JDK-8215483 : Off heap memory accesses should be vectorized
  • Type: Enhancement
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 11,12,13
  • Priority: P3
  • Status: Resolved
  • Resolution: Fixed
  • Submitted: 2018-12-17
  • Updated: 2019-08-15
  • Resolved: 2019-02-05
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
JDK 11 JDK 13
11.0.5-oracleFixed 13 b07Fixed
Related Reports
Duplicate :  
JDK-8198680 - C2 superword optimization crashes with "assert(my_pack(s) == __null) failed: only in one pack"
Description
    void floss(ByteBuffer b, int n) {
        for (int i = 0; i < SIZE; i++) {
            b.putInt(i<<2, n);
        }
    }

for 

ByteBuffer buf = ByteBuffer.allocateDirect(SIZE * 4);

is not vectorized but should be.
Comments
Fix request (11u): Requesting this backport to keep parity with Oracle 11.0.5. Patch applies cleanly. Will be tested @SAP.
24-06-2019
http://cr.openjdk.java.net/~roland/8215483/webrev.00/ Given, in the general case, we have no guarantee on the buffer alignment, this can only work for architectures that support misaligned vector accesses. I change Matcher::misaligned_vectors_ok() so it always reports what the backend is capable of. I'm not sure about ppc: the comment says misaligned vectors are unsupported but the code uses the same pattern that x86 or aarch64 use. My understanding is that the current logic of SWPointer::SWPointer() doesn't properly handle the case where base and adr differ: there's no check for whether adr is loop invariant and the code that performs loop alignment ignores adr. It also seems very rare that adr differs from base for on heap accesses. Instead of handling this case, my change makes superword bailout if adr != base unless it's an off heap access and base == top. For an off heap access, then SuperWord::align_initial_loop_index() tries to align the loop the best it can.
25-01-2019