Re:Mozilla can produce near-native performance on the Web
by azakai on Thursday May 23, 2013 @01:41AM
(#5755519)
Why would you assume they disabled SSE? Compilers generate SSE by default when optimizing, I would be very surprised if Peter Bright disabled it. And not only do normal C/C++ compilers generate SSE, but JS engines generate SSE as well. So SSE was being used on both sides of the comparisons here, I am pretty sure.
What was disabled was C/C++ code that explicitly used SIMD intrinsics or explicitly used SSE function calls in C/C++.
Re:Mozilla can produce near-native performance on the Web
by haberman on Thursday May 23, 2013 @01:50AM
(#5755535)
I must have misinterpreted this remark:
> In general, I took the highest-ranked pure C++ routines for each test. Versions that used, for example, SSE functions were excluded, as were those with explicit multithreading.
For "functions" I interpreted "functionality," but I think now it must mean SSE intrinsics?
Sorry for the confusion.
Re:Mozilla can produce near-native performance on the Web
by DrPizza on Thursday May 23, 2013 @02:43AM
(#5755633)
Yes, sorry, I meant SSE intrinsics. As I think I note later in the article, the compiler I used emitted scalar SSE2 for floating point.
Re:Mozilla can produce near-native performance on the Web
by Gravityloss on Thursday May 23, 2013 @02:16AM
(#5755586)
I'm glad you made the comment, others might read it the same way as you did originally. This clears the potential misunderstanding.
[EDIT: it looks like I may have misinterpreted the article; see replies]
To clarify, I believe they ran the benchmarks with SSE disabled. SSE is more than just SIMD; these days SSE is used for most (all?) floating-point calculations, even non-SIMD. If you disable SSE, you force floating-point ops to use the old x87 FPU, which has been discouraged by the CPU manufacturers for over 10 years.
In other words, disabling SSE is significantly crippling all floating-point performance.
http://www.realworldtech.com/physx87/4/
http://stackoverflow.com/questions/3206101/extended-80-bit-d...
(nothing against asm.js, which I think is very cool)