[llvm-commits] [llvm] r171429 - in /llvm/trunk: lib/Transforms/Vectorize/LoopVectorize.cpp test/Transforms/LoopVectorize/nofloat.ll

[Bob Wilson]

I suppose you're right but I doubt if it would be worth the effort of adding a target hook to indicate whether vector and floating-point values should be treated in similar ways.  Anyway, even for targets that have dedicated integer vector registers, I expect that most of the cases where someone would want to use "noimplicitfloat" would then also require a new "noimplicitvector" attribute.  Until we have a case where there's a good reason to distinguish those, it seems like we should just go with the current behavior that works for most targets.

On Jan 3, 2013, at 12:32 AM, Duncan Sands <baldrick at free.fr> wrote:

> Hi Nadav,
> 
> On 03/01/13 00:54, Nadav Rotem wrote:
>> Author: nadav
>> Date: Wed Jan  2 17:54:43 2013
>> New Revision: 171429
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=171429&view=rev
>> Log:
>> Avoid vectorization when the function has the "noimplicitflot" attribute.
> 
> if the target has dedicated integer vector registers, then by doing
> vectorization you are not necessarily introducing floating point
> operations.  I.e. this seems target specific to me.
> 
> Ciao, Duncan.
> 
>> 
>> Added:
>>     llvm/trunk/test/Transforms/LoopVectorize/nofloat.ll
>> Modified:
>>     llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp
>> 
>> Modified: llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp?rev=171429&r1=171428&r2=171429&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp (original)
>> +++ llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp Wed Jan  2 17:54:43 2013
>> @@ -94,9 +94,17 @@
>>      // Check the function attribues to find out if this function should be
>>      // optimized for size.
>>      Function *F = L->getHeader()->getParent();
>> -    Attribute::AttrKind SzAttr= Attribute::OptimizeForSize;
>> -    bool OptForSize =
>> -      F->getAttributes().hasAttribute(AttributeSet::FunctionIndex, SzAttr);
>> +    Attribute::AttrKind SzAttr = Attribute::OptimizeForSize;
>> +    Attribute::AttrKind FlAttr = Attribute::NoImplicitFloat;
>> +    unsigned FnIndex = AttributeSet::FunctionIndex;
>> +    bool OptForSize = F->getAttributes().hasAttribute(FnIndex, SzAttr);
>> +    bool NoFloat = F->getAttributes().hasAttribute(FnIndex, FlAttr);
>> +
>> +    if (NoFloat) {
>> +      DEBUG(dbgs() << "LV: Can't vectorize when the NoImplicitFloat"
>> +            "attribute is used.\n");
>> +      return false;
>> +    }
>> 
>>      unsigned VF = CM.selectVectorizationFactor(OptForSize, VectorizationFactor);
>> 
>> 
>> Added: llvm/trunk/test/Transforms/LoopVectorize/nofloat.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/nofloat.ll?rev=171429&view=auto
>> ==============================================================================
>> --- llvm/trunk/test/Transforms/LoopVectorize/nofloat.ll (added)
>> +++ llvm/trunk/test/Transforms/LoopVectorize/nofloat.ll Wed Jan  2 17:54:43 2013
>> @@ -0,0 +1,29 @@
>> +; RUN: opt < %s  -loop-vectorize -force-vector-width=4 -dce -instcombine -licm -S | FileCheck %s
>> +
>> +; Make sure that we don't vectorize functions with 'noimplicitfloat' attributes.
>> +
>> +target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
>> +target triple = "x86_64-apple-macosx10.8.0"
>> +
>> + at a = common global [2048 x i32] zeroinitializer, align 16
>> +
>> +;CHECK: @example12
>> +;CHECK-NOT: store <4 x i32>
>> +;CHECK: ret void
>> +define void @example12() noimplicitfloat { ;           <--------- "noimplicitfloat" attribute here!
>> +  br label %1
>> +
>> +; <label>:1                                       ; preds = %1, %0
>> +  %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
>> +  %2 = getelementptr inbounds [2048 x i32]* @a, i64 0, i64 %indvars.iv
>> +  %3 = trunc i64 %indvars.iv to i32
>> +  store i32 %3, i32* %2, align 4
>> +  %indvars.iv.next = add i64 %indvars.iv, 1
>> +  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
>> +  %exitcond = icmp eq i32 %lftr.wideiv, 1024
>> +  br i1 %exitcond, label %4, label %1
>> +
>> +; <label>:4                                       ; preds = %1
>> +  ret void
>> +}
>> +
>> 
>> 
>> _______________________________________________
>> 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