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

src/compiler/translator/TranslatorHLSL.cpp

Log

Author Commit Date CI Message
Shahbaz Youssefi 800e82c6 2021-08-23T11:05:23 Translator: Validate precisions When declaring a variable, a struct field, function parameter etc, there's a precision necessarily applied to the entity being declared. AST Validation is added to enforce this. Intermediate nodes derive their precision from these entities automatically. Consistency of intermediate nodes is not validated. This is because AST transformations replace a node with a transformed one, and that may not have the same precision. Take the following code: mediump float x = ...; mediump float y = ...; ... x + y ... and assume is transformed as such: highp float driver_uniform; ... (x * driver_uniform) + y ... The addition was originally done in mediump, but would seemingly need to be done in highp after transformation. There are a number of options here: - Make sure that when nodes are replaced, the precision is unaffected. This can be intrusive, requiring temp variables. - Bubble up the new precision - Accept the discrepancy ANGLE opts for the last option, which actually respects the original shader's intended precision for operations, even if some transformation needs to temporarily evaluate an expression at a higher precision. Bug: angleproject:4889 Bug: angleproject:6132 Change-Id: Ibcde3a230de159157783b1c6d5ef1cd63ceb4d8f Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3114027 Reviewed-by: Tim Van Patten <timvp@google.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Shahbaz Youssefi 1ce78397 2021-07-30T16:51:45 Remove support for WEBGL_debug_shader_precision This extension was rejected, and the implementation was hacky. This clean up is part of an ongoing work to improve precision handling. Bug: angleproject:6059 Change-Id: If08581ec6f19cf1698ffa3dd6d248dc5e68a1d31 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/3064303 Reviewed-by: Kenneth Russell <kbr@chromium.org> Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Shahbaz Youssefi 382bf288 2020-12-24T23:56:39 Organize AST transforms per backend Most of the AST transforms are written as a workaround to an issue that affects a single backend. This change identifies such transforms and organizes them by backend. They are then only built if the respective backend is. Additionally, about half of the GL transforms are due to mac workarounds, including the large RewriteRowMajorMatrices transform. Mac-specific workarounds are additionally only built on said platform. This change reduces the ANGLE binary size: - 106KB in a Vulkan-only build on Linux - 27KB in a GL-only build on Android (60KB on Linux) Bug: chromium:1084580 Bug: chromium:1161513 Change-Id: I64b334332c0d4f848756c6538af0d8d96864c7e9 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2601346 Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Tim Van Patten <timvp@google.com>
Xinghua Cao 378653f8 2020-11-20T12:24:41 D3D: throw a perf warning for uniform block We had translated an uniform block only containing a large array member into StructuredBuffer instead of cbuffer on D3D backend for slow fxc compile performance issue with dynamic uniform indexing. This patch throw a warning if a uniform block containing a large array member fails to hit the optimization. Bug: angleproject:3682 Change-Id: I33459b559923f16a8dfb70c6f46ec52f68d96e06 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2552365 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jiajia Qin <jiajia.qin@intel.com>
Xinghua Cao dc1c1cb5 2020-08-12T13:30:26 Restrict to translate uniform block to StructuredBuffer We had translated an uniform block only containing a large array member into StructuredBuffer instead of cbuffer on D3D backend for slow fxc compile performance issue with dynamic uniform indexing. Now we add more conditions to restrict the translation. Only indexing operator is allowed to operate on this uniform block variable. And we also restrict the types of uniform block's member. Bug: angleproject:3682 Change-Id: I992b7890d84fcaa6169722af6d7e14785526d48a Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2351728 Commit-Queue: Xinghua Cao <xinghua.cao@intel.com> Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jiajia Qin <jiajia.qin@intel.com>
Tobin Ehlis a1ceb87b 2020-02-25T11:58:19 Vulkan:Refactor precision emulation code GLSL, ESSL, and HLSL translators all had the same code to emulate precision so migrate that code to base class. This is a structural change to prepare for honoring precision qualifiers in the Vulkan translator as well, which is done in a follow-on commit. Bug: angleproject:3078 Change-Id: Id85bba83543189e3bbc4002be92b056e63afc8de Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/2072644 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Tim Van Patten <timvp@google.com> Commit-Queue: Tobin Ehlis <tobine@google.com>
Xinghua Cao 0af8b596 2019-09-03T16:24:45 D3D11: Translate uniform blocks to StructuredBuffer when necessary fxc exhibits slow compile performance with dynamic cbuffer indexing. So when a uniform block contains only one large array member, which is an array of structures, translate this uniform block to a StructuredBuffer instead. Bug: angleproject:3682 TEST=angle_end2end_tests.UniformBufferTest.* Change-Id: Ife80dba8aae65b761737e095895e00a570230f88 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1782046 Commit-Queue: Xinghua Cao <xinghua.cao@intel.com> Reviewed-by: Kenneth Russell <kbr@chromium.org> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
