[vectorIntrinsics+compress] RFR: 8274889: Intrinsify Vector.expand/compress APIs for X86 [v5]

[Sandhya Viswanathan]

On Mon, 11 Oct 2021 12:11:45 GMT, Jatin Bhateja <jbhateja at openjdk.org> wrote:

>> - Common C2 compiler entry point for compress/expand APIs.
>> - Inline expansion handling and C2 graph creation.
>> - For X86 current backed support is added for AVX512 which offers direct compress/expand instructions.
>
> Jatin Bhateja has updated the pull request incrementally with one additional commit since the last revision:
> 
>   8274889: Review comments resolution.

src/hotspot/cpu/x86/x86.ad line 8745:

> 8743: // --------------------------------- Compress/Expand Operations ---------------------------
> 8744: 
> 8745: instruct vcompress_expand_reg(vec dst, vec src, kReg mask) %{

We could name this as evcompress_expand_reg (as it is for evex only).

src/hotspot/share/classfile/vmIntrinsics.hpp line 1097:

> 1095:    do_signature(vector_comexp_op_sig, "(I"                                                                                                     \
> 1096:                                       "Ljava/lang/Class;"                                                                                      \
> 1097:                                       "Ljava/lang/Class;Ljava/lang/Class;"                                                                     \

Why two entries on one line here only?

src/hotspot/share/opto/vectorIntrinsics.cpp line 2710:

> 2708:       elem_klass->const_oop() == NULL || !vlen->is_con()) {
> 2709:     if (C->print_intrinsics()) {
> 2710:       tty->print_cr("  ** missing constant: opr=%s vclass=%s mclass =%s etype=%s vlen=%s",

Why this change?

src/hotspot/share/opto/vectorIntrinsics.cpp line 2743:

> 2741:     }
> 2742:     return false;
> 2743:   }

This check is there in unary op, binary op, etc. Should be kept for consistency?

-------------

PR: https://git.openjdk.java.net/panama-vector/pull/146