Hi!
What happens on this testcase is with the out of bounds rotate we get:
Trying 13 -> 16:
13: r129:SI=r132:DI#0<-<0x20
REG_DEAD r132:DI
16: r123:DI=r129:SI<0
REG_DEAD r129:SI
Successfully matched this instruction:
(set (reg/v:DI 123 [ <retval> ])
(const_int 0 [0]))
during combine. So, perhaps we could also change simplify-rtx.c to punt
if it is out of bounds rather than trying to optimize anything.
Or, but probably GCC11 material, if we decide that ROTATE/ROTATERT doesn't
have out of bounds counts or introduce targetm.rotate_truncation_mask,
we should truncate the argument instead of punting.
Punting is better for backports though.
Bootstrapped/regtested on powerpc64{,le}-linux, preapproved by Segher in the
PR, committed to trunk.
2020-01-30 Jakub Jelinek <[email protected]>
PR middle-end/93505
* combine.c (simplify_comparison) <case ROTATE>: Punt on out of range
rotate counts.
* gcc.c-torture/compile/pr93505.c: New test.
--- gcc/combine.c.jj 2020-01-12 11:54:36.222416290 +0100
+++ gcc/combine.c 2020-01-30 15:08:13.607205111 +0100
@@ -12410,7 +12410,8 @@ simplify_comparison (enum rtx_code code,
bit. This will be converted into a ZERO_EXTRACT. */
if (const_op == 0 && sign_bit_comparison_p
&& CONST_INT_P (XEXP (op0, 1))
- && mode_width <= HOST_BITS_PER_WIDE_INT)
+ && mode_width <= HOST_BITS_PER_WIDE_INT
+ && UINTVAL (XEXP (op0, 1)) < mode_width)
{
op0 = simplify_and_const_int (NULL_RTX, mode, XEXP (op0, 0),
(HOST_WIDE_INT_1U
--- gcc/testsuite/gcc.c-torture/compile/pr93505.c.jj 2020-01-30
15:23:19.919634317 +0100
+++ gcc/testsuite/gcc.c-torture/compile/pr93505.c 2020-01-30
15:22:59.537939346 +0100
@@ -0,0 +1,15 @@
+/* PR middle-end/93505 */
+
+unsigned a;
+
+unsigned
+foo (unsigned x)
+{
+ unsigned int y = 32 - __builtin_bswap64 (-a);
+ /* This would be UB (x << 32) at runtime. Ensure we don't
+ invoke UB in the compiler because of that (visible with
+ bootstrap-ubsan). */
+ x = x << y | x >> (-y & 31);
+ x >>= 31;
+ return x;
+}
Jakub
