On Tue, Jul 25, 2017 at 06:25:49AM -0500, Segher Boessenkool wrote: > On Mon, Jul 24, 2017 at 04:06:39PM -0600, Jeff Law wrote: > > > 2017-07-24 Segher Boessenkool <seg...@kernel.crashing.org> > > > > > > gcc/testsuite/ > > > PR rtl-optimization/81423 > > > * gcc.c-torture/execute/pr81423.c: New testcase. > > I think int32plus just indicates ints are at least 32 bits. But a long > > or long long could still be just 32 bits. so int32plus && long_neq_int, > > to ensure that long/long long are 64 bits? > > Well, long long is required to be 64 bits or more by the C standard. > But some GCC targets do not follow that, with certain options at least. > > It looks like that test actually requires long long to be *exactly* > 64 bits. I'll modify the test to test for that.
So I came up with the following. Is this okay for trunk? (Tested on powerpc64-linux and x86_64-linux, both with both -m32 and -m64, and tested it does fail on x86 without the patches to fix the bug). Segher diff --git a/gcc/testsuite/gcc.c-torture/execute/pr81423.c b/gcc/testsuite/gcc.c-torture/execute/pr81423.c new file mode 100644 index 0000000..731aa8f --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/pr81423.c @@ -0,0 +1,36 @@ +extern void abort (void); + +unsigned long long int ll = 0; +unsigned long long int ull1 = 1ULL; +unsigned long long int ull2 = 12008284144813806346ULL; +unsigned long long int ull3; + +unsigned long long int __attribute__ ((noinline)) +foo (void) +{ + ll = -5597998501375493990LL; + + ll = (5677365550390624949L - ll) - (ull1 > 0); + unsigned long long int ull3; + ull3 = (unsigned int) + (2067854353L << + (((ll + -2129105131L) ^ 10280750144413668236ULL) - + 10280750143997242009ULL)) >> ((2873442921854271231ULL | ull2) + - 12098357307243495419ULL); + + return ull3; +} + +int +main (void) +{ + /* We need a long long of exactly 64 bits for this test. */ + ll--; + if (ll != 0xffffffffffffffffULL) + return 0; + + ull3 = foo (); + if (ull3 != 3998784) + abort (); + return 0; +}