jchen10 b3070102 2019-10-18T16:01:34 Add SH_REMOVE_DYNAMIC_INDEXING_OF_SWIZZLED_VECTOR This is a workaround for the webgl2 conformance test case WebglConformance_conformance2_glsl3_vector_dynamic_indexing_swizzled_lvalue. Dynamic indexing of swizzled lvalue like "v.zyx[i] = 0.0" is problematic on various platforms. This removes the indexing by translating it this way: void dyn_index_write_vec3(inout vec3 base, in int index, in float value){ switch (index) { case (0): (base[0] = value); return ; case (1): (base[1] = value); return ; case (2): (base[2] = value); return ; default: break; } if ((index < 0)) { (base[0] = value); return ; } { (base[2] = value); } } ... dyn_index_write_vec3(v.zyx, i, 0.0); ... Bug: chromium:709351 Change-Id: I971b38eb404209b56e6764af1063878c078a7e88 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1869109 Commit-Queue: Jie A Chen <jie.a.chen@intel.com> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Shahbaz Youssefi 472c74c6 2019-08-19T16:32:13 Translator: Allow tree validation in children of TCompiler This is to be able to perform validation inside TranslatorVulkan, even if it's through ASSERTs. Additionally, every transformation is changed such that they do their validation themselves. TIntermTraverser::updateTree() performs the validation, which indirectly validates many of three tree transformations. Some of the more ancient transformations that don't use this function directly call TCompiler::validateAST. Bug: angleproject:2733 Change-Id: Ie4af029d34e053c5ad1dc8c2c2568eecd625d344 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1761149 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org>
Stuart Morgan 9d737966 2019-08-14T12:25:12 Standardize copyright notices to project style For all "ANGLE Project" copyrights, standardize to the format specified by the style guide. Changes: - "Copyright (c)" and "Copyright(c)" changed to just "Copyright". - Removed the second half of date ranges ("Y1Y1-Y2Y2"->"Y1Y1"). - Fixed a small number of files that had no copyright date using the initial commit year from the version control history. - Fixed one instance of copyright being "The ANGLE Project" rather than "The ANGLE Project Authors" These changes are applied both to the copyright of source file, and where applicable to copyright statements that are generated by templates. BUG=angleproject:3811 Change-Id: I973dd65e4ef9deeba232d5be74c768256a0eb2e5 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1754397 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Clemen Deng f2412bca 2019-07-16T15:47:34 Get rendering to texture working Other small fixes for desktop compatibility Bug: angleproject:3620 Change-Id: I8e75bce1f850fb891c8bb6e16f79302a6d59276c Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1707932 Commit-Queue: Clemen Deng <clemendeng@google.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Lee Salzman 8ba78da0 2019-04-30T23:42:31 add support for EXT_blend_func_extended to D3D11 Change-Id: Id66868851a490d0a68a7e76280720825c4844a45 Reviewed-on: https://chromium-review.googlesource.com/c/angle/angle/+/1591192 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Xinghua Cao f3179a6a 2018-07-12T16:22:06 ES31: Implement bindImageTexture binds a single layer on D3D backend Dynamically generate image2D variables' declaration and function definition in libANGLE. Bug: angleproject:1987 TEST=angle_end2end_tests.ComputeShaderTest.* Change-Id: Idacc756f7bd15f22eccb1d689e18e997f3e74159 Reviewed-on: https://chromium-review.googlesource.com/c/1142885 Commit-Queue: Xinghua Cao <xinghua.cao@intel.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 4e712be2 2019-01-03T13:53:59 Refactor BlockLayoutEncoder APIs for std430. This splits HLSL SSBO access into two steps. First we compute a mapping from the collected SSBO variable names to TField pointers. Then during tree traversal we use a block encoding visitor class that uses the shader names to store BlockMemberInfo structures for the structures and variables. Each nested structure is traversed separately so that the BlockMemberInfo offsets are relative to the structure start rather than the enclosing block. The array stride for a structure is the size of the struct after all the alignment is included. This gives the correct results for the SSBO access chain in the HLSL code. It also will allow us to use the same encoding and visiting logic for SSBOs on the API side. Bug: angleproject:3024 Change-Id: I42b1db0e7547782ae77fe5f64a797f803f203f45 Reviewed-on: https://chromium-review.googlesource.com/c/1352731 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Yuly Novikov <ynovikov@chromium.org>
Qin Jiajia 4622905b 2018-12-10T13:31:00 ES31: Add atomic memory functions for SSBO Due to SSBO is translated to RWByteAddressBuffer in HLSL, the corresponding atomic memory functions atomic* will be translated to RWByteAddressBuffer.Interlocked*. The translation is like below: atomicAdd(instanceName.data[0], 5u); // becomes uint _ssbo_atomicAdd_uint(RWByteAddressBuffer buffer, uint loc, uint value) { uint original_value; buffer.InterlockedAdd(loc, value, original_value); return original_value; } _ssbo_atomicAdd_uint(_instanceName, 0 + 16 * 0, 5); Bug: angleproject:1951 Change-Id: If2af8bedb67a4135b443d2512d43c6058a78888d Reviewed-on: https://chromium-review.googlesource.com/c/1370676 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jiajia Qin <jiajia.qin@intel.com>
Qin Jiajia 81a880aa 2018-11-12T15:53:34 ES31: support ssbo as binary operand This patch will process ssbo as compound assignment binary operand or readonly binary operand. BUG: angleproject:1951 Change-Id: I4a0da77649d719fa08e6bf4c3d9ace58dbfb7aab Reviewed-on: https://chromium-review.googlesource.com/c/1349449 Commit-Queue: Jiajia Qin <jiajia.qin@intel.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill b980c563 2018-11-27T11:34:27 Reformat all cpp and h files. This applies git cl format --full to all ANGLE sources. Bug: angleproject:2986 Change-Id: Ib504e618c1589332a37e97696cdc3515d739308f Reviewed-on: https://chromium-review.googlesource.com/c/1351367 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Shahbaz Youssefi <syoussefi@chromium.org>
