The add expander still contains some expansion code that was required for the
previous prolog/epilog code, but which is no longer needed. I also noticed that
the current version splits off immediates from frame addressing instructions,
which doesn't seem a good idea. Avoiding this resulted in small codesize
improvements.
ChangeLog:
2016-10-24 Wilco Dijkstra <wdijk...@arm.com>
gcc/
* config/aarch64/aarch64.md (add<mode>3): Remove
redundant code. Don't split frame based additions.
--
diff --git a/gcc/config/aarch64/aarch64.md b/gcc/config/aarch64/aarch64.md
index
89e932ae104e9fccb3fbd4af0fbb4a4414e0542e..13a162caae2b93221939bd34a7c01f5da2d4be9e
100644
--- a/gcc/config/aarch64/aarch64.md
+++ b/gcc/config/aarch64/aarch64.md
@@ -1586,25 +1586,11 @@
(match_operand:GPI 2 "aarch64_pluslong_operand" "")))]
""
{
- if (aarch64_pluslong_strict_immedate (operands[2], <MODE>mode))
- {
- /* Give CSE the opportunity to share this constant across additions. */
- if (!cse_not_expected && can_create_pseudo_p ())
- operands[2] = force_reg (<MODE>mode, operands[2]);
-
- /* Split will refuse to operate on a modification to the stack pointer.
- Aid the prologue and epilogue expanders by splitting this now. */
- else if (reload_completed && operands[0] == stack_pointer_rtx)
- {
- HOST_WIDE_INT i = INTVAL (operands[2]);
- HOST_WIDE_INT s = (i >= 0 ? i & 0xfff : -(-i & 0xfff));
- emit_insn (gen_rtx_SET (operands[0],
- gen_rtx_PLUS (<MODE>mode, operands[1],
- GEN_INT (i - s))));
- operands[1] = operands[0];
- operands[2] = GEN_INT (s);
- }
- }
+ /* If the constant is too large for a single instruction and isn't frame
+ based, split off the immediate so it is available for CSE. */
+ if (!aarch64_plus_immediate (operands[2], <MODE>mode)
+ && can_create_pseudo_p () && !REGNO_PTR_FRAME_P (REGNO (operands[1])))
+ operands[2] = force_reg (<MODE>mode, operands[2]);
})
(define_insn "*add<mode>3_aarch64"
