kc3-lang/angle/src/compiler/translator/IntermNode.cpp

src/compiler/translator/IntermNode.cpp

Log

Author Commit Date CI Message
Olli Etuaho e80825ef 2018-02-16T10:24:53 Simplify built-in function node initialization Built-ins with no math op associated with them now have the op code EOpCallBuiltInFunction set. This makes initializing built-in function nodes simpler, since they can always get the op code from the function symbol. We also no longer look for functions in inner scopes, only from the global scope and from built-in functions. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I55a2642f34bb3c8b8f13183c95fa509ec3b9cfdb Reviewed-on: https://chromium-review.googlesource.com/923724 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Nico Weber 41b072b6 2018-02-09T10:01:32 Prepare for -Wimplicit-fallthrough in ANGLE. Disable the warning for flex-generated output, which contains lots of intentional fallthrough. Fixes a bug where GL_SAMPLE_ALPHA_TO_ONE_EXT would fall through to GL_COVERAGE_MODULATION_CHROMIUM and hence behave like that. Fixes a bug in the D3D9 state management where invalidating DIRTY_BIT_POLYGON_OFFSET would also invalidate the stencil bits. One somewhat common incorrect pattern in ANGLE is nested switch statements that look like so: switch (a) { case a1: switch (b) { case b1: ... break; } case a2: ... } The assumption here seems to be that the breakk exits the outer case (here a1), while it in fact only exits the inner switch, so that we fall through to a2. In most places, I fixed this by adding an explicit `break` after the inner switch. This fixes a bug wher GL_PATH_JOIN_STYLE_CHROMIUM would fall through to GL_PATH_MITER_LIMIT_CHROMIUM in validation (but since the join style enum is always > 0, this happened to not have an effect in practice). This also fixes 87 bugs in GetLoadFunctionsMap() where invalid values would previously return an unrelated function map instead of the empty load function map. Bug: chromium:810767 Change-Id: Ib51388c73fbfc229160e2c10f8fb9364cc7c996c Reviewed-on: https://chromium-review.googlesource.com/911529 Commit-Queue: Nico Weber <thakis@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Olli Etuaho fbb1c792 2018-01-19T16:26:59 Store symbol names as a ImmutableString This will enable compile-time initialization of built-in symbols as well as reducing copying strings. Most of the code that deals with names is changed to use ImmutableString where it makes sense to avoid conversions. The lexer/parser now allocate const char pointers into pool memory instead of allocating TStrings. These are then converted to ImmutableString upon entering TParseContext. BUG=angleproject:2267 TEST=angle_unittests, angle_end2end_tests Change-Id: I244d6271ea1ecf7150d4f89dfa388a7745a1150c Reviewed-on: https://chromium-review.googlesource.com/881561 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 95ed1947 2018-02-01T14:01:19 Add a class for function lookups Using a purpose-built class for function lookups instead of using a combination of TFunction and a struct container for the this node and arguments makes the code clearer. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I3f345d836abeaa7f84cc46b4b840fd06c7e2e1a7 Reviewed-on: https://chromium-review.googlesource.com/897363 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 68981eb5 2018-01-23T17:46:12 Track parameter qualifiers of functions in call nodes We now add a reference to TFunction to all TIntermAggregate nodes where it is possible, including built-in ops. We also make sure that internal TFunctions added in traversers have correct parameter qualifiers. This makes TLValueTrackingTraverser much simpler. Instead of storing traversed functions or looking up builtin functions from the symbol table, determining which function parameters are out parameters can now be done simply by looking it up from the function symbol associated with the aggregate node. Symbol instances are no longer deleted when a symbol table level goes out of scope, and TFunction destructor no longer clears the parameters. They're all either statically allocated or pool allocated, so this does not result in leaks. TEST=angle_unittests BUG=angleproject:2267 Change-Id: I57e5570da5b5a69a98a8778da3c2dc82b6284738 Reviewed-on: https://chromium-review.googlesource.com/881324 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 0e99b7a3 2018-01-12T12:05:48 Prevent changing AST expression type from outside TIntermNode classes now contain all the logic for setting node types. Changing the constant values of constant union nodes from outside is also not necessary anymore. BUG=angleproject:2267 TEST=angle_unittests Change-Id: Ic10d41b1e5f93152df440a655057591dc1b783b0 Reviewed-on: https://chromium-review.googlesource.com/863626 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 2c9cc8b6 2018-01-09T16:13:02 Don't duplicate symbol type information in AST nodes Function prototype nodes and symbol nodes already refer to symbols that have type information, so the type doesn't need to be copied to the TInterm* AST node classes. Now type is only stored in those AST node classes that represent other types of expressions. They use a new TIntermExpression base class for this. Since now we may use the TType from builtin symbols directly instead of copying it, building the mangled names of types in the correct memory pool is also required. The code now realizes the types of built-in variables when they get added to the symbol table. BUG=angleproject:2267 TEST=angle_unittests Change-Id: Ic8d7fc912937cb8abb1e306e58c63bb9c146aae9 Reviewed-on: https://chromium-review.googlesource.com/857005 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho ea22b7a5 2018-01-04T17:09:11 Constant fold array indexing and comparison A virtual function to get the constant value of an AST node is added to TIntermTyped. This way a constant value can be retrieved conveniently from multiple different types of nodes. TIntermSymbol nodes pointing to a const variable can return the value associated with the variable, constructor nodes can build a constant value from their arguments, and indexing nodes can index into a constant array. This enables constant folding operations on constant arrays, while making sure that large amounts of data are not duplicated in the output shader. When folding an operation makes sense, the values of the arguments can be retrieved by using the new TIntermTyped::getConstantValue(). When folding an operation would result in duplicating data, the AST can just be left to be written out as is. For example, if the code contains a constant array of arrays, indexing into individual elements of the inner arrays can be folded, but indexing the top level array is left in place and not replaced with duplicated array literals. Constant folding is supported for indexing and comparisons of arrays. In case constant arrays are only referenced through foldable operations, the variable declarations will be pruned from the AST by the RemoveUnreferencedVariables step. BUG=angleproject:2298 TEST=angle_unittests Change-Id: I5b3be237b7e9fdba56aa9bf0a41b691f4d8f01eb Reviewed-on: https://chromium-review.googlesource.com/850973 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho ea78d2bb 2018-01-09T12:55:27 Prune unreferenced variables with a constructor initializer Treat aggregate constructors as having no side effects, which enables pruning declarators that have a constructor initializer. Some logic in RemoveUnreferencedVariables is fixed to make this work correctly for structs. The bugs were previously not exposed since constructors were treated as having side effects, but now that those can be pruned the logic needs to be correct. BUG=angleproject:2298 TEST=angle_unittests Change-Id: I6fbe61a9e82065196baa29c200bf556fc21d8962 Reviewed-on: https://chromium-review.googlesource.com/856499 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 765924f0 2018-01-04T12:48:36 Fold ternary and comma ops only after parsing is done In case folding a ternary op or a comma op would change the qualifier of the expression, the folding is deferred to a separate traversal step. After this there are no more cases where the type of a TIntermSymbol node needs to differ from the type of the variable it is referring to. There are still some cases where some parts of TIntermSymbol type are changed while keeping the TVariable type the same though, like when assigning array size to gl_PerVertex nodes or sanitizing qualifiers of struct declarations. BUG=angleproject:2267 TEST=angle_unittests, angle_end2end_tests Change-Id: I1501c8d361f5f765f43ca810d1b7248d9e2c5986 Reviewed-on: https://chromium-review.googlesource.com/850672 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 8b5e8fdb 2017-12-15T14:59:15 Replace remaining usage of TName with TSymbol TName used to contain just a subset of the information stored in TSymbol. It makes more sense to use TSymbol directly instead of converting it to TName. This also improves type safety a bit by making some functions only take in TVariable or TFunction instead of the more generic TName. BUG=angleproject:2267 TEST=angle_unittests Change-Id: Icb46923c25d33ebbbbc06ddc487da25957dda771 Reviewed-on: https://chromium-review.googlesource.com/829143 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 2768bc8a 2017-12-12T11:51:48 Clean up creating constant folded nodes It's not necessary to set the qualifier on folded nodes separately, they always use the qualifier of the node being folded. BUG=angleproject:1490 TEST=angle_unittests Change-Id: Id2581ef4cae42d7137fbe0caf18c3fcacbf954c6 Reviewed-on: https://chromium-review.googlesource.com/847553 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho bed35d76 2017-12-20T16:36:26 Don't query names of empty symbols This makes it possible to return a reference from TSymbol::name() instead of a pointer. This is safer since it completely avoids the possibility of a nullptr dereference. An assert is making sure that the function is not being called for empty symbols. