[aarch64-port-dev ] RFR(M): 8213134: AArch64: vector shift failed with MaxVectorSize=8

[Yang Zhang (Arm Technology China)]

Hi,

When I implemented AArch64 NEON for Vector API (http://openjdk.java.net/jeps/338), I found an issue about vector shift. I have a patch which could fix this issue. Could anyone help to review this patch?

Webrev: http://cr.openjdk.java.net/~yzhang/8213134/webrev.00/
JBS:  https://bugs.openjdk.java.net/browse/JDK-8213134

This patch is verified both in jdk/jdk master and panama/vectorIntrinsics, and tests are passed.
The only concern is that extra instructions are needed for some test cases.
For example:
  public static void aShiftRImm(int[] a, int b, int[] c) {
    for (int i = 0; i < a.length; i++) {
       c[i] = (int)(a[i] >> b);
    }
  }

Without this patch, this method can be optimized by C2:
  0x0000ffff84e75864: dup          v16.16b, w10
  0x0000ffff84e75868: neg          v16.16b, v16.16b        ------------------------------ only 1 neg instruction

  0x0000ffff84e75878: ldr            q17, [x12, #16]
  0x0000ffff84e7587c: sshl          v17.4s, v17.4s, v16.4s
  0x0000ffff84e75884: str            q17, [x10, #16]
  0x0000ffff84e75888: ldr            q17, [x12, #32]
  0x0000ffff84e7588c: sshl          v17.4s, v17.4s, v16.4s
  0x0000ffff84e75890: str            q17, [x10, #32]
  0x0000ffff84e75894: ldr            q17, [x12, #48]
  0x0000ffff84e75898: sshl          v17.4s, v17.4s, v16.4s
  0x0000ffff84e7589c: str             q17, [x10, #48]
  0x0000ffff84e758a0: ldr            q17, [x12, #64]
  0x0000ffff84e758a4: sshl          v17.4s, v17.4s, v16.4s
  0x0000ffff84e758ac: str             q17, [x10, #64]

With this patch, this method can be optimized by C2:
  0x0000ffff78e708e4: dup          v16.16b, w10
  0x0000ffff78e708f8: ldr             q17, [x12, #16]
  0x0000ffff78e708fc: neg            v18.16b, v16.16b                ------------------------------- 4 neg instructions
  0x0000ffff78e70900: sshl          v17.4s, v17.4s, v18.4s
  0x0000ffff78e70908: str            q17, [x10, #16]
  0x0000ffff78e7090c: ldr            q17, [x12, #32]
  0x0000ffff78e70910: neg          v18.16b, v16.16b
  0x0000ffff78e70914: sshl          v17.4s, v17.4s, v18.4s
  0x0000ffff78e70918: str            q17, [x10, #32]
  0x0000ffff78e7091c: ldr            q17, [x12, #48]
  0x0000ffff78e70920: neg          v18.16b, v16.16b
  0x0000ffff78e70924: sshl          v17.4s, v17.4s, v18.4s
  0x0000ffff78e70928: str            q17, [x10, #48]
  0x0000ffff78e7092c: ldr            q17, [x12, #64]
  0x0000ffff78e70930: neg          v18.16b, v16.16b
  0x0000ffff78e70934: sshl          v17.4s, v17.4s, v18.4s
  0x0000ffff78e7093c: str             q17, [x10, #64]

Regards
Yang