L
Good evening. I was browsing MLIR’s docs and noticed that there are some Open Projects. I am interested in the improving mlir-reduce as it aligns with my interest in Compilers. Will this can be worked on for next year’s GSOC? And are there any potential mentors available to discuss potential project scopes, @mehdi_amini @jpienaar or any other mentors? 1 post - 1 participant Read full topic
L
We are pleased to announce that MLIR will have a devroom at FOSDEM, which takes place on 30 & 31 January 2027 in Brussels, Belgium. The room is tentatively scheduled for the Sunday afternoon, but this could change. This is in addition to the already popular LLVM devroom. As possibly the largest European Open Source conference, FOSDEM attracts more than 600 lectures and over 8000 hackers - many core contributors to the world’s leading open source projects. Complementing the LLVM developer…
L
We are pleased to announce that LLVM will once again have a devroom at FOSDEM, which takes place on 30 & 31 January 2027 in Brussels, Belgium. The room is tentatively scheduled for the Sunday morning session, but this could change. Additionally, this year there will also be a room dedicated to MLIR. As possibly the largest European Open Source conference, FOSDEM attracts more than 600 lectures and over 8000 hackers - many core contributors to the world’s leading open source projects.…
L
[RFC] Running SSAF over whole projects: clang-ssaf-orchestrator-make and clang-ssaf-driver TL;DR We’d like to add two command-line tools: clang-ssaf-orchestrator-make runs an SSAF workflow over a project built with make. clang-ssaf-driver is the user interface for SSAF analyses and transformations. The user picks a transformation, gives it the build command, and gets back the diagnostics or patch and a report respectively. The problem SSAF (RFC) consists of several components. Each of them…
L
I’d like to propose two things, in order. This is an early RFC. A prototype of both exists, nothing is proposed for landing yet, and what I’m after is a judgment on whether the shape is viable. First, declare LLVM’s optimization remark kinds in TableGen and generate one typed builder per kind, so that a remark’s name, outcome, message and arguments are stated once and every channel renders them from that declaration. Second, add a metadata kind, !remark, that attaches such a remark to the…
L
The transcript of the 2026 Kernel Maintainers Summit was the subject of much debate in some circles. Torvalds said that the time has come to make it harder for new people to contribute code to the kernel. People like Kroah-Hartman have spent years trying to teach people how to contribute; it is time to stop doing that, he said. Torvalds added that he does not want to scare off people who want to write code; his message is that you have to either put in the effort to learn, or put a lot of money…
L
This Request for Comments (RFC) proposes an introduction of the LLVM’s mathvec library as yet another -fveclib=... functions provider. Motivation and Background An initial set of the vectorized math routines has became available in the LLVM’s libc recently. The libc/src/mathvec directory tree currently contains a generic vectorized implementation of the following functions: float acosf(float) float acoshf(float) float acospif(float) float asinf(float) float asinhf(float) float asinpif(float)…
L
Nintendo has been developing its own LLVM-based toolchain for game developers since early 2016 (LLVM 3.8). See our previous talk here. We would like to officially introduce our target to LLVM and participate in the community. Please see the following proposed patches for details: Introduce “nintendo” and “nn” as a valid target triple vendor and OS respectively. We officially support our proprietary OS on AArch64. Add basic clang driver and specify the maintainers for our target. Add…
L
Hello everyone, After the last update there is an issue with buildbot scheduling builds. We are looking into this. Sorry for this. And thank you for patience and understanding. 4 posts - 1 participant Read full topic
L
I’m trying to find the C API equivalent for Builder::CreateCall. In Core.h I found, burried under “Miscellaneous instructions”, LLVMBuildCall2; unfortunately without any comment / documentation. Most of the parameters make sense given the prototype of Builder::CreateCall; except for the second one, which is only given by its type: LLVMTypeRef. What is expected here? Maybe the return type of the function? If so, why is there no corresponding parameter in Builder::CreateCall? Or is there another…
L