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I44279f65989dbb828322843fc0216ba84d91dedf Reviewed-on: https://chromium-review.googlesource.com/836894 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho b6af22b5 2017-12-15T14:05:44 Store TVariable* in TIntermSymbol instead of storing id This is an intermediate step to only storing a TVariable * in TIntermSymbol instead of copying the name. This makes it possible to get a constant value out of a TIntermSymbol without doing a symbol table lookup. BUG=angleproject:2267 TEST=angle_unittests Change-Id: Ibff588241a4ad4ac330063296273288b20a072c9 Reviewed-on: https://chromium-review.googlesource.com/829142 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho beb6dc74 2017-12-14T16:03:03 Always use TFunction instead of TFunctionSymbolInfo This reduces unnecessary memory allocations and conversions between different objects containing the same data. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I87316509ab1cd6d36756ff6af7fa2b5c5a76a8ea Reviewed-on: https://chromium-review.googlesource.com/827134 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 1bb8528c 2017-12-14T13:39:53 Remove TFunctionSymbolInfo from TIntermAggregate All the information stored in TFunctionSymbolInfo was duplicated from the TFunction that the aggregate node pointed to. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I1f5574ab0416e5cae00c3dae6fc11d2fe1fa128c Reviewed-on: https://chromium-review.googlesource.com/827065 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 0c37100d 2017-12-13T17:00:25 Always create TFunctions for function call nodes This simplifies code and ensures that nodes get consistent data. In the future function call nodes could have a pointer to the TFunction instead of converting the same information into a different data structure. BUG=angleproject:2267 TEST=angle_unittests, angle_end2end_tests Change-Id: Ic0c24bb86b44b9bcc4a5da7f6b03701081a3af5c Reviewed-on: https://chromium-review.googlesource.com/824606 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho ae4dbf32 2017-12-08T20:49:00 Don't allocate name strings for empty symbols This removes unnecessary memory allocations. BUG=angleproject:2267 TEST=angle_unittests Change-Id: Ide575ea19ab2f8e9fc93092490f1352efa6024a3 Reviewed-on: https://chromium-review.googlesource.com/817415 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 195be942 2017-12-04T23:40:14 Always create TVariables for TIntermSymbol nodes TIntermSymbol nodes are now constructed based on a specific TVariable. This makes sure that all TIntermSymbol nodes that are created to refer to a specific temporary in an AST transform will have consistent data. The TVariable objects are not necessarily added to the symbol table levels - just those variables that can be referred to by their name during parsing need to be reachable through there. In the future this can be taken a step further so that TIntermSymbol nodes just to point to a TVariable instead of duplicating the information. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I4e7bcdb0637cd3b588d3c202ef02f4b7bd7954a1 Reviewed-on: https://chromium-review.googlesource.com/811925 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 54a29ffd 2017-11-28T17:35:20 Refactor TSymbol to follow naming conventions This will make it easier to make TStructure inherit from TSymbol. BUG=angleproject:2267 TEST=angle_unittests Change-Id: I73a5a25234c7f965b6970891821ee9d6f0a23aca Reviewed-on: https://chromium-review.googlesource.com/793812 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho ebee5b3b 2017-11-23T12:56:32 Add GLSL support for runtime-sized arrays in SSBOs The GLSL parser now allows a runtime-sized array as the last member in a shader storage block. Clamping indexing against the memory bounds is done by determining the array length at runtime. Runtime-sized arrays are used in dEQP tests for many compute shader tests, so these now work on the OpenGL backend. BUG=angleproject:1951 TEST=angle_unittests, dEQP-GLES31.functional.shaders.linkage.shader_storage_block.* dEQP-GLES31.functional.shaders.builtin_functions.*compute* Change-Id: Ibecca24623ca8e4723af6f0e0421fe9711ea828d Reviewed-on: https://chromium-review.googlesource.com/787976 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 923ecef6 2017-10-11T12:01:38 Fix switch statement validation corner cases The grammar needs to generate AST nodes even for no-op statements, since they might be the last statement in a switch statement that is required for switch statement validity. Change the grammar to generate nodes from empty blocks and empty declarations. We also need to do some further processing of the AST. This is because PruneEmptyDeclarations will still remove empty declarations, and at least the NVIDIA driver GLSL compiler doesn't accept some types of no-op statements as the last statement inside a switch statement. So after parsing has finished we do rudimentary dead code elimination to remove dead cases from the end of switch statements. BUG=angleproject:2181 TEST=angle_unittests Change-Id: I586f2e4a3ac2171e65f1f0ccb7a7de220e3cc225 Reviewed-on: https://chromium-review.googlesource.com/712574 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho bb2bbfbb 2017-08-24T15:43:33 Refactor GLSL array length method parsing This prepares for accepting arbitrary expressions as the "this" node of the array length method. BUG=angleproject:2142 TEST=angle_unittests Change-Id: I728adb6e76d2779dedbabfaeec7d096872e0d00d Reviewed-on: https://chromium-review.googlesource.com/633945 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 96f6adfa 2017-08-16T11:18:54 Add support for arrays of arrays in AST processing Data concerning arrays of arrays is added in TType. Parsing arrays of arrays and support for arrays of arrays in TPublicType are still left to be implemented later. ShaderVariable interface for arrays of arrays is also left to be implemented later. We rely on existing test coverage to make sure that arrays of arrays are not accidentally exposed. BUG=angleproject:2125 TEST=angle_unittests, angle_end2end_tests, angle_deqp_gles31_tests Change-Id: Ie17d5ac9b8d33958e9126dc0fb40bf1c81ddeec9 Reviewed-on: https://chromium-review.googlesource.com/616146 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 3ec75686 2017-07-05T17:02:55 Collect AST transform utilities to a separate file Collect static functions that are used to create nodes in AST transformations into a single file. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I6f87422988fa088f2f4b48986e378a2909705cb7
