Per our discussion at yesterday's support meeting I ran a comparison of the nightly regression files with -O2 and -O0. These the files with differences on my i7 box: 23-33.PS.pbmraw.72.0 23-33.PS.pdf.pkmraw.300.0 23-33.PS.pdf.ppmraw.300.0 23-33.PS.pdf.ppmraw.72.0 23-33.PS.pgmraw.72.0 23-33.PS.pkmraw.300.0 23-33.PS.pkmraw.300.1 23-33.PS.pkmraw.72.0 23-33.PS.ppmraw.300.0 23-33.PS.ppmraw.300.1 23-33.PS.ppmraw.72.0 30-11.PS.pbmraw.300.0 30-11.PS.pbmraw.300.1 30-11.PS.pbmraw.72.0 30-11.PS.pgmraw.300.0 30-11.PS.pgmraw.300.1 30-11.PS.pgmraw.72.0 30-11.PS.pkmraw.300.0 30-11.PS.pkmraw.300.1 30-11.PS.pkmraw.72.0 30-11.PS.ppmraw.300.0 30-11.PS.ppmraw.300.1 30-11.PS.ppmraw.72.0 Bug688631.pdf.pbmraw.300.0 Bug688631.pdf.pbmraw.300.1 Bug688631.pdf.pbmraw.72.0 Bug688631.pdf.pgmraw.300.0 Bug688631.pdf.pgmraw.300.1 Bug688631.pdf.pgmraw.72.0
Bug688631.pdf appears to be indeterministic, multiple runs with -O2 produce different md5sums (as do multiple -O0 runs).
I'm closing this bug, the three files that show differences between -O0 and -O2 also show up as indeterministic, so it's likely that that there aren't any actual difference based on compiler optimization levels.