[Bug target/114416] New: SPARC V9 struct return with floating-point members violates ABI

["ro at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114416

            Bug ID: 114416
           Summary: SPARC V9 struct return with floating-point members
                    violates ABI
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
                CC: ebotcazou at gcc dot gnu.org
  Target Milestone: ---
            Target: sparc*-sun-solaris2.11

I've been informed of a case where gcc generates code for struct return with
floating-point members that violates the SPARC V9 ABI (SCD 2.4.1): consider the
attached testcase:

$ gcc -m64 -O2 -c fpret.c
$ cc -m64 -O2 caller.c
$ gcc -m64 -o cc-gcc caller.o fpret.o
$ ./cc-gcc
sum2:  x[0] = -nan x[1] = -nan
sum2x: x = 0 y = 1

With gcc, sum2 places the return value into %o0/%o1

sum2()
    sum2:                   9c 03 bf 60  add       %sp, -0xa0, %sp
    sum2+0x4:               90 10 20 00  clr       %o0
    sum2+0x8:               92 10 23 ff  mov       0x3ff, %o1
    sum2+0xc:               93 2a 70 34  sllx      %o1, 0x34, %o1
    sum2+0x10:              81 c3 e0 08  retl
    sum2+0x14:              9c 03 a0 a0  add       %sp, 0xa0, %sp

while cc uses %d0/%d2 instead:

sum2()
    sum2:                   1b 00 00 00  sethi     %hi(0x0), %o5
    sum2+0x4:               81 b0 0c 00  fzerod    %d0
    sum2+0x8:               c1 3b a8 7f  std       %d0, [%sp + 0x87f]
    sum2+0xc:               98 13 60 00  or        %o5, 0x0, %o4
    sum2+0x10:              97 2b 30 0c  sllx      %o4, 0xc, %o3
    sum2+0x14:              c5 1a e0 00  ldd       [%o3], %d2
    sum2+0x18:              81 c3 e0 08  retl
    sum2+0x1c:              c5 3b a8 87  std       %d2, [%sp + 0x887]

I believe this in line with SCD 2.4.1:

* 3P-13 states

  Structure or Union return values

  Structure and union return types up to thirty-two bytes in size are returned
in registers. The registers are assigned as if
  the value was being passed as the first argument to a function with a known
prototype.

* and p. 3P-12 has

  Structure or union types larger than eight bytes, and up to sixteen bytes in
size are assigned to two consecutive
  parameter array words, and align according to the alignment requirements of
the structure or at least to an eight-byte
boundary.

* If I read the table on p. 3P-11 correctly, those parameter array words should
  map to %d0/%d2 for double args, just as cc does.