Olli Etuaho eb7f90fd 2017-07-07T17:25:23 Remove simple Intermediate.h functions Most of the functions were just simple wrappers around node constructors. Dropping this extra redirection makes the code simpler. The fold() functions of node types are simplified, so that if the node can't be folded the pointer to the node itself is returned. This makes the code in ParseContext more straightforward. The few remaining functions in Intermediate are a bit more complex so they should be handled separately, but they'll be removed eventually as well. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I85e11919d1f62358cfba9c011b841e32bc25402f Reviewed-on: https://chromium-review.googlesource.com/563393 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho cccf2b00 2017-07-05T14:50:54 Reorganize AST traversal utility code Define TIntermTraverser and TIntermLValueTrackingTraverser in a separate header file. hash() function is moved out from TIntermTraverser as it is not related to the core functionality of traversing and transforming ASTs. Also reorganize some traversers to follow common conventions: - Intermediate output is now in OutputTree.h/.cpp - Max tree depth check is now in IsASTDepthBelowLimit.h/.cpp BUG=angleproject:1490 TEST=angle_unittests Change-Id: Id4968aa9d4e24d0c5bac90dc147fc9f310de0184 Reviewed-on: https://chromium-review.googlesource.com/559531 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill e72595b4 2017-06-06T15:12:26 Rename EOpFaceForward to EOpFaceforward. This mirrors the spec naming and makes auto-gen a little easier. BUG=chromium:697758 Change-Id: I9bcbc2c874b9a93a6d542aedf2b239f01ee708ce Reviewed-on: https://chromium-review.googlesource.com/526393 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Olli Etuaho a22aa4ed 2017-05-24T18:17:23 Mark some internal functions as not having side effects Precision emulation rounding function calls and vector/matrix dynamic indexing function calls now get a flag that indicates that running the function body does not have side effects. This avoids triggering asserts in OutputHLSL when these internal function calls end up on the right hand side of a non-unfolded logical operator. BUG=chromium:724870 TEST=angle_unittests Change-Id: Id1a2b6b744f6a04c6cdb86a8f4109ccc12bc70b9 Reviewed-on: https://chromium-review.googlesource.com/516705 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 9676d1af 2017-05-16T11:29:24 Handle multiple AST insertions to the same parent in updateTree() Multiple insertions to the same parent can be handled as long as the insertions don't have the same position as well. They're sorted in reverse order so that insertions to greater indices get processed first. This helps to make some AST transformations faster - they don't need multiple tree traversals and updateTree() steps anymore. The SimplifyLoopConditions AST transformation is changed to only use a single traversal. BUG=angleproject:1966 TEST=angle_unittests, angle_end2end_tests, WebGL conformance tests, dEQP-GLES2.functional.shaders.*select_iteration_count* Change-Id: I3183f2644ad58b282926093c77b204fb7e4e9b71 Reviewed-on: https://chromium-review.googlesource.com/506202 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 8fab320c 2017-05-08T18:22:22 Share a single TOperator enum among all constructor AST nodes The code is a lot simpler when the type information is only carried in the TType of the node, instead of being partially duplicated in the enum value. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I956376225ec01e469c7afb7378fa48cc097c0cea Reviewed-on: https://chromium-review.googlesource.com/498768 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho a7ecec38 2017-05-08T17:43:55 GLSL: Simplify constructor parsing Constructor argument checking rules are reorganized to make them easier to understand and constructor node creation is made simpler. This removes usage of constructor op codes from ParseContext. This paves the way for getting rid of constructor op codes entirely, which will remove duplicate information from the AST and simplify lots of code. This refactoring will make adding arrays of arrays slightly easier. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I4053afec55111b629353b4ff7cb0451c1ae3511c Reviewed-on: https://chromium-review.googlesource.com/498767 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 193c0950 2017-05-02T15:51:47 Fix assert when attempting to create a void array node TIntermTyped::CreateZero can be reached with a void array type in an error case. Handle this gracefully instead of asserting. Also remove an assert that wasn't really checking anything in CreateZero. type.isScalar() || type.isVector() || type.isMatrix() can only be false in case of a struct, and struct type was being checked in the condition on the line above. BUG=chromium:717385 TEST=angle_unittests Change-Id: Iff0811d18d399d7b32b2b46deea5df172412eb8c Reviewed-on: https://chromium-review.googlesource.com/492887 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Yunchao He 4f285443 2017-04-21T12:15:49 Refactoring: replace NULL by nullptr for pointers (2nd CL). This CL mainly handles the pointer comparisons (== or !=). BUG=angleproject:2001 Change-Id: I25ac3b61032e7ad91459a1c6541cadc87cf9b160 Reviewed-on: https://chromium-review.googlesource.com/483935 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 8162926b 2017-04-19T11:56:01 Never add declarations without children to the AST When block nodes were being created for loop bodies that didn't have braces in the parsed source, the code didn't check if the loop body was a declaration node without children. Always use appendStatement() for adding statements to a block, so that declaration nodes without children don't end up in the AST. Similarly make sure that loop init nodes aren't declarations without children. BUG=chromium:712550 TEST=angle_end2end_tests Change-Id: I5e79b700fe6158fa2422fcf4cd13818b2bd24863 Reviewed-on: https://chromium-review.googlesource.com/481660 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho f2209f74 2017-04-01T12:45:55 Clean up function name mangling code Fix a few incorrect comments about mangled names, and refactor generating mangled names from function call nodes. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I3ee68c4c0982f1a9c28d8e87aafa19f19559bbf8 Reviewed-on: https://chromium-review.googlesource.com/465826 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho ec9232bd 2017-03-27T17:01:37 Store unmangled function names in the AST This makes the code simpler across the board. There are a few cases where mangled names still need to be generated in AST traversers, but they are outweighed by much leaner output code for all function nodes. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: Id3638e0fca6019bbbe6fc5e1b7763870591da2d8 Reviewed-on: https://chromium-review.googlesource.com/461077 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho fe48632f 2017-03-21T09:30:54 Prefer identifying functions by using symbol ids The shader translator code is now structured in a way that ensures that all function definition, function prototype and function call nodes store the integer symbol id for the function. This is guaranteed regardless of whether the function node is added while parsing or as a result of an AST transformation. TIntermAggregate nodes, which include function calls and constructors can now only be created by calling one of the TIntermAggregate::Create*() functions to ensure they have all the necessary properties. This makes it possible to keep track of functions using integer ids instead of their mangled name strings when generating the call graph and when using TLValueTrackingTraverser. This commit includes a few other small cleanups to the CallDAG class as well. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: Idd1013506cbe4c3380e20d90524a9cd09b890259 Reviewed-on: https://chromium-review.googlesource.com/459603 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho c9da71ff 2017-03-06T16:28:54 Fix translating dynamic indexing of swizzle to HLSL This was broken due to a simple omission in deep copying swizzle AST nodes. It was found that several if not most OpenGL drivers also have some issue related to this, so the end2end test is suppressed on many OpenGL platforms. Intel Windows driver seems to be behaving correctly. BUG=angleproject:1921 TEST=angle_end2end_tests Change-Id: Ieefcedc2f2e36c3d8b607c28e449b696b8ad6892 Reviewed-on: https://chromium-review.googlesource.com/449717 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 74da73fe 2017-02-01T15:37:48 Add ESSL 3.10 ldexp/frexp builtins This adds new built-ins found in ESSL 3.10 section 8.3 Common Functions. This includes constant folding support for ldexp and support for both GLSL and HLSL output. In HLSL these functions need to be emulated. BUG=angleproject:1730 TEST=angle_unittests Change-Id: I1330e69978b0cf53efbc3416150194764414e96c Reviewed-on: https://chromium-review.googlesource.com/435342 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 25aef453 2017-01-29T16:15:44 Add new ESSL 3.10 pack/unpack builtins This adds new built-ins found in ESSL 3.10 section 8.4 Floating-Point Pack and Unpack Functions. This includes constant folding support and support for both GLSL and HLSL output. In HLSL all of these functions need to be emulated. BUG=angleproject:1730 TEST=angle_unittests TEST=dEQP-GLES31.functional.shaders.*pack*norm4x8* Change-Id: Ibed60286a366cd35c4faafd405e79af562a02a06 Reviewed-on: https://chromium-review.googlesource.com/434170 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho af6fc1b4 2017-01-26T17:45:35 Make aggregate node creation more robust Now aggregate nodes are always built with their return type, op and arguments set. They'll determine their qualifier and precision automatically. This fixes setting of gotPrecisionFromChildren in a few cases. This will also make it easier to split TIntermAggregate further into specialized classes if that is desired. