On 03/13/2017 02:01 PM, Richard Biener wrote: > On Mon, 13 Mar 2017, Richard Biener wrote: > >> On Fri, 10 Mar 2017, Martin Liška wrote: >> >>> Hello. >>> >>> As briefly discussed in the PR, there are BB that do not correspond to a >>> real >>> line in source >>> code. profile.c emits locations for all BBs that have a gimple statement >>> belonging to a line. >>> I hope these should be marked in gcov utility and not added in --all-block >>> mode to counts of lines. >>> >>> Patch survives make check RUNTESTFLAGS="gcov.exp". >>> >>> Thanks for review and feedback. >> >> Humm, the patch doesn't seem to change the BBs associated with a line >> but rather somehow changes how we compute counts (the patch lacks a >> description of how you arrived at it). I expected the line-to-BB >> assignment process to be changed (whereever that is...).
Currently, each basic block must belong to a source line. It's how gcov iterates all blocks (via lines). > > Ah, ok, looking at where output_location is called on I see we do not > assign any line to that block. But still why does > line->has_block (arc->src) return true? Good objection! Problematic that 4->5 edge really comes from the same line. <bb 4> [0.00%]: ret_7 = 111; PROF_edge_counter_10 = __gcov0.UuT[0]; PROF_edge_counter_11 = PROF_edge_counter_10 + 1; __gcov0.UuT[0] = PROF_edge_counter_11; <bb 5> [0.00%]: # ret_1 = PHI <ret_5(3), ret_7(4)> goto <bb 7>; [0.00%] Martin > > Richard. >
