[llvm-dev] LLVM GPU News Issue #14, June 18 2021

[Jakub (Kuba) Kuderski via llvm-dev]

Hi folks,

The 14th issue of LLVM GPU News, a bi-weekly newsletter on all
the GPU things under the LLVM umbrella, is out:
<https://llvm-gpu-news.github.io/2021/06/18/issue-14.html>.

I also pasted the content below, in case you prefer to read in your email
client.

-Jakub

======================================================================

# LLVM GPU News Issue #14, June 18 2021
Authors: Jakub Kuderski, Johannes Doerfert

Welcome to LLVM GPU News, a bi-weekly newsletter on all the GPU things
under the LLVM umbrella.
This issue covers the period from June 4 to June 17 2021.

We welcome your feedback and suggestions. Let us know if we missed anything
interesting, or want us to bring attention to your (sub)project, revisions
under review, or proposals. Please see the bottom of the page for details
on how to submit suggestions and contribute.


## Industry News and Conferences

*  The [Game Developers Conference (GDC)](https://gdconf.com/about-gdc) is
a month away (July 19-23 2021). The "Programming" topic sessions include
the ["Advanced Graphics Summit"](
https://schedule.gdconf.com/search/Advanced+graphics+summit).


##  LLVM and Clang

### Discussions

*  Sameer Sahasrabuddhe posted an update on the "Abstracting over SSA form
IRs" RFC. The [newest experimental results](
https://lists.llvm.org/pipermail/llvm-dev/2021-June/150990.html) show
performance comparison of two proposed 'CycleInfo' analysis
implementations: dynamic polymorphism via virtual calls is 5% slower on
average than static polymorphism via templates.
*  Sameer Sahasrabuddhe submitted an [RFC: "Update token semantics with
default convergent attribute"](
https://lists.llvm.org/pipermail/llvm-dev/2021-June/151264.html). The RFC
is designed to look like an [incremental change](
https://reviews.llvm.org/D104504) over Nicolai's original spec for
convergence control intrinsics.

### Commits

*  The [Attributor pass performs heap-2-stack for the GPU correctly now](
https://reviews.llvm.org/D98608).
*  A new AMDGPU target [gfx1013 has been added](
https://reviews.llvm.org/D103663), reverted, and relanded.
*  HIP gained a new flag [`--gpu-bundle-output` to control
bundling/unbundling](https://reviews.llvm.org/D101630) output of HIP device
compilation.


## MLIR

### Discussions

### Commits

*  The SPIR-V dialect now [uses SingleBlock + NoTerminator](
https://reviews.llvm.org/D103265). This allows the removal of the
`spv.mlir.endmodule` op.
*  Conversion of MMA type to NVVM [got simplified](
https://reviews.llvm.org/D103868).


## OpenMP (Target Offloading)

### Discussions

*  To ensure the ["atomicity of updates"](
https://bugs.llvm.org/show_bug.cgi?id=49940) we are working on [fine
grained locking in the mapping table](https://reviews.llvm.org/D104418).


### Commits

*  Dependency on [libelf has been removed in favor of the LLVM ELF
utilities](https://reviews.llvm.org/D103545).
*  A [fine-grained globalization](https://reviews.llvm.org/D97680) scheme
for local variables with [middle-end optimizations](
https://reviews.llvm.org/D97818) will be merged.


## External Compilers

### LLPC

*  Initial patches updating the existing LLPC passes with the New Pass
Manager support have been merged.

### Mesa


-- 
Jakub Kuderski
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210619/e254f9c2/attachment.html>