Hi,
This patch fixes the way 'uses_hard_regs_p' handles paradoxical subregs.
The function is supposed to detect whether a register access of 'x'
overlaps with 'set'. For SUBREGs it should check whether any of the
full multi-register overlaps with 'set'. The former behavior used to
grab the widest mode of the inner/outer registers of a SUBREG and the
inner register, and check all registers from the inner-register onwards
for the given width. For normal SUBREGS this gives you the full
register, for paradoxical SUBREGS however it may give you the wrong set
of registers if the index is not the first of the multi-register set.
The original error reported in PR target/86487 can no longer be
reproduced with the given test, this was due to an unrelated code-gen
change, regardless I believe this should still be fixed as it is simply
wrong behavior by uses_hard_regs_p which may be triggered by a different
test-case or by future changes to the compiler. Also it is useful to
point out that this isn't actually a 'target' issue as this code could
potentially hit any other target using paradoxical SUBREGS. Should I
change the Bugzilla ticket to reflect this is actually a target agnostic
issue in RTL?
There is a gotcha here, I don't know what would happen if you hit the
cases of get_hard_regno where it would return -1, quoting the comment
above that function "If X is not a register or a subreg of a register,
return -1." though I think if we are hitting this then things must have
gone wrong before?
Bootstrapped on aarch64, arm and x86, no regressions.
Is this OK for trunk?
gcc/ChangeLog:
2019-01-07 Andre Vieira <andre.simoesdiasvie...@arm.com>
PR target/86487
* lra-constraints.c(uses_hard_regs_p): Fix handling of
paradoxical SUBREGS.
diff --git a/gcc/lra-constraints.c b/gcc/lra-constraints.c
index c061093ed699620afe2dfda60d58066d6967523a..736b084acc552b75ff4d369b6584bc9ab422e21b 100644
--- a/gcc/lra-constraints.c
+++ b/gcc/lra-constraints.c
@@ -1761,11 +1761,21 @@ uses_hard_regs_p (rtx x, HARD_REG_SET set)
return false;
code = GET_CODE (x);
mode = GET_MODE (x);
+
if (code == SUBREG)
{
+ /* For all SUBREGs we want to check whether the full multi-register
+ overlaps the set. For normal SUBREGs this means 'get_hard_regno' of
+ the inner register, for paradoxical SUBREGs this means the
+ 'get_hard_regno' of the full SUBREG and for complete SUBREGs either is
+ fine. Use the wider mode for all cases. */
+ rtx subreg = SUBREG_REG (x);
mode = wider_subreg_mode (x);
- x = SUBREG_REG (x);
- code = GET_CODE (x);
+ if (mode == GET_MODE (subreg))
+ {
+ x = subreg;
+ code = GET_CODE (x);
+ }
}
if (REG_P (x))
