Skip to content

Releases: intel/mlir-extensions

IMEX v0.3

22 Nov 01:12
@nbpatel nbpatel
v0.3
93d5efd
Compare
Choose a tag to compare
IMEX v0.3 Latest
Latest

Preview release of Intel® Extension for MLIR (IMEX)

Fixes / Improvements

Highlights

  • XeTile Dialect : XeTile dialect supports the tile-based programming model and decomposes the GEMM kernel to large pre-defined tile sizes at the subgroup and workgroup level.
  • XeGPU Dialect: The XeGPU dialect models Xe instructions like DPAS and 2D block load/store.
  • Lowering from subgroup level XeTile to XeGPU VC Mode.
  • Lowering from XeGPU to SPIR-V.
    • XeGPU VC Mode to VC Intrinsics (Covers all XeGPU ops)
    • XeGPU to GenISA Intrinsics & Joint Matrix (Ops supported : create_nd_descriptor, update_nd_offset, load_nd, store_nd, dpas)
  • Dialect/Op, conversion & integration test cases for XeTile & XeGPU.
  • High performance end-to-end GEMM code example based on SPIR-V dialect.
  • RFC summarizing XeTile & XeGPU design

List of all changes

v0.2...v0.3

Dependencies revisions

Project Revision
LLVM Project 49af650

Supported System configurations

  • Ubuntu 22.04 LTS
  • x86 CPU
  • Intel® Data Center GPU Max Series

Supported data types for GPU

  • FP32
  • FP16
  • BF16
  • I32
  • I16
  • I8

Limitations

  • For GEMM end-to-end test cases we only support FP16 & BF16 data types.
  • IMEX v0.3 does not support TF32 & FP64.
  • When the input program uses multiple tile_mma op, tensor used as A matrix in one tile_mma can not be used as B matrix in another.
  • XeTile fusion use case is not fully supported.
  • XeGPU doesn't support SLM use case.

Dependencies for GPU execution

GPU execution supports two different wrapper libraries for interacting with GPU. Level Zero and Sycl wrapper libraries.

Assets 2
Loading

IMEX v0.2

28 Apr 19:06
@drprajap drprajap
v0.2
d31a425
Compare
Choose a tag to compare
IMEX v0.2

Preview release of Intel® Extension for MLIR (IMEX)

Fixes / Improvements

Highlights

  • BF16 Support
  • Performance benchmark for popular models - elementwise, reduction, softmax, transpose, kloop
  • Caching for level-zero/sycl runtime compilation
  • Fix GPU memory leak
  • Change to Synchronous queue for Level-zero runtime to fix sporadic issues with GPU tests

List of all changes

v0.1...v0.2

Dependencies revisions

Project Revision
LLVM Project d2a559f

Supported configurations

  • Ubuntu 20.04 LTS
  • x86 CPU
  • Intel® Data Center GPU Flex Series or Intel® Data Center GPU Max Series

Supported data types for GPU

  • FP32
  • FP16
  • BF16
  • I32
  • I16
  • I8

Dependencies for GPU execution

GPU execution supports two different wrapper libraries for interacting with GPU. Level Zero and Sycl wrapper libraries.

Assets 2
Loading

IMEX v0.1

19 Jan 21:22
@silee2 silee2
v0.1
e020fff
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.
Compare
Choose a tag to compare
IMEX v0.1

Preview release of Intel® Extension for MLIR (IMEX)

This release focuses on functionally enabling Deep Learning workloads with static shaped input tensors written in upstream MLIR Entry dialects (Linalg and other lower level dialects) on select Intel GPUs. The development is test case driven and does not cover all ops in the MLIR entry dialects.

Supported configurations

  • Ubuntu 20.04 LTS
  • x86 CPU
  • Intel® Data Center GPU Flex Series or Intel® Data Center GPU Max Series

Supported data types for GPU

  • Fp32

Dependencies for GPU execution

GPU execution supports two different wrapper libraries for interacting with GPU. Level Zero and Sycl wrapper libraries.

Tested DL workloads

  • Resnet-50 inference
  • Mobilenet-v3 inference
  • Mnist mlp

Tested DL Operations

  • Elementwise
  • Convolution
  • GEMM
  • Reduction
  • Data movement
  • Padding
Assets 2
Loading