[PATCH][AArch64] Implemented vmul/vmux intrinsics

[Ana Pazos]

Hi Tim, Jiangning and reviewers,

This new patch contains:

Implemented aarch64 vmul_lane intrinsics.
Implemented aarch64 vmulx_lane intrinsics,
Implemented aarch64 vmul_n_f64 intrinsic, mapping it to Neon scalar
operation.
Implemented aarch64 vmul_lane_f64 and vmul_laneq_f64 intrinsics, mapping
them to Neon scalar operation.
Added codegen patterns for scalar copy (DUP) with FP types.
Added Scalar Copy (DUP) MOV aliases.

The implementation was straightforward. I was able to create IR patterns for
most cases.

But to force vmul_n_f64, vmul_lane_f64, vmul_laneq_f64 legacy intrinsics to
map to Neon scalar operations I had to add an aarch64 IR intrinsic.


Otherwise the casting of result to float64x1_t might cause non-vector code
to be generated and the Neon scalar instruction cannot be guaranteed.

This has happened before with other Neon Scalar instructions and they were
handled in similar way.

Thanks,

Ana.

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131113/d3c1b91e/attachment.html>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: llvm-vmul-vmulx-int-dup-aliases.txt
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131113/d3c1b91e/attachment.txt>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: clang-vmul-vmulx-int-dup-aliases.txt
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20131113/d3c1b91e/attachment-0001.txt>