[llvm] r210539 - SelectionDAG: Expand SELECT_CC to SELECT + SETCC

[Tom Stellard]

On Wed, Jun 11, 2014 at 09:55:40AM -0700, Arnold Schwaighofer wrote:
> Yes,  I have replied to the wrong commit. git bisect points to 210540.
> 

Ok, that makes a little more sense.  If you could get a reduced test case that would
be great.  A diff of the complied assembly may help too.

-Tom



> git bisect start
> # bad: [9c84db6ba390f4e6ac1d6ce94fa9a6428e0b3007] Use unique_ptr for X86Subtarget pointer members.
> git bisect bad 9c84db6ba390f4e6ac1d6ce94fa9a6428e0b3007
> # good: [45a31492d5475a1744966a0cd5e631694b29f953] Condition codes AL and NV are invalid in the aliases that use inverted condition codes (CINC, CINV, CNEG, CSET, and CSETM).
> git bisect good 45a31492d5475a1744966a0cd5e631694b29f953
> # bad: [81ff9cdedbbe09f699fa8ac81a00231c7737b58f] Use an enum class now that they are available.
> git bisect bad 81ff9cdedbbe09f699fa8ac81a00231c7737b58f
> # bad: [8128a7965f2d24554a4c76db579b2625f9be135b] Remove the uses of AArch64TargetMachine and AArch64Subtarget from AArch64FrameLowering.
> git bisect bad 8128a7965f2d24554a4c76db579b2625f9be135b
> # bad: [102d0f3e3f108fec94cf998e2b1bb2fcf50657c3] SelectionDAG: Don't use MVT::Other to determine legality of ISD::SELECT_CC
> git bisect bad 102d0f3e3f108fec94cf998e2b1bb2fcf50657c3
> # good: [b02d95cb66157aa49bf5e50a8778c182a7b8604e] [PPC64LE] Recognize shufflevector patterns for little endian
> git bisect good b02d95cb66157aa49bf5e50a8778c182a7b8604e
> # bad: [c23f0e1e446426190f563fcb89ef644927f7f438] SelectionDAG: Enable (and (setcc x), (setcc y)) -> (setcc (and x, y)) for vectors
> git bisect bad c23f0e1e446426190f563fcb89ef644927f7f438
> # good: [f586a260ca729c933cb15c6e68cabe95163a6dcb] SelectionDAG: Expand SELECT_CC to SELECT + SETCC
> git bisect good f586a260ca729c933cb15c6e68cabe95163a6dcb
> # first bad commit: [c23f0e1e446426190f563fcb89ef644927f7f438] SelectionDAG: Enable (and (setcc x), (setcc y)) -> (setcc (and x, y)) for vectors
> 
> > On Jun 11, 2014, at 9:51 AM, Tom Stellard <tom at stellard.net> wrote:
> > 
> > On Wed, Jun 11, 2014 at 09:45:04AM -0700, Arnold Schwaighofer wrote:
> >> Okay.
> >> 
> >> Execution time failure (different output).
> >> 
> >> I’ll try to reduce this down more.
> > 
> > Are you sure the regression was caused by: r210539 and not r210540 or r210541?
> > Those last two are much more likely to impact ARM and X86.
> > 
> > -Tom
> >> 
> >>> 
> >>> On Jun 11, 2014, at 9:33 AM, Tom Stellard <tom at stellard.net> wrote:
> >>> 
> >>> On Wed, Jun 11, 2014 at 08:20:05AM -0700, Arnold Schwaighofer wrote:
> >>>> 
> >>>> Please let me know if you have access to spec2006.
> >>>> 
> >>>> 
> >>> 
> >>> I don't have access to spec2006.  What kind of failure is it, a miscompile or a crash?
> >>> 
>