Just changing Int32ToDouble to be predicted SpecInt48 breaks constant folding on that node because of soooper old code that prevented constant folding on mismatched predictions. Kill that code.
Created attachment 211282 [details] the patch
Comment on attachment 211282 [details] the patch r=me
Created attachment 211345 [details] even better patch Introduce setConstant(node, value) method that does both of these things: forNode(node).set(m_graph, value) m_state.setFoundConstants(true)
Comment on attachment 211345 [details] even better patch Attachment 211345 [details] did not pass mac-wk2-ews (mac-wk2): Output: http://webkit-queues.appspot.com/results/1738828 New failing tests: js/dfg-constant-fold-misprediction.html
Created attachment 211347 [details] Archive of layout-test-results from webkit-ews-10 for mac-mountainlion-wk2 The attached test failures were seen while running run-webkit-tests on the mac-wk2-ews. Bot: webkit-ews-10 Port: mac-mountainlion-wk2 Platform: Mac OS X 10.8.4
(In reply to comment #4) > (From update of attachment 211345 [details]) > Attachment 211345 [details] did not pass mac-wk2-ews (mac-wk2): > Output: http://webkit-queues.appspot.com/results/1738828 > > New failing tests: > js/dfg-constant-fold-misprediction.html Because I forgot to upload the test change with this patch like I did with the last one.
Landed in http://trac.webkit.org/changeset/155567
I think this introduced a regression, see https://bugs.webkit.org/show_bug.cgi?id=123664.