Qin Jiajia a602f906 2018-09-11T14:40:24 ES31: Support shader storage buffer in D3D-API side. Bug: angleproject:1951 Test: angle_end2end_tests Change-Id: I0d8a4f8cf00fc7fd2d85315138e2b7457fd0b90c Reviewed-on: https://chromium-review.googlesource.com/1242846 Commit-Queue: Jiajia Qin <jiajia.qin@intel.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Bryan Bernhart c55aefe3 2018-08-28T19:59:00 ES31: Support atomic functions on D3D11 - Part III This patch is the third one to support atomic functions on D3D11. In this patch we enable support for atomic function returns outside of assignments (e.g. part of arithmetic operations or to index into arrays) and when used directly initialize a variable. Note that we are still missing the functionality to be tag loops with [allow_uav_condition] as required by InterlockedCompareExchange. BUG=angleproject:2682 TEST=angle_end2end_tests Change-Id: Ia409ebb10621fd5c514cf6c76f366a320a9d9fc1 Reviewed-on: https://chromium-review.googlesource.com/1208317 Reviewed-by: Bryan Bernhart <bryan.bernhart@intel.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Corentin Wallez <cwallez@chromium.org>
Jiawei Shao 35bc74d6 2018-08-03T14:07:59 ES31: Support atomic functions on D3D11 - Part II This patch adds the support of translating atomicExchange and atomicCompSwap without return value on D3D11 back-ends. As the last parameter of the HLSL intrinsic functions InterlockedExchange and InterlockedCompareExchange is not optional, when there is a call of either atomicExchange or atomicCompSwap without return value, we add a temporary variable for it, so that we can directly translate all of such calls in outputHLSL.cpp. BUG=angleproject:2682 TEST=angle_end2end_tests Change-Id: I7e9c6d3c7d1846c865909b2f5a26592846c82582 Reviewed-on: https://chromium-review.googlesource.com/1161744 Reviewed-by: Geoff Lang <geofflang@chromium.org> Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>
Olli Etuaho 06235df9 2018-07-20T14:26:07 Make HLSL shaders use only one main function Instead of having separate main() and gl_main() functions in HLSL shaders, add initializing outputs and inputs directly to the main function that's in the AST. This works around some HLSL bugs and should not introduce name conflicts inside main() since all the user-defined variables are prefixed. BUG=angleproject:2325 TEST=angle_end2end_tests Change-Id: I5b000c96aac8f321cefe50b6a893008498eac0d5 Reviewed-on: https://chromium-review.googlesource.com/1146647 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho a07b4213 2018-03-22T16:13:13 Move AST transformations to a subdirectory Move AST transformations to compiler/translator/tree_ops. BUG=angleproject:2409 TEST=angle_unittests Change-Id: I9c620e98707d22d005da6192fe7d1b4e8030aadd Reviewed-on: https://chromium-review.googlesource.com/975550 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho c26214de 2018-03-16T10:43:11 Move AST utilities to a subdirectory Move AST related utilities to compiler/translator/tree_util. BUG=angleproject:2409 TEST=angle_unittests Change-Id: I7567c2f6f2710292029263257c7ac26e2a144ac8 Reviewed-on: https://chromium-review.googlesource.com/966032 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 4121799f 2018-03-15T11:15:33 Clean up switch/case pruning code There was some duplicate switch/case pruning in the code in PruneEmptyCases and RemoveNoOpStatementsFromTheEndOfSwitchStatements. Combine the functionality of both AST transformations into PruneEmptyCases and remove the other transformation. The tests are improved to better cover the full functionality. BUG=angleproject:2402 TEST=angle_unittests, angle_end2end_tests --gtest_filter=*Switch* Change-Id: Ib74b6b9b455769ea15650e9653a9c53635342c49 Reviewed-on: https://chromium-review.googlesource.com/964081 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 fa886975 2018-01-18T19:24:54 Fix handling array constructor statements in HLSL output Array constructors that are used as a statement by themselves need to be pruned from the AST before writing HLSL output. This fixes an assert in OutputHLSL. BUG=angleproject:2307 TEST=angle_unittests Change-Id: Ib49461a4be173f3856f5a264ac0af8d818a61798 Reviewed-on: https://chromium-review.googlesource.com/874691 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 89a69a03 2017-10-23T12:20:45 Generate performance warnings in HLSL translation Generate performance warnings for some code that undergoes heavy emulation when translated to HLSL: 1. Dynamic indexing of vectors and matrices. 2. Non-empty fall-through cases in switch/case. The warnings are generated only when code is translated to HLSL. Generating them in the parsing stage would add too much maintenance burden. Improves switch statement fall-through handling in cases where an empty fall-through case follows a non-empty one so that extra performance warnings are not generated. BUG=angleproject:1116 Change-Id: I7c85d78fe7c4f8e6042bda72ceaaf6e37dadfe6c Reviewed-on: https://chromium-review.googlesource.com/732986 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Olli Etuaho 2d88e9bc 2017-07-21T16:52:03 Guarantee that symbol nodes get unique ids The code is refactored so that symbol nodes can only be initialized with an unique id object. This prevents accidentally forgetting to create an id for a symbol node. This opens up possibilities for future optimization: For example the names and types of symbols could be stored in a central location inside the SymbolTable, and TIntermSymbol nodes would only need to store the symbol id. The symbol id could be used to look up the name and type of the node. BUG=angleproject:1490 TEST=angle_unittests Change-Id: Ib8c8675d31493037a5a28c7b36bb9d1113cc10f6 Reviewed-on: https://chromium-review.googlesource.com/580955 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho f6d242ed 2017-10-12T17:21:06 Wrap switch statements in blocks in HLSL If variables are declared inside a GLSL switch statement, they are scoped until the end of the switch statement. This is not compatible with HLSL rules, where the scoping is until the end of the case. To work around this, wrap switch statements