Summary I would like to get feedback on a possible change to the Flang Runtime’s handling of unformatted I/O when endian conversion is required. While investigating the performance of unformatted I/O on the current main branch (24.0.0git), I found that bulk transfer is not used when swapEndianness is true, even for contiguous intrinsic-type data. In such cases, the data is processed element by element. A prototype that allows bulk transfer for contiguous intrinsic-type data and performs endian…
L
The recently written DIL (Data Inspection Language) LLDB documentation, in lldb/docs/resources/dil.md, should probably have a link or page in the lldb.llvm.org web site, but I can’t find any information on how to update that website. Can anybody here tell me how to do that or point me to the documentation? Thanks in advance! 4 posts - 3 participants Read full topic
L
IGC is Intel’s LLVM-based compiler for Intel GPUs. It compiles shaders (DirectX, Vulkan, OpenGL) and compute kernels (OpenCL, SYCL / oneAPI) into native GPU code, and it’s open source. What you’ll work on As an intern you’ll work directly in the LLVM-based compiler stack: Writing and improving LLVM IR passes for GPU-specific optimization Backend code generation for Intel GPU architectures: instruction selection, register allocation, scheduling Performance analysis on real graphics and compute…
L
Authors: Wontae Choi (@wtchoi), Rohan Jacob-Rao (@rohanjr) We propose Spatial Safety Mode, a modular, compiler-based tool for mitigating spatial safety risks in C++ code, including code that interoperates with C. The mode aims to strike a balance among adaptability, security, and performance. The mode should be compiler-based, combining compile-time errors with code instrumentation. Compile-time errors define a language subset in which certain dangerous language features are disallowed. Code…
L
We’ll meet as normal tomorrow, Thu 22nd October at 4pm BST / 8am PDT. If you have more topics for the agenda please drop them in the doc. Note that for the meeting after this one (5th November), daylight savings should have ended in both the US and Europe. The standard meeting time will then be 4pm GMT / 8am PST (i.e. the same local time for most). 3 posts - 2 participants Read full topic
L
Historically, Clang has followed the design taken in LLVM of being a series of composable libraries that build together to make various tools. This design has served us very well and continues to be an important part of what we offer. clangParse links against clangSema which does not link against clangParse, making this a hard library layering boundary where the parser can call into sema, but sema cannot call back into the parser without jumping through hoops. One such hoop is the ParsedType…
L
Hi everyone, I am currently working on a pass in CIRCT (specifically extracting and evaluating Don’t Care bits) and verifying the resulting Verilog against the original design using formal equivalence checking (SEC) with Cadence JasperGold. In my testing pipeline, I want to evaluate how much my pass outperforms standard firtool Dead Code Elimination (DCE). For my baseline (control group), I convert the input Verilog directly into the HW dialect (circt-verilog) and compile it via firtool to…
L
Hi all, I put up a PR for removing the GUI/TUI command from LLDB. If you don’t know what the GUI command is, it’s one of the quickest ways to make LLDB crash open a graphical user interface in LLDB. The PR contains the reasons, but I’ll also repeat them here: The code quality is abysmal. This alone is a good enough justification for removing it. It’s responsible for a flakey test (lldb-api :: commands/gui/spawn-threads/TestGuiSpawnThreads.py failing on AArch64 CI #209874) that no one is working…
L
Hello LLVM Community, I’m the founder of the Localize the Docs organization, and I’m glad to announce that the llvm-libc-docs-l10n project is now published: Preview: llvm-libc-docs-l10n Crowdin: llvm-libc-docs-l10n GitHub: llvm-libc-docs-l10n The goal of this project is to translate The LLVM C Library Documentation into multiple languages. Translations are contributed via the Crowdin platform, automatically synchronized with the GitHub repository, and can be previewed on GitHub Pages. We…
L
Introduction This RFC proposes a way to half the memory usage of TypeSanitizer’s shadow memory on 64-bit systems. This will be accomplished by storing 32-bit indices into lookup tables of type descriptors, rather than 64-bit pointers to these type descriptors themselves. Current State On 64-bit systems, TypeSanitizer uses 8 bytes of shadow memory for each byte of user memory. Shadow memory records the type associated with each byte by storing a pointer to a type descriptor in read-only data.…
L
