[llvm-commits] [llvm] r48542 - in /llvm/trunk: lib/CodeGen/SelectionDAG/DAGCombiner.cpp lib/Target/X86/README.txt lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/field-extract-use-trunc.ll

Evan Cheng evan.cheng at apple.com
Wed Mar 19 02:06:32 PDT 2008


I don't think this is the right fix. isTruncFree is checked to  
determine whether the xform is profitable. It shouldn't be a  
correctness issue. Perhaps you need to check whether sext of the type  
is legal?

Evan

On Mar 19, 2008, at 1:30 AM, Christopher Lamb <christopher.lamb at gmail.com 
 > wrote:

> Author: clamb
> Date: Wed Mar 19 03:30:06 2008
> New Revision: 48542
>
> URL: http://llvm.org/viewvc/llvm-project?rev=48542&view=rev
> Log:
> Fix X86's isTruncateFree to not claim that truncate to i1 is free.  
> This fixes Bill's testcase that failed for r48491.
>
> Modified:
>    llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
>    llvm/trunk/lib/Target/X86/README.txt
>    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
>    llvm/trunk/test/CodeGen/X86/field-extract-use-trunc.ll
>
> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=48542&r1=48541&r2=48542&view=diff
>
> === 
> === 
> === 
> =====================================================================
> --- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
> +++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Wed Mar 19  
> 03:30:06 2008
> @@ -2383,6 +2383,31 @@
>                          DAG.getConstant(Sum, N1C->getValueType(0)));
>     }
>   }
> +
> +  // fold sra (shl X, m), result_size - n
> +  // -> (sign_extend (trunc (shl X, result_size - n - m))) for
> +  // result_size - n != m. If truncate is free for the target sext 
> (shl) is
> +  // likely to result in better code.
> +  if (N0.getOpcode() == ISD::SHL) {
> +    // Get the two constanst of the shifts, CN0 = m, CN = n.
> +    const ConstantSDNode *N01C = dyn_cast<ConstantSDNode> 
> (N0.getOperand(1));
> +    if (N01C && N1C) {
> +      // Determine if the truncate type's bitsize would correspond to
> +      // an integer type for this target.
> +      unsigned VTValSize = MVT::getSizeInBits(VT);
> +      MVT::ValueType TruncVT = MVT::getIntegerType(VTValSize - N1C- 
> >getValue());
> +      unsigned ShiftAmt = N1C->getValue() - N01C->getValue();
> +
> +      // If the shift wouldn't be a noop, the truncated type is an  
> actual type,
> +      // and the truncate is free, then proceed with the transform.
> +      if (ShiftAmt != 0 && TLI.isTruncateFree(VT, TruncVT)) {
> +        SDOperand Amt = DAG.getConstant(ShiftAmt,  
> TLI.getShiftAmountTy());
> +        SDOperand Shift = DAG.getNode(ISD::SRL, VT, N0.getOperand 
> (0), Amt);
> +        SDOperand Trunc = DAG.getNode(ISD::TRUNCATE, TruncVT, Shift);
> +        return DAG.getNode(ISD::SIGN_EXTEND, N->getValueType(0),  
> Trunc);
> +      }
> +    }
> +  }
>
>   // Simplify, based on bits shifted out of the LHS.
>   if (N1C && SimplifyDemandedBits(SDOperand(N, 0)))
>
> Modified: llvm/trunk/lib/Target/X86/README.txt
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/README.txt?rev=48542&r1=48541&r2=48542&view=diff
>
> === 
> === 
> === 
> =====================================================================
> --- llvm/trunk/lib/Target/X86/README.txt (original)
> +++ llvm/trunk/lib/Target/X86/README.txt Wed Mar 19 03:30:06 2008
> @@ -1647,47 +1647,3 @@
> if it commuted the addl in LBB1_1.
>
> // 
> === 
> --------------------------------------------------------------------- 
> ===//
> -
> -These two functions perform identical operations:
> -
> -define i32 @test(i32 %f12) {
> -    %tmp7.25 = lshr i32 %f12, 16
> -    %tmp7.26 = trunc i32 %tmp7.25 to i8
> -    %tmp78.2 = sext i8 %tmp7.26 to i32
> -    ret i32 %tmp78.2
> -}
> -
> -define i32 @test2(i32 %f12) {
> -    %f11 = shl i32 %f12, 8
> -    %tmp7.25 = ashr i32 %f11, 24
> -    ret i32 %tmp7.25
> -}
> -
> -but the first compiles into significantly better code on x86-32:
> -
> -_test:
> -    movsbl    6(%esp), %eax
> -    ret
> -_test2:
> -    movl    4(%esp), %eax
> -    shll    $8, %eax
> -    sarl    $24, %eax
> -    ret
> -
> -and on x86-64:
> -
> -_test:
> -    shrl    $16, %edi
> -    movsbl    %dil, %eax
> -    ret
> -_test2:
> -    shll    $8, %edi
> -    movl    %edi, %eax
> -    sarl    $24, %eax
> -    ret
> -
> -I would like instcombine to canonicalize the first into the second  
> (since it is
> -shorter and doesn't involve type width changes) but the x86 backend  
> needs to do
> -the right thing with the later sequence first.
> -
> -// 
> === 
> --------------------------------------------------------------------- 
> ===//
>
> Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=48542&r1=48541&r2=48542&view=diff
>
> === 
> === 
> === 
> =====================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Wed Mar 19  
> 03:30:06 2008
> @@ -5662,7 +5662,7 @@
>     return false;
>   unsigned NumBits1 = Ty1->getPrimitiveSizeInBits();
>   unsigned NumBits2 = Ty2->getPrimitiveSizeInBits();
> -  if (NumBits1 <= NumBits2)
> +  if (NumBits1 <= NumBits2 || NumBits2 < 8)
>     return false;
>   return Subtarget->is64Bit() || NumBits1 < 64;
> }
> @@ -5673,7 +5673,7 @@
>     return false;
>   unsigned NumBits1 = MVT::getSizeInBits(VT1);
>   unsigned NumBits2 = MVT::getSizeInBits(VT2);
> -  if (NumBits1 <= NumBits2)
> +  if (NumBits1 <= NumBits2 || NumBits2 < 8)
>     return false;
>   return Subtarget->is64Bit() || NumBits1 < 64;
> }
>
> Modified: llvm/trunk/test/CodeGen/X86/field-extract-use-trunc.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/field-extract-use-trunc.ll?rev=48542&r1=48541&r2=48542&view=diff
>
> === 
> === 
> === 
> =====================================================================
> --- llvm/trunk/test/CodeGen/X86/field-extract-use-trunc.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/field-extract-use-trunc.ll Wed Mar  
> 19 03:30:06 2008
> @@ -1,6 +1,5 @@
> ; RUN: llvm-as < %s | llc -march=x86 | grep sar | count 1
> ; RUN: llvm-as < %s | llc -march=x86-64 | not grep sar
> -; XFAIL: *
>
> define i32 @test(i32 %f12) {
>    %tmp7.25 = lshr i32 %f12, 16
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits



More information about the llvm-commits mailing list