_split)));
  369. } else {
  370. Error(consumer(), nullptr, {},
  371. "--loop-fission must have a positive integer argument");
  372. return false;
  373. }
  374. } else if (pass_name == "loop-fusion") {
  375. int max_registers_per_loop =
  376. (pass_args.size() > 0) ? atoi(pass_args.c_str()) : -1;
  377. if (max_registers_per_loop > 0) {
  378. RegisterPass(
  379. CreateLoopFusionPass(static_cast<size_t>(max_registers_per_loop)));
  380. } else {
  381. Error(consumer(), nullptr, {},
  382. "--loop-fusion must have a positive integer argument");
  383. return false;
  384. }
  385. } else if (pass_name == "loop-unroll") {
  386. RegisterPass(CreateLoopUnrollPass(true));
  387. } else if (pass_name == "upgrade-memory-model") {
  388. RegisterPass(CreateUpgradeMemoryModelPass());
  389. } else if (pass_name == "vector-dce") {
  390. RegisterPass(CreateVectorDCEPass());
  391. } else if (pass_name == "loop-unroll-partial") {
  392. int factor = (pass_args.size() > 0) ? atoi(pass_args.c_str()) : 0;
  393. if (factor > 0) {
  394. RegisterPass(CreateLoopUnrollPass(false, factor));
  395. } else {
  396. Error(consumer(), nullptr, {},
  397. "--loop-unroll-partial must have a positive integer argument");
  398. return false;
  399. }
  400. } else if (pass_name == "loop-peeling") {
  401. RegisterPass(CreateLoopPeelingPass());
  402. } else if (pass_name == "loop-peeling-threshold") {
  403. int factor = (pass_args.size() > 0) ? atoi(pass_args.c_str()) : 0;
  404. if (factor > 0) {
  405. opt::LoopPeelingPass::SetLoopPeelingThreshold(factor);
  406. } else {
  407. Error(consumer(), nullptr, {},
  408. "--loop-peeling-threshold must have a positive integer argument");
  409. return false;
  410. }
  411. } else if (pass_name == "ccp") {
  412. RegisterPass(CreateCCPPass());
  413. } else if (pass_name == "code-sink") {
  414. RegisterPass(CreateCodeSinkingPass());
  415. } else if (pass_name == "O") {
  416. RegisterPerformancePasses();
  417. } else if (pass_name == "Os") {
  418. RegisterSizePasses();
  419. } else if (pass_name == "legalize-hlsl") {
  420. RegisterLegalizationPasses();
  421. } else {
  422. Errorf(consumer(), nullptr, {},
  423. "Unknown flag '--%s'. Use --help for a list of valid flags",
  424. pass_name.c_str());
  425. return false;
  426. }
  427. return true;
  428. }
  429. void Optimizer::SetTargetEnv(const spv_target_env env) {
  430. impl_->target_env = env;
  431. }
  432. bool Optimizer::Run(const uint32_t* original_binary,
  433. const size_t original_binary_size,
  434. std::vector<uint32_t>* optimized_binary) const {
  435. return Run(original_binary, original_binary_size, optimized_binary,
  436. OptimizerOptions());
  437. }
  438. bool Optimizer::Run(const uint32_t* original_binary,
  439. const size_t original_binary_size,
  440. std::vector<uint32_t>* optimized_binary,
  441. const ValidatorOptions& validator_options,
  442. bool skip_validation) const {
  443. OptimizerOptions opt_options;
  444. opt_options.set_run_validator(!skip_validation);
  445. opt_options.set_validator_options(validator_options);
  446. return Run(original_binary, original_binary_size, optimized_binary,
  447. opt_options);
  448. }
  449. bool Optimizer::Run(const uint32_t* original_binary,
  450. const size_t original_binary_size,
  451. std::vector<uint32_t>* optimized_binary,
  452. const spv_optimizer_options opt_options) const {
  453. spvtools::SpirvTools tools(impl_->target_env);
  454. tools.SetMessageConsumer(impl_->pass_manager.consumer());
  455. if (opt_options->run_validator_ &&
  456. !tools.Validate(original_binary, original_binary_size,
