[llvm-commits] Patch: Floating point optimizations for SimplifyInstruction

[Michael Ilseman]

Here's two patches to follow the prior two. The first one removes the optimizations that are made redundant in InstCombine, and has InstCombine call the Simplify helpers for FAdd/FSub/FMul just like all the other parts of InstCombine. The second updates SimplifyBinOp to call the new helpers (using default, or zero, fast-math flags).

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-Remove-redunant-optimizations-from-InstCombine-inste.patch
Type: application/octet-stream
Size: 4155 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20121206/59c756c5/attachment.obj>
-------------- next part --------------

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0002-Have-SimplifyBinOp-call-the-new-FAdd-FSub-FMul-helpe.patch
Type: application/octet-stream
Size: 1561 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20121206/59c756c5/attachment-0001.obj>
-------------- next part --------------



On Dec 5, 2012, at 4:34 PM, Michael Ilseman <milseman at apple.com> wrote:

> Here's a second round of patches, incorporating the review feedback Duncan had given me earlier but I had spaced. The biggest difference is an extension to PatternMatch.h for pattern matching ConstantFP values of interest.
> 
> Patch1: Pattern matching for more floating point values: -0, +/-0, 1.0, and ConstantFPs
> 
> Patch2: Added a slew of SimplifyInstruction floating-point optimizations, many of which take advantage of fast-math flags. Test cases included.
> fsub X, 0 ==> X
> fsub 0, (fsub 0, X) ==> X
> fsub X, -0 ==> X
>   when X is known non-negative-zero, e.g. the fsub is nsz.
> fsub nnan ninf x, x ==> 0.0
> fadd X, -0 ==> X
> fadd X, 0 ==> X
>   when X is known non-negative-zero, e.g. the fadd is nsz.
> fadd [nnan ninf] X, (fsub [nnan ninf] 0, X) ==> 0
>   where nnan and ninf have to occur at least once somewhere in this expression
> fmul X, 1.0 ==> X
> 
> 
> 
> <0001-Pattern-matching-for-more-floating-point-values-0-0-.patch><0002-Added-a-slew-of-SimplifyInstruction-floating-point-o.patch>
> 
> 
> 
> On Dec 4, 2012, at 3:19 PM, Michael Ilseman <milseman at apple.com> wrote:
> 
>> Attached is a patch adding some floating-point optimizations to SimplifyInstruction, some of which take advantage of fast-math flags. Test cases included.
>> 
>> The optimizations themselves:
>> fsub X, 0 ==> X
>> 
>> fsub 0, (fsub 0, X) ==> X
>> 
>> fadd nsz X, 0 ==> X
>> 
>> fadd [nnan ninf] X, (fsub [nnan ninf] 0, X) ==> 0
>>  where nnan and ninf have to occur at least once somewhere in this expression
>> 
>> fsub nnan ninf x, x ==> 0.0
>> 
>> <0001-Added-a-slew-of-SimplifyInstruction-floating-point-o.patch>
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits