Request for reviews (M): 7181494: cleanup avx and vectors code
vladimir.kozlov at oracle.com
Mon Jul 16 12:02:59 PDT 2012
Thank you, Christian
Christian Thalinger wrote:
> Looks good. -- Chris
> On Jul 13, 2012, at 7:35 PM, Vladimir Kozlov wrote:
>> 7181494: cleanup avx and vectors code
>> Use shorter (non-ladder) form for XMM registers definition in x86.ad.
>> Do renaming (vsubF_reg --> subF_reg_reg) for mach nodes which use AVX (3-operands) scalar instructions to avoid confusing with vector nodes.
>> Add integer vectors shuffling instructions since SIMD Integer execution unit is separate from Float unit in SSE - there is at least one cycle penalty for moving data between them. Use them in replicate integer values mach nodes.
>> Tested with compiler regression tests and CTW.
More information about the hotspot-compiler-dev