  457. &opt_options->val_options_)) {
  458. return false;
  459. }
  460. std::unique_ptr<opt::IRContext> context = BuildModule(
  461. impl_->target_env, consumer(), original_binary, original_binary_size);
  462. if (context == nullptr) return false;
  463. context->set_max_id_bound(opt_options->max_id_bound_);
  464. auto status = impl_->pass_manager.Run(context.get());
  465. if (status == opt::Pass::Status::SuccessWithChange ||
  466. (status == opt::Pass::Status::SuccessWithoutChange &&
  467. (optimized_binary->data() != original_binary ||
  468. optimized_binary->size() != original_binary_size))) {
  469. optimized_binary->clear();
  470. context->module()->ToBinary(optimized_binary, /* skip_nop = */ true);
  471. }
  472. return status != opt::Pass::Status::Failure;
  473. }
  474. Optimizer& Optimizer::SetPrintAll(std::ostream* out) {
  475. impl_->pass_manager.SetPrintAll(out);
  476. return *this;
  477. }
  478. Optimizer& Optimizer::SetTimeReport(std::ostream* out) {
  479. impl_->pass_manager.SetTimeReport(out);
  480. return *this;
  481. }
  482. Optimizer::PassToken CreateNullPass() {
  483. return MakeUnique<Optimizer::PassToken::Impl>(MakeUnique<opt::NullPass>());
  484. }
  485. Optimizer::PassToken CreateStripDebugInfoPass() {
  486. return MakeUnique<Optimizer::PassToken::Impl>(
  487. MakeUnique<opt::StripDebugInfoPass>());
  488. }
  489. Optimizer::PassToken CreateStripReflectInfoPass() {
  490. return MakeUnique<Optimizer::PassToken::Impl>(
  491. MakeUnique<opt::StripReflectInfoPass>());
  492. }
  493. Optimizer::PassToken CreateEliminateDeadFunctionsPass() {
  494. return MakeUnique<Optimizer::PassToken::Impl>(
  495. MakeUnique<opt::EliminateDeadFunctionsPass>());
  496. }
  497. Optimizer::PassToken CreateSetSpecConstantDefaultValuePass(
  498. const std::unordered_map<uint32_t, std::string>& id_value_map) {
  499. return MakeUnique<Optimizer::PassToken::Impl>(
  500. MakeUnique<opt::SetSpecConstantDefaultValuePass>(id_value_map));
  501. }
  502. Optimizer::PassToken CreateSetSpecConstantDefaultValuePass(
  503. const std::unordered_map<uint32_t, std::vector<uint32_t>>& id_value_map) {
  504. return MakeUnique<Optimizer::PassToken::Impl>(
  505. MakeUnique<opt::SetSpecConstantDefaultValuePass>(id_value_map));
  506. }
  507. Optimizer::PassToken CreateFlattenDecorationPass() {
  508. return MakeUnique<Optimizer::PassToken::Impl>(
  509. MakeUnique<opt::FlattenDecorationPass>());
  510. }
  511. Optimizer::PassToken CreateFreezeSpecConstantValuePass() {
  512. return MakeUnique<Optimizer::PassToken::Impl>(
  513. MakeUnique<opt::FreezeSpecConstantValuePass>());
  514. }
  515. Optimizer::PassToken CreateFoldSpecConstantOpAndCompositePass() {
  516. return MakeUnique<Optimizer::PassToken::Impl>(
  517. MakeUnique<opt::FoldSpecConstantOpAndCompositePass>());
  518. }
  519. Optimizer::PassToken CreateUnifyConstantPass() {
  520. return MakeUnique<Optimizer::PassToken::Impl>(
  521. MakeUnique<opt::UnifyConstantPass>());
  522. }
  523. Optimizer::PassToken CreateEliminateDeadConstantPass() {
  524. return MakeUnique<Optimizer::PassToken::Impl>(
  525. MakeUnique<opt::EliminateDeadConstantPass>());
  526. }
  527. Optimizer::PassToken CreateDeadVariableEliminationPass() {
  528. return MakeUnique<Optimizer::PassToken::Impl>(
  529. MakeUnique<opt::DeadVariableElimination>());
  530. }
  531. Optimizer::PassToken CreateStrengthReductionPass() {
  532. return MakeUnique<Optimizer::PassToken::Impl>(
  533. MakeUnique<opt::StrengthReductionPass>());
  534. }
  535. Optimizer::PassToken CreateBlockMergePass() {
  536. return MakeUnique<Optimizer::PassToken::Impl>(
