[LLVMbugs] [Bug 16140] Instrcombine increases register usage significantly

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=16140

Justin Holewinski <justin.holewinski at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
                 CC|                            |justin.holewinski at gmail.com
         Resolution|---                         |FIXED

--- Comment #2 from Justin Holewinski <justin.holewinski at gmail.com> ---
The crash is now fixed since we are back to emitting virtual registers in PTX
(so spilling is not possible).  And the standard compilation passes are not
showing any regression in register usage:

(no optimization)
# llc -march=nvptx:

ptxas info    : 0 bytes gmem
ptxas info    : Compiling entry function 'main' for 'sm_35'
ptxas info    : Function properties for main
    0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
ptxas info    : Used 43 registers, 348 bytes cmem[0], 4 bytes cmem[2]


(-O3 including instcombine)
# opt -O3 | llc -march=nvptx:

ptxas info    : 0 bytes gmem
ptxas info    : Compiling entry function 'main' for 'sm_35'
ptxas info    : Function properties for main
    0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
ptxas info    : Used 40 registers, 348 bytes cmem[0], 4 bytes cmem[2]


# opt -instcombine | llc -march=nvptx

ptxas info    : 0 bytes gmem
ptxas info    : Compiling entry function 'main' for 'sm_35'
ptxas info    : Function properties for main
    0 bytes stack frame, 0 bytes spill stores, 0 bytes spill loads
ptxas info    : Used 43 registers, 348 bytes cmem[0], 4 bytes cmem[2]


Instcombine is an interesting case here, though.  Running it on the attached
kernel takes a long time (currently 10 minutes and counting...).  Frank, if
you're concerned with instcombine compilation time, please open a different
bug.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20130624/ad0baa58/attachment.html>