public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Manolis Tsamis <manolis.tsamis@vrull.eu>
To: gcc-patches@gcc.gnu.org
Cc: Robin Dapp <rdapp@linux.ibm.com>,
	Jiangning Liu <jiangning.liu@amperecomputing.com>,
	Philipp Tomsich <philipp.tomsich@vrull.eu>,
	Richard Sandiford <richard.sandiford@arm.com>,
	Jakub Jelinek <jakub@redhat.com>,
	Manolis Tsamis <manolis.tsamis@vrull.eu>
Subject: [PATCH v4 2/3] [RFC] ifcvt: Allow more operations in multiple set if conversion
Date: Tue, 23 Apr 2024 12:47:39 +0200	[thread overview]
Message-ID: <20240423104740.4027243-3-manolis.tsamis@vrull.eu> (raw)
In-Reply-To: <20240423104740.4027243-1-manolis.tsamis@vrull.eu>

Currently the operations allowed for if conversion of a basic block with
multiple sets are few, namely REG, SUBREG and CONST_INT (as controlled by
bb_ok_for_noce_convert_multiple_sets).

This commit allows more operations (arithmetic, compare, etc) to participate
in if conversion. The target's profitability hook and ifcvt's costing is
expected to reject sequences that are unprofitable.

This is especially useful for targets which provide a rich selection of
conditional instructions (like aarch64 which has cinc, csneg, csinv, ccmp, ...)
which are currently not used in basic blocks with more than a single set.

gcc/ChangeLog:

	* ifcvt.cc (try_emit_cmove_seq): Modify comments.
	(noce_convert_multiple_sets_1): Modify comments.
	(bb_ok_for_noce_convert_multiple_sets): Allow more operations.

gcc/testsuite/ChangeLog:

	* gcc.target/aarch64/ifcvt_multiple_sets_arithm.c: New test.

Signed-off-by: Manolis Tsamis <manolis.tsamis@vrull.eu>
---

Changes in v4:
        - Remove unnecessary hardcoded list of allowed ops in
        bb_ok_for_noce_convert_multiple_sets.
        - Set need_cmov based on BB live_out instead of REG_DEAD notes.
        - Fix preexisting issues and improve the code that sets read_comparison.

 gcc/ifcvt.cc                                  | 34 +++-----
 .../aarch64/ifcvt_multiple_sets_arithm.c      | 79 +++++++++++++++++++
 2 files changed, 92 insertions(+), 21 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/aarch64/ifcvt_multiple_sets_arithm.c

diff --git a/gcc/ifcvt.cc b/gcc/ifcvt.cc
index 763a25f816e..dc00042be81 100644
--- a/gcc/ifcvt.cc
+++ b/gcc/ifcvt.cc
@@ -3432,13 +3432,13 @@ try_emit_cmove_seq (struct noce_if_info *if_info, rtx temp,
 /* We have something like:
 
      if (x > y)
-       { i = a; j = b; k = c; }
+       { i = EXPR_A; j = EXPR_B; k = EXPR_C; }
 
    Make it:
 
-     tmp_i = (x > y) ? a : i;
-     tmp_j = (x > y) ? b : j;
-     tmp_k = (x > y) ? c : k;
+     tmp_i = (x > y) ? EXPR_A : i;
+     tmp_j = (x > y) ? EXPR_B : j;
+     tmp_k = (x > y) ? EXPR_C : k;
      i = tmp_i;
      j = tmp_j;
      k = tmp_k;
@@ -3839,11 +3839,10 @@ noce_convert_multiple_sets_1 (struct noce_if_info *if_info,
 
 
 
-/* Return true iff basic block TEST_BB is comprised of only
-   (SET (REG) (REG)) insns suitable for conversion to a series
-   of conditional moves.  Also check that we have more than one set
-   (other routines can handle a single set better than we would), and
-   fewer than PARAM_MAX_RTL_IF_CONVERSION_INSNS sets.  While going
+/* Return true iff basic block TEST_BB is suitable for conversion to a
+   series of conditional moves.  Also check that we have more than one
+   set (other routines can handle a single set better than we would),
+   and fewer than PARAM_MAX_RTL_IF_CONVERSION_INSNS sets.  While going
    through the insns store the sum of their potential costs in COST.  */
 
 static bool
@@ -3869,20 +3868,13 @@ bb_ok_for_noce_convert_multiple_sets (basic_block test_bb, unsigned *cost)
       rtx dest = SET_DEST (set);
       rtx src = SET_SRC (set);
 
