gcc -march=armv8-a generates LSE instructions but not -march=armv8.2-a or -mcpu=neoverse-n1 about aws-graviton-getting-started

[AGSaidi]

The symbols you're searching for (e.g. __aarch64_cas4_acq_rel) are symbols used by -moutline-atomics to determine at run-time if a LSE (e.g. casal or a load-store exclusive e.g.ldaxr/stlxr are used)

0000000000400650 <__aarch64_cas4_acq_rel>:
  400650:	90000110 	adrp	x16, 420000 <getauxval@GLIBC_2.17>
  400654:	39409610 	ldrb	w16, [x16, #37]
  400658:	34000070 	cbz	w16, 400664 <__aarch64_cas4_acq_rel+0x14>
  40065c:	88e0fc41 	casal	w0, w1, [x2]
  400660:	d65f03c0 	ret
  400664:	2a0003f0 	mov	w16, w0
  400668:	885ffc40 	ldaxr	w0, [x2]
  40066c:	6b10001f 	cmp	w0, w16
  400670:	54000061 	b.ne	40067c <__aarch64_cas4_acq_rel+0x2c>  // b.any
  400674:	8811fc41 	stlxr	w17, w1, [x2]
  400678:	35ffff91 	cbnz	w17, 400668 <__aarch64_cas4_acq_rel+0x18>
  40067c:	d65f03c0 	ret

When you compile with -mno-outline-atomics or a compiler that didn't make -moutline-atomics the default and -march=armv8-a the compile will just inline the ldaxr/stlxr in the calling function the symbol you reference above won't be present. (note: GCC 7 in Amazon Linux2 includes back ported patches making -moutline-atomics the default.

When you compile with -march=armv8.2-a or with -mcpu=neoverse-n1 similarly a casal or other LSE atomic will simply be inlined in the caller instead of jumping to the symbols that come from libgcc which includes the above mentioned __aarcch64_* functions.

[FranckPachot]

Thanks, got it.