Suppress some clang-tidy errors in kpv-cpr-1568 #446
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
Prevent DAGCombiner from narrowing loads via TLI::shouldReduceLoadWidth, inhibit [zs]ext(load) -> [zs]extload transformation.
Fat pointer type is to used to distinguish with regular i256 type, which does not require special instruction handling to maintain its tagged status within VM. Pointers to generic address space will be interpreted as fat pointers in the backend. It will be assigned a physical register which could be overlapped with i256 -- this is conformant to fat pointer semantics.
Co-authored-by: Alex Z <a.zarudnyy@matterlabs.dev>
Farcalls now return resulting fat pointer in R1 and success flag in R2.
With the help of free flag setting, we could try to fold a comparison with 0 (part of select or branching) with previous instruction.
Also make the soft toolchain requirements hard. This allows us to use C++17 features in LLVM now. If we find patterns with C++17 that improve readability it should be recommended in the coding standards. Reviewed By: jhenderson, cor3ntin, MaskRay Differential Revision: https://reviews.llvm.org/D130689 Conflicts: llvm/docs/ReleaseNotes.rst
Our previous value 10.9 causes an error with our build setup: > 'shared_mutex' is unavailable: introduced in macOS 10.12 There are ways around this; see https://reviews.llvm.org/D66313, but testing MacOS 11, 12, and 13 seems reasonable.
Add ourselves to Windows as not supporting file permission error-checks, though it’s actually that our CI runs too permissively.
We should not allow ISEL failure because of cannot select. For all unsupported types they must all be marked as `Expand` or explicitly fail with a reason. * expand remaining ISEL nodes that are not supported on our target * support of 256-bit CTPOP expansion by reusing 128-bit parallel counting algorithm. * custom lower CTPOP and BSWAP for 256bit
Remove unsupported from lit config for generic codegen tests and handle every single case on it's own: * Relax requirement on global constants to be i256. * Wrap block address and external symbol to materialize it in a register before using. * Wrap global address in calling conventions. * Disable bit tests clustering lowering for switches. * Fix handling of negative wrt to FrameIndex offsets. * Mark tests that have floats, vectors, inline asm, EH but cleanups as UNSUPPORTED. * Classify and mark as TODO still unhandled failures.
We do not distinguish conditional jumps and unconditional jumps, but in fact conditional jumps are not barriers. This patch separates conditional jumps and unconditional jumps and update analyzeBranch to properly return analyzed results.
* in our architecture, CALLSTACK take CImm operands instead of imm * CALLs should not be modeled as terminators * also update the test file to enable testing these changes
Because fat pointers are not numeric, LLVM doesn't know how to assert alignment, so trust the FE on that respect.
When a global array is accessed by a constant index LLVM refers to it as symbol+offset where offset is in bytes. The patch converts offset to cells. The transformation works however only for global variable arrays (i.e. stack address space). Our handling of constant arrays (code address space) require isel pattern rework (CPR-940). So the patch disables SimplifyCFG's logic that convert switches to const arrays since it causes code generation issues.
Which will cause `--verify-regalloc` to fail due to incompatible type of operand. Use `SyncVM::R0` instead of `0`.
This is needed to fix the `--verify-twice` option, as it will run the compilation pipeline twice. The problem was that we did not reset cached constant pool entries after finishing emitting.
To be squashed with f7198c1
Manually squash single file, as git says CONFLICT (content): Merge conflict in .github/workflows/compiler-integration-tests.yml error: could not apply 9b2e437
* Export commit ID getting API as LLVMPrintCommitIDTo. * Print commit id in LLVM tool --version message.
INVOKE is eventually converted to a near call, which should not be a terminator by itself. The throwing and error handling is already done in other parts. This patch fixes the following test: ``` llc --verify-regalloc --verify-machineinstrs llvm/test/CodeGen/SyncVM/cxa_throw.ll ```
before trying fixes for some Pylint warnings.
warnings, which seemed to silence similar warnings during execution, and might conceivably help with false positives and failed suppressions.
(Didn’t work completely locally, e.g. Driver.cpp:2078:29 clang-analyzer-core.NullDereference)
This reverts commit 4139547.
command-line for otherwise apparently unsuppressible failures.
trying to restrict the former.
…ompiler-llvm into FlashSheridan-kpv-cpr-1568
|
Benchmark results: |
PavelKopyl
force-pushed
the
kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker
branch
from
e55e78d
to
d237304
Compare
PavelKopyl
force-pushed
the
kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker
branch
2 times, most recently
from
31248da
to
c4b7b65
Compare
hedgar2017
force-pushed
the
kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker
branch
from
c4b7b65
to
2010039
Compare
PavelKopyl
force-pushed
the
kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker
branch
2 times, most recently
from
c879900
to
f443a5f
Compare
PavelKopyl
force-pushed
the
kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker
branch
2 times, most recently
from
760de6c
to
7c76b3e
Compare
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.
Purpose
Suppress some clang-tidy errors in branch kpv-cpr-1568-adding-support-of-evm-arch-to-lld-linker; I’m proposing to add these changes to that branch while it‘s in development. (Some of the debugging output in compiler-integration-tests.yml may not be suitable for main.)
Ticket Number
CPR-1600
Implementation
clang-tidy-diff_Zegar.pycommand-line for otherwise apparently unsuppressible failures.lld/.clang-tidyby having one positive check (clang-analyzer-osx.cocoa.MissingSuperCall, which is unlikely to conflict with anything.)lld/unittests.clang-tidy-diff_Zegar.py, and silence some Pylint (and probably runtime) warnings.clang-tidy-15r/t genericclang-tidy.This does not fix the unsuppressible breakage from “llvm/llvm/tools/llvm-driver/llvm-driver.cpp:20:10: error: 'LLVMDriverTools.def' file not found [clang-diagnostic-error]”, which is in Chris Bieneman’s code, not ours, and which I don’t understand the CMake setup well enough to address.
There are remaining warnings in, for instance,
lld/MinGW/Driver.cpp, many of them due to Alexandre Ganea; I hope that theWarningsAsErrors: ''inlld/.clang-tidywill prevent these from breaking the GitHub Action once the[clang-diagnostic-error]s are fixed. (I agree with that we should ignore them rather than fix them.) If not, or if there‘s too much noise, reverting e91ad1f (“Remove the universal suppression from lld, now that it's in lld/unittests”) may help.