public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Zhao Wei Liew <zhaoweiliew@gmail.com>
To: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: [PATCH] tree-optimization: [PR103855] Fold (type)X / (type)Y
Date: Tue, 22 Feb 2022 11:57:19 +0800	[thread overview]
Message-ID: <CALHvHFV6kWN1gdRf6HR6gr5qz1LOopXDJ8JtcBNkxZKmspYb5A@mail.gmail.com> (raw)

[-- Attachment #1: Type: text/plain, Size: 512 bytes --]

Hi,

This is a partial optimization for PR103855.

Initially, I looked into optimizing RTL generation or a more complex
GIMPLE transformation so that we could optimize other cases as well,
such as ((unsigned long long) short / int).

However, that is a bit too complex for now. While I continue to look
into that change, I've decided to implement this simpler match.pd
transformation.

Greatly appreciate any feedback on this patch or guidance for
implementing the more advanced optimizations!

Thanks,
Zhao Wei

[-- Attachment #2: 0001-tree-optimization-PR103855-Fold-type-X-type-Y.patch --]
[-- Type: application/octet-stream, Size: 3754 bytes --]

From dd3bb05cd7be72d080598cb693549ac74d5cb02d Mon Sep 17 00:00:00 2001
From: Zhao Wei Liew <zhaoweiliew@gmail.com>
Date: Sat, 19 Feb 2022 16:28:38 +0800
Subject: [PATCH] tree-optimization: [PR103855] Fold (type)X / (type)Y

This pattern converts (trunc_div (convert a) (convert b)) to
(convert (trunc_div a b)) when:

1. type, a, and b all have unsigned integeral types
2. a and b have the same type precision
3. type has type precision at least as larger as a and b

This is useful as wider divisions are typically more expensive.

To illustrate the effects, consider the following code snippet:

unsigned long long f(unsigned int a, unsigned int b) {
	unsigned long long all = a;
	return all / b;
}

Without the pattern, g++ -std=c++20 -O2 generates the following
assembly:

f(unsigned int, unsigned int):
	mov eax, edi
	mov esi, esi
	xor edx, edx
	div rsi
	ret

With the pattern, it generates this:

f(unsigned int, unsigned int):
	mov eax, edi
	xor edx, edx
	div esi
	ret

This is identical to what clang++ -std=c++20 -O2 generates.

Bootstrapped and regression tested on x86_64-pc-linux-gnu.

Signed-off-by: Zhao Wei Liew <zhaoweiliew@gmail.com>

	PR tree-optimization/103855

gcc/ChangeLog:

	* match.pd: Add pattern for (type)X / (type)Y.

gcc/testsuite/ChangeLog:

	* gcc.dg/tree-ssa/divide-8.c: New test.
	* gcc.dg/tree-ssa/divide-9.c: New test.
---
 gcc/match.pd                             | 15 +++++++++++++++
 gcc/testsuite/gcc.dg/tree-ssa/divide-8.c |  9 +++++++++
 gcc/testsuite/gcc.dg/tree-ssa/divide-9.c |  9 +++++++++
 3 files changed, 33 insertions(+)
 create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/divide-8.c
 create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/divide-9.c

diff --git a/gcc/match.pd b/gcc/match.pd
index 10f62284862..393b43756dd 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -684,6 +684,21 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
  (if (INTEGRAL_TYPE_P (type) || VECTOR_INTEGER_TYPE_P (type))
   (convert (trunc_mod @0 @1))))
 
+/* (type)X / (type)Y -> (type)(X / Y)
+   when the resulting type is at least precise as the original types
+   and when all the types are unsigned integral types. */
+(simplify
+ (trunc_div (convert @0) (convert @1))
+ (if (INTEGRAL_TYPE_P (type)
+      && INTEGRAL_TYPE_P (TREE_TYPE (@0))
+      && INTEGRAL_TYPE_P (TREE_TYPE (@1))
+      && TYPE_UNSIGNED (type)
+      && TYPE_UNSIGNED (TREE_TYPE (@0))
+      && TYPE_UNSIGNED (TREE_TYPE (@1))
+      && TYPE_PRECISION (TREE_TYPE (@0)) == TYPE_PRECISION (TREE_TYPE (@1))
+      && TYPE_PRECISION (type) >= TYPE_PRECISION (TREE_TYPE (@0)))
+  (convert (trunc_div @0 @1))))
+
 /* x * (1 + y / x) - y -> x - y % x */
 (simplify
  (minus (mult:cs @0 (plus:s (trunc_div:s @1 @0) integer_onep)) @1)
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/divide-8.c b/gcc/testsuite/gcc.dg/tree-ssa/divide-8.c
new file mode 100644
index 00000000000..dc3dc9ca769
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/divide-8.c
@@ -0,0 +1,9 @@
+/* PR tree-optimization/103855 */
+/* { dg-options "-O -fdump-tree-optimized" } */
+
+unsigned int f(unsigned int a, unsigned int b) {
+    unsigned long long all = a;
+    return all / b;
+}
+
+/* { dg-final { scan-tree-dump-not "\\\(long long unsigned int\\\)" "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/divide-9.c b/gcc/testsuite/gcc.dg/tree-ssa/divide-9.c
new file mode 100644
index 00000000000..6986b5484e4
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/divide-9.c
@@ -0,0 +1,9 @@
+/* PR tree-optimization/103855 */
+/* { dg-options "-O -fdump-tree-optimized" } */
+
+unsigned long long f(unsigned int a, unsigned int b) {
+    unsigned long long all = a;
+    return all / b;
+}
+
+/* { dg-final { scan-tree-dump-times "\\\(long long unsigned int\\\)" 1 "optimized" } } */
-- 
2.35.1


             reply	other threads:[~2022-02-22  3:57 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-02-22  3:57 Zhao Wei Liew [this message]
2022-02-22  4:00 ` Zhao Wei Liew
2022-02-22  7:53   ` Richard Biener
2022-03-17  0:49     ` [PATCH][v2] tree-optimization: Fold (type)X / (type)Y [PR103855] Zhao Wei Liew
2022-07-09 18:19       ` Jeff Law
  -- strict thread matches above, loose matches on Subject: below --
2022-02-19  9:36 [PATCH] tree-optimization: [PR103855] Fold (type)X / (type)Y Zhao Wei Liew
2022-02-19 10:05 ` Zhao Wei Liew

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=CALHvHFV6kWN1gdRf6HR6gr5qz1LOopXDJ8JtcBNkxZKmspYb5A@mail.gmail.com \
    --to=zhaoweiliew@gmail.com \
    --cc=gcc-patches@gcc.gnu.org \
    /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).