-
Notifications
You must be signed in to change notification settings - Fork 559
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Validate location assignments #3308
Validate location assignments #3308
Conversation
The glslang tests appear to because the common test runner in glslang doesn't perform the IO mapping to actually assign auto-mapped locations. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks really good.
I think the main issue is stripping away the arrayness needs to be conditional on stage and some other decorations.
source/val/validate_interfaces.cpp
Outdated
// This function assumes a base location has been determined already. As such | ||
// any further location decorations are invalid. | ||
// TODO: if this code turns out to be slow, there is an opportunity to cache | ||
// the for a given type id. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Cache the ... result?
std::tie(is_int, is_const, value) = | ||
_.EvalInt32IfConst(type->GetOperandAs<uint32_t>(2)); | ||
if (is_int && is_const) *num_locations *= value; | ||
break; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should return an error if is not a constant?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That should be validated elsewhere. Also, the validator only really evaluates 32-bit integer constants so is_int
would be false for other integer sizes.
break; | ||
} | ||
default: | ||
break; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Should return an error here? E.g. pointer type.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That validation belongs on the variable, but it might currently be missing.
@@ -114,6 +419,16 @@ spv_result_t ValidateInterfaces(ValidationState_t& _) { | |||
} | |||
} | |||
|
|||
if (spvIsVulkanEnv(_.context()->target_env)) { | |||
for (auto& inst : _.ordered_instructions()) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
nit: Could exit as soon as you see OpTypeVoid. That way we don't scan the entire module.
has_index = true; | ||
index = dec.params()[0]; | ||
} else if (dec.dec_type() == SpvDecorationBuiltIn) { | ||
// Don't check built-ins. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Or should the validator error out if builtins have any kind of location annotation?
Vulkan 14.1.1: "Built-ins must not have any Location or Component decorations."
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I thought about that, but I'm not sure this is the best place to do it. Ideally the validator would check for all levels of variable/struct and I didn't want to do all that searching here.
source/val/validate_interfaces.cpp
Outdated
auto ptr_type = _.FindDef(ptr_type_id); | ||
auto type_id = ptr_type->GetOperandAs<uint32_t>(2); | ||
auto type = _.FindDef(type_id); | ||
// Unpack optional arrayness. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This only applies to the ones that are required to be arrays:
tessellation control and mesh per-vertex shader outputs
tessellation evaluation, tessellation control, geometry shader per-vertex inputs.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For "mesh per-vertex output", that means on Task shader the variables with PerTaskNV decoration.
http://htmlpreview.github.io/?https://github.com/KhronosGroup/SPIRV-Registry/blob/master/extensions/NV/SPV_NV_mesh_shader.html
For the inputs cases noted above, see GLSL 4.5:
Tessellation control, evaluation, and geometry shader input variables get the per-vertex values written out by output variables of the same names in the previous active shader stage. For these inputs, centroid and interpolation qualifiers are allowed, but have no effect. Since tessellation control, tessellation evaluation, and geometry shaders operate on a set of vertices, each input variable (or input block, see interface blocks below) needs to be declared as an array.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For tessellation control outputs, outputs are partitioned into per-vertex and per-patch. Per-patch variables are decorated with Patch. Per-vertex variables are always arrayed, and an invocation can only write to its own lane, i.e. with index gl_InvocationID.
Example:
#version 460
layout(vertices=5) out;
layout(location=0) out uint arr[];
layout(location=1) patch out uint count;
void main() {
const uint c = arr.length();
count = c;
arr[gl_InvocationID] = c;
}
In the SPIR-V. arr is an array and c is not:
OpCapability Tessellation
%1 = OpExtInstImport "GLSL.std.450"
OpMemoryModel Logical GLSL450
OpEntryPoint TessellationControl %main "main" %count %arr %gl_InvocationID
OpExecutionMode %main OutputVertices 5
OpSource GLSL 460
OpName %main "main"
OpName %count "count"
OpName %arr "arr"
OpName %gl_InvocationID "gl_InvocationID"
OpDecorate %count Patch
OpDecorate %count Location 1
OpDecorate %arr Location 0
OpDecorate %gl_InvocationID BuiltIn InvocationId
%void = OpTypeVoid
%3 = OpTypeFunction %void
%uint = OpTypeInt 32 0
%_ptr_Output_uint = OpTypePointer Output %uint
%count = OpVariable %_ptr_Output_uint Output
%uint_5 = OpConstant %uint 5
%_arr_uint_uint_5 = OpTypeArray %uint %uint_5
%_ptr_Output__arr_uint_uint_5 = OpTypePointer Output %_arr_uint_uint_5
%arr = OpVariable %_ptr_Output__arr_uint_uint_5 Output
%int = OpTypeInt 32 1
%_ptr_Input_int = OpTypePointer Input %int
%gl_InvocationID = OpVariable %_ptr_Input_int Input
%main = OpFunction %void None %3
%5 = OpLabel
OpStore %count %uint_5
%16 = OpLoad %int %gl_InvocationID
%17 = OpAccessChain %_ptr_Output_uint %arr %16
OpStore %17 %uint_5
OpReturn
OpFunctionEnd
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, what I implemented is slightly more permissive (since variables will consume fewer locations).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I've made a bunch of changes related to this. The validator now accounts for these implicitly arrayed types and also accounts for component decorations getting used on arrayed types to interleave variables in the location component space. @dneto0 PTAL.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for your patience.
