Add experimental half precison support (!332) · Merge requests · pycodegen / pystencils · GitLab

Snippets Groups Projects

GitLab now enforces expiry dates on tokens that originally had no set expiration date. Those tokens were given an expiration date of one year later. Please review your personal access tokens, project access tokens, and group access tokens to ensure you are aware of upcoming expirations. Administrators of GitLab can find more information on how to identify and mitigate interruption in our documentation.

An error occurred while fetching the assigned milestone of the selected merge_request.

Merged Add experimental half precison support

holzer/pystencils:halfP into master

Merged Markus Holzer requested to merge holzer/pystencils:halfP into master 1 year ago

With this MR experimental half-precision support is added

Activity

Markus Holzer added feature label 1 year ago

added feature label
Markus Holzer requested review from @kohl 1 year ago

requested review from @kohl
Markus Holzer assigned to @holzer 1 year ago

assigned to @holzer
Markus Holzer added 1 commit 1 year ago
added 1 commit

6c583c13 - Small correction

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

edd4fe27 - Add half precision support for CPU

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

6b8859d3 - Corrected is_half checker

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

5a5f13dc - Correct header detection

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

6ceafb9c - test number type

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

5ef0dd37 - remove typed numbers etc

Compare with previous version
Nils Kohl approved this merge request 1 year ago

approved this merge request
Michael Kuron @kuron started a thread on an old version of the diff 1 year ago

Resolved 1 year ago by Michael Kuron
Michael Kuron @kuron started a thread on an old version of the diff 1 year ago

Resolved 1 year ago by Michael Kuron
Michael Kuron @kuron · 1 year ago

Maintainer

Resolved 1 year ago by Michael Kuron

Any plans to support CPU vectorization? Half precision is supported by Sapphire Rapids' AVX512 implementation (_m512h types), ARM Neon (float16x4_t), and ARM SVE (svfloat16_t), while the RISC-V "Zvfh" extension (vfloat16m1_t) is currently in the process of being ratified.

Edited 1 year ago by Michael Kuron

Last reply by Markus Holzer 1 year ago
Michael Kuron requested review from @kuron and removed review request for @kohl 1 year ago

requested review from @kuron and removed review request for @kohl
Michael Kuron @kuron started a thread on an old version of the diff 1 year ago

Resolved 1 year ago by Michael Kuron
Markus Holzer added 1 commit 1 year ago
added 1 commit

e57837b1 - Pythonise half precison header

Compare with previous version
Markus Holzer added 1 commit 1 year ago
added 1 commit

287a4c79 - Rename cudajit to gpujit

Compare with previous version
Michael Kuron @kuron started a thread on the diff 1 year ago

Resolved 1 year ago by Michael Kuron
Michael Kuron @kuron started a thread on an old version of the diff 1 year ago

Resolved 1 year ago by Michael Kuron
Markus Holzer added 1 commit 1 year ago
added 1 commit

533afcdf - Add more warnings and tests

Compare with previous version

Please register or sign in to reply