in a block that declares the variables in HLSL. This is done after most other transformations done to the AST are complete, since some of the other transformations may introduce temporary variables. BUG=angleproject:2179 TEST=angle_end2end_tests Change-Id: Id0bb89affe103177fd3d6a6b2f3619b5e1ada0a6 Reviewed-on: https://chromium-review.googlesource.com/716381 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho bb5a7e29 2017-08-30T13:03:12 Allow length() on arbitrary array expressions This is required to pass some dEQP GLES 3.1 tests for arrays of arrays, and WebGL conformance tests were also recently fixed to require this behavior. The intent of the GLSL ES spec was not to restrict usage of length(). In practice GL drivers don't implement array length() on expressions with side effects correctly in all cases. HLSL doesn't have an array length operator either. Because of this we always remove array length ops from the AST before output. BUG=angleproject:2142 TEST=angle_unittests, angle_end2end_tests, WebGL conformance tests Change-Id: I863a92e83ac5315b013af9a5626348482bad72b3 Reviewed-on: https://chromium-review.googlesource.com/643190 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Jiajia Qin 9b11ea4f 2017-07-11T16:50:08 Gather UniformBlock and ShaderStorageBlock separately Refactor InterfaceBlocks since it only stands for UniformBlock before ES31. But for ES31, uniform block and shader storage block both belong to interface block. This CL will add GetUniformBlocks and GetShaderStorageBlocks in ShaderLang.h. Meanwhile, keep GetInterfaceBlocks which can return all the interface blocks together. BUG=angleproject:1951 Change-Id: I3036e201aadfbd490575ed03538c81bcc3793ff3 Reviewed-on: https://chromium-review.googlesource.com/582546 Commit-Queue: Geoff Lang <geofflang@chromium.org> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Olli Etuaho a5e693af 2017-07-13T16:07:26 Make unique id counter a member of TSymbolTable This makes unique id counting thread-safe. BUG=angleproject:624 TEST=angle_unittests Change-Id: Ie0f2c7e574470b39750d37d2181c790bc874b275 Reviewed-on: https://chromium-review.googlesource.com/570419 Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho 4dd06d5d 2017-07-05T12:41:06 Set proper symbol ids on temporary symbol nodes Temporary symbols used to all have symbol id 0. Now they get assigned unique symbol ids. This makes it possible to keep track of them according to the symbol id instead of their name, paving way to more robust AST handling in the future. BUG=angleproject:1490 TEST=angle_unittests Change-Id: I292e2e483cc39173524fd30a30b48c4c808442e5 Reviewed-on: https://chromium-review.googlesource.com/559335 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 5df2b9d2 2017-05-18T12:15:28 Clean up AddDefaultReturnStatements It doesn't need to use a traverser, since all function definitions can be found simply by iterating over the children of the root node. BUG=angleproject:2040 TEST=angle_end2end_tests Change-Id: I380942f90a0e73152f296b98d1fb027762d913b0 Reviewed-on: https://chromium-review.googlesource.com/508689 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 1212bcac 2016-11-23T13:44:05 translator: separate declarations after rewriting loops Otherwise when trying to add the declarations back, things might fail because the loop initialization is a sequence and not a block. BUG=668028 Change-Id: I8d84a25c25765e9655c16ce56604ae08f0f8176c Reviewed-on: https://chromium-review.googlesource.com/414305 Commit-Queue: Corentin Wallez <cwallez@chromium.org> Reviewed-by: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill acb4b81a 2016-11-07T13:50:29 translator: Put ShaderLang APIs in "sh" namespace. Working with glslang in Vulkan means we are static linking libANGLE with functions that have the same name as our translator APIs. We can fix this by scoping our APIs. We don't need to scope the types of the file, since they don't conflict. This will require a follow-up patch to remove the unscoped APIs once we switch over Chromium. We also scope TCompiler and some related classes to avoid multiply defined link errors with glslang. BUG=angleproject:1576 Change-Id: I729b19467d2ff7d374a82044b16dbebdf2dc8f16 Reviewed-on: https://chromium-review.googlesource.com/408337 Reviewed-by: Geoff Lang <geofflang@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jiawei-Shao f979524a 2016-09-21T15:19:00 Remove redundant code in RewriteTexelFetchOffset Remove useTemporaryIndex() in function RewriteTexelFetchOffset since no temporary variables will be created in this function. BUG=angleproject:1469 Change-Id: Ibed56f0ba5c89b6ae1ab1611de39514c3fbb600f Reviewed-on: https://chromium-review.googlesource.com/387198 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Jiawei-Shao e292e902 2016-09-07T10:49:01 Workaround the unary minus operator issue on Intel On some Intel D3D drivers, evaluating unary minor operator on an integer variable may get wrong answer in vertex shader. This patch works around this bug by replacing -(int) with ~(int)+1 on Windows Intel. BUG=chromium:644033 Change-Id: I0af719e84d618a33f25bcb33bde0c381fb462a31 Reviewed-on: https://chromium-review.googlesource.com/381675 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Qiankun Miao 7ebb97fc 2016-09-08T18:01:50 Use 64-bits compile options BUG=chromium:645071 Change-Id: I31825123bf4cb45fb37a93f538e8936487beb5ff Reviewed-on: https://chromium-review.googlesource.com/382712 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Jamie Madill <jmadill@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>
