procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fold __builtin_signbit to nonzero instead of 1.

Browse source 
After Joseph's comment wrt __builtin_signbit, I have been unable to
convince myself that arbitrarily folding __builtin_signbit () of a
negative number to 1 is correct.

For example, on the true side of x < -5.0 we know the sign bit is set,
but on the false side, we know nothing because X may be a NAN.  I
don't want to put ourselves in a position where the same call to
__builtin_signbit can return two different things (1 or negative
whatever), so I'm going to return nonzero which is correct across the
board until someone can convince me otherwise.

Setting the range to nonzero still allows us to fold conditionals
checking __fold_builtin, while not actually propagating a 1.  Zero
propagation still works.

That being said, I still think we should be folding
__builtin_signbit's of negative numbers to whatever the hardware
returns, instead of 1/0 like what the front ends do.

gcc/ChangeLog:

	* gimple-range-fold.cc
	(fold_using_range::range_of_builtin_int_call): Fold a set signbit
	in __builtin_signbit to nonzero.

gcc/testsuite/ChangeLog:

	* gcc.dg/tree-ssa/vrp-float-signbit-2.c: New test.
This commit is contained in:
Aldy Hernandez 2022-09-05 07:36:25 +02:00
parent 810d981524
commit 1de9aa4552
2 changed files with 25 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
gcc/gimple-range-fold.cc
View file

@ -1032,10 +1032,7 @@ fold_using_range::range_of_builtin_int_call (irange &r, gcall *call,
if (tmp.get_signbit ().varying_p ())
return false;
if (tmp.get_signbit ().yes_p ())
{
tree one = build_one_cst (type);
r.set (one, one);
}
r.set_nonzero (type);
else
r.set_zero (type);
return true;

24
gcc/testsuite/gcc.dg/tree-ssa/vrp-float-signbit-2.c Normal file
View file

@ -0,0 +1,24 @@
// { dg-do compile }
// { dg-options "-O2 -fdump-tree-evrp" }
// Test that the only thing we know about the signbit about negative number is
// that it's not 0.
void link_error ();
int num;
void func(float x)
{
if (x < -5.0)
{
num = __builtin_signbit (x);
// We may not know the exact signbit, but we know it's not 0.
if (!__builtin_signbit (x))
link_error ();
}
}
// { dg-final { scan-tree-dump-not "num = \[-0-9\];" "evrp" } }
// { dg-final { scan-tree-dump-not "link_error" "evrp" } }