https://gcc.gnu.org/g:07a0baa7609a5a78d8b4a3c17897768fb2119605
commit r16-3237-g07a0baa7609a5a78d8b4a3c17897768fb2119605
Author: Artemiy Volkov <artem...@acm.org>
Date: Sat Aug 16 20:40:28 2025 -0600
ira: tie output allocnos for fused instruction pairs
Some of the instruction pairs recognized as fusible by a preceding
invocation of the dep_fusion pass require that both components of a pair
have the same hard register output for the fusion to work in hardware.
(An example of this would be a multiply-add operation, or a zero-extract
operation composed of two shifts.)
For all such pairs, the following conditions will hold:
(a) Both insns are single_sets
(b) Both insns have a register destination
(c) The pair has been marked as fusible by setting the second insn's
SCHED_GROUP flag
(d) Additionally, post-RA, both instructions' destination regnos are
equal
(All of these conditions are encapsulated in the newly created
single_output_fused_pair_p () predicate.)
During IRA, if conditions (a)-(c) above hold, we need to tie the two
instructions' destination allocnos together so that they are allocated
to the same hard register. We do this in add_insn_allocno_copies () by
adding a constraint conflict to the output operands of the two
instructions.
gcc/ChangeLog:
* ira-conflicts.cc (add_insn_allocno_copies): Handle fused insn
pairs.
* rtl.h (single_output_fused_pair_p): Declare new function.
* rtlanal.cc (single_output_fused_pair_p): Define it.
Diff:
---
gcc/ira-conflicts.cc | 12 ++++++++++--
gcc/rtl.h | 1 +
gcc/rtlanal.cc | 20 ++++++++++++++++++++
3 files changed, 31 insertions(+), 2 deletions(-)
diff --git a/gcc/ira-conflicts.cc b/gcc/ira-conflicts.cc
index d8f7c1e1c379..e9ab16a5d20a 100644
--- a/gcc/ira-conflicts.cc
+++ b/gcc/ira-conflicts.cc
@@ -448,7 +448,7 @@ process_reg_shuffles (rtx_insn *insn, rtx reg, int op_num,
int freq,
static void
add_insn_allocno_copies (rtx_insn *insn)
{
- rtx set, operand, dup;
+ rtx set = single_set (insn), operand, dup;
bool bound_p[MAX_RECOG_OPERANDS];
int i, n, freq;
alternative_mask alts;
@@ -456,7 +456,15 @@ add_insn_allocno_copies (rtx_insn *insn)
freq = REG_FREQ_FROM_BB (BLOCK_FOR_INSN (insn));
if (freq == 0)
freq = 1;
- if ((set = single_set (insn)) != NULL_RTX
+
+ /* Tie output register operands of two consecutive single_sets
+ marked as a fused pair. */
+ if (single_output_fused_pair_p (insn))
+ process_regs_for_copy (SET_DEST (set),
+ SET_DEST (single_set (prev_nonnote_nondebug_insn (insn))),
+ true, NULL, freq);
+
+ if (set != NULL_RTX
&& REG_SUBREG_P (SET_DEST (set)) && REG_SUBREG_P (SET_SRC (set))
&& ! side_effects_p (set)
&& find_reg_note (insn, REG_DEAD,
diff --git a/gcc/rtl.h b/gcc/rtl.h
index 5bd0bd4d168b..9684b45f2a5b 100644
--- a/gcc/rtl.h
+++ b/gcc/rtl.h
@@ -3670,6 +3670,7 @@ extern bool contains_symbol_ref_p (const_rtx);
extern bool contains_symbolic_reference_p (const_rtx);
extern bool contains_constant_pool_address_p (const_rtx);
extern void add_auto_inc_notes (rtx_insn *, rtx);
+extern bool single_output_fused_pair_p (rtx_insn *);
/* Handle the cheap and common cases inline for performance. */
diff --git a/gcc/rtlanal.cc b/gcc/rtlanal.cc
index 87332ffebcec..19b66456e45c 100644
--- a/gcc/rtlanal.cc
+++ b/gcc/rtlanal.cc
@@ -6976,6 +6976,26 @@ add_auto_inc_notes (rtx_insn *insn, rtx x)
}
}
+/* Return true if INSN is the second element of a pair of macro-fused
+ single_sets, both of which having the same register output as another. */
+bool
+single_output_fused_pair_p (rtx_insn *insn)
+{
+ rtx set, prev_set;
+ rtx_insn *prev;
+
+ return INSN_P (insn)
+ && SCHED_GROUP_P (insn)
+ && (prev = prev_nonnote_nondebug_insn (insn))
+ && (set = single_set (insn)) != NULL_RTX
+ && (prev_set = single_set (prev))
+ != NULL_RTX
+ && REG_P (SET_DEST (set))
+ && REG_P (SET_DEST (prev_set))
+ && (!reload_completed
+ || REGNO (SET_DEST (set)) == REGNO (SET_DEST (prev_set)));
+}
+
/* Return true if X is register asm. */
bool
