Red Hat Compiler Developer Working On Compiler-Assisted Performance Analysis For GCC

[phoronix]

Phoronix: Red Hat Compiler Developer Working On Compiler-Assisted Performance Analysis For GCC

Longtime GNU toolchain developer at Red Hat, David Malcolm, has announced the work he is pursuing on compiler-assisted performance analysis with GCC...

http://www.phoronix.com/scan.php?page=news_item&px=GCC-Comp-Assist-Perf-Analysis

[bregma]

This is very very nice. Hopefully it can be backported to GCC 8 so my customers can take advantage of it before 2023.

The next step will be to feed the output of that effort into some deep-learning robot and get suggestions on how to rewrite your code for minimal footprint, maximal throughput, or improved safety. It will not be a long wait.

[pyler]

So many cool things in 8.1 and planned for 9.0.

Also GCC 8.1 in my tests generates far better optimized code than Clang 7 SVN. But! Also GCC now compiles programs way faster than Clang, they did really good work to improve compile time.

Clang needs to bring some really cool things to catch up GCC. GCC focuses on the compiler, LLVM on many tools, and compiler progress is slow, no cool new optimizations like in GCC

[pyler]

Cool new optimizations in GCC 8.1/9.0 but Clang is lagging here. Also GCC compiles code now faster than Clang

[gufide]

GCC is nice, but 8.1 had many regressions with sfinae. There is one I reported and got fixed, but another one more important I wasn't able to isolate. I'll report as soon as I have a minimal code that triggers the bug.

[slacka]

So many cool things in 8.1 and planned for 9.0.

Also GCC 8.1 in my tests generates far better optimized code than Clang 7 SVN.

(

If you build with the New GVN by using the "-mllvm -enable-newgvn" option, real world apps like OpenArena sees about a 5% improvement. And with many compiler benchmarks, clang once again takes the lead.

[pyler]

I tested both with Spec2006 (int) and GCC scored 33,2 vs clang 31,4

[timofonic]

I just hope this is the start of a new beginning in GCC.

GCC is a mature project, but needs to renew itself and evolve a lot more faster. I remember nostalgically the EGCS vs GCCdays.

I wish this could evolve beyond Compiler-Assisted Performance Analysis: Compute-parsable output, able to use it with tools that automate performance analysis tasks beyond LLVM and others could do (even Intel tools, why not?).

PS: I also wish a FOSS IDE that doesn't suck too. It's OK to use emacs, vim/neovim and others, but it's not going to be a mainstream solution. I'm waiting for one with the good aspects of the Visual Studio + JetBrains ecosystem, both Microsoft and JetBrains do many things excellently in the software development technologies (despite they're proprietary, they have the merit). KDevelop, Eclipse, Qt Creator, GNOME Builder and others are crappy compared to Visual Studio/CLion. Refactoring is one of the aspects where ALL FOSS IDEs suck, for example. Maybe having better reusable FOSS technologies could make the FOSS ecosystem better in development solutions and FOSS IDEs become "magically" better by using them. But this is another story