- 51
- 44 203
MLIR
เข้าร่วมเมื่อ 6 พ.ค. 2021
Modern Multi-Level Compiler Infrastructure!
Open MLIR Meeting 9-12-2024: XLA Indexing Maps
In this meeting, Alex talks about XLA Indexing maps, which are affine maps + domain information. XLA:GPU relies on the indexing maps for coalescing analysis, tiling, defining thread/block mappings and code generation.
Discuss on LLVM forums: discourse.llvm.org/t/open-mlir-meeting-9-12-2024-xla-indexing-maps/81166
Slides: mlir.llvm.org/OpenMeetings/2024-09-12-XLA-Indexing-Analysis.pdf
More Tech talks on #MLIR: mlir.llvm.org/talks/
Discuss on LLVM forums: discourse.llvm.org/t/open-mlir-meeting-9-12-2024-xla-indexing-maps/81166
Slides: mlir.llvm.org/OpenMeetings/2024-09-12-XLA-Indexing-Analysis.pdf
More Tech talks on #MLIR: mlir.llvm.org/talks/
มุมมอง: 133
วีดีโอ
Open MLIR Meeting 8-22-2024: Upstream SPIR-V Conversion
มุมมอง 238หลายเดือนก่อน
This is a presention about a generic MLIR lowering pass to SPIR-V, with the goal of having a better test coverage of SPIR-V compilation upstream, and enabling writing simple kernels by hand. The presentation includes a breakdown of the stages (i.e., function signature conversion, vector unrolling, dialect conversions, etc.), IR examples, limitations and future plans. Discuss on LLVM forums: dis...
Open MLIR Meeting 2-15-2024: OpenMP GPU Target Offloading
มุมมอง 8047 หลายเดือนก่อน
In this meeting, Fabian led a discussion on the [RFC] about adding support for OpenMP GPU target offload and the on-going work on revamping our offloading infrastructure in MLIR. Discuss on LLVM forums: discourse.llvm.org/t/open-mlir-meeting-2-15-2024-discussion-on-openmp-gpu-target-offloading-infrastructure/76970 Slides: mlir.llvm.org/OpenMeetings/2023-12-21-PyDSL.pdf More Tech talks on #MLIR:...
Open MLIR Meeting 12-21-2023: PyDSL, a subset of Python for constructing affine & transform dialects
มุมมอง 1.1K9 หลายเดือนก่อน
Kai Ting Wang and Kevin Lee present PyDSL: a small subset of the Python language aiming to assist users in constructing MLIR particularly in the affine and transform dialects. They have developed several in-house optimizations in the affine dialect, and they wish to expose these optimizations to the kernel developers via usage of the transform ops. Programming in MLIR’s Python binding is quite ...
Open MLIR Meeting 12-14-2023: Vector dialect, reshape, and handling of unit dimensions
มุมมอง 6499 หลายเดือนก่อน
This is a design discussion on the topic of handling reshapes in the vector dialect. Here is the high-level agenda: - Formalizing the definition of a virtual vector - How to position various reshape operations (collapse/expand/squeeze/unsqueeze/shape_cast) within the dialect - How do we expect to target SPIR-V/LLVM/Scalable Vectors without each being so sensitive to small canonicalization/patte...
Open MLIR Meeting 11-16-2023: Targeting H100 with NVGPU and NVVM Dialects
มุมมอง 1.3K10 หลายเดือนก่อน
In this talk, Guray presents the developments within the NVGPU and NVVM dialects to target H100. To unlock NVIDIA H100 full potential, leveraging new features is essential. The focus will be on illustrating the implementation of key elements such as the Tensor Memory Accelerator (TMA), warp-group level tensor core instructions, and transactional barriers. Discuss on LLVM forums: discourse.llvm....
Open MLIR Meeting 11-02-2023: A MPI Dialect Proposal
มุมมอง 39711 หลายเดือนก่อน
MPI is at the heart of most modern massively parallel workloads, serving as the fundamental building blocks on which high-performance codes are built. With MLIR having proven itself well as a compiler for single-node workloads already, there is a lot interest in seeing MLIR scale up as a compiler for more massively parallel workloads. We are interested in bringing an MPI dialect into MLIR to se...
Open MLIR Meeting 10-26-2023: A Polynomial Dialect Proposal
มุมมอง 24011 หลายเดือนก่อน
This is a presentation of the proposal for a "poly" dialect for polynomial arithmetic. This is motivated by the use of such (high-degree) polynomials in a variety of post-quantum cryptography, including Fully Homomorphic Encryption (FHE). At a high level, poly will contain a type representing a polynomial, attributes specifying the domain of the polynomials (e.g., integer or real coefficients),...
