Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.compilers > #294 > unrolled thread

benchmarks for global value numbering

Started byZjzzjz <jianzhou.zh@gmail.com>
First post2011-10-20 06:14 -0400
Last post2011-11-16 07:25 -0800
Articles 3 — 3 participants

Back to article view | Back to comp.compilers


Contents

  benchmarks for global value numbering Zjzzjz <jianzhou.zh@gmail.com> - 2011-10-20 06:14 -0400
    Re: benchmarks for global value numbering Nikolaos Kavvadias <nikolaos.kavvadias@gmail.com> - 2011-10-26 14:31 -0700
    Re: benchmarks for global value numbering andreyb <andreybokhanko@gmail.com> - 2011-11-16 07:25 -0800

#294 — benchmarks for global value numbering

FromZjzzjz <jianzhou.zh@gmail.com>
Date2011-10-20 06:14 -0400
Subjectbenchmarks for global value numbering
Message-ID<11-10-012@comp.compilers>
Hi everyone,

I was wondering what are good benchmarks for a global value numbering
optimization. The GVN should optimize both temporaries and memory
operations. For testing the GVN, should I run the standard spec
benchmark first or are there any typical 'small' functions that I can
try and see how GVN can change performance? I think for developing,
small programs may help. Thanks.

[toc] | [next] | [standalone]


#300

FromNikolaos Kavvadias <nikolaos.kavvadias@gmail.com>
Date2011-10-26 14:31 -0700
Message-ID<11-10-018@comp.compilers>
In reply to#294
Hi there

> I was wondering what are good benchmarks for a global value numbering
> optimization.

I recall a technical report called "Optimizer Evaluation" by Preston
Briggs. You could find this from around the web.

This report is/was accompanied by a set of C benchmarks, with each one
exercising a different compilation phase/aspect.

Nowadays, the "OptimizerEval" benchmarks are part of the LLVM
Multisource benchmark suite.

Kind regards
Nikolaos Kavvadias

[toc] | [prev] | [next] | [standalone]


#338

Fromandreyb <andreybokhanko@gmail.com>
Date2011-11-16 07:25 -0800
Message-ID<11-11-036@comp.compilers>
In reply to#294
You *really* should use SPECs to evaluate your algorithm /
implementation.

Small tests are articifial at best. If you need them as an aid for
development, you can easily write them yourself.

Sorry for general (and, frankly, not very useful) answer, but the
things are as they are.

Andrey

[toc] | [prev] | [standalone]


Back to top | Article view | comp.compilers


csiph-web