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I1fbe0c75679c517a22d44dfc1ea160ad7a7fdfda Reviewed-on: https://chromium-review.googlesource.com/433468 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 1ecd14b8 2017-01-26T13:54:15 Fold user-definedness of function nodes into TOperator Whether a function call is user-defined is not orthogonal to TOperator associated with the call node - other ops than function calls can't be user-defined. Because of this it makes sense to store the user- definedness by having different TOperator enums for different types of calls. This patch also tags internal helper functions that have a raw definition outside the AST with a separate TOperator enum. This way they can be handled with logic that is easy to understand. Before this, function calls like this left the user-defined bit unset, despite not really being built-ins either. The EmulatePrecision traverser uses this. This is also something that could be used to clean up built-in emulation in the future. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I597fcd9789d0cc22b689ef3ce5a0cc3f621d4859 Reviewed-on: https://chromium-review.googlesource.com/433443 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Corentin Wallez 17a5c062 2017-01-22T15:20:53 IntermTyped::CreateZero: handle non-basictypes CreateZero is called in ParseContext so it should handle types which don't necessarily make sense to call it with. BUG=chromium:680961 Change-Id: I8627850e49eb9a4f4ecde61ca2d68371ea6a8dd6 Reviewed-on: https://chromium-review.googlesource.com/431001 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 9250cb24 2017-01-21T10:51:27 Add ESSL 3.10 integer math built-ins This adds built-ins found in ESSL 3.10 section 8.8 Integer functions. This includes constant folding support for functions that may be constant folded, and support for both GLSL and HLSL output. In HLSL several of the functions need to be emulated. The precision qualification for the return value of some of these functions is determined by special rules, that are now part of type promotion for TIntermUnary nodes and determining the type of TIntermAggregate nodes. BUG=angleproject:1730 TEST=angle_unittests TEST=dEQP-GLES31.functional.shaders.builtin_functions.integer.* Change-Id: Ib0056c17671c42b6496c2f0ef059b99f8f25c122 Reviewed-on: https://chromium-review.googlesource.com/431310 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 51182ab9 2017-01-22T00:12:29 Clean up TIntermAggregate folding The division between 2- and 3-parameter built-ins is superfluous and can be removed. Also folding support for ESSL 3.00 is complete, so the comments about needing to add folding support for more functions can be removed. BUG=angleproject:1730 TEST=angle_unittests Change-Id: I5ca93a957a98a7740eca506e44af48e7b4b83bad Reviewed-on: https://chromium-review.googlesource.com/431108 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 8ad9e757 2017-01-16T19:55:20 Always store function headers in TIntermFunctionPrototype nodes TIntermFunctionDefinition nodes now have a TIntermFunctionPrototype child that stores the function signature, instead of having a separate type and an aggregate child that stores the parameters. This makes parsing functions simpler, and paves the way for further simplifications of function parsing, like reducing conversions between symbol table structures and AST structures. TIntermAggregate is now only used for function calls. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: Ib56a77b5ef5123b142963a18499690bf37fed987 Reviewed-on: https://chromium-review.googlesource.com/427945 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 16c745a3 2017-01-16T17:02:27 Split TIntermFunctionPrototype from TIntermAggregate Function prototypes now have their own class TIntermFunctionPrototype. It's only used for prototypes, not function parameter lists. TIntermAggregate is still used for parameter lists and function calls. BUGS=angleproject:1490 TEST=angle_unittests Change-Id: I6e246ad00a29c2335bd2ab7f61cf73fe463b74bb Reviewed-on: https://chromium-review.googlesource.com/427944 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho d68924e5 2017-01-02T17:34:40 Use GetOperatorString when writing GLSL unary built-in calls GetOperatorString is now used when writing GLSL for built-in calls that fall under TIntermUnary. Component-wise not TOperator enum is renamed for consistency. This also cleans up some unnecessary creation of string objects when writing built-in functions. BUG=angleproject:1682 TEST=angle_unittests, angle_end2end_tests, WebGL conformance tests Change-Id: I89b2ef222bf5af479d4977417f320789b58ace85 Reviewed-on: https://chromium-review.googlesource.com/424552 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho e180559f 2017-01-02T16:41:20 Use GetOperatorString when writing GLSL built-in function calls GetOperatorString is now used when writing GLSL for built-in calls that fall under TIntermAggregate. Component wise and not component wise TOperator enums are disambiguated from each other. BUG=angleproject:1682 TEST=angle_unittests, angle_end2end_tests, WebGL conformance tests Change-Id: I861f1e94eb695eb712592df99705848b442ef07b Reviewed-on: https://chromium-review.googlesource.com/424532 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 09b04a2f 2016-12-15T13:30:26 Add shader translator support for OVR_multiview The OVR_multiview and OVR_multiview2 extensions add gl_ViewID_OVR to shaders. gl_ViewID_OVR can be translated either as is in GLSL output or as a uniform by setting the SH_TRANSLATE_VIEWID_OVR_AS_UNIFORM compiler flag. If WebGL output is selected, the shaders will be validated according to proposed rules in the WEBGL_multiview spec. BUG=angleproject:1669 TEST=angle_unittests Change-Id: I19ea3a6c8b4edb78be03f1a50a96bfef018870d0 Reviewed-on: https://chromium-review.googlesource.com/422848 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 4de340ac 2016-12-16T09:32:03 Remove extraInfo parameter from compiler diagnostic functions This makes error messages more consistent. It was not clear what was supposed to go to the extraInfo parameter, and previously it was mostly being misused, resulting in poorly formatted error messages. Sometimes the order of parameters to the diagnostic functions like error() and warning() was wrong altogether. The diagnostics API is simpler when there's only the "reason" and "token" parameters that have clear meaning and that are separated by consistent punctuation in the output. Fixes error messages like "redifinition interface block member" to be grammatically reasonable like the rest of the error messages. For other error messages, punctuation is added to make them clearer. Example: "invalid layout qualifier location requires an argument" is changed to "invalid layout qualifier: location requires an argument". Extra spaces are also removed from the beginning of error messages. BUG=angleproject:1670 BUG=angleproject:911 TEST=angle_unittests Change-Id: Id5fb1a1f2892fad2b796aaef47ffb07e9d79759c Reviewed-on: https://chromium-review.googlesource.com/420789 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho bf4e1b73 2016-12-09T11:30:15 Split TIntermInvariantDeclaration from TIntermAggregate This change is pure refactoring and doesn't fix bugs related to invariant declarations. Invariant declarations are supposed to accept a list of identifiers, but this refactoring keeps the current behavior of only accepting a single identifier in an invariant declaration. When the bug will be fixed, the new TIntermInvariantDeclaration class that now has only a single child node can be changed so that it may have multiple children. TIntermAggregate is still used for function calls, function prototypes and function parameter lists. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I3e22092c87e1c06445fd7e123d9922c2fcb59428 Reviewed-on: https://chromium-review.googlesource.com/419415 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill d7b1ab58 2016-12-12T14:42:19 Fix up translator style. Using git cl format. BUG=angleproject:650 Change-Id: I7d3f98d2b0dcfb0a8de6c35327db74e55c28d761 Reviewed-on: https://chromium-review.googlesource.com/419059 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Corentin Wallez 36fd100d 2016-12-08T11:30:44 GLSLTest: test interaction of declaration splitting with other passes In TranslatorHLSL a number of AST simplifications are done, that must happen in a precise order for things to be correct: - First for-loops must be split - Then multideclarations must be split - Finally comma operators must be split This adds tests for interaction between this passes to make sure they are done in the right order. BUG=668028 Change-Id: I306915b51011bb5467d117352becfd60cbe77be4 Reviewed-on: https://chromium-review.googlesource.com/417989 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 2d73665d 2016-11-30T10:37:49 Handle constant folding arithmetic involving infinity Constant folding arithmetic operations that involve infinity are now handled correctly in the cases where the result is infinity or zero. The implementation mostly relies on C++ to implement IEEE float arithmetic correctly so that unnecessary overhead is avoided. Constant folding arithmetic operations that result in overflow now issue a warning but result in infinity. This is not mandated by the spec but is a reasonable choice since it is the behavior of the default IEEE rounding mode. Constant folding arithmetic operations that result in NaN in IEEE will generate a warning but the NaN is kept. This is also not mandated by the spec, but is among the allowed behaviors. There's no special handling for ESSL 1.00. ESSL 1.00 doesn't really have the concept of NaN, but since it is not feasible to control generating NaNs at shader run time either way, it should not be a big issue if constant folding may generate them as well. TEST=angle_unittests BUG=chromium:661857 Change-Id: I06116c6fdd02f224939d4a651e4e62f2fd4c98a8 Reviewed-on: https://chromium-review.googlesource.com/414911 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 45bcc784 2016-11-07T13:58:48 translator: Scope all classes with "sh". I was seeing an odd problem with our PoolAlloc conflicting with the glslang/Vulkan TIntermNode, so the fix was to move everything to a separate namespace. The bison grammars are also regenerated. No functional changes. BUG=angleproject:1576 Change-Id: I959c7afe4c092f0d458432c07b4dcee4d39513f3 Reviewed-on: https://chromium-review.googlesource.com/408267 Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 60e6edfa 2016-10-31T12:17:19 Make ASSERT reference the conditional expression. This should prevent further unexpected bot breakage due to unreferenced variables in the ASSERT expression. Also remove the no longer needed variable referencing macro. BUG=angleproject:1586 Change-Id: I127695165bdfe39c51fe8d17e00daf6bf2fa8252 Reviewed-on: https://chromium-review.googlesource.com/404948 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 13389b66 2016-10-16T11:48:18 Split TIntermDeclaration from TIntermAggregate The new class TIntermDeclaration is now used for struct, interface block and variable declarations. TIntermDeclaration nodes do not have a type - rather the type is stored in each child node. The types may differ in case the declaration is a series of array declarators with mismatching sizes. TIntermAggregate is still used for function calls, function prototypes, function parameter lists and invariant declarations. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I0457188f354481470855f61ac1c878fc2579b1d1 Reviewed-on: https://chromium-review.googlesource.com/400023 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 4db7ded5 2016-10-13T12:23:11 Change comma nodes to TIntermBinary Comma nodes always have just two parameters. If there's an expression with several commas in the middle, it's parsed as a tree of comma operations. It makes more sense to represent it as a binary node rather than an aggregate node. After this patch, TIntermAggregate is still used for function prototypes, function parameter lists, function calls, and variable and invariant declarations. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: I66be10624bf27bcf25987b4d93958d4a07600771 Reviewed-on: https://chromium-review.googlesource.com/397320 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 336b1470 2016-10-05T16:37:55 Split TIntermFunctionDefinition from TIntermAggregate This makes the code easier to understand. Function definition nodes always have just two children, the parameters node and the function body node, so there was no proper reason why they should be aggregate nodes. As a part of this change, intermediate output is modified to print symbol table ids of functions so that debugging function id related functionality will be easier in the future. After this patch, TIntermAggregate is still used for function prototypes, function parameter lists, function calls, variable and invariant declarations and the comma (sequence) operator. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: Ib88b4ca5d21abd5f126836ca5900d0baecabd19e Reviewed-on: https://chromium-review.googlesource.com/394707 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Olli Etuaho bd674557 2016-10-06T13:28:42 Separate function info from TIntermAggregate This change will make it easier to split types of TIntermAggregate nodes representing functions and function calls into different node classes. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I730aa7858fe31fda86218fc685980c6ad486f5e0 Reviewed-on: https://chromium-review.googlesource.com/394706 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 6d40bbdd 2016-09-30T13:49:38 Split TIntermBlock from TIntermAggregate The new TIntermBlock node class replaces TIntermAggregate nodes with the EOpSequence op. It represents the root node of the tree which is a list of declarations and function definitions, and any code blocks that can be denoted by curly braces. These include function and loop bodies, and if-else branches. This change enables a bunch of more compile-time type checking, and makes the AST code easier to understand and less error-prone. The PostProcess step that used to be done to ensure that the root node is TIntermAggregate is removed in favor of making sure that the root node is a TIntermBlock in the glslang.y parsing code. Intermediate output formatting is improved to print the EOpNull error in a clearer way. After this patch, TIntermAggregate is still used for function definitions, function prototypes, function parameter lists, function calls, variable and invariant declarations and the comma (sequence) operator. BUG=angleproject:1490 TEST=angle_unittests, angle_end2end_tests Change-Id: I04044affff979a11577bc1fe75d747e538b799c8 Reviewed-on: https://chromium-review.googlesource.com/393726 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 32db19b7 2016-10-04T14:43:16 Ensure that if-else branches are always sequence nodes This mainly affects RewriteElseBlocks, which was the only piece of code still adding TIntermIfElse nodes directly as children of other TIntermIfElse nodes. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I5b25c2fb9c642424417cd6c29e37c20482c6ffaf Reviewed-on: https://chromium-review.googlesource.com/392847 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho b6fa043d 2016-09-28T16:28:05 Split vector swizzle AST nodes into a different node class This avoids creating a weird aggregate node with a sequence of constant union nodes to store the offsets. They're stored neatly inside a vector instead. This makes code that needs to iterate over the swizzle offsets much simpler. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I156b95723529ee05a94d30295ffb6d0952a98564 Reviewed-on: https://chromium-review.googlesource.com/390832 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 42fad76d 2016-09-28T10:06:29 Handle negation of minimum representable integer Negating the minimum representable integer overflows, so it has undefined behavior in C++. Handle this as a special case in the code. BUG=chromium:637050 TEST=angle_unittests Change-Id: Ic6e6d638faddad9b70b5d1637bb4b42ef4f43784 Reviewed-on: https://chromium-review.googlesource.com/390551 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho d4453573 2016-09-27T13:21:46 Fix integer division constant folding corner cases ESSL has undefined behavior when the integer modulus operator is used on negative operands. Generate a warning and fold the result to zero in this case. In case the minimum representable signed integer is divided by -1, the result is defined to be either the maximum or minimum representable value. We choose to fold the calculation to the maximum representable value in this case. BUG=angleproject:1537 TEST=angle_unittests Change-Id: I57fac6b54a3553b7a0f0e36cc6ba0ed59a88eea9 Reviewed-on: https://chromium-review.googlesource.com/390251 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 596018ce 2016-09-21T12:57:03 translator: Refactor Constant Union shift ops. In preparation for making them robust. BUG=chromium:648135 Change-Id: I88fc87d8887064fda04087c56de05d8725a6fe5f Reviewed-on: https://chromium-review.googlesource.com/387469 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 5db69f57 2016-09-15T12:47:32 Add robust math to constant folding. Previously our multiplication and other operators could do overflows, which can lead to security bugs. BUG=chromium:637050 Change-Id: Icee22a87909e205b71bda1c5bc1627fcf5e26e90 Reviewed-on: https://chromium-review.googlesource.com/382678 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 5796127e 2016-09-14T13:57:46 Rename TIntermSelection to TIntermIfElse Now that ternary nodes are not represented by TIntermSelection any more, TIntermIfElse is an easier name to understand for newcomers to the code. BUG=angleproject:1490 TEST=angle_unittests Change-Id: Ia1e04e356ab93409400245092a84533d7dfd129d Reviewed-on: https://chromium-review.googlesource.com/385416 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 47cb73ab 2016-09-09T11:41:44 Refactor TConstantUnion. In preparation for constant folding fixes. BUG=chromium:637050 Change-Id: I9ea49ce96b34c6ac3d2f0478b8fc6732c59e28be Reviewed-on: https://chromium-review.googlesource.com/373741 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Olli Etuaho d0bad2c7 2016-09-09T18:01:16 Split ternary node class from TIntermSelection Ternary operator nodes are typed parts of expressions, they always have two children and the children are also guaranteed to be TIntermTyped. "If" selection nodes can't be a part of an expression, they can have either one or two children and the children are code blocks. Due to all of these differences it makes sense to store these using two different AST node classes. