[llvm] r174152 - Optimize shift lefts of a constant by a value plus constant into a single shift.

[Nuno Lopes]

Hi Nadav,

I don't think this transformation is correct when the 'add' overflows.
E.g. (2 bits):

01 << (01 + 11)  == 01 << 00 == 01

(01 << 11) << 01 == 00 << 01 == 00

Nuno


Quoting Nadav Rotem <nrotem at apple.com>:

> Author: nadav
> Date: Fri Feb  1 00:45:40 2013
> New Revision: 174152
>
> URL: http://llvm.org/viewvc/llvm-project?rev=174152&view=rev
> Log:
> Optimize shift lefts of a constant by a value plus constant into a  
> single shift.
>
>
> Modified:
>     llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp
>     llvm/trunk/test/Transforms/InstCombine/shift.ll
>
> Modified: llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp
> URL:  
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp?rev=174152&r1=174151&r2=174152&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp (original)
> +++ llvm/trunk/lib/Transforms/InstCombine/InstCombineShifts.cpp Fri  
> Feb  1 00:45:40 2013
> @@ -709,6 +709,12 @@ Instruction *InstCombiner::visitShl(Bina
>        match(I.getOperand(1), m_Constant(C2)))
>      return BinaryOperator::CreateShl(ConstantExpr::getShl(C1, C2), A);
>
> +  // shl (c1 , add(y , c2)) -> (shl (shl(c1, c2)), y)
> +  if (match(I.getOperand(0), m_Constant(C1)) &&
> +      match(I.getOperand(1), m_Add(m_Value(A), m_Constant(C2)))) {
> +    return BinaryOperator::CreateShl(ConstantExpr::getShl(C1, C2), A);
> +  }
> +
>    return 0;
>  }
>
>
> Modified: llvm/trunk/test/Transforms/InstCombine/shift.ll
> URL:  
> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/shift.ll?rev=174152&r1=174151&r2=174152&view=diff
> ==============================================================================
> --- llvm/trunk/test/Transforms/InstCombine/shift.ll (original)
> +++ llvm/trunk/test/Transforms/InstCombine/shift.ll Fri Feb  1 00:45:40 2013
> @@ -745,3 +745,23 @@ define i32 @test62(i32 %x) {
>  ; CHECK: @test62
>  ; CHECK: ashr exact i32 %x, 3
>  }
> +
> +; CHECK: @test63
> +; CHECK: shl <4 x i32> <i32 1, i32 2, i32 4, i32 8>, %B
> +define <4 x i32> @test63(i32 %n) {
> +entry:
> +  %K = insertelement <4 x i32> undef, i32 %n, i32 0
> +  %B = shufflevector <4 x i32> %K, <4 x i32> undef, <4 x i32>  
> zeroinitializer
> +  %A = add <4 x i32> %B, <i32 0, i32 1, i32 2, i32 3>
> +  %T = shl <4 x i32> <i32 1, i32 1, i32 1, i32 1>, %A
> +  ret <4 x i32> %T
> +}
> +
> +; CHECK: @test64
> +; CHECK: shl i32 524288, %n
> +define i32 @test64(i32 %n) {
> +entry:
> +  %A = add i32 %n, 19
> +  %T = shl i32 1 , %A
> +  ret i32 %T
> +}
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits