Firefox Developers Continue Tuning ASM.js Performance

[phoronix]

Phoronix: Firefox Developers Continue Tuning ASM.js Performance

Mozilla developers continue to optimize everything they can with ASM.js to deliver the greatest performance possible inside Firefox...

http://www.phoronix.com/vr.php?view=MTU1MDM

[plonoma]

Responsibity matters sometimes more than performance.

Seriously the responsivity issues are more pressing than the perofrmance of ASM.js is now.

[iniudan]

Responsibity matters sometimes more than performance.

Seriously the responsivity issues are more pressing than the perofrmance of ASM.js is now.

Your forgetting asm.js is extremely useful for the low end device powered by FirefoxOS, has all the application on it, are web application.

[balouba]

Responsibity matters sometimes more than performance.

Seriously the responsivity issues are more pressing than the perofrmance of ASM.js is now.

faster means more responsive anyway. task finishes faster means less time locking up.
only way to fix responsivity issues with poorly made js is multiprocess which they also work on

[mark45]

faster means more responsive anyway. task finishes faster means less time locking up.
only way to fix responsivity issues with poorly made js is multiprocess which they also work on

Asm.js isn't meant to be used by hand, it's too low level, it only solves the performance issue of porting certain apps from native languages.
It's not a cure, it's a painkiller, and hence has limited future and also widely depends on how Dart is gonna do.

[markg85]

I find the speed tests from mozilla very false.
They compare against clang 3.2 as "native" whereas their same gcc (3.6) tests are outperforming clang in all but one test.

Then you still have the version matter. Clang improved quite a bit in 3.3. Gcc also improved quite a bit in 3.7 and 3.8. Specifically if you have intel with all those new optimizations.

So yeah, false results of mozilla.

[azakai]

I find the speed tests from mozilla very false.
They compare against clang 3.2 as "native" whereas their same gcc (3.6) tests are outperforming clang in all but one test.

Then you still have the version matter. Clang improved quite a bit in 3.3. Gcc also improved quite a bit in 3.7 and 3.8. Specifically if you have intel with all those new optimizations.

So yeah, false results of mozilla.

As I wrote in a comment on the blog, yes, the version can matter. I didn't test the very latest LLVM and gcc versions because the goal wasn't to see the very limit of performance, but get a more broad picture of what results people can get by using typical native compilers (most applications people use are not compiled with bleeding edge compilers).

So I used the gcc that came with my linux distro - it's the one that built practically every single app I run on my machine - and for LLVM, I used the same version used by emscripten, which means it has identical frontend optimizations and the comparison of native to JS is more apples-to-apples. But again, I agree that testing the very latest is interesting as well, just depends on what your purpose is.

[liam]

I find the speed tests from mozilla very false.
They compare against clang 3.2 as "native" whereas their same gcc (3.6) tests are outperforming clang in all but one test.

Then you still have the version matter. Clang improved quite a bit in 3.3. Gcc also improved quite a bit in 3.7 and 3.8. Specifically if you have intel with all those new optimizations.

So yeah, false results of mozilla.

Yeah... no. You should've read the blog post. They said, specifically, that they also wanted to highlight the notion of"native". The speed of native being very much dependent on the compiler. Moreover, emscripten is currently using that particular version of llvm/clang so you can do an apples to apples comparison.

[smitty3268]

I find the speed tests from mozilla very false.
They compare against clang 3.2 as "native" whereas their same gcc (3.6) tests are outperforming clang in all but one test.

Then you still have the version matter. Clang improved quite a bit in 3.3. Gcc also improved quite a bit in 3.7 and 3.8. Specifically if you have intel with all those new optimizations.

So yeah, false results of mozilla.

Where's your proof that the most recent version of compilers make a difference for these micro-benchmarks?

It's not like they're using ancient compilers. I'd be pretty surprised if it made a huge difference.

And compiling with some march flags to recent cpus is what would be cheating, because the point of this code is to run across common computers like it would be compiled for a common distro.

[Thaodan]

Your forgetting asm.js is extremely useful for the low end device powered by FirefoxOS, has all the application on it, are web application.

you WASTE 50% speed on an already low powered device?