Commits · fbb12e1932ac24583652c263049c7a4a90997fee · Chen Yisong / swiftshader

02 Jun, 2020 3 commits

SpirvShaderDebugger: Don't display SSA values with DebugInfo · fbb12e19

authored May 27, 2020

If the shader contains `OpenCL.DebugInfo.100` rich debug info instructions, don't also include raw SSA values in the watch window.

Bug: b/148401179
Change-Id: Id70a702bbb24fb0bf79faa55cd22c0a87c973f62
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45350
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

fbb12e19

Make ManagedStatic registration thread-safe · 141d5bdf

authored May 25, 2020

As part of chromium:944811, LLVM_ENABLE_THREADS was turned off, but
since https://swiftshader-review.googlesource.com/c/SwiftShader/+/33484
we may invoke Reactor JIT-compilation on multiple threads. Instead of
re-enabling LLVM_ENABLE_THREADS wholesale, just make ManagedStatic
thread-safe.

Since sys::Mutex usage is not thread-safe without LLVM_ENABLE_THREADS,
use standard C++ synchronization primitives instead.

Note this change only affects the Subzero JIT backend, not the LLVM JIT.

Bug: chromium:950343
Change-Id: Ie70be4adb2c395459cb9bbd4f6699e0b04969c74
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45328
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

141d5bdf

Regres: Test changes from external contributors if reviewed by Googler · 2fbd57b6

authored Jun 02, 2020

Previously only changes authored by Googlers would be tested, leaving
external contributions untested until the next 'daily' run.

This change enables testing of external contributions when a Googler
added a Code-Review+1, Code-Review+2, or Presubmit-Ready+1 vote.

Change-Id: I2f5c327848ad507d75967c9fd7f16085d0729a1a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45508Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

2fbd57b6

01 Jun, 2020 2 commits

Remove generation of OpenCLDebugInfo100.h · 11dd7183

authored May 25, 2020

This header is now pre-built as part of third_party/SPIRV-Headers
It was first added by https://swiftshader-review.googlesource.com/c/SwiftShader/+/43035

Bug: b/150936967
Change-Id: I2931cb6475fd2d99cdc015d1a2e3bd759f9255d6
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45309Tested-by: Nicolas Capens <nicolascapens@google.com>
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Ben Clayton <bclayton@google.com>

11dd7183

Fix deallocation of uninitialized pointer · 6f5ff495

authored Jun 01, 2020

When pCreateInfo->setLayoutCount is 0, descriptorSets[0].bindings would
not be assigned the initial bindingStorage value, which is the start of
the additional memory that is allocated for the PipelineLayout object.
It is used in PipelineLayout::destroy() for deallocating said memory,
which leads to attempting to free an initialized pointer.

This was a regression caused by https://swiftshader-review.googlesource.com/c/SwiftShader/+/44328

Bug: b/154522740
Change-Id: I98f63d58e9d64b36f15d966afd1bc70c1ec97f59
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45448
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Trevor Black <vantablack@google.com>

6f5ff495

28 May, 2020 1 commit

Don't inline rr::Variable methods · 471c120a

authored May 28, 2020

This is not a template class, so its methods don't have to be inlined.
Nor are they trivial enough to save us anything in performance or
binary size if they were inlined. Instead, this could reduce binary
size a bit because there are a large number of rr::Variable uses.

Bug: chromium:1086501
Change-Id: I1a6d078628ed7e634e0d90d6d8455e7d2938d134
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45388
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