  537. MakeUnique<opt::BlockMergePass>());
  538. }
  539. Optimizer::PassToken CreateInlineExhaustivePass() {
  540. return MakeUnique<Optimizer::PassToken::Impl>(
  541. MakeUnique<opt::InlineExhaustivePass>());
  542. }
  543. Optimizer::PassToken CreateInlineOpaquePass() {
  544. return MakeUnique<Optimizer::PassToken::Impl>(
  545. MakeUnique<opt::InlineOpaquePass>());
  546. }
  547. Optimizer::PassToken CreateLocalAccessChainConvertPass() {
  548. return MakeUnique<Optimizer::PassToken::Impl>(
  549. MakeUnique<opt::LocalAccessChainConvertPass>());
  550. }
  551. Optimizer::PassToken CreateLocalSingleBlockLoadStoreElimPass() {
  552. return MakeUnique<Optimizer::PassToken::Impl>(
  553. MakeUnique<opt::LocalSingleBlockLoadStoreElimPass>());
  554. }
  555. Optimizer::PassToken CreateLocalSingleStoreElimPass() {
  556. return MakeUnique<Optimizer::PassToken::Impl>(
  557. MakeUnique<opt::LocalSingleStoreElimPass>());
  558. }
  559. Optimizer::PassToken CreateInsertExtractElimPass() {
  560. return MakeUnique<Optimizer::PassToken::Impl>(
  561. MakeUnique<opt::SimplificationPass>());
  562. }
  563. Optimizer::PassToken CreateDeadInsertElimPass() {
  564. return MakeUnique<Optimizer::PassToken::Impl>(
  565. MakeUnique<opt::DeadInsertElimPass>());
  566. }
  567. Optimizer::PassToken CreateDeadBranchElimPass() {
  568. return MakeUnique<Optimizer::PassToken::Impl>(
  569. MakeUnique<opt::DeadBranchElimPass>());
  570. }
  571. Optimizer::PassToken CreateLocalMultiStoreElimPass() {
  572. return MakeUnique<Optimizer::PassToken::Impl>(
  573. MakeUnique<opt::LocalMultiStoreElimPass>());
  574. }
  575. Optimizer::PassToken CreateAggressiveDCEPass() {
  576. return MakeUnique<Optimizer::PassToken::Impl>(
  577. MakeUnique<opt::AggressiveDCEPass>());
  578. }
  579. Optimizer::PassToken CreatePropagateLineInfoPass() {
  580. return MakeUnique<Optimizer::PassToken::Impl>(
  581. MakeUnique<opt::ProcessLinesPass>(opt::kLinesPropagateLines));
  582. }
  583. Optimizer::PassToken CreateRedundantLineInfoElimPass() {
  584. return MakeUnique<Optimizer::PassToken::Impl>(
  585. MakeUnique<opt::ProcessLinesPass>(opt::kLinesEliminateDeadLines));
  586. }
  587. Optimizer::PassToken CreateCommonUniformElimPass() {
  588. return MakeUnique<Optimizer::PassToken::Impl>(
  589. MakeUnique<opt::CommonUniformElimPass>());
  590. }
  591. Optimizer::PassToken CreateCompactIdsPass() {
  592. return MakeUnique<Optimizer::PassToken::Impl>(
  593. MakeUnique<opt::CompactIdsPass>());
  594. }
  595. Optimizer::PassToken CreateMergeReturnPass() {
  596. return MakeUnique<Optimizer::PassToken::Impl>(
  597. MakeUnique<opt::MergeReturnPass>());
  598. }
  599. std::vector<const char*> Optimizer::GetPassNames() const {
  600. std::vector<const char*> v;
  601. for (uint32_t i = 0; i < impl_->pass_manager.NumPasses(); i++) {
  602. v.push_back(impl_->pass_manager.GetPass(i)->name());
  603. }
  604. return v;
  605. }
  606. Optimizer::PassToken CreateCFGCleanupPass() {
  607. return MakeUnique<Optimizer::PassToken::Impl>(
  608. MakeUnique<opt::CFGCleanupPass>());
  609. }
  610. Optimizer::PassToken CreateLocalRedundancyEliminationPass() {
  611. return MakeUnique<Optimizer::PassToken::Impl>(
  612. MakeUnique<opt::LocalRedundancyEliminationPass>());
  613. }
  614. Optimizer::PassToken CreateLoopFissionPass(size_t threshold) {
  615. return MakeUnique<Optimizer::PassToken::Impl>(
  616. MakeUnique<opt::LoopFissionPass>(threshold));
  617. }
  618. Optimizer::PassToken CreateLoopFusionPass(size_t max_registers_per_loop) {
  619. return MakeUnique<Optimizer::PassToken::Impl>(
  620. MakeUnique<opt::LoopFusionPass>(max_registers_per_loop));
