LLVM and SPIRV-LLVM-Translator pulldown (WW10 2024) #12939
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Seemingly I either missed this on a buildbot or otherwise didn't cover this :(
This patch addresses an oversight in `ProcessEventDataTest::SetUp` unittest to ensure the Debugger is initialized properly. Signed-off-by: Med Ismail Bennani <ismail@bennani.ma>
When a call to `getFreelyInvertedImpl` with a select/phi node fails, `DoesConsume` should not be changed. Fixes llvm/llvm-project#82877.
Add `liblldb` dependency and use correct extension for compiled Lua module. Replace 'Python' with 'Lua' in install path name. Fixes #55075.
We can use bitwise arithmetic to implement these, making them considerably faster than legalization via promotion.
…#83884) This work corrects a few inappropriate uses of HasImageInsts predicate in vimage instruction definitions. In MnemonicAlias for VIMAGE_Atomic_gfx12_Renamed, we also need HasImageInsts to be in the "Requires" predicate list for the alias to depend on whether or not the GPU has image instruction support. For nested uses of "let OtherPredicates = ..." around vimage instruction definitions, the inner assignment will override the outer one. This makes the outermost "let OtherPredicates = [HasImageInsts]" unused when we have an inner assignment. As a result, HasImageInsts is not actually used for some vimage instructions. To resove this issue, we propogate the predicates in an outer assignment into the inner one. We should avoid using nested "let SubtargetPredicate = ...". However, we can always put the predicate into OtherPtredicates list.
Fixes typo in documentation for clang-format Fixes #83207.
Either: - I forgot my alphabet (that E comes before F). - My juvenile inner brain finds unsigned literal constants with the sequence FU funny. ¿Por qué no los dos?
So far, all the work we've done for compute constructs has only used 'parallel'. This patch does the work to enable the same logic for 'serial' and 'kernels' constructs as well, since they are the same semantic behavior.
Addresses the `third-party/benchmark` part of #81859 (by happening to remove `requirements.txt`)
Several profdata tests pass the byte 012 to printf. This causes these tests to fail when using GnuWin32's version of printf because printf will detect that 012 is the LF character and will prepend the byte 015 (CR) in front of LF. This change is required after llvm/llvm-project#82711 which bumped the version number.
This reverts commit 2e93ee6. buildbot failures, e.g. `/third-party/benchmark/cmake/pthread_affinity.cpp`
Add SPIR-V backend support for the HLSL SV_DispatchThreadID semantic attribute, which is lowered to a @llvm.dx.thread.id intrinsic in LLVM IR. In the SPIR-V backend, this is now correctly translated to a `GlobalInvocationId` builtin variable. Fixes #82534
Given a scalable VF of the form <NumElts * VScale>, this patch adds the ability to discharge a backedge test for a loop whose trip count is between (NumElts, MinVScale*NumElts). A couple of notes on this: * Annoyingly, I could not figure out to write a test for this case. My attempt is checked in as test32_i8 in f67ef1a, but LV uses a fixed vector in that case, and ignored the force flags. * This depends on 9eb5f94 to avoid appearing like a regression. Since SCEV doesn't know any upper bound on vscale without the vscale_range attribute (it doesn't query TTI), the ranges overflow on the multiply. Arguably, this is fixing a bug in the current LV code since in theory vscale can be large enough to overflow for real, but no actual target is going to see that case.
This was copy+pasted from count_ones without updating the test name completely.
This reverts commit aec6a04. (google/benchmark still at hash 1576991177ba97a4b2ff6c45950f1fa6e9aa678c as it was in #83488. Also reapplied same extra local diffs) Verified locally.
If a gep has only one phi as one of its operands and the remaining indexes are constant, we can unfold `gep ptr, (phi idx1, idx2)` to `phi ((gep ptr, idx1), (gep ptr, idx2))`. Take care not to unfold recursive phis. Followup to #80983. This was initially was #83087. Initial PR did not handle allocas in entry block that weren't at the beginning of the function, causing GEPs to be inserted after the first chunk of allocas but potentially before an alloca not at the beginning. Insert GEPs at the end of the entry block instead since constants/arguments/static allocas can all be used there.
More information is more testing! Also adjusts already migrated integration tests
This mirrors the scalar version.
…ssumptionCache; NFC
…ConditionCache This helps cover some missing cases in both and hopefully serves as creating an easier framework for extending general condition based analysis. Closes #83161
Fixed-point arithmetic support is targeted towards baremetal targets.
gfx940 does not allow abs/sext/neg on v_cvt_fp8/bf8 & pk variants. Fixes SWDEV-447468
When a variadic argument is expected but not provided the compilation fails later with a difficult to follow compilation error. Add a simple check to catch one such case. This is not yet general as it doesn't yet check leaf nodes.