471c120a

26 May, 2020 3 commits

Update SPIR-V Tools to 55af3902f · 9e718f96

authored May 26, 2020

Changes:
55af3902fc2 Fix function use (#3372)
9cb2571a184 spirv-val: allow DebugInfoNone for DebugTypeComposite.Size (#3374)
4386ef4234e Add validation support for ImageGatherBiasLodAMD (#3363)
b0264b87ffb Fix validation failure on OpDecorationGroup (#3365)
4410272bdda Remove deprecated interfaces from instrument passes (#3361)
50b15578866 Preserve debug info in inline pass (#3349)
4dbe18b0c86 Reject folding comparisons with unfoldable types. (#3370)
55193b06e5e Improve build instructions for fuzzer (#3364)
3c47dac2820 Add unrolling to performance passes (#3082)
2b987c49a4e Handle OpConstantNull in ssa-rewrite (#3362)
95df4c9643c Add in a bunch of missed files to the BUILD.gn (#3360)
90930cb3115 Remove stale entries from BUILD.gn (#3358)
18ba3d9a353 allow cross compiling for Windows Store, UWP, etc. (#3330)
2f69ea849aa spirv-fuzz: Remove FuzzerPassAddUsefulConstructs (#3341)
522561619a9 Add support for StorageBuffer (#3348)
b75dbf82a69 Prevent Effcee install his things when build spirv-tools with testing enabled (#3256)
85c7e7956bf Don't register edges twice in merge return (#3350)
bd0a2da946c Revert "Revert "[spirv-opt] refactor inlining pass (#3328)" (#3342)" (#3345)
31182763704 spirv-reduce: Remove unused struct members (#3329)
a6b0e132ecc Add adjust branch weights transformation (#3336)
d4fac3451b7 Revert "[spirv-opt] refactor inlining pass (#3328)" (#3342)
233246bc9c5 [spirv-opt] refactor inlining pass (#3328)
2992386ebea spirv-reduce: Remove unused uniforms and similar (#3321)
a9f2a145e65 spirv-fuzz: Fix to fact manager (#3339)
045a26e6e37 spirv-fuzz: Get rid of unnecessary template method (#3340)
63fa9114a93 Do merge return if the return is not at the end of the function. (#3337)
c8590c18bd0 Preserve debug info for wrap-opkill (#3331)
d2b48621949 Validate ShaderCallKHR memory scope (#3332)
2e1d208ed9d spirv-fuzz: Do not allow adding stores to read-only pointers (#3316)
54fb17b2d30 reduce: increase default step limit (#3327)
49842b88eec Generalize IsReadOnlyVariable() to apply to pointers (#3325)
49ca250b44c Delete nullptr in function bb list immedietly (#3326)
d0a87194f7b Set DebugScope for termination instructions (#3323)
f278b467dfd spirv-fuzz: Do not outline regions that end with a loop header (#3312)
23d68608b00 vscode: Handle '|' chains on BitEnum / ValueEnum (#3309)
42268740c95 Add debug information analysis (#3305)
eed48ae479d Add spvtools::opt::Operand::AsLiteralUint64 (#3320)
94d6002dc53 spirv-fuzz: Pass on validator options during shrinking (#3317)
88faf63ad3c spirv-fuzz: Clamp statically out-of-bounds accesses in code donation (#3315)
b74199a22d4 spirv-fuzz: Fix memory management in the fact manager (#3313)
d158ffe5405 spirv-fuzz: Do not replace the Sample argument in OpImageTexelPointer (#3311)
5547553a0c7 Allow various validation options to be passed to spirv-opt (#3314)
30ffe62e257 typo fix: in README.md exectuable->executable (#3306)
67f4838659f spirv-fuzz: Make handling of synonym facts more efficient (#3301)
61b7de3c39f Remove unreachable code. (#3304)
ed96301c6c4 spirv-fuzz: Fix to outliner (#3302)
c018fc6ae66 spirv-fuzz: Do not outline regions that produce pointer outputs (#3291)
f460cca9dca spirv-fuzz: Handle OpRuntimeArray when replacing ids with synonyms (#3292)
2f180468a71 spirv-fuzz: Handle image storage class in donation (#3290)
f82d47003e7 spirv-fuzz: Respect rules for OpSampledImage (#3287)
7ce2db1763b spirv-fuzz: Fix comment. (#3300)
7d65bce0bbe Sampled images as read-only storage (#3295)
2a2bdbd5d72 Remove implicit fallthrough (#3298)
49566448944 Add tests for recently added command line option (#3297)
ca5751590ed If SPIRV-Headers is in our tree, include it as subproject (#3299)
e70d25f6fa5 Struct CFG analysus and single block loop (#3293)
000040e707a Preserve debug info in eliminate-dead-functions (#3251)
c531099eb34 Update acorn version (#3294)
34be23373b9 Handle more cases in dead member elim (#3289)
d0490ef080c Fix pch macro to ignore clang-cl (#3283)
538512e8e89 spirv-fuzz: Improve the handling of equation facts (#3281)
183e3242a36 spirv-fuzz: Handle more general SPIR-V in donation (#3280)
4af38c49bfe spirv-fuzz: Improve support for compute shaders in donation (#3277)
e95fbfb1f50 spirv-fuzz: Transformation to add OpConstantNull (#3273)
5d491a7ed66 spirv-fuzz: Handle isomorphic types property in composite construction (#3262)
bfd25ace084 spirv-fuzz: Limit adding of new variables to 'basic' types (#3257)
f28cdeff16f spirv-fuzz: Only replace regular ids with synonyms (#3255)
8d4261bc440 spirv-fuzz: Introduce TransformationContext (#3272)
2fdea57d19d spirv-fuzz: Add validator options (#3254)
af01d57b5e3 Update dominates to check for null nodes (#3271)
f20c0d7971c Set wrapped kill basic block's parent (#3269)
c37c94929bf Validate Buffer and BufferBlock apply only to struct types (#3259)