We are happy to announce that LLVM 23.1.4 is now released! This includes the main LLVM project, and its subprojects including clang, lld, libc++, and MLIR. Huge thanks to everyone that contributed, reviewed, provided support and in any other way contributed to this release! Please note the subproject tarballs have been removed after RFC: Do “something” with the subproject tarballs in the release page Download Find sources on GitHub. A note on binaries Official binaries will not be available…
L
Title: [SimplifyCFG] Does speculating a global load introduce UB under the documented noalias semantics? I have a small reproducer where SimplifyCFG speculates a load from a global variable. Under my reading of the documented noalias semantics, this appears to introduce undefined behavior on an execution that was previously defined. This seems related to the earlier discussion, Interaction of noalias and dereferenceable. Here, there is no explicit dereferenceable attribute; the speculated load…
L
Hi I want to be able to restrict existing exports on WIndows, similarly to how I can build a custom export linker script for ld or macho-ld Is this possible? I see that there is clang equivalent of the Visual Studio LINK.EXE which supports the exact same functionality, but sadly that is insufficient for me. (For bonus points, I would also like to use wildcards in my linker script, but that is a nice to have Thanks, Noel. 7 posts - 4 participants Read full topic
L
We recently ran into an issue with arith.remsi. Division by zero is already UB, but we noticed that the lowered llvm.srem op has an additional UB case: srem(INT_MIN, -1). From a mathematical point of view, INT_MIN % -1 is well-defined. The current lowering to LLVM is broken because it turns a program without UB into one that may have UB. My personal preference is to keep the existing arith.remsi semantics and fix the lowering to LLVM with an additional runtime check (PR), which could (maybe)…
L
Authors: @amsen20, @davidzhengyes, @ajhexer, @aryanv2003 Intro The MC layer is the highest level at which LLVM can process inline asm blocks, assembly files, and the lowered machine instructions emitted by the compiler. In the MC layer, the program is viewed as a linear stream of section switches, labels, data, machine instructions, CFI directives, and other assembler directives. Each machine instruction is represented as an opcode with a list of operands, each of which is an immediate, a…
L
Board Meeting Minutes Date: September 4, 2026 Time: 9AM pacific time Location: Video conferencing, multiple locations Attendees: Tanya Lattner, Kristof Beyls, Chris Bieneman, Anna Zaks, Reid, Dave Blaikie, Wei Wu, Anshu Dasgupta, Anton Korobeynikov Agenda Approve previous meeting minutes Approved June 2026 Approved August 2026 Board Member Introductions Welcoming new member: Dave Blaikie 2026 US LLVM Dev Meeting Updates ~1.5 months out. Speakers getting registered. Early bird registration…
L
LLVM Weekly - #668, October 19th 2026 If you prefer, you can read the canonical version of this issue at https://llvmweekly.org/issue/668. Welcome to the six hundred and sixty-eighth issue of LLVM Weekly, a weekly newsletter (published every Monday) covering developments in LLVM, Clang, and related projects. LLVM Weekly is brought to you by Alex Bradbury. Subscribe to future issues at https://llvmweekly.org and pass it on to anyone else you think may be interested. Please send any tips or…
L
TL;DR: When small code references a large section through a 32-bit relocation, we propose to demote that section to a small one. Demotion clears SHF_X86_64_LARGE and drops the l prefix from the section name, so the linker places it together with the small sections. This makes it practical to link translation units (TUs) compiled with the small and large code models into the same binary. Background The x86-64 psABI doesn’t fully specify how code models can be mixed. LLVM supports mixing them in…
L
We (Sony) are starting to open patches (starting here) that change the representation of source locations in LLVM to use “Function-Local Metadata”, significantly reducing the memory usage for source locations (currently represented by DILocation); more detail can be found here. This introduces some breaking changes to LLVM’s API for DebugLoc and DILocation, and deprecates several existing functions in LLVM’s C-API. A dev implementation, representative of the overall design (but with some…
L
The LLDB test suite is unfortunately known for being prone to randomly failing/flakey tests. This post is a short update about what’s going on and what we’re doing to address this. Why are the tests randomly failing? There is a variety of reasons for why the LLDB tests are randomly failing: The use of various short timeouts within LLDB, which are exceeded on slow/overloaded CI bots. Interactions with the OS/hardware that are unreliable. Sometimes these are the fault of LLDB, sometimes there are…