LGTM
6a6e715
to
bb956a1
Compare
@dneto0 this latest patch fixes up some arrayed interfaces checks after digging into the remaining glslang failures. |
* Add validation that input/output locations are assigned correctly * Many tests
* Account for 4 components per location and track the combined coordinate
* Track index1 for outputs separately
* handle specifically arrayed variables (WIP) * update a test * fix comment * early exit
* Arrayed variables that specify a component get locations determined per index of the array for the sub type * Added tests that check locations and components can be assigned to interleave an array's locations and components
* Fix up which interfaces are allowed to be arrayed for various shader stages based on glslang * rebase on master
bb956a1
to
90e779f
Compare
Roll third_party/glslang/ 8397044..f5ed7a6 (30 commits) KhronosGroup/glslang@8397044...f5ed7a6 $ git log 8397044..f5ed7a6 --date=short --no-merges --format='%ad %ae %s' 2020-07-03 marcin.slusarz Add --quiet option. 2020-07-05 ShabbyX gn: Fix dawn tests in Chromium 2020-07-05 ShabbyX gn: Fix `gn gen --check` by adding missing dependency 2020-07-03 bclayton Add GLSLANG_BUILD_PIC CMake flag 2020-07-03 ShabbyX gn: Optionally disable optimizations and HLSL 2020-07-03 bclayton Don't use add_link_options() on old CMake versions 2020-07-03 bclayton License headers: s/Google/The Khronos Group 2020-07-03 bclayton Kokoro: Correct the `build_file' path to build.sh 2020-07-02 bclayton Add config for license-checker and Kokoro scripts. 2020-07-02 bclayton Fix GLSLANG_IS_SHARED_LIBRARY define 2020-07-01 bclayton Add missing copyright headers 2020-07-02 cepheus Bump revision. 2020-07-01 cepheus SPIRV-Tools and tests: Update to location-validation in SPIRV-Tools. 2020-07-01 cepheus Tests: More broadly use automapping binding/location. 2020-07-01 bclayton Add additional licenses in use to LICENSE.txt 2020-07-01 cepheus HLSL: Catch error cases earlier, preventing a later assert. 2020-06-29 bclayton glslang: Only export public interface for SOs 2020-06-29 bclayton CMake: break up glslang into smaller static libs 2020-06-30 cepheus SPV: RelaxedPrecision: use the result precision for texture sampling. 2020-06-30 cepheus SPV: RelaxedPrecision: Generalize fix KhronosGroup#2293 to cover more operations. 2020-06-24 e.proydakov Fixed GCC -Wunused-parameter in hlslParseables.cpp. 2020-06-29 bclayton CMake: Compile with -fPIC when building SOs 2020-06-29 bclayton CMake: Error on unresolved symbols 2020-06-29 bclayton Remove root kokoro/linux-*-cmake configs 2020-06-26 cepheus SPV: Fix KhronosGroup#2293: keep relaxed precision on arg passed to relaxed param 2020-06-26 cepheus SPV: Partially address KhronosGroup#2293: correct "const in" precision matching. 2020-06-25 lriki.net Add pack_matrix test 2020-06-12 lriki.net HLSL: Fix #pragma pack_matrix(row_major) not work on global uniforms 2020-06-24 bclayton Kokoro: Split linux cmake cfgs into static/shared 2020-06-23 e.proydakov Fixed msvc 2019 nmake compiler warnings with RTTI. By default cmake generates cxx_flags with `/GR` parameter. I updated CMAKE_CXX_FLAGS string and replaced `/GR` -> `/GR-` Created with: roll-dep third_party/glslang Roll third_party/googletest/ c6e309b26..356f2d264 (10 commits) google/googletest@c6e309b...356f2d2 $ git log c6e309b26..356f2d264 --date=short --no-merges --format='%ad %ae %s' 2020-07-01 absl-team Googletest export 2020-06-26 absl-team Googletest export 2020-06-25 absl-team Googletest export 2020-06-24 absl-team Googletest export 2020-06-24 absl-team Googletest export 2020-06-19 mayur.shingote Updated googletest issue tracker url. 2020-06-10 rharrison Fix build issue for MinGW 2020-02-21 nini16041988-gitbucket Add missing call for gtest_list_output_unittest_ unitTest. Add unitTest for fixed TEST_P line number. Use CodeLocation TestInfo struct. 