Commands:
./third_party/update-spirvtools.sh

Bug: b/123642959
Change-Id: Id033a5049da5a8b7c7fbf5f1896a985abfbd4ad8

9e718f96

Squashed 'third_party/SPIRV-Tools/' changes from fd773eb50d6..55af3902fc2 · 38e46911

authored May 26, 2020

55af3902fc2 Fix function use (#3372)
9cb2571a184 spirv-val: allow DebugInfoNone for DebugTypeComposite.Size (#3374)
4386ef4234e Add validation support for ImageGatherBiasLodAMD (#3363)
b0264b87ffb Fix validation failure on OpDecorationGroup (#3365)
4410272bdda Remove deprecated interfaces from instrument passes (#3361)
50b15578866 Preserve debug info in inline pass (#3349)
4dbe18b0c86 Reject folding comparisons with unfoldable types. (#3370)
55193b06e5e Improve build instructions for fuzzer (#3364)
3c47dac2820 Add unrolling to performance passes (#3082)
2b987c49a4e Handle OpConstantNull in ssa-rewrite (#3362)
95df4c9643c Add in a bunch of missed files to the BUILD.gn (#3360)
90930cb3115 Remove stale entries from BUILD.gn (#3358)
18ba3d9a353 allow cross compiling for Windows Store, UWP, etc. (#3330)
2f69ea849aa spirv-fuzz: Remove FuzzerPassAddUsefulConstructs (#3341)
522561619a9 Add support for StorageBuffer (#3348)
b75dbf82a69 Prevent Effcee install his things when build spirv-tools with testing enabled (#3256)
85c7e7956bf Don't register edges twice in merge return (#3350)
bd0a2da946c Revert "Revert "[spirv-opt] refactor inlining pass (#3328)" (#3342)" (#3345)
31182763704 spirv-reduce: Remove unused struct members (#3329)
a6b0e132ecc Add adjust branch weights transformation (#3336)
d4fac3451b7 Revert "[spirv-opt] refactor inlining pass (#3328)" (#3342)
233246bc9c5 [spirv-opt] refactor inlining pass (#3328)
2992386ebea spirv-reduce: Remove unused uniforms and similar (#3321)
a9f2a145e65 spirv-fuzz: Fix to fact manager (#3339)
045a26e6e37 spirv-fuzz: Get rid of unnecessary template method (#3340)
63fa9114a93 Do merge return if the return is not at the end of the function. (#3337)
c8590c18bd0 Preserve debug info for wrap-opkill (#3331)
d2b48621949 Validate ShaderCallKHR memory scope (#3332)
2e1d208ed9d spirv-fuzz: Do not allow adding stores to read-only pointers (#3316)
54fb17b2d30 reduce: increase default step limit (#3327)
49842b88eec Generalize IsReadOnlyVariable() to apply to pointers (#3325)
49ca250b44c Delete nullptr in function bb list immedietly (#3326)
d0a87194f7b Set DebugScope for termination instructions (#3323)
f278b467dfd spirv-fuzz: Do not outline regions that end with a loop header (#3312)
23d68608b00 vscode: Handle '|' chains on BitEnum / ValueEnum (#3309)
42268740c95 Add debug information analysis (#3305)
eed48ae479d Add spvtools::opt::Operand::AsLiteralUint64 (#3320)