Shao 11e43ece 2016-08-11T09:54:08 Add a workaround for Intel drivers on glsl function texelfetchoffset GLSL function texelfetchoffset will be translated into texture.Load in ANGLE. In D3D there is a note that When one or more of the coordinates in Location exceeds the u, v, or w mipmap level dimensions of the texture, Load returns zero in all components, but in glsl there is no such restriction, which will cause the WebGL 2 dEQP test deqp/functional/gles3/shadertexturefunction/texelfetchoffset.html fail on Windows with Intel GPU. Adapted from ExpandIntegerPowExpressions.cpp, this patch adds a translation from texelFetchOffset into texelFetch to work around this issue. BUG=angleproject:1469 Change-Id: Iecfb9570472036acf5960789bdb1a63f191316be Reviewed-on: https://chromium-review.googlesource.com/367883 Commit-Queue: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Jamie Madill 666f65a1 2016-08-26T01:34:37 Revert "Revert "Cover vector dynamic indexing case in SplitSequenceOperator"" This reverts commit d2f59bb6dda4f3548e158a09540829f9ff56bba4. Change-Id: If2842bce17a0c085e2bc913ff120083fbe90497c Reviewed-on: https://chromium-review.googlesource.com/376189 Reviewed-by: Jamie Madill <jmadill@chromium.org> Commit-Queue: Jamie Madill <jmadill@chromium.org>
Kenneth Russell bccc65d3 2016-07-19T16:48:43 Flatten "#pragma STDGL invariant(all)" into varying variables. This is implemented as a compiler option which is enabled by default when outputting to desktop GLSL version 130 and greater, which does not support this #pragma in fragment shaders. As a workaround, and for better compatibility on desktop OpenGL drivers, this pragma is also flattened into the outputs of vertex shaders, and the inputs of ESSL 1.00 fragment shaders. TEST=conformance/glsl/misc/shaders-with-invariance.html with --enable-unsafe-es3-apis BUG=629622, angleproject:1293 Change-Id: Ib040230915e639971505ed496d26e804c9d64e68 Reviewed-on: https://chromium-review.googlesource.com/361792 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Yuly Novikov <ynovikov@chromium.org> Commit-Queue: Kenneth Russell <kbr@chromium.org>
Jamie Madill d2f59bb6 2016-08-17T11:50:52 Revert "Cover vector dynamic indexing case in SplitSequenceOperator" This CL was causing inverted rendering in a WebGL application. This reverts commit 7da9850643f55335a13a4663d226c73d0ac4d3b1. Vectors or matrices that are dynamically indexed as a part of an l-value generate new statements in the RemoveDynamicIndexing AST transformation step. SplitSequenceOperator needs to detect this case and split the sequence operator before statements are generated from its operands to ensure the correct order of execution. BUG=angleproject:1341 TEST=angle_end2end_tests Change-Id: I854f8cce2d46107afa62f48edf3d32c6d5c97eda Reviewed-on: https://chromium-review.googlesource.com/371643 Reviewed-by: Kenneth Russell <kbr@chromium.org> Commit-Queue: 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 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>
Geoff Lang 156d7197 2016-07-21T16:11:00 HLSL: Insert return statements into functions that are missing them. It's allowed to not have all code paths return a value in ESSL but the HLSL compiler detects this and generates an error. Work around this by adding dummy return statements at the end of each function that doesn't have one. TEST=deqp/data/gles2/shaders/functions.html BUG=angleproject:1015 BUG=478572 Change-Id: I2913f90f0994d4caf25cc43b16b9fc4e9efb19a5 Reviewed-on: https://chromium-review.googlesource.com/362085 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Commit-Queue: Geoff Lang <geofflang@chromium.org>
Olli Etuaho 7da98506 2016-07-20T18:45:09 Cover vector dynamic indexing case in SplitSequenceOperator Vectors or matrices that are dynamically indexed as a part of an l-value generate new statements in the RemoveDynamicIndexing AST transformation step. SplitSequenceOperator needs to detect this case and split the sequence operator before statements are generated from its operands to ensure the correct order of execution. BUG=angleproject:1341 TEST=angle_end2end_tests Change-Id: I84e41a59c88fb5d0111669cab60312b930531a22 Reviewed-on: https://chromium-review.googlesource.com/361695 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho e1d199bb 2016-07-19T17:14:27 Split sequence operator when necessary Split sequence operators if some of their operands generate statements in subsequent AST transformations to guarantee the right order of execution. For now, this is supported for expressions that return arrays and unfolded short-circuiting operators, which is enough to get WebGL 2 tests passing. A trickier corner case with dynamic indexing of vectors as an l-value is left to be addressed later. BUG=angleproject:1341 TEST=angle_end2end_tests, WebGL 2 conformance test: conformance2/glsl3/array-in-complex-expression.html Change-Id: I9301edd3366be7607a8aa4c42a5ec13928749e10 Reviewed-on: https://chromium-review.googlesource.com/361694 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho b741c761 2016-06-29T15:49:22 Support precision emulation on HLSL Re-submit with missing virtual destructor and angle::NonCopyable added. Add precision emulation support to HLSL 4.1 output. This makes it possible for developers to test their shaders for precision issues easily on Chrome on Windows without having to use the GL backend. The patch has been verified with Chrome on Windows to reproduce some precision bugs in real-world WebGL content, including old versions of the babylon.js library. The EmulatePrecision AST transformation still relies on writing out raw shader code for the rounding functions, with raw HLSL code added alongside pre-existing GLSL and ESSL code. In some ways it would be nicer to do the EmulatePrecision step as a pure AST transformation, but on the other hand the raw code is much more readable and easier to optimize. To better support multiple output languages in EmulatePrecision, add a RoundingHelperWriter class that has different subclasses for writing the rounding functions in different languages. The unit tests are expanded to cover the HLSL output of precision emulation. The parts of the tests that require the HLSL output are only active on Windows where ANGLE_ENABLE_HLSL define is added to the unit tests. Putting the HLSL tests in an entirely separate file is a worse alternative, since