Hi,
i want to speak out a very big laud to the .NET compiler teams. Especially to the C++ compiler team. Beside some bugs and issues that all beta soft has i tested the speed and asm listings of mathematical code that the C++ compiler of Whidbey produces on x64 platform and i must say the code is really fast compared to hand optimized code.
My hand optimized code is just 10 - 15 % faster than the C++ compiler code ( mean ) and i have to think over whether i code some speed critical functions in asm, because of the very little speed advantage.
The compiler only produces SSE and SSE2 code by default on x64 and it is very fast.
On x86 hand optimzed asm codes are by default much faster than the code written by the compiler, because SSE is not used by default.
CYA
Martin
BIG LAUD TO THE C++ COMPILER TEAM ( .NET compiler generally )
Hot Topic
- Function pointer for GC
- Where is C99 round function?
- VC++ 8 Show-Stopper?
- linking problems (invalid or corrupt file)
- native members of managed classes?
- cannot open file 'comdlg.lib'
- Public String Constants in .NET Assemblies
- Problems with CCmdTarget.h
- Why is VC8 code much slower than VC7.1?
- Linking-Error with VSC++ 8.0
Visual C++
- Error C2440: cannot convert from type to type
- Coding
- IOStream Bug on Visual Studio 2005 when open Chinese style file name's text file???
- Lost all visual style on UI controls after importing to VS2005
- Getting a certain Line of a string
- Is there any thing equivalent to NotOverridable in C++?
- Debugging inside a DLL
- How to initialize an independent display using CreateDC
- Making dlls with Visual C++ Express Edition Beta 2
- ClassView - missing classes
BIG LAUD TO THE C++ COMPILER TEAM ( .NET compiler generally )
Sven Beneke
Thanks a lot for you comment and feedback!
I will make sure to forward it to the rest of VC++ folks.
BTW, on x86, you can generate SSE & SSE2 asm by using the /arch:SSE & /arch:SSE2 compiler switches respectively. The compiler in such cases doesn't generate fully SSE[2] instructions but a mix of x86 & SSE[2] which could be faster in some cases.
Once more, thanks for your feedback Martin!
Thanks,
Ayman Shoukry
VC++ Team