This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH][AArch64][GCC 6] PR target/79041: Correct -mpc-relative-literal-loads logic in aarch64_classify_symbol
- From: Kyrill Tkachov <kyrylo dot tkachov at foss dot arm dot com>
- To: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Cc: Richard Earnshaw <Richard dot Earnshaw at arm dot com>, James Greenhalgh <james dot greenhalgh at arm dot com>, Marcus Shawcroft <marcus dot shawcroft at arm dot com>
- Date: Wed, 11 Jan 2017 16:32:45 +0000
- Subject: [PATCH][AArch64][GCC 6] PR target/79041: Correct -mpc-relative-literal-loads logic in aarch64_classify_symbol
- Authentication-results: sourceware.org; auth=none
Hi all,
In this PR we generated ADRP/ADD instructions with :lo12: relocations on symbols even though -mpc-relative-literal-loads
is used. This is due to the confusing double-negative logic of the
nopcrelative_literal_loads aarch64 variable and its relation to the aarch64_nopcrelative_literal_loads global variable
in the GCC 6 branch.
Wilco fixed this on trunk as part of a larger patch (r237607) and parts of that patch were backported, but other parts weren't and
that patch now doesn't apply cleanly to the branch.
The actual bug here is that aarch64_classify_symbol uses nopcrelative_literal_loads instead of the correct aarch64_nopcrelative_literal_loads.
nopcrelative_literal_loads gets set to 1 if the user specifies -mpc-relative-literal-loads(!) whereas aarch64_nopcrelative_literal_loads gets
set to false, so that is the variable we want to check.
So this is the minimal patch that fixes this.
Bootstrapped and tested on aarch64-none-linux-gnu on the GCC 6 branch.
Ok for the branch?
Thanks,
Kyrill
2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
PR target/79041
* config/aarch64/aarch64.c (aarch64_classify_symbol): Use
aarch64_nopcrelative_literal_loads instead of
nopcrelative_literal_loads.
2017-01-11 Kyrylo Tkachov <kyrylo.tkachov@arm.com>
PR target/79041
* gcc.target/aarch64/pr79041.c: New test.
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index 83dbd57..fa61289 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -9324,7 +9324,7 @@ aarch64_classify_symbol (rtx x, rtx offset)
/* This is alright even in PIC code as the constant
pool reference is always PC relative and within
the same translation unit. */
- if (nopcrelative_literal_loads
+ if (aarch64_nopcrelative_literal_loads
&& CONSTANT_POOL_ADDRESS_P (x))
return SYMBOL_SMALL_ABSOLUTE;
else
diff --git a/gcc/testsuite/gcc.target/aarch64/pr79041.c b/gcc/testsuite/gcc.target/aarch64/pr79041.c
new file mode 100644
index 0000000..a23b1ae
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/pr79041.c
@@ -0,0 +1,26 @@
+/* PR target/79041. Check that we don't generate the LO12 relocations
+ for -mpc-relative-literal-loads. */
+/* { dg-do compile } */
+/* { dg-options "-O2 -mcmodel=large -mpc-relative-literal-loads" } */
+
+extern int strcmp (const char *, const char *);
+extern char *strcpy (char *, const char *);
+
+static struct
+{
+ char *b;
+ char *c;
+} d[] = {
+ {"0", "000000000000000"}, {"1", "111111111111111"},
+};
+
+void
+e (const char *b, char *c)
+{
+ int i;
+ for (i = 0; i < 1; ++i)
+ if (!strcmp (d[i].b, b))
+ strcpy (c, d[i].c);
+}
+
+/* { dg-final { scan-assembler-not ":lo12:" } } */