94d6002dc53 spirv-fuzz: Pass on validator options during shrinking (#3317)
88faf63ad3c spirv-fuzz: Clamp statically out-of-bounds accesses in code donation (#3315)
b74199a22d4 spirv-fuzz: Fix memory management in the fact manager (#3313)
d158ffe5405 spirv-fuzz: Do not replace the Sample argument in OpImageTexelPointer (#3311)
5547553a0c7 Allow various validation options to be passed to spirv-opt (#3314)
30ffe62e257 typo fix: in README.md exectuable->executable (#3306)
67f4838659f spirv-fuzz: Make handling of synonym facts more efficient (#3301)
61b7de3c39f Remove unreachable code. (#3304)
ed96301c6c4 spirv-fuzz: Fix to outliner (#3302)
c018fc6ae66 spirv-fuzz: Do not outline regions that produce pointer outputs (#3291)
f460cca9dca spirv-fuzz: Handle OpRuntimeArray when replacing ids with synonyms (#3292)
2f180468a71 spirv-fuzz: Handle image storage class in donation (#3290)
f82d47003e7 spirv-fuzz: Respect rules for OpSampledImage (#3287)
7ce2db1763b spirv-fuzz: Fix comment. (#3300)
7d65bce0bbe Sampled images as read-only storage (#3295)
2a2bdbd5d72 Remove implicit fallthrough (#3298)
49566448944 Add tests for recently added command line option (#3297)
ca5751590ed If SPIRV-Headers is in our tree, include it as subproject (#3299)
e70d25f6fa5 Struct CFG analysus and single block loop (#3293)
000040e707a Preserve debug info in eliminate-dead-functions (#3251)
c531099eb34 Update acorn version (#3294)
34be23373b9 Handle more cases in dead member elim (#3289)
d0490ef080c Fix pch macro to ignore clang-cl (#3283)
538512e8e89 spirv-fuzz: Improve the handling of equation facts (#3281)
183e3242a36 spirv-fuzz: Handle more general SPIR-V in donation (#3280)
4af38c49bfe spirv-fuzz: Improve support for compute shaders in donation (#3277)
e95fbfb1f50 spirv-fuzz: Transformation to add OpConstantNull (#3273)
5d491a7ed66 spirv-fuzz: Handle isomorphic types property in composite construction (#3262)
bfd25ace084 spirv-fuzz: Limit adding of new variables to 'basic' types (#3257)
f28cdeff16f spirv-fuzz: Only replace regular ids with synonyms (#3255)
8d4261bc440 spirv-fuzz: Introduce TransformationContext (#3272)
2fdea57d19d spirv-fuzz: Add validator options (#3254)
af01d57b5e3 Update dominates to check for null nodes (#3271)
f20c0d7971c Set wrapped kill basic block's parent (#3269)
c37c94929bf Validate Buffer and BufferBlock apply only to struct types (#3259)

git-subtree-dir: third_party/SPIRV-Tools
git-subtree-split: 55af3902fc24db889b0ef8010a83efca04a6352c

38e46911

Document helper invocations being considered active · 79dcb2a5

authored May 25, 2020

See https://swiftshader-review.googlesource.com/c/SwiftShader/+/42088/7#message-18bed01232babd6cf5b5770b138d9034073fec95
for rationale.