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I913ab1d806e3cdb5c21106f078cc9c0b6c72ac54 Reviewed-on: https://chromium-review.googlesource.com/384512 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho a223430c 2016-08-31T12:05:39 Promote unary nodes automatically Unary nodes now get their type set automatically based on the operation and operand. The operand should only be changed to another of the same type after the node is constructed. The operation can't be changed on unary and binary nodes after they've been constructed. BUG=angleproject:1490 TEST=angle_unittests Change-Id: Ib1ea3dcb1162261966c02d5f03d8091cf647fac1 Reviewed-on: https://chromium-review.googlesource.com/378935 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 3272a6d3 2016-08-29T17:54:50 Promote and fold indexing nodes similarly to other binary ops Indexing nodes now get their type set in TIntermBinary::promote, same as math and logic ops. They are also constant folded through TIntermBinary::fold() instead of having special functions for constant folding them in ParseContext. Index nodes for struct and interface block member access now always have integer type, instead of sometimes having the type of the field they were used to access. Usage of TIntermBinary constructor is cleaned up so only the constructor that takes in left and right operands is used. The type of TIntermBinary nodes is always determined automatically. Together these changes make the code considerably cleaner. Note that the code for constant folding for array indexing is actually never hit because constant folding array constructors is still intentionally disabled in the code. BUG=angleproject:1490 TEST=angle_unittests Change-Id: Ifcec45257476cdb0d495c7d72e3cf2f83388e8c5 Reviewed-on: https://chromium-review.googlesource.com/377961 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho c955058b 2016-08-29T17:56:22 Constant-qualify nodes in a consistent way Rely on that constant qualification of binary and unary nodes comes from promote(). BUG=angleproject:1490 TEST=angle_unittests Change-Id: Ie8d1d4df3c82ae5a2de8cc536e47016d13a4fd3d Reviewed-on: https://chromium-review.googlesource.com/377960 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho f119a263 2016-08-19T15:54:22 Clean up unary and aggregate math folding Prefer asserts instead of adding internal errors to the compiler log when types of arguments are not as expected or the folding function runs into an operation it can't handle. Neither of these cases should be possible, the checks for correct argument types are solid at this point. In the future, when new built-in functions are added, constant folding support for them should be added as well. foldUnaryWithDifferentReturnType and foldUnaryWithSameReturnType are renamed to foldUnaryNonComponentWise and foldUnaryComponentWise respectively. These names better reflect what these functions are doing. The info sink member is removed from TIntermediate, since TDiagnostics is now passed into the functions that may generate warnings instead. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I6a08abbe29cf23f3a318032fdc46dd3dbaf4410e Reviewed-on: https://chromium-review.googlesource.com/377959 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Corentin Wallez 509e4560 2016-08-25T14:55:44 compiler: Work around a HLSL compiler aliasing opt bug. BUG=angleproject:1448 Change-Id: I7d5bcbd100069152cea0cb03bc4fa6af1044460b Reviewed-on: https://chromium-review.googlesource.com/376020 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Olli Etuaho d5da505d 2016-08-29T13:16:55 Fix constant folding non-square outerProduct Use all the vector elements correctly when constant folding non-square outerProduct. Previously the code used to discard some elements of the smaller outerProduct operand. Also clear up confusion about matrix rows/columns in matrix constant folding code in general. BUG=angleproject:1482 TEST=angle_unittests Change-Id: I7cba8f97a92b875de01e57255d54258cdfd92a47 Reviewed-on: https://chromium-review.googlesource.com/377298 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 63e1ec5c 2016-08-18T22:05:12 Move the rest of the validation out of TIntermBinary::promote TIntermBinary::promote now has a single purpose of determining the type resulting from a binary math operation. The TIntermBinary constructor taking the left and right nodes can now also call promote automatically, and promote is made into a private member of TIntermBinary. Validation of binary math operand types is done inside ParseContext. BUG=angleproject:952 TEST=angle_unittests Change-Id: I52a409f680c8d4120b757193972d03aed34c6895 Reviewed-on: https://chromium-review.googlesource.com/372624 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 1dded803 2016-08-18T18:13:13 Check multiplication validity in ParseContext This improves separation of responsibilities in the code: ParseContext should handle operand type validation, while TIntermBinary::promote should ideally only determine the type of the node based on the operation and operands. BUG=angleproject:952 TEST=angle_unittests Change-Id: I9a8d8ede21cdf35de631623a62194c0da5c604d2 Reviewed-on: https://chromium-review.googlesource.com/372622 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 3fdec919 2016-08-18T15:08:06 Refactor binary node creation 1. Simplify code by using asserts instead of adding internal errors to log. 2. Add a TIntermBinary constructor that takes left and right operand nodes as parameters. 3. Remove TIntermediate functions with trivial functionality. BUG=angleproject:952 TEST=angle_unittests Change-Id: I2e0e52160c9377d8efcf15f14fd59f01cb41bd83 Reviewed-on: https://chromium-review.googlesource.com/372720 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 3cbb27a1 2016-07-14T11:55:48 Simplify loop conditions so that they won't generate statements Introduce an AST traverser that can move the evaluation of certain types of loop conditions and loop expressions inside the loop. This way subsequent AST transformations don't have to worry about cases where they have to insert new statements to implement a loop condition or expression. This includes the revert of "Unfold short-circuiting operators in loop conditions correctly". The new traverser covers the loop cases that used to be handled in UnfoldShortCircuitToIf. BUG=angleproject:1465 TEST=WebGL conformance tests, dEQP-GLES2.functional.shaders.