On 29/06/17 09:57, Toebs Douglass wrote:
> There are as far as I can tell in the entire disassembly (and certainly
> not in the function call above) any use of the "X" type load/store
> instructions, which are the type for load-linked/store-conditional.
This is what's in libatomic:
0000000000003048 <__atomic_compare_exchange_16>:
3048: a9bc7bfd stp x29, x30, [sp,#-64]!
304c: 910003fd mov x29, sp
3050: a90153f3 stp x19, x20, [sp,#16]
3054: aa0003f4 mov x20, x0
3058: aa0103f3 mov x19, x1
305c: a9025bf5 stp x21, x22, [sp,#32]
3060: f9001bf7 str x23, [sp,#48]
3064: aa0203f6 mov x22, x2
3068: aa0303f7 mov x23, x3
306c: 97fffd97 bl 26c8 <libat_lock_1>
3070: a9401263 ldp x3, x4, [x19]
3074: a9400a81 ldp x1, x2, [x20]
3078: ca010060 eor x0, x3, x1
307c: ca020085 eor x5, x4, x2
3080: aa050000 orr x0, x0, x5
3084: f100001f cmp x0, #0x0
3088: 1a9f17f5 cset w21, eq
308c: eb01007f cmp x3, x1
3090: 54000140 b.eq 30b8 <__atomic_compare_exchange_16+0x70>
3094: a9000a61 stp x1, x2, [x19]
3098: aa1403e0 mov x0, x20
309c: 97fffd91 bl 26e0 <libat_unlock_1>
30a0: f9401bf7 ldr x23, [sp,#48]
30a4: 2a1503e0 mov w0, w21
30a8: a94153f3 ldp x19, x20, [sp,#16]
30ac: a9425bf5 ldp x21, x22, [sp,#32]
30b0: a8c47bfd ldp x29, x30, [sp],#64
30b4: d65f03c0 ret
30b8: eb02009f cmp x4, x2
30bc: 54fffec1 b.ne 3094 <__atomic_compare_exchange_16+0x4c>
30c0: a9005e96 stp x22, x23, [x20]
30c4: aa1403e0 mov x0, x20
30c8: 97fffd86 bl 26e0 <libat_unlock_1>
30cc: f9401bf7 ldr x23, [sp,#48]
30d0: 2a1503e0 mov w0, w21
30d4: a94153f3 ldp x19, x20, [sp,#16]
30d8: a9425bf5 ldp x21, x22, [sp,#32]
30dc: a8c47bfd ldp x29, x30, [sp],#64
30e0: d65f03c0 ret
I think I now properly understand Richard Earnshaw's point: that we
*do* support a full set of atomic primitives for 16-byte types via
libatomic, but for them to work as a sequentially-consistent set we
must use the same locking scheme for all of them. It's ugly, and
horrible for anyone who simply wants a double-word CAS, but it is what
it is. We can't use LDXP because it isn't atomic on its own, and the
ARM manual is quite explicit about this. Anyone who wants to use
a real compare-and-swap-16 is on their own.
--
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671