it would require either a lot of code duplication or add a lot of boilerplate to the individual tests. BUG=angleproject:1437 TEST=angle_unittests Change-Id: I47d501037c206f4bd8b976d3acab9b21c717084c Reviewed-on: https://chromium-review.googlesource.com/360152 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 3fdaf6f2 2016-07-13T15:07:41 Revert "Support precision emulation on HLSL" Forgot to add virtual destructors. BUG=angleproject:1437 This reverts commit a42e8b2cb9d0857f53c0490b5be3bf25b4e1f827. Change-Id: If33fecfeca9947deedf4668c64dbadf25a5dc5eb Reviewed-on: https://chromium-review.googlesource.com/360122 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho a42e8b2c 2016-06-29T15:49:22 Support precision emulation on HLSL Add precision emulation support to HLSL 4.1 output. This makes it possible for developers to test their shaders for precision issues easily on Chrome on Windows without having to use the GL backend. The patch has been verified with Chrome on Windows to reproduce some precision bugs in real-world WebGL content, including old versions of the babylon.js library. The EmulatePrecision AST transformation still relies on writing out raw shader code for the rounding functions, with raw HLSL code added alongside pre-existing GLSL and ESSL code. In some ways it would be nicer to do the EmulatePrecision step as a pure AST transformation, but on the other hand the raw code is much more readable and easier to optimize. To better support multiple output languages in EmulatePrecision, add a RoundingHelperWriter class that has different subclasses for writing the rounding functions in different languages. The unit tests are expanded to cover the HLSL output of precision emulation. The parts of the tests that require the HLSL output are only active on Windows where ANGLE_ENABLE_HLSL define is added to the unit tests. Putting the HLSL tests in an entirely separate file is a worse alternative, since it would require either a lot of code duplication or add a lot of boilerplate to the individual tests. BUG=angleproject:1437 TEST=angle_unittests Change-Id: Ia4ba0374cd415908f16f34752321af1cb93525a5 Reviewed-on: https://chromium-review.googlesource.com/358473 Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 9696316d 2016-03-21T11:54:33 Support ESSL structs containing samplers on D3D Since HLSL can't natively handle samplers in structs, samplers need to be extracted out of structs into separate variables in the translated shader code. In HLSL 4.1, samplers that were in structs go into the normal sampler arrays and are identified by index constants. In other HLSL versions, samplers that were in structs are translated as uniform variables. These transformations are done inside the HLSL output classes, not as tree transformations. This helps to keep the uniform API provided by the shader translator intact. Wherever a struct containing samplers is passed into a user-defined function, the translated HLSL code passes the separate sampler variables alongside a struct where the samplers have been removed. The D3D backend in libANGLE queries the uniform registers of any samplers that were in uniform structs, and adds them to the register maps, so that correct sampler state gets assigned to them. The extracted sampler variables are prefixed with "angle_" instead of the usual "_" to prevent any name conflicts between them and regular variables. BUG=angleproject:504 TEST=angle_end2end_tests, dEQP-GLES*.functional.shaders.struct.uniform.* (all pass), dEQP-GLES*.functional.uniform_api.* (most now pass) Change-Id: Ib79cba2fa0ff8257a973d70dfd917a64f0ca1efb Reviewed-on: https://chromium-review.googlesource.com/333743 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Corentin Wallez <cwallez@chromium.org> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 9b4e8626 2015-12-22T15:53:22 Redesign samplers in shaders on D3D11 Translation of samplers to HLSL on D3D11 is changed as follows: Instead of passing around HLSL sampler and HLSL texture references in shaders, all references to ESSL samplers are converted to constant indices within the shader body. Each ESSL sampler is identified by an unique index. In the code generated to implement ESSL texture functions, these indices are used to index arrays of HLSL samplers and HLSL textures to get the sampler and texture to use. HLSL textures and samplers are grouped into arrays by their types. Each unique combination of a HLSL texture type + HLSL sampler type gets its own array. To convert a unique sampler index to an index to one of these arrays, a constant offset is applied. In the most common case of a 2D texture and a regular (non-comparison) sampler, the index offset is always zero and is omitted. The end goal of this refactoring is to make adding extra metadata for samplers easier. The unique sampler index can be used in follow-up changes to index an array of metadata passed in uniforms, which can contain such things as the base level of the texture. This does not solve the issues with samplers in structs. The interface from the point of view of libANGLE is still exactly the same, the only thing that changes is how samplers are handled inside the shader. On feature level 9_3, the D3D compiler has a bug where it can report that the maximum sampler index is exceeded when in fact it is not. This can happen when an array of samplers is declared in the shader. Because of this the new approach can't be used on D3D11 feature level 9_3, but it will continue using the old approach instead. BUG=angleproject:1261 TEST=angle_end2end_tests, dEQP-GLES3.functional.shaders.texture_functions.* (no regressions) dEQP-GLES3.functional.texture.units.* (no regressions) Change-Id: I5fbb0c4280000202dc2795a628b56bd8194ef96f Reviewed-on: https://chromium-review.googlesource.com/320571 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com> Commit-Queue: Olli Etuaho <oetuaho@nvidia.com> Tryjob-Request: Olli Etuaho <oetuaho@nvidia.com>
