11-Way Intel Ivy Bridge Compiler Comparison

[smitty3268]

You've been under a rock perhaps? The binaries by ICC will suck on AMD, VIA and anything else non-Intel x86. See Agner's site for insightful explanations, including benchmarks where he changes his Via to introduce as Intel.

I'm pretty sure one of the conditions of the lawsuit Intel settled with AMD was that they removed any special de-optimizations for AMD hardware in their compiler.

I'm sure they still don't go to any special lengths like scheduling instructions properly for AMD hardware, but it should properly take advantage of all the SSE/AVX codepaths now when doing so requires no extra work and Intel was specifically turning it off for no reason before.

[curaga]

I'm pretty sure one of the conditions of the lawsuit Intel settled with AMD was that they removed any special de-optimizations for AMD hardware in their compiler.

I'm sure they still don't go to any special lengths like scheduling instructions properly for AMD hardware, but it should properly take advantage of all the SSE/AVX codepaths now when doing so requires no extra work and Intel was specifically turning it off for no reason before.

Yes, they removed the de-optimization, but their automatic dispatch still works based on a "GenuineIntel" CPUID string. That means that on a non-Intel CPU, ICC-compiled binaries will choose the x87 path on 32-bit and SSE2 on x64, no better.

[kukreknecmi]

Actually Intel Compilers produce fastest code for AMD too in Windows enviroment, if you compare with other compilers avaible.

On topic: I wonder why c-ray scores are so different from previous tests, which were around 27-32 secs for i7 3770k.