  621. }
  622. Optimizer::PassToken CreateLoopInvariantCodeMotionPass() {
  623. return MakeUnique<Optimizer::PassToken::Impl>(MakeUnique<opt::LICMPass>());
  624. }
  625. Optimizer::PassToken CreateLoopPeelingPass() {
  626. return MakeUnique<Optimizer::PassToken::Impl>(
  627. MakeUnique<opt::LoopPeelingPass>());
  628. }
  629. Optimizer::PassToken CreateLoopUnswitchPass() {
  630. return MakeUnique<Optimizer::PassToken::Impl>(
  631. MakeUnique<opt::LoopUnswitchPass>());
  632. }
  633. Optimizer::PassToken CreateRedundancyEliminationPass() {
  634. return MakeUnique<Optimizer::PassToken::Impl>(
  635. MakeUnique<opt::RedundancyEliminationPass>());
  636. }
  637. Optimizer::PassToken CreateRemoveDuplicatesPass() {
  638. return MakeUnique<Optimizer::PassToken::Impl>(
  639. MakeUnique<opt::RemoveDuplicatesPass>());
  640. }
  641. Optimizer::PassToken CreateScalarReplacementPass(uint32_t size_limit) {
  642. return MakeUnique<Optimizer::PassToken::Impl>(
  643. MakeUnique<opt::ScalarReplacementPass>(size_limit));
  644. }
  645. Optimizer::PassToken CreatePrivateToLocalPass() {
  646. return MakeUnique<Optimizer::PassToken::Impl>(
  647. MakeUnique<opt::PrivateToLocalPass>());
  648. }
  649. Optimizer::PassToken CreateCCPPass() {
  650. return MakeUnique<Optimizer::PassToken::Impl>(MakeUnique<opt::CCPPass>());
  651. }
  652. Optimizer::PassToken CreateWorkaround1209Pass() {
  653. return MakeUnique<Optimizer::PassToken::Impl>(
  654. MakeUnique<opt::Workaround1209>());
  655. }
  656. Optimizer::PassToken CreateIfConversionPass() {
  657. return MakeUnique<Optimizer::PassToken::Impl>(
  658. MakeUnique<opt::IfConversion>());
  659. }
  660. Optimizer::PassToken CreateReplaceInvalidOpcodePass() {
  661. return MakeUnique<Optimizer::PassToken::Impl>(
  662. MakeUnique<opt::ReplaceInvalidOpcodePass>());
  663. }
  664. Optimizer::PassToken CreateSimplificationPass() {
  665. return MakeUnique<Optimizer::PassToken::Impl>(
  666. MakeUnique<opt::SimplificationPass>());
  667. }
  668. Optimizer::PassToken CreateLoopUnrollPass(bool fully_unroll, int factor) {
  669. return MakeUnique<Optimizer::PassToken::Impl>(
  670. MakeUnique<opt::LoopUnroller>(fully_unroll, factor));
  671. }
  672. Optimizer::PassToken CreateSSARewritePass() {
  673. return MakeUnique<Optimizer::PassToken::Impl>(
  674. MakeUnique<opt::SSARewritePass>());
  675. }
  676. Optimizer::PassToken CreateCopyPropagateArraysPass() {
  677. return MakeUnique<Optimizer::PassToken::Impl>(
  678. MakeUnique<opt::CopyPropagateArrays>());
  679. }
  680. Optimizer::PassToken CreateVectorDCEPass() {
  681. return MakeUnique<Optimizer::PassToken::Impl>(MakeUnique<opt::VectorDCE>());
  682. }
  683. Optimizer::PassToken CreateReduceLoadSizePass() {
  684. return MakeUnique<Optimizer::PassToken::Impl>(
  685. MakeUnique<opt::ReduceLoadSize>());
  686. }
  687. Optimizer::PassToken CreateCombineAccessChainsPass() {
  688. return MakeUnique<Optimizer::PassToken::Impl>(
  689. MakeUnique<opt::CombineAccessChains>());
  690. }
  691. Optimizer::PassToken CreateUpgradeMemoryModelPass() {
  692. return MakeUnique<Optimizer::PassToken::Impl>(
  693. MakeUnique<opt::UpgradeMemoryModel>());
  694. }
  695. Optimizer::PassToken CreateInstBindlessCheckPass(uint32_t desc_set,
  696. uint32_t shader_id) {
  697. return MakeUnique<Optimizer::PassToken::Impl>(
  698. MakeUnique<opt::InstBindlessCheckPass>(desc_set, shader_id));
  699. }
  700. Optimizer::PassToken CreateCodeSinkingPass() {
  701. return MakeUnique<Optimizer::PassToken::Impl>(
  702. MakeUnique<opt::CodeSinkingPass>());
  703. }
  704. } // namespace spvtools