On Dec 14, 2009, at 1:49 PM, Eli Friedman wrote: > Two issues here: > > 1. The second operand of psrldq is in bytes, not bits.
While that's true of the x86 instruction, that's not true of the LLVM intrinsic for the instruction. You can check the output of the testcase that got checked in if you'd like. > 2. It would be better to simply emit this as a shuffle; we're > generally trying to get rid of unnecessary intrinsics, not add more > uses of them. I did not add any intrinsics not already in clang. If you would like to remove the sse2 psrldq intrinsic and replace it with a shuffle, please go for it. Nate _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