Bug: b/151137030
Change-Id: Ieef8636734be1dc19ced8f572daea4990231106b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45288
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>

79dcb2a5

25 May, 2020 4 commits

Move the Reactor 'Sample' unit test back to the top · d2af84ff

authored May 14, 2020

Made 'Sample' come first again and thereby made it easier to use for
quick experiments and smoke testing.

Bug: b/156739423
Change-Id: Ic821aba66afbac97556b072750e24747af1010de
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45208
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

d2af84ff

Unconditionally define the RValue copy constructor · 04e6b3b2

authored May 13, 2020

Previously we relied on the default copy constructor to be generated
when ENABLE_RR_DEBUG_INFO is disabled. This is not necessary as we can
just use the straightforward manual definition of it in either case.
Note RR_DEBUG_INFO_EMIT_VAR() is a no-op when ENABLE_RR_DEBUG_INFO is
disabled.

Bug: b/155302798
Change-Id: I04c020a795b4333e3c3d0a7ed375d21d9382e3d1
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45152
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

04e6b3b2

Call non-template base methods through 'this' pointer · 3b655b65

authored May 13, 2020

From https://gcc.gnu.org/onlinedocs/gcc-9.3.0/gcc/Name-lookup.html:
"The C++ standard prescribes that all names that are not dependent on
template parameters are bound to their present definitions when parsing
a template function or class.5 Only names that are dependent are looked
up at the point of instantiation.
...
To make the code valid either use this->f(), or Base<T>::f()"

It is best to use the this->f() solution because it could be a virtual
method.

Also use the type-safe load() and store() methods where possible.

Bug: b/155302798
Change-Id: Id636bfb0b1aae9529bf7f3f58213ec7823d9eddd
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45150
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

3b655b65

Eliminate the array size from rr::Variable · 5e16bc45

authored May 13, 2020

We only need to know the array size during materialization of Array<>
types, so store it only for that class and make the allocation a
virtual method.

This change removes the REACTOR_MATERIALIZE_LVALUES_ON_DEFINITION
option, since it requires knowing the full size of Variable at
construction, which would require Array<> to still pass that into the
Variable constructor. If we need an debugging feature like this again,
we could also materialize variables on first use instead of at
construction.

Bug: b/155302798
Change-Id: I65b02c1258643dcbd3ae18f3f097cb815efa0830
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45148
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

5e16bc45

20 May, 2020 1 commit

LLVMReactor: set alignment when allocating stack variables · cc5cda0f

authored May 19, 2020

Bug: none
Change-Id: I1e9d97e95fea1e9c2863211c711feda6d794a825
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45268
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

cc5cda0f

16 May, 2020 6 commits

Encapsulate the RValue<T> value field · b6e8c3f0

authored May 01, 2020

This will enable us to lazily create the underlying Value* in subsequent
changes.

Also make Argument<T> encapsulate its value, and make it type-safe.

Bug: b/155302798
Change-Id: I50eb16be06bc83e6c87b11f11effed33bee7107a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44671Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

b6e8c3f0

Add type-safe Reference<T> load method · b4e4f11f

authored May 01, 2020

This methods returns an RValue<T> type so it's safer than 'loadValue()'.

Bug: b/155302798
Change-Id: Ib48819bd0a1b64ae2d22eda4910ad0b551b1eed7
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44670
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

b4e4f11f

Add type-safe LValue<T> load and store methods · 5f77c5e5

authored May 01, 2020

These methods operate on RValue<T> types so they're safer than
'loadValue()' and 'storeValue()'. They also enable overloading
the latter in subsequent changes.

