[llvm-commits] [Please Review] [PATCH] optimization of -(x != 0) to use neg in X86

[Evan Cheng]

The first chunk seems right to me.

+// sub 0, reg                                                                                                                                                                                                        
+def : Pat<(X86sub_flag 0, GR8 :$src), (NEG8r  GR8 :$src)>;                                                                                                                                                           
+def : Pat<(X86sub_flag 0, GR16:$src), (NEG16r GR16:$src)>;                                                                                                                                                           
+def : Pat<(X86sub_flag 0, GR32:$src), (NEG32r GR32:$src)>;                                                                                                                                                           
+def : Pat<(X86sub_flag 0, GR64:$src), (NEG64r GR64:$src)>;                                                                                                                                                           
+                                                                                                                                                                                                                     

I'm very skeptical this would ever match. I think we handle it with c++ code in isel. Or am I confusing the issue? See rdar://6172640 and rdar://11069732.
+// sub 0, mem                                                                                                                                                                                                        
+def : Pat<(store (X86sub_flag 0, (loadi8 addr:$dst)), addr:$dst),                                                                                                                                                    
+          (NEG8m addr:$dst)>;                                                                                                                                                                                        
+def : Pat<(store (X86sub_flag 0, (loadi16 addr:$dst)), addr:$dst),                                                                                                                                                  
+          (NEG16m addr:$dst)>;                                                                                                                                                                                       
+def : Pat<(store (X86sub_flag 0, (loadi32 addr:$dst)), addr:$dst),                                                                                                                                                  
+          (NEG32m addr:$dst)>;                                                                                                                                                                                       
+def : Pat<(store (X86sub_flag 0, (loadi64 addr:$dst)), addr:$dst),                                                                                                                                                  
+          (NEG64m addr:$dst)>;

Evan

On May 6, 2012, at 8:20 AM, Manman Ren wrote:

> 
> Hi Duncan,
> 
> I created a ISD::SUB node with two outputs and used the 2nd output for X86ISD::SETCC_CARRY.
> 
> ISD::SUB is not supposed to have two outputs.
> Now I used X86ISD::SUB with two outputs and added patterns to match X86ISD::SUB(0, r).
> 
> Thanks,
> Manman
> 
> On May 5, 2012, at 4:15 AM, Duncan Sands wrote:
> 
>> Hi Manman,
>> 
>>> PING
>> 
>> why was it failing before, i.e. what did you fix?
>> 
>> Ciao, Duncan.
>> 
>>> 
>>> On May 3, 2012, at 7:14 PM, manman ren wrote:
>>> 
>>>> 
>>>> X86: optimization for -(x != 0)
>>>> 
>>>> This patch will optimize -(x != 0) on X86
>>>> FROM
>>>> cmpl	$0x01,%edi
>>>> sbbl	%eax,%eax
>>>> notl	%eax
>>>> TO
>>>> negl %edi
>>>> sbbl %eax %eax
>>>> 
>>>> In order to generate negl, I added patterns in Target/X86/X86InstrCompiler.td:
>>>> def : Pat<(X86sub_flag 0, GR32:$src), (NEG32r GR32:$src)>;
>>>> 
>>>> This is my first experience with tablegen, please review.
>>>> The initial version of this patch (r155853) was the root cause of PR12720 and I did verify this patch against PR12720.
>>>> 
>>>> Thanks,
>>>> Manman
>>>> 
>>>> <sext_ne.patch>
>>>> 
>>>> 
>>>> 
>>>> On Apr 30, 2012, at 11:38 AM, manman ren wrote:
>>>> 
>>>>>> 
>>>>>> 
>>>>>> On Apr 27, 2012, at 6:03 PM, manman ren wrote:
>>>>>> 
>>>>>>> 
>>>>>>> This patch will optimize -(x != 0) on X86
>>>>>>> FROM
>>>>>>> cmpl	$0x01,%edi
>>>>>>> sbbl	%eax,%eax
>>>>>>> notl	%eax
>>>>>>> TO
>>>>>>> negl %edi
>>>>>>> sbbl %eax %eax
>>>>>>> 
>>>>>>> Please review&  provide feedback,
>>>>>>> Thanks,
>>>>>>> Manman
>>>>>>> 
>>>>>>> <sext_ne.patch>
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> llvm-commits mailing list
>>>>>>> llvm-commits at cs.uiuc.edu
>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>>>> 
>>>>> 
>>>> 
>>> 
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits