Nothing major. Two minor changes:
Edge frequencies are now compared with BB frequencies in verify_cgraph_node(). This showed an inconsistency in the edge created in ipa_tm_insert_gettmclone_call(). Fixed. The function free_lang_data():tree.c no longer runs for all front-ends, so lang_hooks.decl_printable_name is not set to gimple_decl_printable_name which demangled names. This causes IPA dumps to no longer have prettified "transaction clone for func()". I've changed the test depending on this behavior to match the actual mangled assembler symbol. No regressions. Committing merge to branch. Aldy
