Janez Atmapuri Makovsek found some #XE6 array optimizations
Inline declared setter and getter functions can yield up to six times the performance on one-dimensional arrays and up to four times the performance on 2D arrays, compared to XE5.
Spotted at the EMBT forums.
Well now, that is great news indeed.
Word on the street is that this inline optimization has broken AtomicIncrement where the increment value is not a constant. So that’s not so good news…
We use MtxVec for years (since version 1.5 on D7), highly recommended (although we didn’t try yet the OpenCL specific functions)
Asbjørn Heid Is the AtomicIncrement issue documented somewhere?
Leif Uneus I only heard it via-via, but from reliable source. I presumed it would get logged in QC (due to source) but i can’t seem to find it.