2020-02-18 nini16041988-gitbucket Fix: shadow member 2020-02-18 nini16041988-gitbucket Add correct line number to TEST_P test cases for gtest_output. Created with: roll-dep third_party/googletest Roll third_party/re2/ 14d319322..fe8a81adc (3 commits) google/re2@14d3193...fe8a81a $ git log 14d319322..fe8a81adc --date=short --no-merges --format='%ad %ae %s' 2020-07-05 junyer Bump SONAME, which missing ')' versus unexpected ')' needed. 2020-07-03 courbet Make the compiler inline the hot RE2::DFA loop. 2020-06-25 Shikugawa change bazel cpu symbol from wasm to wasm32 Created with: roll-dep third_party/re2 Roll third_party/spirv-cross/ f9ae06512..559b21c6c (14 commits) KhronosGroup/SPIRV-Cross@f9ae065...559b21c $ git log f9ae06512..559b21c6c --date=short --no-merges --format='%ad %ae %s' 2020-07-06 dsinclair Roll deps. 2020-07-01 post MSL: Do not emit swizzled writes in packing fixups. 2020-07-01 post MSL: Workaround broken vector -> scalar access chain in MSL. 2020-07-06 post MSL: Use input attachment index directly for resource index fallback. 2020-07-06 post GLSL: Support I/O flattening with arrays as final type. 2020-07-03 post GLSL: Support multi-level struct flattening for I/O. 2020-07-01 post Run format_all.sh. 2020-07-01 post test: Use --hlsl-dx9-compatible when attempting to compile SM 3.0 shaders. 2020-06-30 post GLSL: Fix nested legacy switch workarounds. 2020-06-29 post GLSL: Implement switch on ESSL 1.0. 2020-06-29 post GLSL: Use for-loop fallback instead of do/while for legacy ESSL. 2020-06-29 post Implement context-sensitive expression read tracking. 2020-06-29 post Fix bug with control dependent expression tracking. 2020-06-23 post HLSL: Workaround FXC bugs with degenerate switch blocks. Created with: roll-dep third_party/spirv-cross Roll third_party/spirv-headers/ 11d7637..308bd07 (1 commit) KhronosGroup/SPIRV-Headers@11d7637...308bd07 $ git log 11d7637..308bd07 --date=short --no-merges --format='%ad %ae %s' 2020-06-26 dj2 Register the Tint compiler Created with: roll-dep third_party/spirv-headers Roll third_party/spirv-tools/ d4b9f57..6a4da9d (16 commits) KhronosGroup/SPIRV-Tools@d4b9f57...6a4da9d $ git log d4b9f57..6a4da9d --date=short --no-merges --format='%ad %ae %s' 2020-07-06 jaebaek Debug info preservation in copy-prop-array pass (KhronosGroup#3444) 2020-07-03 vasniktel spirv-fuzz: TransformationInvertComparisonOperator (KhronosGroup#3475) 2020-07-02 vasniktel Fix regression (KhronosGroup#3481) 2020-07-02 vasniktel spirv-fuzz: Add fuzzerutil::FindOrCreate* (KhronosGroup#3479) 2020-06-30 vasniktel spirv-fuzz: Add FuzzerPassAddCopyMemoryInstructions (KhronosGroup#3391) 2020-06-30 vasniktel spirv-fuzz: Add one parameter at a time (KhronosGroup#3469) 2020-06-29 jaebaek Fix ADCE pass bug for mulitple entries (KhronosGroup#3470) 2020-06-26 ehsannas Add gl_BaseInstance to the name mapper. (KhronosGroup#3462) 2020-06-26 andreperezmaselco.developer Implement the OpMatrixTimesScalar linear algebra case (KhronosGroup#3450) 2020-06-25 jaebaek Clear debug information for kill and replacement (KhronosGroup#3459) 2020-06-25 alanbaker Validate location assignments (KhronosGroup#3308) 2020-06-23 ehsannas Support OpCompositeExtract pattern in desc_sroa (KhronosGroup#3456) 2020-06-23 vasniktel spirv-fuzz: Implement FuzzerPassAddParameters (KhronosGroup#3399) 2020-06-23 vasniktel spirv-fuzz: Add GetParameters (KhronosGroup#3454) 2020-06-23 vasniktel spirv-fuzz: Permute OpPhi instruction operands (KhronosGroup#3421) 2020-06-22 rharrison Add support for different default/trunks in roll-deps (KhronosGroup#3442) Created with: roll-dep third_party/spirv-tools
Validate that location, component and index decorations do not cause an overlap in any interface.
Checks:
Tests include common corners such as a scalar being packed after 3-component 64-bit vector.
Tested against CTS.