https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104815

            Bug ID: 104815
           Summary: [nvptx] Use bitbucket operand when REG_UNUSED
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vries at gcc dot gnu.org
  Target Milestone: ---

Consider this source code:
...
enum memmodel
{
  MEMMODEL_RELAXED = 0
};

unsigned long long int *p64;
unsigned long long int v64;

int
main()
{
  __atomic_fetch_add (p64, v64, MEMMODEL_RELAXED);

  return 0;
}
...

It results in this code:
...
{
        .reg.u32 %value;
        .reg.u64 %r25;
        .reg.u64 %r26;
        .reg.u64 %r27;
                ld.global.u64   %r25, [p64];
                ld.global.u64   %r27, [v64];
                atom.add.u64    %r26, [%r25], %r27;
                mov.u32 %value, 0;
        st.param.u32    [%value_out], %value;
        ret;
}
...

We can however do:
...
                atom.add.u64    _, [%r25], %r27;
...
because %r26 is not used anywhere.

We can detect this situation in the insn by finding a reg-note:
...
/* Identifies a register set in this insn and never used.  */
REG_NOTE (UNUSED)
...

Currently testing this code:
...
diff --git a/gcc/config/nvptx/nvptx.cc b/gcc/config/nvptx/nvptx.cc
index c6cec0c27c22..c7223737cca5 100644
--- a/gcc/config/nvptx/nvptx.cc
+++ b/gcc/config/nvptx/nvptx.cc
@@ -538,7 +538,15 @@ output_reg (FILE *file, unsigned regno, machine_mode
inner_mode,

       if (HARD_REGISTER_NUM_P (regno))
        fprintf (file, "%s", reg_names[regno]);
       else
-       fprintf (file, "%%r%d", regno);
+       {
+         rtx reg = regno_reg_rtx[regno];
+         if (current_output_insn
+             && (find_reg_note (current_output_insn, REG_UNUSED, reg)
+                 != NULL_RTX))
+           fprintf (file, "_");
+         else
+           fprintf (file, "%%r%d", regno);
+       }
     }
   else if (subreg_offset >= 0)
     {
...

Reply via email to