Bug: b/155302798
Change-Id: I30903fc70e353dfd668873122ad7d8f74e857420
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44669
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

5f77c5e5

Retrieve Variable type through an abstract method · 67cdce99

authored May 01, 2020

This is just a refactoring which avoids storing the type as a Variable
fields, since the 'LValue<T>' subclass knows the type statically and we
only need it during materialization.

This gives us a vtable for overloading other methods in subsequent
changes, at no additional storage cost for Variable.

Bug: b/155302798
Change-Id: I36f6ef8f5f4d2b50ba04d539dc8b9735481c816a
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44668
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

67cdce99

Rename Reactor static getType() methods to type() · 519cf22b

authored May 08, 2020

This is just a refactoring which will allow adding a virtual getType()
method in a subsequent change without overload naming conflicts.

'isVoid()' has been removed since it appears unused now.

Bug: b/155302798
Change-Id: I16f5cbd947da9581e0f20837497f6b61cfeb7a21
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45008
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

519cf22b

Fix processing OpExecutionMode for the requested entry point · 57eb48ae

authored May 15, 2020

Previously any SPIR-V OpExecutionMode instruction would override the
execution mode parameters, so a shader module with multiple entry points
with different execution modes would not work as intended.

Bug: b/155550449
Change-Id: I86211e66a3cef5a3bbf9e535e9be8415b5d69d4b
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45249
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

57eb48ae

15 May, 2020 1 commit

Regres: Reverse full test run order · 660c3bc9

authored May 13, 2020

Doing a test run with LLVM after the one with Subzero results in various
tests timing out (including some 'UNSUPPORTED' ones). The Subzero test
run doesn't exhibit this issue. It's unclear at this point why this
happens, but reversing the order appears to stabilize it.

Bug: b/156728197
Change-Id: I6450d2d95d6043ef206d09f0020313fbf4b3f2e2
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45149Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

660c3bc9

14 May, 2020 1 commit

Fix asserts due to missing texture formats · c9625f16

authored Mar 20, 2020

In has16bitTextureFormat(), we were missing several formats. Also this
function was poorly named, as it made it seem like we were just
duplicating the work of the bytes() function but giving less
informations. What has16bitTextureFormat() is really checking for is if
the texture format is 16 bit Packed. So it's been renamed to
has16bitPackedTextureFormat()

Bug: b/150076646
Change-Id: I8082e8e9d3845cfa3b5c3ec4a269454457d558ce
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42688Tested-by: Sean Risser <srisser@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

c9625f16

13 May, 2020 3 commits

Fix missing border update after Image blit · d748e165

authored May 08, 2020

ANGLE uses image blit to generate mipmaps so, for cubemaps,
a border update is required after we blit into an image so
that the cube's border are still seamless when using lower
mip levels.

Fixes WebGL Conformance Test:
conformance/textures/misc/texture-size.html

Bug: b/154620294
Change-Id: I22daf60afd9143c52a57c67ae0039eb49be4f1ca
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44990
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

d748e165

Wrap Xcb reply call for error handling · c66681be

authored May 11, 2020

The reply function for xcb_get_geometry was called without handling errors,
causing a null-dereference in the fuzzer. These calls should be wrapped and
handled safely.

Bug: chromium:1080987
Change-Id: Ie1377909b68375fa9bc6eb54bdcbfda0627fd680
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45068
Presubmit-Ready: Jonah Ryan-Davis <jonahr@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Jonah Ryan-Davis <jonahr@google.com>

c66681be

Fix sRGB color clear · 64da65bd

authored May 12, 2020

According to the Vulkan spec:

2.9.2. Conversion from Floating-Point to Normalized Fixed-Point
"The conversion from a floating-point value f to the corresponding
 unsigned normalized fixed-point value c is defined by first
 clamping f to the range [0,1], then computing ..."

Also, in section 18.3. Clear Values:
"float32 are the color clear values when the format of the image or
 attachment is one of the formats in the Interpretation of Numeric
 Format table other than signed integer (SINT) or unsigned integer
 (UINT). Floating point values are automatically converted to the
 format of the image, with the clear value being treated as linear
 if the image is sRGB"

Added a clamp to sRGB values to avoid getting out of bound values
for the sRGB color space.

