[Bug target/107102] New: SVE function fails to realize it doesn't need the frame-pointer in the tail call.

[Bug target/107102] New: SVE function fails to realize it doesn't need the frame-pointer in the tail call.

[tnfchris at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107102

            Bug ID: 107102
           Summary: SVE function fails to realize it doesn't need the
                    frame-pointer in the tail call.
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tnfchris at gcc dot gnu.org
  Target Milestone: ---
            Target: aarch64-*

The following example:

#include <arm_sve.h>

void foo(svbool_t p, svbool_t q);

void bar() {
    foo(svptrue_b32(), svptrue_b8());
}

generates at -O2 -march=armv8-a+sve2:

bar:
        stp     x29, x30, [sp, -16]!
        ptrue   p1.b, all
        ptrue   p0.s, all
        mov     x29, sp
        bl      foo
        ldp     x29, x30, [sp], 16
        ret

The tail-call gets de-optimized and GCC fails to realize it doesn't need a
frame pointer nor the stack pointer at all.

In RTL before frame layout there is no stack usage at all however this looks to
happen because in the expansion there's a clobber on the LR:

(call_insn 7 6 10 2 (parallel [
            (call (mem:DI (symbol_ref:DI ("foo") [flags 0x41]  <function_decl
0x7fac39ef9e00 foo>) [0 foo S8 A8])
                (const_int 0 [0]))
            (unspec:DI [
                    (const_int 2 [0x2])
                ] UNSPEC_CALLEE_ABI)
            (clobber (reg:DI 30 x30))
        ]) "example.c":6:5 46 {*call_insn}

and it, it looks like the expansion missed the sibcall?

[Bug target/107102] SVE function fails to realize it doesn't need the frame-pointer in the tail call.

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107102

ktkachov at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ktkachov at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-10-04

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Confirmed, clang tail-calls this:
bar:                                    // @bar
        ptrue   p1.b
        ptrue   p0.s
        and     p0.b, p1/z, p1.b, p0.b
        b       foo