RFR: 8073093: AARCH64: C2 generates poor code for ByteBuffer accesses

[Andrew Haley]

On 02/18/2015 09:15 AM, Andrew Haley wrote:
> On 18/02/15 09:14, Florian Weimer wrote:
>> Wow, looks nice.  What OpenJDK build did you use?  I want to see if this
>> happens on x86_64, too.
> 
> I'm working on JDK9.  You don't have this code yet.  I'll do an x86
> build.

  0x00007f2948acbf8c: mov    0xc(%rdx),%r10d    ;*synchronization entry
                                                ; - java.nio.HeapByteBuffer::<init>@-1 (line 84)
                                                ; - java.nio.ByteBuffer::wrap at 7 (line 373)
                                                ; - java.nio.ByteBuffer::wrap at 4 (line 396)
                                                ; - bytebuffertests.ByteBufferTests3::getLong at 1 (line 23)
                                                ; implicit exception: dispatches to 0x00007f2948acbff5
  ;; B2: #      B5 B3 <- B1  Freq: 0.999999

  ;; MEMBAR-release ! (empty encoding)

  0x00007f2948acbf90: test   %ecx,%ecx
  0x00007f2948acbf92: jl     0x00007f2948acbfb5  ;*iflt
                                                ; - java.nio.Buffer::checkIndex at 1 (line 545)
                                                ; - java.nio.HeapByteBuffer::getLong at 18 (line 465)
                                                ; - bytebuffertests.ByteBufferTests3::getLong at 5 (line 23)

  ;; B3: #      B6 B4 <- B2  Freq: 0.999999

  0x00007f2948acbf94: mov    %r10d,%ebp
  0x00007f2948acbf97: sub    %ecx,%ebp          ;*isub
                                                ; - java.nio.Buffer::checkIndex at 10 (line 545)
                                                ; - java.nio.HeapByteBuffer::getLong at 18 (line 465)
                                                ; - bytebuffertests.ByteBufferTests3::getLong at 5 (line 23)

  0x00007f2948acbf99: cmp    $0x8,%ebp
  0x00007f2948acbf9c: jl     0x00007f2948acbfd5  ;*if_icmple
                                                ; - java.nio.Buffer::checkIndex at 11 (line 545)
                                                ; - java.nio.HeapByteBuffer::getLong at 18 (line 465)
                                                ; - bytebuffertests.ByteBufferTests3::getLong at 5 (line 23)

  ;; B4: #      N95 <- B3  Freq: 0.999998

  0x00007f2948acbf9e: movslq %ecx,%r10
  0x00007f2948acbfa1: mov    0x10(%rdx,%r10,1),%rax
  0x00007f2948acbfa6: bswap  %rax               ;*invokestatic reverseBytes
                                                ; - java.nio.Bits::swap at 1 (line 61)
                                                ; - java.nio.HeapByteBuffer::getLong at 41 (line 466)
                                                ; - bytebuffertests.ByteBufferTests3::getLong at 5 (line 23)

So, just the same except that there is no explicit fence instruction
to remove.  It's a shame for AArch64 because that fence really kills
performance but it's bad for x86 too.  Even on machines that don't
emit fence instructions the fence still acts as a compiler barrier.

Andrew.