Hi Andrew! On 2022-10-28T10:38:11+0200, "Stubbs, Andrew" <andrew_stu...@mentor.com> wrote: >> -----Original Message----- >> Looking into commit r13-3225-gbd9a05594d227cde79a67dc715bd9d82e9c464e9 >> "amdgcn: vector testsuite tweaks" for a moment, I also did wonder about >> the following changes, because for 'vect_multiple_sizes' (for example, >> x86_64-pc-linux-gnu) that seems to lose more specific testing; >> previously: 'scan-tree-dump-times' exactly once, now: 'scan-tree-dump' >> any number of times. But I've no clue about that myself, so just >> mentioning this, in case somebody else has an opinion. ;-) > > When vect_multiple_sizes is true the number of times the pattern appears will > be greater that normal. Most likely the pattern will appear once for each > vector size. In the case of GCN, a pattern that normally appears 4 times now > appears 24 times. > > The alternative would be to have a whole set of patterns for each > configuration of each target that can have the multiple sizes. That or > change the implementation of 'scan-tree-dump-times' to support expressions of > some kind, but even then the expressions would get hairy.
I guess my confusion is why this then hasn't already previously be FAILing for example for x86_64-pc-linux-gnu, which also is 'vect_multiple_sizes'? Anyway: "I've no clue about that myself". Grüße Thomas ----------------- Siemens Electronic Design Automation GmbH; Anschrift: Arnulfstraße 201, 80634 München; Gesellschaft mit beschränkter Haftung; Geschäftsführer: Thomas Heurung, Frank Thürauf; Sitz der Gesellschaft: München; Registergericht München, HRB 106955
