LLVM邮件列表2018年4月内容选辑

版权声明:本文为博主原创文章,未经博主允许不得转载。 https://blog.csdn.net/snsn1984/article/details/80315139

缘起:

LLVM的邮件列表关注了很久,最闲的时候每篇都读,也回过几篇邮件。忙的时候,可能好几个月不怎么读,积累一大堆未读,只能直接清理了。后来,有人做了llvmweekly,好了很多,每周花点时间看一看,就知道个大概,方便很多。转眼之间,llvmweekly已经做到了200多期,历时4年多了,我从中获益很多。一直在想,除了经常记录点自己学习的感受和零碎的知识点,我还能做点什么?最近忽然有想法,我能不能做个月度的邮件列表内容选辑,选几条我自己觉得最值得读的邮件列出来,做个大致的分类,方便关注这些而又没有时间读周报的人读一下自己关心的内容。于是有了这个尝试,希望能对LLVM的关注者有用。我会连续做几个月,然后看效果决定是否做下去。

内容选辑遵循以下几个原则:1、会根据当月的内容做出一些分类,每个分类如果不是优质内容特别集中,最多选三条,少则不限;2、内容选辑会参考llvmweekly的选取范围;3、这个内容选辑的重点主要在月报和分类,这也是和llvmweekly的区别之处;4、做的过程之中,会根据实际情况不断的做调整;5、欢迎大家对这个内容选辑提出自己的意见和见解。


LLVM邮件列表2018年4月内容选辑

一、前端:

1、Mostyn Bramley-Moore shared work on supporting 'Jumbo'/unity builds with Clang. The proof of concept clang plugin gives unique name to anonymous namespaces and undefs macros defined in each top-level file. This reduces the modifications needed to a large codebase like Clang to support a Jumbo build. Much of the ensuing discussions focus on the merits of Jumbo/unity builds vs Clang header modules.

2、Katya Romanova has started a discussion on unifying the bitcode file generation phase for ThinLTO and FullLTO, producing a 'unified' bitcode format. Mehdi Aminiproposed a change in the compilation flow that might meet some of Katya's requirements.

3、Manoj Gupta kicked off an RFC discussion on implementing -fno-delete-null-pointer-checks in Clang. This flag is used when building the Linux kernel. This flag has the semantics "Assume that programs cannot safely dereference null pointers, and that no code or data element resides at address zero". The proposal has triggered quite a lot of discussion so far.



扫描二维码关注公众号,回复: 3509446 查看本文章

二、后端:

1、Alex Susu reports that source code is now publicly available for an LLVM backened for the research Connex SIMD processor.

2、Simon Tatham is interested in contributing a JSON backend for TableGen, providing a machine-readable alternative to -print-records. An initial patch is already up for review.

3、Jin Lin has an RFC on representing target device information in LLVM IR, which is needed for OpenMP backend outlining.



三、优化:

1、Chad Rosier proposes adding function attributes to represent the compile-time optimisation level.

2、Jessica Paquette proposes turning the MachineOutliner on by default in AArch64 under -Oz. As Jessica explains, the hope is that the same outlining algorithm and interface can be used for the proposed IR-level outlining pass. On CTMark, outlining is giving an extra 4.4% geomean text size reduction for the cost of a geomean 1.1% compile-time overhead.



四、RISC-V

1、Robin Kruppe has written an RFC on supporting the proposed RISC-V Vector extension in LLVM. This requires representing a hardware vector length that is unknown at compile time, and may change at runtime. Robin is presenting a lightning talk based on this proposal at EuroLLVM and is keen to discuss further.


五、企业:

1、Alex L announces that Apple have decided to switch focus from the libclang-based tooling infrastructure in order to join forces on Clangd development. They intend to contribute towards making Clangd faster, more capable, and more efficient.

2、The Google Summer of Code community bonding period is starting up. Welcome to Alexander Polyakov who will be working on LLDB-mi and to Anastasis Grammenoswho will be working on improving debugging of optimized code.


六、扩展:

1、Joel E. Denny is looking for feedback on a change to lit to have it report the line number of a failed RUN command.

2、Rui Ueyama has shared an RFC on adding a new --check-library-dependencycommand line option to lld, which can be used to get symbol resolution semantics that are closer to the GNU linker.

3、Ciro Ceissler shares news of the HardCloud OpenMP extension for offloading to FPGA accelerators.

4、Following on from his lightning talk at EuroLLVM, Kristof Beyls has shared a script for matching up open Phabricator reviews to potential reviewers and is seeking feedback.

猜你喜欢

转载自blog.csdn.net/snsn1984/article/details/80315139
今日推荐