*select_iteration_count* Change-Id: I88e50e007e924d5884a217117690ac7fa2f96d38 Reviewed-on: https://chromium-review.googlesource.com/362570 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 03d863c8 2016-07-27T18:15:53 translator: Refactor node replacement APIs. BUG=angleproject:851 Change-Id: I50c3b3a4f00b27fed85f09509738513a441c7b5b Reviewed-on: https://chromium-review.googlesource.com/363990 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jamie Madill 1048e43f 2016-07-23T18:51:28 D3D: Work around HLSL integer pow folding bug. BUG=angleproject:851 Change-Id: I68a47b8343a29e42c0a69ca3f2a6cb5054d03782 Reviewed-on: https://chromium-review.googlesource.com/362775 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Olli Etuaho dcf12c70 2016-06-28T15:03:06 Fix validating non-square matrix compound multiplication The validation previously checked that two matrices in compound multiplication are the exact same size, which isn't correct for non-square matrices introduced in ESSL 3.00. Instead, check that the matrix multiplication is valid and that the resulting value has the same number of columns as the lvalue. The number of rows in the result is taken from the lvalue so it doesn't need to be checked. BUG=angleproject:1431 TEST=angle_unittests Change-Id: I6f32b7dc037d72c3c5cfdfffcda5d996e8450283 Reviewed-on: https://chromium-review.googlesource.com/356411 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho d4f4c11b 2016-04-15T15:11:24 Fix deferring global array initialization The initial implementation of DeferGlobalInitializers did not take HLSL corner cases into account. In particular, in case there was a const-qualified array variable with an initializer that contained elements that weren't constant folded, initialization would not be deferred and the global scope of HLSL output would contain a call to angle_construct_into_*(). On the other hand, deferring global initializers was also done in cases where it wasn't necessary. Initializers of non-const qualified array variables that could be written as HLSL literals by HLSL output were unnecessarily deferred. This patch fixes both of these issues: Now all global initializers are potential candidates for deferral instead of just those where the symbol has the EvqGlobal qualifier, and initializers that are constructors taking only constant unions as parameters are not unnecessarily deferred. BUG=angleproject:1205 BUG=541551 TEST=angle_end2end_tests Change-Id: I4027059e0e5f39c8a5a48b5c97a3fceaac6b6f8a Reviewed-on: https://chromium-review.googlesource.com/339201 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 40d9edf1 2015-11-12T17:30:34 Fix structure comparison constant folding objectSize() will return the size of all data in the structure, and simply iterating over the data will work for determining whether two structures are equal. The earlier complex and broken approach where the structure was traversed recursively is not needed. BUG=angleproject:1211 TEST=angle_unittests Change-Id: I0e5c5ccbb767d44ef6acb0f1f25f27dfc42866e1 Reviewed-on: https://chromium-review.googlesource.com/312490 Tryjob-Request: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 5c0e023c 2015-11-11T15:55:59 Qualify stored constant union data with const This prevents accidentally changing data that may be shared between multiple TIntermConstantUnion nodes. Besides making the code less prone to bugs in general, this will make it easier to implement constant folding of array constructors. BUG=541551 TEST=angle_unittests, WebGL conformance tests Change-Id: I4f3059f70b841d9dd0cf20fea4d37684da9cd47e Reviewed-on: https://chromium-review.googlesource.com/312440 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tryjob-Request: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 3fed4306 2015-11-02T12:26:02 Unfold short-circuiting operators in loop conditions correctly Sometimes short-circuiting operators need to be unfolded to if statements. If the unfolded operator is inside a loop condition or expression, it needs to be evaluated repeatedly inside the loop. Add logic to UnfoldShortCircuitToIf that can move or copy the unfolded part of loop conditions or expressions to inside the loop. The exact changes that need to be done depend on the type of the loop. For loops may require also moving the initializer to outside the loop. The unfolded expression inside a loop condition or expression is moved or copied to inside the loop on the first traversal of the loop node, and unfolding to if is deferred until a second traversal. This keeps the code relatively simple. BUG=angleproject:1167 TEST=WebGL 2 conformance tests, dEQP-GLES2.functional.shaders.*select_iteration_count* Change-Id: Ieffc0ea858186054378d387dca9aa64a5fa95137 Reviewed-on: https://chromium-review.googlesource.com/310230 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 1d122789 2015-11-06T15:35:17 Fix constructor constant folding The previous solution for constant folding constructors was significantly overengineered and partially incorrect. Switch to a much simpler constructor folding function that does not use an AST traverser, but simply iterates over the constant folded parameters of the constructor and doesn't do any unnecessary checks. It also reuses some code for constant folding other built-in functions. This fixes issues with initializing constant matrices with only a single parameter. Instead of copying the first component of the constructor parameter all over the matrix, passing a vec4 or matrix argument now assigns the values correctly. BUG=angleproject:1193 TEST=angle_unittests, WebGL conformance tests Change-Id: I50b10721ea30cb15843fba892c1b1a211f1d72e5 Reviewed-on: https://chromium-review.googlesource.com/311191 Tryjob-Request: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 7c3848e5 2015-11-04T13:19:17 Allow constant folding some non-constant expressions This requires removing the assumption that constant folding implies constness in the constant expression sense from various places in the code. This particularly benefits ternary operators, which can now be simplified if just the condition is a compile-time constant. In the future, the groundwork that is laid here could be used to implement more aggressive constant folding of user-defined functions for example. TEST=angle_unittests BUG=angleproject:851 Change-Id: I0eede806570d56746c3dad1e01aa89a91d66013d Reviewed-on: https://chromium-review.googlesource.com/310750 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Zhenyao Mo <zmo@chromium.org>
Olli Etuaho b1edc4f5 2015-11-02T17:20:03 Accept const array initialization in shader parsing Array constructors are not folded, unlike all other constant expressions. Change initializer parsing path so that it accepts constant initializers whether they are folded or not. Some parts need to be adapted to work with expressions that are qualified as constant but that are not necessarily folded: 1. Identifier parsing 2. Indexing parsing 3. Field selection parsing 4. HLSL output for variable declarations 5. Determining unary operator result type 6. Determining binary operator result type 7. Determining built-in function call result type 8. Determining ternary operator result type Corner cases that are not supported yet: 1. Using array constructors inside case labels 2. Using array constructors inside array size expressions 3. Detecting when a negative constant expression containing an array constructor is used to index an array In these cases being able to constant fold the expression is essential to validating that the code is correct, so they require a more sophisticated solution. For now we keep the old code that rejects the shader if ANGLE hasn't been able to constant fold the case label or array size. In case of indexing an array with a negative constant expression containing an array constructor, ANGLE will simply treat it as a non-constant expression. BUG=541551 BUG=angleproject:1094 TEST=dEQP-GLES3.functional.shaders.constant_expressions.* (all pass), angle_unittests Change-Id: I0cbc47afd1651a4dece3d68acf7ec72a01fdf047 Reviewed-on: https://chromium-review.googlesource.com/310231 Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 5d91dda9 2015-06-18T15:47:46 Remove dynamic indexing of matrices and vectors in HLSL Re-re-relanding after clang warning fix. Re-re-landing with fix to setting qualifiers on generated nodes. The previous version failed when a uniform was indexed, because it would set the uniform qualifier on some of the generated nodes and that interfered with the operation of UniformsHLSL. Re-landing after fixing D3D9 specific issues. HLSL doesn't support dynamic indexing of matrices and vectors, so replace that with helper functions that unroll dynamic indexing into switch/case and static indexing. Both the indexed vector/matrix expression and the index may have side effects, and these will be evaluated correctly. If necessary, index expressions that have side effects will be written to a temporary variable that will replace the index. Besides dEQP tests, this change is tested by a WebGL 2 conformance test. In the case that a dynamic index is out-of-range, the base ESSL 3.00 spec allows undefined behavior. KHR_robust_buffer_access_behavior adds the requirement that program termination should not occur and that out-of-range reads must return either a value from the active program's memory or zero, and out-of-range writes should only affect the active program's memory or do nothing. This patch clamps out-of-range indices so that either the first or last item of the matrix/vector is accessed. The code is not transformed in case the it fits within the limited subset of ESSL 1.00 given in Appendix A of the spec. If the code isn't within the restricted subset, even ESSL 1.00 shaders may require this workaround. BUG=angleproject:1116 TEST=dEQP-GLES3.functional.shaders.indexing.* (all pass after change) WebGL 2 conformance tests (glsl3/vector-dynamic-indexing.html) Change-Id: I9f6d7c7ecda8ac4dc3c30b39e15a9a0b5381c5a8 Reviewed-on: https://chromium-review.googlesource.com/310010 Reviewed-by: Geoff Lang <geofflang@chromium.org> Tested-by: Geoff Lang <geofflang@chromium.org>
