[llvm-commits] [llvm] r133452 - in /llvm/trunk: lib/Target/X86/X86InstrSSE.td test/CodeGen/X86/bitcast2.ll test/CodeGen/X86/vec_set-8.ll test/CodeGen/X86/vec_set-9.ll test/CodeGen/X86/vec_set-C.ll test/CodeGen/X86/vec_shuffle-14.ll test/CodeGen/X86/vec_shuffle-17.ll test/MC/X86/x86-64.s test/MC/X86/x86_64-avx-encoding.s

[Bob Wilson]

On Jun 20, 2011, at 5:13 PM, Charles Davis wrote:

> On 6/20/11 12:33 PM, Nick Lewycky wrote:
>> Author: nicholas
>> Date: Mon Jun 20 13:33:26 2011
>> New Revision: 133452
>> 
>> URL: http://llvm.org/viewvc/llvm-project?rev=133452&view=rev
>> Log:
>> Emit movq for 64-bit register to XMM register moves, but continue to accept
>> movd when assembling.
> This breaks compiling compiler-rt on Mac OS X like so:
> 
> /tmp/cc-E8HNBz.s:7:suffix or operands invalid for `movq'
> clang: error: assembler command failed with exit code 1 (use -v to see
> invocation)
> make[5]: ***
> [.../llvm/build1/tools/clang/runtime/compiler-rt/clang_darwin/cc_kext/x86_64/SubDir.lib/fixdfti.o]
> Error 1
> 
> There are several more errors like that.
> 
> Xcode 4.0's assembler still has the movq bug. We still have to produce
> movd on Darwin when not using the integrated assembler.

Yes, it's also breaking llvm-gcc building.  I've reverted that change for now.