On Thu, Jul 11, 2024 at 1:03 AM Jeff Law <jeffreya...@gmail.com> wrote: > > > > On 6/3/24 5:34 AM, Manolis Tsamis wrote: > > This is an extension of what was done in PR106590. > > > > Currently if a sequence generated in noce_convert_multiple_sets clobbers the > > condition rtx (cc_cmp or rev_cc_cmp) then only seq1 is used afterwards > > (sequences that emit the comparison itself). Since this applies only from > > the > > next iteration it assumes that the sequences generated (in particular seq2) > > doesn't clobber the condition rtx itself before using it in the > > if_then_else, > > which is only true in specific cases (currently only register/subregister > > moves > > are allowed). > > > > This patch changes this so it also tests if seq2 clobbers cc_cmp/rev_cc_cmp > > in > > the current iteration. This makes it possible to include arithmetic > > operations > > in noce_convert_multiple_sets. > > > > It also makes the code that checks whether the condition is used outside of > > the > > if_then_else emitted more robust. > > > > gcc/ChangeLog: > > > > * ifcvt.cc (check_for_cc_cmp_clobbers): Use modified_in_p instead. > > (noce_convert_multiple_sets_1): Don't use seq2 if it clobbers cc_cmp. > > Refactor the code that sets read_comparison. > > > > Signed-off-by: Manolis Tsamis <manolis.tsa...@vrull.eu> > I'm assuming you've bootstrapped and regression tested on x86, aarch64 > or some other primary platform. If that assumption is true, then this > is fine for the trunk. > Hi Jeff,
Since some time had passed from when I had submitted this I rerun the tests and bootstrapped again on x86-64 and AArch64. The x64 bootstrap failed and after debugging the issue I fixed a remaining flags-related bug due to allowing all the new arithmetic operations in multi statement ifcvt. I have submitted the fixed version as v5 of this patch. I have bootstrapped and regtested the new patch on the two platforms and tested various benchmarks and everything looks clean now. Thanks, Manolis > Jeff >
