https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107041
Bug ID: 107041
Summary: [13 Regression] C '-Wenum-int-mismatch' diagnostic for
OpenACC 'acc_on_device'
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Keywords: openacc
Severity: normal
Priority: P3
Component: libgomp
Assignee: unassigned at gcc dot gnu.org
Reporter: tschwinge at gcc dot gnu.org
CC: burnus at gcc dot gnu.org, jakub at gcc dot gnu.org,
mpolacek at gcc dot gnu.org, vries at gcc dot gnu.org
Target Milestone: ---
Clean for C++:
$ printf '#include <openacc.h>\nint foo(){ return
acc_on_device(acc_device_host); }\n' | build-gcc/gcc/xgcc -Bbuild-gcc/gcc
-Isource-gcc/libgomp -x c++ - -fopenacc -S -Wall
..., but not anymore for C:
$ printf '#include <openacc.h>\nint foo(){ return
acc_on_device(acc_device_host); }\n' | build-gcc/gcc/xgcc -Bbuild-gcc/gcc
-Isource-gcc/libgomp -x c - -fopenacc -S -Wall
In file included from <stdin>:1:
source-gcc/libgomp/openacc.h:103:5: warning: conflicting types for
‘acc_on_device’ due to enum/integer mismatch; have ‘int(acc_device_t)’
[-Wenum-int-mismatch]
103 | int acc_on_device (acc_device_t __arg) __GOACC_NOTHROW;
| ^~~~~~~~~~~~~
That's commit r13-627-g7da9a089608b0ca09683332ce014fb6184842724 "c: Implement
new -Wenum-int-mismatch warning [PR105131]".
This causes excess errors (only!) for
'libgomp.oacc-c/../libgomp.oacc-c-c++-common/host_data-1.c' (when its
prerequisites are met); that's one of the few libgomp test cases where '-Wall'
is in effect. (I'd "complained" about lack of '-Wall' testing before.)
OpenACC 'acc_on_device' is implemented as follows:
libgomp/openacc.h-#ifdef __cplusplus
libgomp/openacc.h:int acc_on_device (int __arg) __GOACC_NOTHROW;
libgomp/openacc.h-#else
libgomp/openacc.h:int acc_on_device (acc_device_t __arg) __GOACC_NOTHROW;
libgomp/openacc.h-#endif
libgomp/openacc.h-#ifdef __cplusplus
libgomp/openacc.h-[...]
libgomp/openacc.h-
libgomp/openacc.h-/* Forwarding function with correctly typed arg. */
libgomp/openacc.h-
libgomp/openacc.h-#pragma acc routine seq
libgomp/openacc.h:inline int acc_on_device (acc_device_t __arg)
__GOACC_NOTHROW
libgomp/openacc.h-{
libgomp/openacc.h: return acc_on_device ((int) __arg);
libgomp/openacc.h-}
libgomp/openacc.h-#endif
gcc/omp-builtins.def:DEF_GOACC_BUILTIN_COMPILER (BUILT_IN_ACC_ON_DEVICE,
"acc_on_device",
gcc/omp-builtins.def- BT_FN_INT_INT,
ATTR_CONST_NOTHROW_LEAF_LIST)
libgomp/oacc-init.c:/* For -O and higher, the compiler always attempts to
expand acc_on_device, but
libgomp/oacc-init.c- if the user disables the builtin, or calls it via a
pointer, we'll need this
libgomp/oacc-init.c- version.
libgomp/oacc-init.c-
libgomp/oacc-init.c- Compile this with optimization, so that the compiler
expands
libgomp/oacc-init.c- this, rather than generating infinitely recursive
code.
libgomp/oacc-init.c-
libgomp/oacc-init.c: The function just forwards its argument to
__builtin_acc_on_device. It does
libgomp/oacc-init.c- not verify that the argument is a valid acc_device_t
enumeration value. */
libgomp/oacc-init.c-
libgomp/oacc-init.c-int __attribute__ ((__optimize__ ("O2")))
libgomp/oacc-init.c:acc_on_device (acc_device_t dev)
libgomp/oacc-init.c-{
libgomp/oacc-init.c: return __builtin_acc_on_device (dev);
libgomp/oacc-init.c-}
libgomp/oacc-init.c-
libgomp/oacc-init.c:ialias (acc_on_device)
If I remember correctly, the reason for the "strange" prototyping/forwarding is
that GCC builtins don't allow for 'enum' arguments? (I may be misremembering.)
---
Maybe a fix for this would fall out of Tom's patch for PR82391 "[openacc]
acc_on_device not folded at -O0"? (I've paged out all state of that one,
unfortunately.)