Thanks,
Interesting paper, and it illustrates the issue in an easily reproducible way.
It is just one more validation of the key underlying principle:
https://en.wikipedia.org/wiki/Everything_You_Know_Is_WrongDave
On Sun, Feb 24, 2019 at 11:01:21AM -0800, tim Rowledge wrote:
>
>
http://sci-hub.tw/https://dl.acm.org/citation.cfm?id=1508275>
> "For example, we had arrived at the conclusion: Changing UNIX environment size causes a change in start address of the stack which causes a change in O3 speedup."
>
> As in, decades ago when Eliot & I were doing stuff based on his BrouHaHa VM I had a dynamic-compilation bitblt for ARM. It was wonderful, dozens of times faster than a basic blt written in C. BUT the performance could vary wildly from build to build and eventually I tracked it down to the alignment of the generated code. And to make things even weirder, remember that in those days the ARM had no caches to worry about aligning with - not even an instruction prefetch queue.
>
>
> tim
> --
> tim Rowledge;
[hidden email];
http://www.rowledge.org/tim> Strange OpCodes: FSRA: Forms Skip and Run-Away
>
>
Locked