![Open MLIR Meeting 09-28-2023: [RFC] Sharding Framework for Device Mesh](/img/n.gif)
Open MLIR Meeting 09-28-2023: [RFC] Sharding Framework for Device Mesh
มุมมอง 538ปีที่แล้ว
Nowadays, a robust sharding infrastructure is imperative for distributed computing, in particular for ML and Large Language Models, regardless of whether sharding is determined manually or automatically. Meanwhile, mesh-like deep learning clusters are being extensively employed. For instance, the TPU 3D torus is explicitly defined as a mesh-like cluster, and several DGX boxes could be regarded ...
Open MLIR Meeting 07-13-2023: mlir-query: Tool to query MLIR IR dynamically
มุมมอง 439ปีที่แล้ว
In this presentation Devajith introduces mlir-query: a tool to query MLIR IR dynamically. The tool has a REPL (or command-line) interface to enable users to query various properties of MLIR code, such as “isConstant”. The tool is intended to be similar to clang-query, which allows developers to match AST expressions in C code using a TUI with autocomplete and other features. The presentation wi...
Open MLIR Meeting 07-06-2023: RFC on Pattern Matching for Library Rewriting
มุมมอง 449ปีที่แล้ว
In this presentation, Vinicius Espindola presents an approach to Pattern Matching for Library and Acceleration Instruction Rewriting. The motivation for this work is to support modern ISA extensions and hardware accelerators which created the demand for raising program idioms to acceleration instructions or optimized library calls. Although recent works have proposed different ways of raising c...
Open MLIR Meeting 06-22-2023: Targeting ARM SME from MLIR and SME Dialect
มุมมอง 873ปีที่แล้ว
This is a review and discussion of the RFC on creating a ArmSME Dialect. We will first have an introduction into Arm’s Scalable Matrix Extension, including the MLIR codegen perspective, the work done so far to target SME from MLIR, and discuss what’s the best way to target “CPU accelerators” from MLIR. Then we will have a second part to detail the proposal for this dialect. More info on Discour...
Open MLIR Meeting 06-08-2023: Update on Numba/MLIR
มุมมอง 447ปีที่แล้ว
Ivan Butygin presents the progress on python compiler based on Numba and MLIR. The goal of this project is to create compiler for numeric python code which provides efficient code generation for CPUs and portable and efficient offload capabilities for GPUs. Numba is used as a frontend and mapped to a set of custom MLIR dialects: Ntensor - dialect for numpy arrays and functions, which supports m...
Open MLIR Meeting 05-25-2023: Distinct Attributes: Modeling LLVM's distinct metadata in MLIR
มุมมอง 301ปีที่แล้ว
In this session, Tobias Gysi introduces the problem of modeling LLVM disctinct metadata in MLIR and explore various design options, before discussing collectively the tradeoffs. More info on Discourse: discourse.llvm.org/t/open-mlir-meeting-5-24-2023-rfc-discussion-about-distinct-attribute-in-mlir/70856 Slides: mlir.llvm.org/OpenMeetings/2023-05-25-Distinct_Attributes.pdf More Tech talks on #ML...
Open MLIR Meeting 05-04-2023: Catalyst, an AOT/JIT compiler for hybrid quantum programs
มุมมอง 549ปีที่แล้ว
In this presentation, David Ittah presents Catalyst: an AOT/JIT compiler for hybrid classical-quantum programs, with integration into the Python ML ecosystem and the PennyLane quantum programming framework. A new quantum MLIR dialect (on top of existing MLIR dialects) provides the core IR for the compiler, and enables classical and quantum optimization, code generation, and automatic differenti...
