[QUOTE=Amateurprofi;1528396]
Excellent work.
I have few suggestion:
1) Switch from using Integer to NativeUInt or NativeInt, this will pay in x64, as the compiler will not have to insert resizing instructions like movzx and will have the ability to use full register operation.
2) Replace that EnsureRange with simple old fashion if-statement, saving a needless branch.
3) I wouldn't trust the compiler to generate fast div every time when the division is by 2^n, proof this by replacing them with shr n, so div 16 can be shr 4.
4) This is the meat of this and i think it should pay on low cache CPU's or big images or very busy CPU, instead of getting the last line which have the index 0 then go backward "PP:=Bmp.ScanLine[0];" replace with getting the first line and move forward, also for X there is no point of walking backward, see, with huge images, and walking backward the cache lines will continuously be read in backward causing violation and request to update, while the CPU request its cache lines in bulk forward most the time, so accessing the memory backward with thrash the cache and waste time and cycles waiting for memory.
Have tested your suggestions.
1) NativeInt instead of Integers
Doubles the used time.
2) Replacing EnsureRange bei If then ..,
Adds another few ms
3) shr instead of Div 16
Commented earlier
4) Starting with the Bottom Line
Not tested, would expect advantages
4) X forward
Not tested would expect disadvantages (see the "if X<>..." in the X-Loop)
The measured runtimes
1339 With Integers
2559 With NativeInts
2606 EnsureRange replaced by If-Then