On Wed, Nov 26, 2014 at 7:55 AM, Matt Turner <[email protected]> wrote: > On Thu, Nov 20, 2014 at 8:09 PM, Zhigang Gong <[email protected]> wrote: >> Register spilling awlays cost much more than fallback to simd8 >> which could avoid register spilling or at least reduce the spilled >> registers. > > For what it's worth, we made the same decision in the i965 Mesa driver. > > There has been some conjecture that a spilling SIMD16 program *could* > potentially be faster than a non-spilling SIMD8 program, but I don't > know of any hard evidence.
Thanks for the comment. I thought if a kernel is totally computational intensity and most of the computation occurs in a loop and we just spill a few registers out of any loops. Then it may be faster than a non-spilling simd8 kernel. But all the real case we have could only prove that spilling simd16 is much slower than non-spilling simd8. > _______________________________________________ > Beignet mailing list > [email protected] > http://lists.freedesktop.org/mailman/listinfo/beignet _______________________________________________ Beignet mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/beignet