Bug: b/151469361
Change-Id: I22615b807384904ca4fa887e93529d5a4b185d04
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/45090
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

64da65bd

11 May, 2020 1 commit

Add high precision filtering extension · d9ba4b7d

authored May 04, 2020

This lets users specify the level of sampling precision they want.
Chromium needs this to pass WebGL tests with SWANGLE.

Bug: b/146423360
Bug: b/154620295

Change-Id: I83575823b5909d836c3d4c02803ba7909ce08935
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42268Tested-by: Antonio Maiorano <amaiorano@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

d9ba4b7d

08 May, 2020 1 commit

Regres: Limit the Gerrit comment message length · 7b87548e

authored May 08, 2020

This prevents '400 Bad Request' responses. The limit of 16000 was
determined through experimentation.

Bug: b/150284316
Change-Id: I5507efdcfbaac2c09c2dffde9bf173f4f9219a72
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44989Tested-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

7b87548e

07 May, 2020 3 commits

Revert "Produce MSVC error when virtual class misses virtual destructor" · 2dd86447

authored May 07, 2020

This reverts commit cd09f8e9.

Reason for revert: Kokoro build failures

Bug: b/130335507
Change-Id: I5479cc40723ae3a95e3d9fb7b74de8506f188e08
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44928Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

2dd86447

Remove typeinfo from Vulkan exported symbols · 45be4684

authored May 04, 2020

Exporting all typeinfo was copied over from the GLES version scripts
which addressed crbug.com/737384. However, this can lead to ODR
violations, and we have several other solutions in place for avoiding
UBSan failures.

Remove it entirely for now. If this causes a regression and we still
have to export some typeinfo, we should do so selectively. For example
to only export typeinfo for objects in the vk namespace:
_ZTS*vk*;
_ZTI*vk*;
Alternatively, or in addition, we can try using
'-mllvm -asan-use-private-alias=1'.

Bug: b/155441530
Change-Id: I07303ac01b05d02b879fff829b6a9945c16fdf82
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44689
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

45be4684

Produce MSVC error when virtual class misses virtual destructor · cd09f8e9

authored May 05, 2020

This is already a warning treated as an error by GCC/Clang:
-Werror=delete-non-virtual-dtor

Bug: b/130335507
Change-Id: I9d44f668168e2a77da8839017871e469f5d53bcb
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44788Tested-by: Nicolas Capens <nicolascapens@google.com>
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

cd09f8e9

06 May, 2020 4 commits

Also use out of bound detection for image stores · 08afdde7

authored May 06, 2020

Followup to:
https://swiftshader-review.googlesource.com/c/SwiftShader/+/43950

Fixes dEQP.KHR_GLES31/core_compute_shader_copyimage.
This dEQP test was passing by chance, because load and store
were paired properly together, being equally incorrect with
regards to the OpenGL ES specs, which mentions:

GLES 3.1 spec, section 8.22:
If the individual texel identified for an image load or store
operation doesn’t exist, the access is treated as invalid.
Invalid image loads will return a vector where the value of
R, G, and B components is 0 and the value of the A component
is undefined. Invalid image stores will have no effect.

Bug: b/150464740
Change-Id: I5aa508ed7f08e234387478d3a6faee958ca9dd97
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44868
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>

08afdde7

CMakeLists: Fix building of llvm-with-cov · 8d1f8716

authored May 06, 2020

`target_compile_options` should have been `target_compile_definitions`

Also remove the global `set_cpp_flag` for these defines which are now unnecessary.

Bug: b/145758253
Change-Id: Ie03b0ad196bd1ee155e3f4953d920d6b502db51c
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44848Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

8d1f8716

Insert __msan_unposion for masked/scattered stores · 1d9d6451

authored May 04, 2020

Fixes more MSan LLVM JIT false positives.

