[llvm-commits] [llvm] r114843 - /llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

[Dale Johannesen]

On Sep 28, 2010, at 2:09 AMPDT, Duncan Sands wrote:

> Hi Dale,
> 
>> Don't try to make a vector of x86mmx; this won't work,
>> and asserts.
> 
> what is x86mmx?  

An abstraction of an MMX register; the point of using it instead of v2i32 etc. is to inhibit optimizations that create MMX operations where there weren't any.  We can't do this because of the EMMS problem.

> If it isn't a scalar or vector type, should bitcast
> be allowed for it?  

Chris says yes.

> If bitcast is to be allowed, it would be better
> to create a new predicate for testing whether a type can be used as
> the element type of a vector, and use that rather than hard-coding
> x86mmx here.

I'm not generally big on extra levels of abstraction, but it could be done this way.

>  Also, what's the testcase?

Lots of testcases are coming later in the week.

> Ciao,
> 
> Duncan.
> 
>> 
>> 
>> Modified:
>>     llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
>> 
>> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp?rev=114843&r1=114842&r2=114843&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp (original)
>> +++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp Mon Sep 27 12:29:14 2010
>> @@ -1572,7 +1572,8 @@
>> 
>>    unsigned WidenSize = WidenVT.getSizeInBits();
>>    unsigned InSize = InVT.getSizeInBits();
>> -  if (WidenSize % InSize == 0) {
>> +  // x86mmx is not an acceptable vector element type, so don't try.
>> +  if (WidenSize % InSize == 0&&  InVT != MVT::x86mmx) {
>>      // Determine new input vector type.  The new input vector type will use
>>      // the same element type (if its a vector) or use the input type as a
>>      // vector.  It is the same size as the type to widen to.
>> @@ -2049,7 +2050,8 @@
>>    // Check if we can convert between two legal vector types and extract.
>>    unsigned InWidenSize = InWidenVT.getSizeInBits();
>>    unsigned Size = VT.getSizeInBits();
>> -  if (InWidenSize % Size == 0&&  !VT.isVector()) {
>> +  // x86mmx is not an acceptable vector element type, so don't try.
>> +  if (InWidenSize % Size == 0&&  !VT.isVector()&&  VT != MVT::x86mmx) {
>>      unsigned NewNumElts = InWidenSize / Size;
>>      EVT NewVT = EVT::getVectorVT(*DAG.getContext(), VT, NewNumElts);
>>      if (TLI.isTypeSynthesizable(NewVT)) {
>> 
>> 
>> _______________________________________________
>> 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