-      /* We can possibly relax this, but for now only handle REG to REG
-	 (including subreg) moves.  This avoids any issues that might come
-	 from introducing loads/stores that might violate data-race-freedom
-	 guarantees.  */
-      if (!REG_P (dest))
-	return false;
-
-      if (!((REG_P (src) || CONSTANT_P (src))
-	    || (GET_CODE (src) == SUBREG && REG_P (SUBREG_REG (src))
-	      && subreg_lowpart_p (src))))
+      /* Do not handle anything involving memory loads/stores since it might
+	 violate data-race-freedom guarantees.  */
+      if (!REG_P (dest) || contains_mem_rtx_p (src))
 	return false;
 
-      /* Destination must be appropriate for a conditional write.  */
-      if (!noce_operand_ok (dest))
+      /* Destination and source must be appropriate.  */
+      if (!noce_operand_ok (dest) || !noce_operand_ok (src))
 	return false;
 
       /* We must be able to conditionally move in this mode.  */
diff --git a/gcc/testsuite/gcc.target/aarch64/ifcvt_multiple_sets_arithm.c b/gcc/testsuite/gcc.target/aarch64/ifcvt_multiple_sets_arithm.c
new file mode 100644
index 00000000000..d977f4d62ec
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/ifcvt_multiple_sets_arithm.c
@@ -0,0 +1,79 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-rtl-ce1" } */
+
+void sink2(int, int);
+void sink3(int, int, int);
+
+void cond1(int cond, int x, int y)
+{
+  if (cond)
+    {
+      x = x << 4;
+      y = 1;
+    }
+
+  sink2(x, y);
+}
+
+void cond2(int cond, int x, int y)
+{
+  if (cond)
+    {
+      x++;
+      y++;
+    }
+
+  sink2(x, y);
+}
+
+void cond3(int cond, int x1, int x2, int x3)
+{
+  if (cond)
+    {
+      x1++;
+      x2++;
+      x3++;
+    }
+
+  sink3(x1, x2, x3);
+}
+
+void cond4(int cond, int x, int y)
+{
+  if (cond)
+    {
+      x += 2;
+      y += 3;
+    }
+
+  sink2(x, y);
+}
+
+void cond5(int cond, int x, int y, int r1, int r2)
+{
+  if (cond)
+    {
+      x = r1 + 2;
+      y = r2 - 34;
+    }
+
+  sink2(x, y);
+}
+
+void cond6(int cond, int x, int y)
+{
+  if (cond)
+    {
+      x = -x;
+      y = ~y;
+    }
+
+  sink2(x, y);
+}
+
+/* { dg-final { scan-assembler-times "cinc\t" 5 } } */
+/* { dg-final { scan-assembler-times "csneg\t" 1 } } */
+/* { dg-final { scan-assembler-times "csinv\t" 1 } } */
+/* { dg-final { scan-assembler "csel\t" } } */
+
+/* { dg-final { scan-rtl-dump-times "if-conversion succeeded through noce_convert_multiple_sets" 6 "ce1" } } */
\ No newline at end of file
-- 
2.34.1


  parent reply	other threads:[~2024-04-23 10:47 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-04-23 10:47 [PATCH v4 0/3] ifcvt: Allow if conversion of arithmetic in basic blocks with multiple sets Manolis Tsamis
2024-04-23 10:47 ` [PATCH v4 1/3] [RFC] ifcvt: handle sequences that clobber flags in noce_convert_multiple_sets Manolis Tsamis
2024-04-23 10:47 ` Manolis Tsamis [this message]
2024-04-24 23:40   ` [PATCH v4 2/3] [RFC] ifcvt: Allow more operations in multiple set if conversion Hans-Peter Nilsson
2024-04-30 14:54     ` Manolis Tsamis
2024-04-23 10:47 ` [PATCH v4 3/3] [RFC] ifcvt: Handle multiple rewired regs and refactor noce_convert_multiple_sets Manolis Tsamis
2024-05-14 12:06 ` [PATCH v4 0/3] ifcvt: Allow if conversion of arithmetic in basic blocks with multiple sets Manolis Tsamis

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20240423104740.4027243-3-manolis.tsamis@vrull.eu \
    --to=manolis.tsamis@vrull.eu \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=jakub@redhat.com \
    --cc=jiangning.liu@amperecomputing.com \
    --cc=philipp.tomsich@vrull.eu \
    --cc=rdapp@linux.ibm.com \
    --cc=richard.sandiford@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).