Change-Id: I0579e2cc71b089424fe168cc60fed63c4b431f91
Bug: b/140204746
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44708
Presubmit-Ready: James Price <jrprice@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: James Price <jrprice@google.com>

1d9d6451

Detect out of bounds image pointers · 60727de1

authored Apr 15, 2020

When the out of bounds behavior is defined and any of the
image coordinates is out of bounds, this cl manually pushes
the pointer out of bounds in order to trigger the out of
bounds behavior even if the resulting pointer would still
be within memory bounds.

Bug: b/150464740
Change-Id: I685973dfac440a8bca830810cf974819cfbc0f33
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/43950
Presubmit-Ready: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>

60727de1

05 May, 2020 3 commits

Regres: Update test lists @ · baa10d7b

authored May 05, 2020

Reactor backend: LLVM

Change-Id: Idacca75cfd2a18886dc9f86f3d9a60d9d40d8bd3
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/42930Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

baa10d7b

Regres: Don't nuke the daily test dir before using the test results · 098cdda2

authored May 05, 2020

`runDailyTest()` calls `test.cleanup()` before returning, removing the entire checkout directory, along with coverage data, and the git repo.

This breaks both `postDailyResults()` and `postCoverageResults()`.

Instead use a callback to handle the daily results. This way we can keep the `defer` to clean up even in case of error.

Change-Id: I9730d7dd8ac9c3a19d82d07a68325afdf38bfd40
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44748Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

098cdda2

CMake: Only build turbo-cov with clang + llvm 10 · 6ce626b4

authored May 05, 2020

turbo-cov can only process clang coverage information, so there's no point building it for other toolchains.

turbo-cov depends on the llvm-10 libs and headers, and does not build against llvm-7.

Bug: b/152192800
Change-Id: I522ecb3adfbd448aa47673c452a83083134a36ca
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44728
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Nicolas Capens <nicolascapens@google.com>
Tested-by: Ben Clayton <bclayton@google.com>

6ce626b4

04 May, 2020 1 commit

Remove typeinfo from GLES exported symbols · c4bbd378

authored May 04, 2020

This typeinfo was added for UBSan failures reported in crbug.com/737384,
but it didn't address other 'Incorrect-function-pointer-type' failures
for crbug.com/746914. We've suppressed sanitization for these functions
with the NO_SANITIZE_FUNCTION macro, refactored interfaces used across
library boundaries into purely abstract ones for crbug.com/732667, and
use '-mllvm -asan-use-private-alias=1' to address b/128551743.

It's unclear whether this typeinfo still serves ay purpose We shouldn't
have to export typeinfo for objects used only inside the shared library,
to avoid ODR violations.

Remove it entirely for now. If this causes a regression and we still
have to export any typeinfo, we should do so more selectively. For
example to only export typeinfo for objects in the egl namespace:
_ZTS*egl*;
_ZTI*egl*;
Alternatively, or in addition, we can try using
'-mllvm -asan-use-private-alias=1' for more build systems.

Bug: b/155441530
Change-Id: Ia966c40dfe45817f356d11725910afef1bb94d6e
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44688
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Reviewed-by: Alexis Hétu <sugoi@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>

c4bbd378

02 May, 2020 2 commits

Enable the shaderStorageImageArrayDynamicIndexing feature · b798cfa3

authored Apr 29, 2020

Bug: b/146166966
Bug: angleproject:4071
Change-Id: Ib2c69de7fb00ae1917b657840f7c456792f804f6
Tests: dEQP-VK.*
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44531Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>

b798cfa3

Enable the shaderStorageBufferArrayDynamicIndexing feature · fe636858

authored Apr 29, 2020

Bug: b/146166966
Bug: angleproject:4071
Change-Id: Ifbfcf2d68c8789b0f91fef48db95bd9dc5a7576a
Tests: dEQP-VK.*
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/44530Reviewed-by: Alexis Hétu <sugoi@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Kokoro-Result: kokoro <noreply+kokoro@google.com>

fe636858