|
Written by Daniel P
|
|
Thursday, 22 June 2006 |
|
I rewrote Peter Warden's wave video effect in C#. My first attempt was a straigthforward
cut at it and the result wasn't pretty. I managed to treat 45-48 frames out of a total
of 292. I looked at the original C++ code and it treated 261-262 frames. So I used
unsafe code and rewrote my code in C# but "in a C-style". The result was much better.
256-258 frames were treated with my newer version or within 1-3% of the C/C++ equivalent.
In fact, unsafe code with pointers in C# reminds me a lot of the inline assembly feature of
VC++, for example.
Then I searched Google and found an interesting collection of articles by
Mr. Net Performance, Rico Mariani, which stated: "The managed code got a very good
result for hardly any effort".
|
|
Written by Daniel P
|
|
Thursday, 15 June 2006 |
|
While working on my FreeFrame wrapper I encountered an interesting situation. The
only function exported by the dll of a FreeFrame video effect has a signature that
only uses "unsigned int" in order to ease portability. But this function is used to
return (single precision) floating-point values (stored in an unsigned int). Unsigned
integers in the .Net world are not CLS-compliant so I created a COM interface in which
these functions return float values and I just make the translation in the COM component
using the familiar C/C++ code:
float * fptr = (float *)&tmpResult; //tmpResult is unsigned int
*finalResult = *fptr;
Then I realized that I could use the same code with C# in an unsafe method. I tried
and I checked that the assembly code for the C/C++ version and the C# version is identical
(a "lea" instruction followed by a "mov"). Decidedly, we need C/C++ in fewer and fewer
places.
|
|
|
<< Start < Prev 1 2 3 4 5 6 7 8 9 10 Next > End >>
|
| Results 37 - 40 of 81 |
