[Bug debug/86455] var-tracking mishandles pre_dec

[vries at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86455

--- Comment #2 from Tom de Vries <vries at gcc dot gnu.org> ---
Created attachment 44391
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44391&action=edit
Reproducer patch, splits push

Reproduced without the fkeep-vars-live patch.

Consider this test-case:
... 
static volatile int vv = 1;
static volatile int z;

static long __attribute__((noinline, noclone))
foo (long x)
{
  z = x;
}

int
main ()
{
  long x = vv;
  foo (x);
  foo (x + 1);
  return 0;
}
...

which at -O1 -g -fpeephole2 contains a pushq at the start of main:
...
        .cfi_startproc
        pushq   %rbx
        .cfi_def_cfa_offset 16
        .cfi_offset 3, -16
...

Using this debug patch:
...
diff --git a/gcc/var-tracking.c b/gcc/var-tracking.c
index 8e800960b6d..0a2ac16b914 100644
--- a/gcc/var-tracking.c
+++ b/gcc/var-tracking.c
@@ -10213,7 +10213,11 @@ vt_initialize (void)
                    }

                  cselib_hook_called = false;
+                 fprintf (stderr, "ORIGINAL: \n");
+                 debug_rtx (insn);
                  adjust_insn (bb, insn);
+                 fprintf (stderr, "ADJUSTED: \n");
+                 debug_rtx (insn);
                  if (DEBUG_MARKER_INSN_P (insn))
                    {
                      reemit_marker_as_note (insn);
...

we can observe the var-tracking translation of the push:
...
ORIGINAL:
(insn/f 26 3 27 2 (set (mem:DI (pre_dec:DI (reg/f:DI 7 sp)) [0  S8 A8])
        (reg:DI 3 bx)) "clztest.c":12 61 {*pushdi2_rex64}
     (expr_list:REG_DEAD (reg:DI 3 bx)
        (nil)))
ADJUSTED:
(insn/f 26 3 27 2 (parallel [
            (set (mem:DI (plus:DI (reg/f:DI 16 argp)
                        (const_int -24 [0xffffffffffffffe8])) [0  S8 A8])
                (reg:DI 3 bx))
            (set (reg/f:DI 7 sp)
                (plus:DI (reg/f:DI 16 argp)
                    (const_int -24 [0xffffffffffffffe8])))
        ]) "clztest.c":12 61 {*pushdi2_rex64}
     (expr_list:REG_DEAD (reg:DI 3 bx)
        (nil)))
...

Now, using the reproducer patch, we split up the push:
...
       .cfi_startproc
        leaq    -8(%rsp), %rsp
        .cfi_def_cfa_offset 16
        movq    %rbx, (%rsp)
        .cfi_offset 3, -16
...

and once more observe the var-tracking translation, now of the split up push:
...
ORIGINAL:
(insn/f 34 3 35 2 (set (reg/f:DI 7 sp)
        (plus:DI (reg/f:DI 7 sp)
            (const_int -8 [0xfffffffffffffff8]))) "clztest.c":12 228 {*leadi}
     (nil))
ADJUSTED:
(insn/f 34 3 35 2 (set (reg/f:DI 7 sp)
        (plus:DI (reg/f:DI 16 argp)
            (const_int -16 [0xfffffffffffffff0]))) "clztest.c":12 228 {*leadi}
     (nil))
ORIGINAL:
(insn/f 35 34 27 2 (set (mem:DI (reg/f:DI 7 sp) [0  S8 A8])
        (reg:DI 3 bx)) "clztest.c":12 85 {*movdi_internal}
     (nil))
ADJUSTED:
(insn/f 35 34 27 2 (set (mem:DI (plus:DI (reg/f:DI 16 argp)
                (const_int -16 [0xfffffffffffffff0])) [0  S8 A8])
        (reg:DI 3 bx)) "clztest.c":12 85 {*movdi_internal}
     (nil))
...

Interestingly, the argp offset changed from -24 to -16.