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>
Corentin Wallez d4b5054d 2015-09-28T12:19:26 compiler: Rewrite do-while loops as while loops This works around a Mac driver shader compiler bug that makes many do-while loops cause GPU-hangs when ran. BUG=angleproject:891 Change-Id: I29828d6ea9e887ad0ed0c577f1deb41fb632a900 Reviewed-on: https://chromium-review.googlesource.com/302465 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Corentin Wallez <cwallez@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>
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>
Olli Etuaho 4d59f3c6 2015-05-28T17:33:53 Make false blocks produced by RewriteElseBlocks sequence nodes All child nodes of selection should be sequence nodes, so that they will output braces and extra braces can be removed from HLSL output. Also make RewriteElseBlocks to reuse common IntermTraverser functionality to simplify the code. TEST=WebGL conformance tests on D3D9 BUG=angleproject:1013 Change-Id: Iafdc05468b22d110abcd020cf52c646dd98fb4a0 Reviewed-on: https://chromium-review.googlesource.com/273608 Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org>
Olli Etuaho 4f1af784 2015-05-25T11:55:07 Handle function calls where returned array is not used This is done by declaring a temporary variable which is passed as the array out parameter defined by ArrayReturnValueToOutParameter. SeparateExpressionsReturningArrays takes care of transforming the rest of the cases where a function call returns an array into form that ArrayReturnValueToOutParameter can handle. BUG=angleproject:971 TEST=WebGL 2 conformance tests, dEQP-GLES3.functional.shaders.arrays.* Change-Id: I70c07712ba5cd91efb4c2e575ecc49b9ef71bfd7 Reviewed-on: https://chromium-review.googlesource.com/273111 Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho a5316a18 2015-05-15T15:25:16 Separate expressions returning arrays for HLSL output Complex array expressions need to be broken down in HLSL output so that they are built out of simple combinations of operations and symbols. In practice this means that array constructors, array assignments and functions that return arrays inside complex expressions need to be replaced by symbols. After this change, ANGLE passes all dEQP-GLES3.functional.shaders.arrays tests. The old SimplifyArrayAssignment stub is removed, the new name SeparateExpressionsReturningArrays more closely reflects what the function needs to do. BUG=angleproject:971, angleproject:941 TEST=dEQP-GLES3.functional.shaders.arrays.*, WebGL 2 conformance tests Change-Id: Iab8dde72b1126dc2f958ffb5b1b830fe3ce25912 Reviewed-on: https://chromium-review.googlesource.com/272122 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho d4f303ee 2015-05-20T17:09:06 Refactoring: Make creating temporary symbols in AST traversal reusable Temporary symbols will also be needed to store temporary arrays when complex array expressions are unfolded. Also clear tree update related structures at the end of updateTree(), so that the traverser can be reused for several rounds of replacement more easily, and remove unnecessary InVisit step from UnfoldShortCircuitToIf. BUG=angleproject:971 TEST=angle_end2end_tests, WebGL conformance tests Change-Id: Iecdd3008d43f01b02fe344ccde8614f70e6c0c65 Reviewed-on: https://chromium-review.googlesource.com/272121 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho a6f22096 2015-05-08T18:31:10 Make UnfoldShortCircuit to change AST instead of writing output This is needed to make way for further AST transformations to handle array expressions that need to work correctly together with unfolding short- circuiting operators. This also improves the maintainability of HLSL output by isolating the unfolding into a separate compilation step. The new version of UnfoldShortCircuit traverser will traverse the tree until an expression that needs to be unfolded is encountered. It then unfolds it and gets reset. The traverser will be run repeatedly until no more operations to unfold are found. This helps with keeping the traverser's design relatively simple. All declarations are separated to single declarations before short-circuit unfolding is run. Previously OutputHLSL already output every declaration separately. BUG=angleproject:960 TEST=WebGL conformance tests, angle_unittests, angle_end2end_tests Change-Id: Id769be396adbd4c0223e418980dc464dd855f019 Reviewed-on: https://chromium-review.googlesource.com/270460 Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Jamie Madill <jmadill@chromium.org>
Olli Etuaho a8c414ba 2015-04-16T15:51:03 Add basic support for arrays as return values in HLSL output In HLSL output, user-defined functions that have an array as their return value get changed so that they have the array as an out parameter instead. Still missing: support for calling a function that has array as a return value without assigning the array. Also support for assignments inside complex expressions. TEST=dEQP-GLES3.functional.shaders.arrays.return.* BUG=angleproject:941 Change-Id: I79f5170139116a3dcfb2be2df5f0f79a3d955ca8 Reviewed-on: https://chromium-review.googlesource.com/266003 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho 822fa84e 2015-04-16T14:26:10 Support array initialization in HLSL output Do this by separating each array initialization into a declaration and an assignment. Array assignment is already supported in HLSL output by replacing it with a function call. The functionality is tested by the struct array constructor tests in dEQP. BUG=angleproject:941 TEST=dEQP-GLES3.functional.shaders.arrays.constructor.