Jamie Madill a5f64de7 2015-10-30T12:31:00 Revert "Remove dynamic indexing of matrices and vectors in HLSL" Failing build on Clang-win: ..\..\third_party\angle\src\compiler\translator\RemoveDynamicIndexing.cpp(128,43) : error: expected '(' for function-style cast or type construction fieldType.setPrimarySize(unsigned char(indexedType.getRows())); ~~~~~~~~ ^ BUG=angleproject:1116 This reverts commit 7535b761dd4740c8e76b888d7c58c7cbeefd2083. Change-Id: I7b502e3dcd45e17b7ed88fec18be702614d9ac65 Reviewed-on: https://chromium-review.googlesource.com/309772 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 7535b761 2015-06-18T15:47:46 Remove dynamic indexing of matrices and vectors in HLSL Re-re-landing with fix to setting qualifiers on generated nodes. The previous version failed when a uniform was indexed, because it would set the uniform qualifier on some of the generated nodes and that interfered with the operation of UniformsHLSL. Re-landing after fixing D3D9 specific issues. HLSL doesn't support dynamic indexing of matrices and vectors, so replace that with helper functions that unroll dynamic indexing into switch/case and static indexing. Both the indexed vector/matrix expression and the index may have side effects, and these will be evaluated correctly. If necessary, index expressions that have side effects will be written to a temporary variable that will replace the index. Besides dEQP tests, this change is tested by a WebGL 2 conformance test. In the case that a dynamic index is out-of-range, the base ESSL 3.00 spec allows undefined behavior. KHR_robust_buffer_access_behavior adds the requirement that program termination should not occur and that out-of-range reads must return either a value from the active program's memory or zero, and out-of-range writes should only affect the active program's memory or do nothing. This patch clamps out-of-range indices so that either the first or last item of the matrix/vector is accessed. The code is not transformed in case the it fits within the limited subset of ESSL 1.00 given in Appendix A of the spec. If the code isn't within the restricted subset, even ESSL 1.00 shaders may require this workaround. BUG=angleproject:1116 TEST=dEQP-GLES3.functional.shaders.indexing.* (all pass after change) WebGL 2 conformance tests (glsl3/vector-dynamic-indexing.html) Change-Id: I16119f9092360fb72798f9550a6f4d3cfffdc92f Reviewed-on: https://chromium-review.googlesource.com/308790 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Zhenyao Mo <zmo@chromium.org>
Jamie Madill b066669d 2015-10-26T10:38:18 Revert "Remove dynamic indexing of matrices and vectors in HLSL" This reverts commit 3766a40d6fda7e7190514ab7838a3f37169d863f. This CL was causing crashes in UniformHLSL.cpp, where an internal uniform "base" was attempted to be declared in HLSL. Was crashing on an external WebGL 3D canvas page (http://www.taccgl.org/?dbg=t). BUG=546686 Original commit message: Re-landing after fixing D3D9 specific issues. HLSL doesn't support dynamic indexing of matrices and vectors, so replace that with helper functions that unroll dynamic indexing into switch/case and static indexing. Both the indexed vector/matrix expression and the index may have side effects, and these will be evaluated correctly. If necessary, index expressions that have side effects will be written to a temporary variable that will replace the index. Besides dEQP tests, this change is tested by a WebGL 2 conformance test. In the case that a dynamic index is out-of-range, the base ESSL 3.00 spec allows undefined behavior. KHR_robust_buffer_access_behavior adds the requirement that program termination should not occur and that out-of-range reads must return either a value from the active program's memory or zero, and out-of-range writes should only affect the active program's memory or do nothing. This patch clamps out-of-range indices so that either the first or last item of the matrix/vector is accessed. The code is not transformed in case the it fits within the limited subset of ESSL 1.00 given in Appendix A of the spec. If the code isn't within the restricted subset, even ESSL 1.00 shaders may require this workaround. BUG=angleproject:1116 TEST=dEQP-GLES3.functional.shaders.indexing.* (all pass after change) WebGL 2 conformance tests (glsl3/vector-dynamic-indexing.html) Change-Id: I1d4b2e3888e91af7d5eebf743d12778698b6b903 Reviewed-on: https://chromium-review.googlesource.com/308770 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 3766a40d 2015-06-18T15:47:46 Remove dynamic indexing of matrices and vectors in HLSL Re-landing after fixing D3D9 specific issues. HLSL doesn't support dynamic indexing of matrices and vectors, so replace that with helper functions that unroll dynamic indexing into switch/case and static indexing. Both the indexed vector/matrix expression and the index may have side effects, and these will be evaluated correctly. If necessary, index expressions that have side effects will be written to a temporary variable that will replace the index. Besides dEQP tests, this change is tested by a WebGL 2 conformance test. In the case that a dynamic index is out-of-range, the base ESSL 3.00 spec allows undefined behavior. KHR_robust_buffer_access_behavior adds the requirement that program termination should not occur and that out-of-range reads must return either a value from the active program's memory or zero, and out-of-range writes should only affect the active program's memory or do nothing. This patch clamps out-of-range indices so that either the first or last item of the matrix/vector is accessed. The code is not transformed in case the it fits within the limited subset of ESSL 1.00 given in Appendix A of the spec. If the code isn't within the restricted subset, even ESSL 1.00 shaders may require this workaround. BUG=angleproject:1116 TEST=dEQP-GLES3.functional.shaders.indexing.* (all pass after change) WebGL 2 conformance tests (glsl3/vector-dynamic-indexing.html) Change-Id: I66a5e5a8d7f4267da0045f1cc2ba6b0dc7eb3f5d Reviewed-on: https://chromium-review.googlesource.com/296671 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tryjob-Request: Jamie Madill <jmadill@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org>
Olli Etuaho 80ecac9e 2015-08-25T15:37:48 Remove unused mOptimize/mDebug flags from TIntermAggregate These flags were written but they were never read. TEST=compile BUG=angleproject:1116 Change-Id: I41e3e89f13861ebda4828c76c753ca17c74c4358 Reviewed-on: https://chromium-review.googlesource.com/294931 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 3c192a78 2015-08-26T20:32:53 Revert "Remove dynamic indexing of matrices and vectors in HLSL" Seems to be failing a WebGL/ES2 CTS test in D3D9: conformance/ogles/GL/mat3/mat3_001_to_006 BUG=angleproject:1116 BUG=525188 This reverts commit 83f3411da456faac8570892e3dd7d76edf4095e5. Change-Id: Ic186f51240dbdd96ccab3f5470329cdc9727c618 Reviewed-on: https://chromium-review.googlesource.com/295730 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 83f3411d 2015-06-18T15:47:46 Remove dynamic indexing of matrices and vectors in HLSL HLSL doesn't support dynamic indexing of matrices and vectors, so replace that with helper functions that unroll dynamic indexing into switch/case and static indexing. Both the indexed vector/matrix expression and the index may have side effects, and these will be evaluated correctly. If necessary, index expressions that have side effects will be written to a temporary variable that will replace the index. Besides dEQP tests, this change is tested by a WebGL 2 conformance test. In the case that a dynamic index is out-of-range, the base ESSL 3.00 spec allows undefined behavior. KHR_robust_buffer_access_behavior adds the requirement that program termination should not occur and that out-of-range reads must return either a value from the active program's memory or zero, and out-of-range writes should only affect the active program's memory or do nothing. This patch clamps out-of-range indices so that either the first or last item of the matrix/vector is accessed. The code is not transformed in case the it fits within the limited subset of ESSL 1.00 given in Appendix A of the spec. If the code isn't within the restricted subset, even ESSL 1.00 shaders may require this workaround. BUG=angleproject:1116 TEST=dEQP-GLES3.functional.shaders.indexing.* (all pass after change) WebGL 2 conformance tests (glsl3/vector-dynamic-indexing.html) Change-Id: I024722ef4ca1e14d5ad47fdc540397e18858bed6 Reviewed-on: https://chromium-review.googlesource.com/290515 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
kmx.io     mail     discord kmxgit v0.4.0