This patch regroups declarations in `Sema` based on the file they are implemented in (e.g. `SemaChecking.cpp`). This allows to logically split `Sema` in 42 groups. No physical separation is done (e.g. splitting `Sema` into multiple classes). Table of contents added at the very beginning of `Sema`. Grouping is reflected in Doxygen commands, so structure of API reference of `Sema` is also significantly improved ([example from official documentation](https://www.doxygen.nl/manual/examples/memgrp/html/class_memgrp___test.html), [comparison of Sema API reference](llvm/llvm-project#82217 (comment))). While grouping is intentional, as well as each group consisting of `public` declarations followed by `private` ones (without changing access in-between), exact contents and order of declarations of each group is partially carried over from old structure, partially accidental due to time constrains to do the regrouping over the weekend (`Sema` is just enormously big). Data members and inline function definitions in `Sema.h` complicate the matter, since it's not obvious which group they belong to. Further work is expected to refine contents and order of declarations. What is also intentional is some kind of layering, where Concepts group follows template groups, and ObjC, code completion, CUDA, HLSL, OpenACC, OpenMP, and SYCL are all placed at the end of the file, after C and C++ parts of `Sema`. I used `clang-query` to verify that access specifiers were preserved during the process (https://gcc.godbolt.org/z/9johffY9T, thank you @ilya-biryukov). Only the following 3 member types were converted from `private` to `public` because of limitations of the new grouping: `DeclareTargetContextInfo`, `TypoExprState`, `SatisfactionStackEntryTy`. Member initializer list of `Sema` in `Sema.cpp` is rewritten to reflect new order of data members in order to avoid `-Wreorder-ctor`. Since this patch touches almost every line in `Sema.h`, it was considered appropriate to run clang-format on the whole file, and not just on changed lines.
CONFLICT (content): Merge conflict in clang/lib/Driver/Driver.cpp
This merge cause two tests fail: SYCL-Unit :: Extensions/CommandGraph/./CommandGraphExtensionTests/32/46 SYCL-Unit :: Extensions/CommandGraph/./CommandGraphExtensionTests/33/46 CONFLICT (content): Merge conflict in clang/include/clang/Sema/Sema.h CONFLICT (content): Merge conflict in clang/lib/Sema/Sema.cpp
The Headers for this extension were published so we should use them instead: KhronosGroup/SPIRV-Headers@b73e168 Original commit: KhronosGroup/SPIRV-LLVM-Translator@7d7e0ac5303f93d
If SPV_KHR_bit_instructions is not enabled lower llvm.bitreverse.* to a function in LLVM IR. Signed-off-by: Lu, John <john.lu@intel.com> Original commit: KhronosGroup/SPIRV-LLVM-Translator@08d939609f186a4
Use unordered_map instead of map for better performance. Signed-off-by: Lu, John <john.lu@intel.com> Original commit: KhronosGroup/SPIRV-LLVM-Translator@56538038eda11b7
This patch fixes verification of Get/Async Capacity literals making translator accept zero values which are valid by spec. Original commit: KhronosGroup/SPIRV-LLVM-Translator@22f9e3e67b36b36
Fix the following compiler warning:
.../SPIRVNameMapEnum.h:704:61: warning: extra ';' [-Wpedantic]
Original commit:
KhronosGroup/SPIRV-LLVM-Translator@ff0206f025b03bd
In translation from __spirv_AtomicCompareExchange to OpenCL builtin atomic_compare_exchange_strong_explicit, a new alloca `expected` is created and read/written in the OpenCL builtin. The OpenCL builtin call can't have tail marker since the marker requires that callee doesn't access alloca from the caller. Otherwise llvm alias analysis deduces that the alloca isn't accessed by the call, and instcombine pass replaces the load from the alloca after the call with the value stored to the alloca before the call. Original commit: KhronosGroup/SPIRV-LLVM-Translator@1ff4a764cd0f97c
The SPIR-V to LLVM conversion would bail out when encountering an `OpVectorShuffle` whose vector operands differ in size. SPIR-V allows differing vector sizes, but LLVM's `shufflevector` does not. Remove the assert and insert an additional `shufflevector` to align the vector operands when needed. Original commit: KhronosGroup/SPIRV-LLVM-Translator@3df5e38250a6d7c
sys-ce-bb
added
the
disable-lint
|
@bader @intel/llvm-gatekeepers This is ready for merge. No review required. Thanks. |
|
/merge |
bb-sycl
approved these changes
Mar 7, 2024
|
Thu 07 Mar 2024 04:50:19 PM UTC --- Start to merge the commit into sycl branch. It will take several minutes. |
|
Thu 07 Mar 2024 04:59:59 PM UTC --- Merge the branch in this PR to base automatically. Will close the PR later. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
LLVM: llvm/llvm-project@ac74d9e
SPIRV-LLVM-Translator: KhronosGroup/SPIRV-LLVM-Translator@3df5e38