* Change-Id: Ida84fc343b767bea8b2d04e91c60cb8197d39039 Reviewed-on: https://chromium-review.googlesource.com/266002 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho fc0e2bc0 2015-04-16T13:39:56 Put each array declarator into a separate declaration in HLSL output Since HLSL doesn't support arrays as l-values, HLSL output needs to split declarators that initialize arrays to variable declaration and assignment implemented via a function call. To prepare for this, it is necessary that each declarator has its own declaration. BUG=angleproject:941 TEST=angle_end2end_tests, WebGL conformance tests Change-Id: I43dee487578561c01dbde90c2f55a93dda2f057a Reviewed-on: https://chromium-review.googlesource.com/266001 Reviewed-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Olli Etuaho b2d6a9be 2015-03-30T16:00:53 Stub simplifying array assignment for HLSL output Add an AST traverser that rejects shaders with expressions that use the return value of an assignment where an array is assigned to another array. In the future this should be made into a tool that can simplify these expressions so that return values of array assignments are not used. In its current form this code prevents OutputHLSL from producing garbage code when the original code contains expressions like a = (b = c); where a, b, and c are all arrays. BUG=angleproject:960 Change-Id: I11353d7ed7160c853e58a0ef3471ca439cb314c8 Reviewed-on: https://chromium-review.googlesource.com/263070 Reviewed-by: Jamie Madill <jmadill@chromium.org> Tested-by: Olli Etuaho <oetuaho@nvidia.com> Reviewed-by: Geoff Lang <geofflang@chromium.org>
Olli Etuaho a3a5cc6a 2015-02-13T13:12:22 Expose the IntermNode tree generated in the compiler for testing This refactoring makes it possible for tests to access the IntermNode tree produced by compilation by calling compileTree(). Removing ParseContext usage from OutputHLSL has the additional benefit of better separation between parsing and output. BUG=angle:916 Change-Id: Ib40954832316328772a5c1dcbbe6b46b238e4e65 Reviewed-on: https://chromium-review.googlesource.com/249723 Reviewed-by: Olli Etuaho <oetuaho@nvidia.com> Tested-by: Olli Etuaho <oetuaho@nvidia.com>
Jamie Madill 54ad4f81 2014-09-03T09:40:46 Use the CollectVariables path on the HLSL translator. This approach consolidates our two methods, and lets us reuse the same code for both methods of variable collection. BUG=angle:466 Change-Id: Ie92f76ff0b6d0d0dbfd211a234d0ab86290fa798 Reviewed-on: https://chromium-review.googlesource.com/213504 Tested-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Nicolas Capens <capn@chromium.org>
Jamie Madill 9fe25e9e 2014-07-18T10:33:08 Add a uniform register query to the translator. This returns the uniform index that we assigned for default uniforms. All the dependent structure offsets can be determined from the base register, so we won't have to store uniform information in the shader variable. BUG=angle:466 Change-Id: I0dd05251e8dba00c20d09fd865dfb150de56738e Reviewed-on: https://chromium-review.googlesource.com/207254 Tested-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Nicolas Capens <capn@chromium.org>
Jamie Madill 4e1fd412 2014-07-10T17:50:10 Store a map of interface block registers. The shader translator can return the assigned register for a block via a new API. This will let us delete the member variable in interface blocks for the register -- a nice thing for GLSL. BUG=angle:466 Change-Id: I9bc38e0cd031e32f90787be42c2324fc7c79dbf9 Reviewed-on: https://chromium-review.googlesource.com/206828 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Nicolas Capens <capn@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill ed27c729 2014-07-02T15:31:23 Consolidate shader variable storage in Compiler. The Compiler base class now stores all the shader variables and interface block information, instead of duplicating the information in the HLSL translator. BUG=angle:466 Change-Id: Ia69079fde64fbd6b0cbfc66defd5e37d99ee3e6e Reviewed-on: https://chromium-review.googlesource.com/206020 Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Nicolas Capens <capn@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 183bde55 2014-07-02T15:31:19 Return shader variable information using GLenum values. Instead of duplicating GL header define values, explictly return GLenum for variable queries in the shader inspection API. This reduces the duplicate defines in the shader compiler header. BUG=angle:466 Change-Id: Iddaaff597b188251fa2e546f352bf77ab3ac43bc Reviewed-on: https://chromium-review.googlesource.com/205860 Tested-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org> Reviewed-by: Nicolas Capens <capn@chromium.org>
Jamie Madill ce505553 2014-06-26T14:56:33 Revert "Return variable type queries using GLenum values." Breaks the FYI bots until we get Chromium patched. BUG=angle:466 This reverts commit 53221f5a0382887155d90f7f286e41190d4f5bfb. Change-Id: Ib28548df5c10a6f76f46e4cf8f2013dca5cf0ee2 Reviewed-on: https://chromium-review.googlesource.com/205850 Reviewed-by: Shannon Woods <shannonwoods@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Jamie Madill 53221f5a 2014-06-25T16:04:59 Return variable type queries using GLenum values. Instead of duplicating GL header define values, explictly return GLenum for variable queries in the shader inspection API. This reduces the duplicate defines in the shader compiler header. BUG=angle:466 Change-Id: If631b20ce68747297a946e1371def7709027a613 Reviewed-on: https://chromium-review.googlesource.com/204937 Tested-by: Jamie Madill <jmadill@chromium.org> Reviewed-by: Nicolas Capens <nicolascapens@chromium.org> Reviewed-by: Zhenyao Mo <zmo@chromium.org>
Jamie Madill 68fe74aa 2014-05-27T12:56:01 Add a compiler query for the translator output type. This is useful for determining if we are compiling to a D3D9 or D3D11 shader outside of the internal translator classes. BUG=angle:656 Change-Id: Ib1c1d3de569edaa2b65c24c09d05aa4dd229d3e4 Reviewed-on: https://chromium-review.googlesource.com/201564 Reviewed-by: Nicolas Capens <nicolascapens@chromium.org> Reviewed-by: Shannon Woods <shannonwoods@chromium.org> Tested-by: Jamie Madill <jmadill@chromium.org>
Geoff Lang 17732823 2013-08-29T13:46:49 Moved the compiler source files into directories based on their project and added a compiler.gypi to generate the compiler projects.
kmx.io     mail     discord kmxgit v0.4.0