Open MLIR Meeting 3-30-2023: Representing Dataflow with MLIR
มุมมอง 1.3Kปีที่แล้ว
Open MLIR Meeting 3-30-2023: Representing Dataflow with MLIR
Open MLIR Meeting 3-23-2023: Nelli, a lightweight,Pythonic, frontend for MLIR
มุมมอง 1.8Kปีที่แล้ว
Open MLIR Meeting 3-23-2023: Nelli, a lightweight,Pythonic, frontend for MLIR
Open MLIR Meeting 3-9-2023: Convolution Optimization to Improve Performance Beyond Im2Col+GEMM
มุมมอง 1.2Kปีที่แล้ว
Open MLIR Meeting 3-9-2023: Convolution Optimization to Improve Performance Beyond Im2Col GEMM
Open MLIR Meeting 2-23-2023: Actions: Tracing and Debugging MLIR-based Compilers
มุมมอง 919ปีที่แล้ว
Open MLIR Meeting 2-23-2023: Actions: Tracing and Debugging MLIR-based Compilers
Open MLIR Meeting 2-9-2023: Deep Dive on MLIR Internals, Operation&Attribute, towards Properties
มุมมอง 2.5Kปีที่แล้ว
Open MLIR Meeting 2-9-2023: Deep Dive on MLIR Internals, Operation&Attribute, towards Properties
Open MLIR Meeting 1-26-2023: Controllable Transformations in MLIR with the `transforms` dialect
มุมมอง 1.4Kปีที่แล้ว
Open MLIR Meeting 1-26-2023: Controllable Transformations in MLIR with the `transforms` dialect
Open MLIR Meeting 12-15-2022: Embedded Reactive Programming in MLIR
มุมมอง 422ปีที่แล้ว
Open MLIR Meeting 12-15-2022: Embedded Reactive Programming in MLIR
Open MLIR Meeting 11-03-2022: Poison Semantics in MLIR
มุมมอง 317ปีที่แล้ว
Open MLIR Meeting 11-03-2022: Poison Semantics in MLIR
Open MLIR Meeting 10-13-2022: Discussion about broadcasting in TCP (Tensor Compute Primitives)
มุมมอง 281ปีที่แล้ว
Open MLIR Meeting 10-13-2022: Discussion about broadcasting in TCP (Tensor Compute Primitives)
MLIR Open Meeting 2022-10-06: Coordinate Transformations in AMD’s rocMLIR
มุมมอง 489ปีที่แล้ว
MLIR Open Meeting 2022-10-06: Coordinate Transformations in AMD’s rocMLIR
MLIR Open Meeting 2022-08-18: Discussion about a proposal for a high-level ML dialect in MLIR
มุมมอง 8972 ปีที่แล้ว
MLIR Open Meeting 2022-08-18: Discussion about a proposal for a high-level ML dialect in MLIR
MLIR Open Meeting 2022-08-11: MLIR-based IR for Accelerator Design with Decoupled Customizations
มุมมอง 1K2 ปีที่แล้ว
MLIR Open Meeting 2022-08-11: MLIR-based IR for Accelerator Design with Decoupled Customizations
MLIR Open Meeting 2022-06-02: IRDL, a dialect to represent IR definitions
มุมมอง 9052 ปีที่แล้ว
MLIR Open Meeting 2022-06-02: IRDL, a dialect to represent IR definitions
MLIR Open Meeting 2022-03-17: Interfaces and dialects for precise IR transformation control
มุมมอง 1.3K2 ปีที่แล้ว
MLIR Open Meeting 2022-03-17: Interfaces and dialects for precise IR transformation control
MLIR Open Meeting 2022-02-24: Discussion on "ML program" dialect proposal
มุมมอง 3442 ปีที่แล้ว
MLIR Open Meeting 2022-02-24: Discussion on "ML program" dialect proposal
great talk, max understands that his situation is far more common than the case of “compiler expert” many people look at stuff like mlir or llvm and stop because it’s not beginner friendly
Is this work open-sourced?
Awesome introduction to the Linalg dialect. Thank Yi and great respect to your presentation!
20” to birth a ‘why i did it’s …. .. 🤦♂️
Impressive talk, thanks !
What is non-value semantic tensor and value semantic tensor (!torch.vtensor)?
Hi, the link to slides is wrong. mlir.llvm.org/OpenMeetings/2023-06-22-Targeting-SME.pdf
How can we generate a perfetto profile / some method to look through the call-stack of methods?
Really helpful, great talk!
Expressive talk, really helpful!
Thanks!
Great talk, thank you Yi!
This is very helpful. Would be great to see more newcomer videos about MLIR internals.
Is it only on my system that there are black bar artifacts flashing in and out at the top of the recorded video?
yes it's the issue with the video.
I like Max’s style
for the love of all things sacred and pure, please change TCP acronym!!
My first thoughts were "why not just use UDP?" and "Did I miss the RFC?". Then it turned out to be a lecture on "Tensor Compute Primitives (TCP)", and not "Transmission Control Protocol (TCP)", with RFCs having nothing to do with the IETF. Sometimes tech acronyms collide in peculiar ways (and youtube suggestions have no way to distinguish). I went though the rabbit hole, checking abstracts of a few docs linked in the description. Learned something new today. Thanks!
Ah indeed, and "broadcasting" in the same sentence adds to the confusion. Well sorry for the confusion but glad you learnt something about another kind of "broadcast" and "TCP" :)
Thank you for the great contents. Can bufferization do multi-level memory bufferization?
I would advise to ask technical questions on our Discourse forum: discourse.llvm.org/c/mlir/31
Have you shared the code? If you do, could you please tell me where the code base is?
the slides link is this one: mlir.llvm.org/OpenMeetings/2022-01-27-Intro-to-Linalg.pdf
Really enjoyed this meeting. Some super insightful conversation
How can I use this lsp in my vscode? I can't find it on my extension market.
Sorry, we have been a bit slowed down on this for "reasons". One of them is just creating the repo to host the extension and make it available in VSCode market without having to build it yourself, stay tuned! I'll update here when we get news on this.