forked from OSchip/llvm-project
[CodeGen] improve coverage for float (32-bit) type of NAN; NFC
Goes with D88238
This commit is contained in:
parent
bdc85292fb
commit
81921ebc43
|
@ -5,18 +5,28 @@
|
|||
|
||||
// Run a variety of targets to ensure there's no target-based difference.
|
||||
|
||||
// The builtin always produces a 64-bit (double).
|
||||
// An SNaN with no payload is formed by setting the bit after the
|
||||
// the quiet bit (MSB of the significand).
|
||||
|
||||
// CHECK: float 0x7FF8000000000000, float 0x7FF4000000000000
|
||||
// CHECK: double 0x7FF8000000000000, double 0x7FF4000000000000
|
||||
|
||||
float f[] = {
|
||||
__builtin_nanf(""),
|
||||
__builtin_nansf(""),
|
||||
};
|
||||
|
||||
|
||||
// Doubles are created and converted to floats.
|
||||
|
||||
// CHECK: float 0x7FF8000000000000, float 0x7FF4000000000000
|
||||
|
||||
float converted_to_float[] = {
|
||||
__builtin_nan(""),
|
||||
__builtin_nans(""),
|
||||
};
|
||||
|
||||
// CHECK: double 0x7FF8000000000000, double 0x7FF4000000000000
|
||||
|
||||
double d[] = {
|
||||
__builtin_nan(""),
|
||||
__builtin_nans(""),
|
||||
|
|
Loading…
Reference in New Issue