[Bug target/82717] New: [RISCV] Default value of the -mabi option doesn't match documentation

[asb at lowrisc dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82717

            Bug ID: 82717
           Summary: [RISCV] Default value of the -mabi option doesn't
                    match documentation
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: asb at lowrisc dot org
  Target Milestone: ---

The documentation for the -mabi command-line option for RISC-V states:

"""
@item -mabi=@var{ABI-string}
@opindex mabi
Specify integer and floating-point calling convention.  This defaults to the
natural calling convention: e.g.@ LP64 for RV64I, ILP32 for RV32I, LP64D for
RV64G.
"""

By my reading, this indicates that -march=rv32imafd would imply -mabi=ilp32d as
the "natural" calling convention. This is not the case. Note how the soft-float
calling convention is used in the example below:

$ cat t.c
double callee(float a, double b, int c) {
  return a+b+c;
}

double caller(void) {
  callee(1.0, 2.0, 3);
}

$ ./riscv32-unknown-elf-gcc -march=rv32imafdc t.c -c -S -o -
        .file   "t.c"
        .option nopic
        .text
        .align  1
        .globl  callee
        .type   callee, @function
callee:
        add     sp,sp,-48
        sw      s0,44(sp)
        add     s0,sp,48
        sw      a0,-20(s0)
        sw      a1,-32(s0)
        sw      a2,-28(s0)
        sw      a3,-24(s0)
        flw     fa5,-20(s0)
        fcvt.d.s        fa4,fa5
        fld     fa5,-32(s0)
        fadd.d  fa4,fa4,fa5
        lw      a5,-24(s0)
        fcvt.d.w        fa5,a5
        fadd.d  fa5,fa4,fa5
        fsd     fa5,-40(s0)
        lw      a0,-40(s0)
        lw      a1,-36(s0)
        lw      s0,44(sp)
        add     sp,sp,48
        jr      ra
        .size   callee, .-callee
        .align  1
        .globl  caller
        .type   caller, @function
caller:
        add     sp,sp,-32
        sw      ra,28(sp)
        sw      s0,24(sp)
        add     s0,sp,32
        lui     a5,%hi(.LC0)
        addi    a5,a5,%lo(.LC0)
        fld     fa4,0(a5)
        lui     a5,%hi(.LC1)
        flw     fa5,%lo(.LC1)(a5)
        li      a3,3
        fsd     fa4,-24(s0)
        lw      a1,-24(s0)
        lw      a2,-20(s0)
        fmv.x.s a0,fa5
        call    callee
        nop
        fsd     fa5,-24(s0)
        lw      a0,-24(s0)
        lw      a1,-20(s0)
        lw      ra,28(sp)
        lw      s0,24(sp)
        add     sp,sp,32
        jr      ra
        .size   caller, .-caller
        .section        .rodata
        .align  3
.LC0:
        .word   0
        .word   1073741824
        .align  2
.LC1:
        .word   1065353216
        .ident  "GCC: (GNU) 8.0.0 20171025 (experimental)"



I would also note that the documentation could be improved by better detailing
the accepted ABI strings and giving valid examples (ILP32 isn't accepted as it
is uppercase).