From: Kyrylo Tkachov <[email protected]>
The PR71636 fold turns
x & ((1U << b) - 1)
into
x & ~(~0U << b)
but only when the mask type is unsigned. Signed source and vector forms keep
the longer expression.
int f (int x, int b)
{
return x & ((1 << b) - 1);
}
aarch64 -O2 before:
f:
mov w2, 1
lsl w2, w2, w1
sub w2, w2, #1
and w0, w2, w0
ret
After:
f:
mov w2, -1
lsl w2, w2, w1
bic w0, w0, w2
ret
Build a signed mask in the corresponding unsigned type and convert it back.
This makes the all-ones shift defined and exposes the shorter form. The
preceding vector constant canonicalization lets the vector spelling reach the
same addition-of-minus-one rule.
The signed form is not valid when the source subtraction can trap or is
instrumented for overflow. It can also remove the signed shift-base check for
the top-bit count. Keep these cases. After GIMPLE lowering, an explicit
shift sanitizer check remains visible, so the fold is safe again.
Bootstrapped and tested on aarch64-none-linux-gnu and x86_64-pc-linux-gnu.
Ok for trunk?
Thanks,
Kyrill
gcc/ChangeLog:
* match.pd (x & ((1 << b) - 1)): Handle signed scalar and vector
types.
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/pr71636-signed-1.c: New test.
* gcc.dg/tree-ssa/pr71636-signed-vector-1.c: Likewise.
* gcc.dg/tree-ssa/pr71636-signed-trap-1.c: Likewise.
* gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c: Likewise.
* gcc.dg/tree-ssa/pr71636-signed-shift-ubsan-1.c: Likewise.
Signed-off-by: Kyrylo Tkachov <[email protected]>
---
gcc/match.pd | 14 ++++++++---
.../gcc.dg/tree-ssa/pr71636-signed-1.c | 24 +++++++++++++++++++
.../tree-ssa/pr71636-signed-shift-ubsan-1.c | 11 +++++++++
.../gcc.dg/tree-ssa/pr71636-signed-trap-1.c | 10 ++++++++
.../gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c | 10 ++++++++
.../gcc.dg/tree-ssa/pr71636-signed-vector-1.c | 24 +++++++++++++++++++
6 files changed, 90 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-1.c
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-shift-ubsan-1.c
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-trap-1.c
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c
create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-vector-1.c
diff --git a/gcc/match.pd b/gcc/match.pd
index 2c88e56b366..3fb22c12917 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -1577,11 +1577,19 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(convert @0)
(convert @1)))))
-/* PR71636: Transform x & ((1U << b) - 1) -> x & ~(~0U << b); */
+/* PR71636: Transform x & ((1U << b) - 1) -> x & ~(~0U << b). For signed
+ types, build the mask in the corresponding unsigned type, where shifting
+ all ones left is defined. Preserve signed overflow and shift checks. */
(simplify
(bit_and:c @0 (plus:s (lshift:s integer_onep @1) integer_minus_onep))
- (if (TYPE_UNSIGNED (type))
- (bit_and @0 (bit_not (lshift { build_all_ones_cst (type); } @1)))))
+ (with { tree etype = VECTOR_TYPE_P (type) ? TREE_TYPE (type) : type; }
+ (if (TYPE_UNSIGNED (etype)
+ || (!TYPE_OVERFLOW_TRAPS (etype)
+ && !TYPE_OVERFLOW_SANITIZED (etype)
+ && (GIMPLE || !sanitize_flags_p (SANITIZE_SHIFT_BASE))))
+ (with { tree utype = unsigned_type_for (type); }
+ (bit_and @0 (convert
+ (bit_not (lshift { build_all_ones_cst (utype); } @1))))))))
/* PR112533: Canonicalize boolean comparisons of masked pow2 bits into
xor-mask tests.
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-1.c
new file mode 100644
index 00000000000..9db533fdf64
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-1.c
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+
+int
+f_signed (int x, int b)
+{
+ return x & ((1 << b) - 1);
+}
+
+unsigned int
+f_unsigned (unsigned int x, int b)
+{
+ return x & ((1U << b) - 1U);
+}
+
+long
+f_long (long x, int b)
+{
+ return x & ((1L << b) - 1L);
+}
+
+/* { dg-final { scan-tree-dump-not "1 <<" "optimized" } } */
+/* { dg-final { scan-tree-dump-not " \\+ -1;" "optimized" } } */
+/* { dg-final { scan-tree-dump-times "= ~" 3 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-shift-ubsan-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-shift-ubsan-1.c
new file mode 100644
index 00000000000..3abee74264c
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-shift-ubsan-1.c
@@ -0,0 +1,11 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fsanitize=shift-base -fdump-tree-optimized" } */
+
+int
+f (int x, int b)
+{
+ return x & ((1 << b) - 1);
+}
+
+/* { dg-final { scan-tree-dump-times
"__builtin___ubsan_handle_shift_out_of_bounds" 1 "optimized" } } */
+/* { dg-final { scan-tree-dump-times "= ~" 1 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-trap-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-trap-1.c
new file mode 100644
index 00000000000..13a1a5b4cf2
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-trap-1.c
@@ -0,0 +1,10 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -ftrapv -fdump-tree-optimized" } */
+
+int
+f (int x, int b)
+{
+ return x & ((1 << b) - 1);
+}
+
+/* { dg-final { scan-tree-dump-times " \\+ -1;" 1 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c
new file mode 100644
index 00000000000..68c1b92f18b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-ubsan-1.c
@@ -0,0 +1,10 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fsanitize=signed-integer-overflow -fdump-tree-optimized"
} */
+
+int
+f (int x, int b)
+{
+ return x & ((1 << b) - 1);
+}
+
+/* { dg-final { scan-tree-dump-times "\\.UBSAN_CHECK_SUB" 1 "optimized" } } */
diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-vector-1.c
b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-vector-1.c
new file mode 100644
index 00000000000..51ba0b35334
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr71636-signed-vector-1.c
@@ -0,0 +1,24 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+/* { dg-require-effective-target vect_int } */
+/* { dg-require-effective-target vect_var_shift } */
+
+typedef int v4si __attribute__ ((vector_size (16)));
+typedef unsigned int v4ui __attribute__ ((vector_size (16)));
+
+v4si
+f_signed (v4si x, v4si b)
+{
+ v4si one = { 1, 1, 1, 1 };
+ return x & ((one << b) - one);
+}
+
+v4ui
+f_unsigned (v4ui x, v4ui b)
+{
+ v4ui one = { 1, 1, 1, 1 };
+ return x & ((one << b) - one);
+}
+
+/* { dg-final { scan-tree-dump-not "\\{ 1, 1, 1, 1 \\} <<" "optimized" } } */
+/* { dg-final { scan-tree-dump-times "= ~" 2 "optimized" } } */
--
2.50.1 (Apple Git-155)
