https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121573
Benjamin Schulz <schulz.benjamin at googlemail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|INVALID |---
Status|RESOLVED |UNCONFIRMED
--- Comment #3 from Benjamin Schulz <schulz.benjamin at googlemail dot com> ---
I want to note is that this is potentially dangerous if a source which looks
standars compilant, gets such different outputs with ngc++ and gcc
nvc++ output:
on host: a=42 b=3.140000
on nvidia
on parallel region, should be 42; dev_var.a=0
on parallel region, should be 3.14: dev_var.b=0.000000
Device-resident struct copied back, should be 42 and 3.14: a=0 b=0.000000
gcc output:
on host: a=42 b=3.140000
on nvidia
on parallel region, should be 42; dev_var.a=42
on parallel region, should be 3.14: dev_var.b=3.140000
Device-resident struct copied back, should be 42 and 3.14: a=42 b=3.140000
Both compilers create a device variable and a host variable for something that
should, according to the spec, reside on device only.
And they understand the initialization differently. GCC initializes on host and
device, and initializes it with these values, while nvc++ forgets the
initialization on the device...
I do not think that this is resolved invalid. since some problem is there if
one declares the variable above the pragma...
