Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Implement SIMD matrix multiplications #47

Draft
wants to merge 16 commits into
base: master
Choose a base branch
from
Draft

Implement SIMD matrix multiplications #47

wants to merge 16 commits into from

Conversation

demianmnave
Copy link
Owner

Improve performance of specific functions by implementing SIMD versions. The scope will be modest for this pull request:

  • 4x4 double- and single-precision matrix-matrix products using AVX2
  • 4x1 double- and single-precision matrix-vector products using AVX2

After getting some practice with the basics, it may be prudent to use an existing library to extend support for other instruction sets. For example:

Fixes #20

@demianmnave
Use CMake policy 0091 to ensure MSVC uses a static runtime for buildi…
11330d2 
…ng tests
@demianmnave
Timed naive mxm for 4x4 C arrays
29a0433
@demianmnave
Split out uniform_random_rotation_4
710ec33
@demianmnave
Split out rotation matrix pair generator
b287126
@demianmnave
Add get(...) overloads with mutable return types
ec57665
@demianmnave
Use an array of output matrices to prevent elision of the mxm code
07f3863
@demianmnave
Fix bugs in the random rotation matrix generator
fd05cb9
@demianmnave
Make the mxm implementation configurable via CMake
9047b83
@demianmnave
Add CML type system specializations for a 4x4 C-array (for the timing…
106bc39 
… code only)
@demianmnave
Minor cleanup of common/traits.h
9cb9f03
@demianmnave
Complete transition to configurable timing code (for 4x4, anyway), in…
8314e5c 
…cluding ET-based mxm
@demianmnave
Minor CMakeLists.txt cleanup to prepare for SIMD timing codes
1be481e
@demianmnave
Add SIMD skeleton with AVX and start of SSE4.2
c9c8218
@demianmnave
Add iostream output for a 4x4 C-style matrix, and make N a command li…
86b8a7c 
…ne parameter to matrix_mxm_timing.cpp
@demianmnave
Merge branch 'master' into add-mxm-timing-tests
b8721b6
@demianmnave
Implement experimental SIMD 4x4 matrix products with AVX2
34b6e4f 
- The new functions are enabled only when compiling with MSVC and "/arch:AVX2".
@demianmnave demianmnave self-assigned this Dec 1, 2024
@demianmnave demianmnave added this to the vNext milestone Dec 1, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees

@demianmnave demianmnave

Labels
enhancement
Projects
None yet
Milestone
vNext
Development

Successfully merging this pull request may close these issues.

CML matrix multiplication is terribly inefficient
1 participant
@demianmnave