[llvm-commits] [llvm] r151877 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/sibcall-5.ll

[Chad Rosier]

On Mar 1, 2012, at 6:50 PM, Chad Rosier wrote:

> Author: mcrosier
> Date: Thu Mar  1 20:50:46 2012
> New Revision: 151877
> 
> URL: http://llvm.org/viewvc/llvm-project?rev=151877&view=rev
> Log:
> Prevent obscure and incorrect tail-call optimization.
> 
> In this instance we are generating the tail-call during legalizeDAG.  The 2nd
> floor call can't be a tail call because it clobbers %xmm1, which is defined by
> the first floor call.  The first floor call can't be a tail-call because it's
> not in the tail position.  The only reasonable way I could think to fix this
> in a target-independent manner was to check for glue logic on the copy reg.

And by target-independent I of course meant target dependent. :/

> rdar://10930395
> 
> Modified:
>    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
>    llvm/trunk/test/CodeGen/X86/sibcall-5.ll
> 
> Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=151877&r1=151876&r2=151877&view=diff
> ==============================================================================
> --- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
> +++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Thu Mar  1 20:50:46 2012
> @@ -1589,6 +1589,11 @@
>       Copy->getOpcode() != ISD::FP_EXTEND)
>     return false;
> 
> +  // If anything is glued to the copy, then we can't safely perform a tail call.
> +  if (Copy->getOpcode() == ISD::CopyToReg &&
> +      Copy->getNumOperands() == 4)
> +    return false;
> +
>   bool HasRet = false;
>   for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
>        UI != UE; ++UI) {
> 
> Modified: llvm/trunk/test/CodeGen/X86/sibcall-5.ll
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/sibcall-5.ll?rev=151877&r1=151876&r2=151877&view=diff
> ==============================================================================
> --- llvm/trunk/test/CodeGen/X86/sibcall-5.ll (original)
> +++ llvm/trunk/test/CodeGen/X86/sibcall-5.ll Thu Mar  1 20:50:46 2012
> @@ -1,5 +1,6 @@
> ; RUN: llc < %s -mtriple=i386-apple-darwin8 -mattr=+sse2  | FileCheck %s --check-prefix=X32
> ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+sse2 | FileCheck %s --check-prefix=X64
> +; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=-sse3 | FileCheck %s --check-prefix=X64_BAD
> 
> ; Sibcall optimization of expanded libcalls.
> ; rdar://8707777
> @@ -29,3 +30,31 @@
> declare float @sinf(float) nounwind readonly
> 
> declare double @sin(double) nounwind readonly
> +
> +; rdar://10930395
> +%0 = type opaque
> +
> +@"\01L_OBJC_SELECTOR_REFERENCES_2" = external hidden global i8*, section "__DATA, __objc_selrefs, literal_pointers, no_dead_strip"
> +
> +define hidden { double, double } @foo2(%0* %self, i8* nocapture %_cmd) uwtable optsize ssp {
> +; X64_BAD: foo
> +; X64_BAD: call
> +; X64_BAD: call
> +; X64_BAD: call
> +  %1 = load i8** @"\01L_OBJC_SELECTOR_REFERENCES_2", align 8, !invariant.load !0
> +  %2 = bitcast %0* %self to i8*
> +  %3 = tail call { double, double } bitcast (i8* (i8*, i8*, ...)* @objc_msgSend to { double, double } (i8*, i8*)*)(i8* %2, i8* %1) optsize
> +  %4 = extractvalue { double, double } %3, 0
> +  %5 = extractvalue { double, double } %3, 1
> +  %6 = tail call double @floor(double %4) optsize
> +  %7 = tail call double @floor(double %5) optsize
> +  %insert.i.i = insertvalue { double, double } undef, double %6, 0
> +  %insert5.i.i = insertvalue { double, double } %insert.i.i, double %7, 1
> +  ret { double, double } %insert5.i.i
> +}
> +
> +declare i8* @objc_msgSend(i8*, i8*, ...)
> +
> +declare double @floor(double) optsize
> +
> +!0 = metadata !{}
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits