Hi Luc, On 10 August 2017 at 00:51, Luc Van Oostenryck <luc.vanoostenryck@xxxxxxxxx> wrote: > On Thu, Aug 10, 2017 at 1:32 AM, Dibyendu Majumdar > <mobile@xxxxxxxxxxxxxxx> wrote: >> >> Yes I suppose that is what it is. LLVM complains that 'instruction >> does not dominate all uses'. >> The unsimplified IR works fine >> (https://github.com/dibyendumajumdar/dmr_c/blob/master/tests/bugs/simplifybug.lin). >> The simplified IR has an issue >> (https://github.com/dibyendumajumdar/dmr_c/blob/master/tests/bugs/simplifybug_O1.lin). > > Yes, I saw now. It's hopeless. > It's a problem with the SSA construction (happen between linearization > and simplification), basically, it puts the phi-nodes where there is a load > but it's often too late, they need to be put where paths join. > >>> You may try the new SSA construction at : >>> https://github.com/lucvoo/sparse/tree/sssa >>> It will help a lot. >>> >> >> Is that a standalone change I can apply? > > Arf ... in a sense yes but it's really a 40 patches series > (but most changes are isolated, maybe not too much work for you > I don't know). > I am merging the current changes in RC5 to my repository. When that is done, I would like to try out your new SSA construction approach. Please let me know if this is complete (i.e. functional). I had a look at the branch above, but not all commits appeared to be related to the SSA construction, so any pointer to the list of changes I should pick would be very helpful. I can test the new approach in my repository and report back. Regards Dibyendu -- To unsubscribe from this list: send the line "unsubscribe linux-sparse" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
