Replace brittle gcc3_linux_aarch64 vtableSlotCall with pure assembler code ...
* For one, as discussed in the comment at <https://gerrit.libreoffice.org/c/
core/+/91978/2#message-97da6c6ece7ae7bd49e9eb4843be333192fcc057> "Port to
FreeBSD aarch64":
"So it looks like Clang does not treat those register asm as we expect it would,
at least on aarch64.
"Witness a local Linux recent master --with-distro=LibreOfficeAndroidAarch64
[i.e., using Clang] build's
> $ llvm-objdump --disassemble instdir/program/libgcc3_uno.a
[...]
> 0000000000000000 <vtableSlotCall>:
> 0: ff 43 03 d1 sub sp, sp, #208
> 4: f4 4f 0b a9 stp x20, x19, [sp, #176 ]
> 8: fd 7b 0c a9 stp x29, x30, [sp, #192 ]
> c: fd 03 03 91 add x29, sp, #192
> 10: a8 43 00 91 add x8, x29, #16
> 14: bf 83 1d f8 stur xzr, [x29, #-40]
> 18: e0 07 04 a9 stp x0, x1, [sp, #64 ]
> 1c: e2 0f 05 a9 stp x2, x3, [sp, #80 ]
> 20: e4 17 06 a9 stp x4, x5, [sp, #96 ]
> 24: e6 1f 07 a9 stp x6, x7, [sp, #112 ]
> 28: e0 07 00 6d stp d0, d1, [sp]
> 2c: e2 0f 01 6d stp d2, d3, [sp, #16 ]
> 30: e4 17 02 6d stp d4, d5, [sp, #32 ]
> 34: e6 1f 03 6d stp d6, d7, [sp, #48 ]
> 38: a8 03 1c f8 stur x8, [x29, #-64]
> 3c: a0 43 5e b8 ldur w0, [x29, #-28]
> 40: a1 03 5e b8 ldur w1, [x29, #-32]
> 44: a5 83 5e f8 ldur x5, [x29, #-24]
> 48: e4 03 08 aa mov x4, x8
> 4c: e2 03 01 91 add x2, sp, #64
> 50: e3 03 00 91 mov x3, sp
> 54: f3 03 01 91 add x19, sp, #64
> 58: f4 03 00 91 mov x20, sp
> 5c: 00 00 00 94 bl 0x5c <vtableSlotCall+0x5c>
> 60: 60 06 40 a9 ldp x0, x1, [x19]
> 64: 80 06 40 6d ldp d0, d1, [x20]
> 68: 82 0e 41 6d ldp d2, d3, [x20, #16 ]
> 6c: fd 7b 4c a9 ldp x29, x30, [sp, #192 ]
> 70: f4 4f 4b a9 ldp x20, x19, [sp, #176 ]
> 74: ff 43 03 91 add sp, sp, #208
> 78: c0 03 5f d6 ret
[...]
vs. [this commit's vtableslotcall.s; see below for details].
"And also latest Clang 12 trunk still does e.g.
> $ cat test.c
> void f(long);
> void g() {
> register long volatile a asm ("x8");
> f(a);
> }
> $ clang --target=unknown-linux-aarch64 -S -O2 test.c
> $ cat test.s
> .text
> .file "test.c"
> .globl g // -- Begin function g
> .p2align 2
> .type g,@function
> g: // @g
> // %bb.0: // %entry
> sub sp, sp, #16 // =16
> ldr x0, [sp, #8 ]
> add sp, sp, #16 // =16
> b f
> .Lfunc_end0:
> .size g, .Lfunc_end0-g
> // -- End function
> .ident "clang version 12.0.0 (git@github.com:llvm/llvm-project eb31ddd71eb44d53ebe12a09c9587198bb6f2a2e)"
> .section ".note.GNU-stack","",@progbits
> .addrsig
"(This is probably also the underlying issue that
eb15ac837edddb527db1https://gerrit.libreoffice.org/c/core/+/99660
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
252aba41e5