[PATCH] arm: Low overhead loop handle long range branches [PR98931]

[Andrea Corallo via Gcc-patches]

Hi all,

this is to fix PR98931 where the LE (loop end, Armv8.1-M low overhead
loops) instruction cannot cover sufficently long branches.

In this case we emit as an alternative:

        subs    lr, #1
        bne     label

arm-none-eabi regtested, arm-none-linux-gnueabihf boostrapped.

Okay for trunk?

Regards

  Andrea
  

>From c8216ed1313d670e79b28141dadd644e698c83cf Mon Sep 17 00:00:00 2001
From: Andrea Corallo <andrea.cora...@arm.com>
Date: Wed, 3 Feb 2021 15:21:54 +0100
Subject: [PATCH] arm: Low overhead loop handle long range branches [PR98931]

gcc/ChangeLog

2021-02-05  Andrea Corallo  <andrea.cora...@arm.com>

        * config/arm/thumb2.md: Generate alternative sequence to
        handle long range branches.

gcc/testsuite/Changelog

2021-02-08  Andrea Corallo  <andrea.cora...@arm.com>

         * gcc.target/arm/pr98931.c: New testcase.
---
 gcc/config/arm/thumb2.md               | 13 ++++++++++++-
 gcc/testsuite/gcc.target/arm/pr98931.c | 17 +++++++++++++++++
 2 files changed, 29 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/gcc.target/arm/pr98931.c

diff --git a/gcc/config/arm/thumb2.md b/gcc/config/arm/thumb2.md
index bd53bf320de..2646926d3c1 100644
--- a/gcc/config/arm/thumb2.md
+++ b/gcc/config/arm/thumb2.md
@@ -1719,7 +1719,18 @@
               (set (reg:SI LR_REGNUM)
                    (plus:SI (reg:SI LR_REGNUM) (const_int -1)))])]
   "TARGET_32BIT && TARGET_HAVE_LOB"
-  "le\t%|lr, %l0")
+  "*
+  if (get_attr_length (insn) == 4)
+    return \"le\\t%|lr, %l0\";
+  else
+    return \"subs\\t%|lr, #1\;bne\\t%l0\";
+  "
+  [(set (attr "length")
+        (if_then_else
+            (lt (minus (pc) (match_dup 0)) (const_int 1024))
+           (const_int 4)
+           (const_int 6)))
+   (set_attr "type" "branch")])
 
 (define_expand "doloop_begin"
   [(match_operand 0 "" "")
diff --git a/gcc/testsuite/gcc.target/arm/pr98931.c 
b/gcc/testsuite/gcc.target/arm/pr98931.c
new file mode 100644
index 00000000000..313876a3912
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/pr98931.c
@@ -0,0 +1,17 @@
+/* { dg-do assemble } */
+/* { dg-skip-if "avoid conflicting multilib options" { *-*-* } { "-marm" 
"-mcpu=*" } } */
+/* { dg-options "-march=armv8.1-m.main -O3 
--param=max-completely-peeled-insns=1300 --save-temps" } */
+
+extern long long a[][20][26][26][22];
+
+void
+foo ()
+{
+  for (short d = 0; d + 1; d++)
+    for (unsigned e = 0; e < 25; e += 4)
+      for (unsigned f = 0; f < 25; f += 4)
+        for (int g = 0; g < 21; g += 4)
+          a[4][d][e][f][g] = 0;
+}
+
+/* { dg-final { scan-assembler-not {le\slr,\s\S*} } } */
-- 
2.20.1