[Bug target/81663] New: nvptx and fstack-check

[vries at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81663

            Bug ID: 81663
           Summary: nvptx and fstack-check
           Product: gcc
           Version: 8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vries at gcc dot gnu.org
  Target Milestone: ---

For nvptx, we run into this failure (specifically, a failure to produce a
warning):
...
Executing on host: nvptx-none-gcc
src/gcc-mainline/gcc/testsuite/gcc.dg/pr70017.c  -fno-diagnostics-show-caret
-fdiagnostics-color=never   -fstack-check=generic -S    -o pr70017.s   
(timeout = 300)
FAIL: gcc.dg/pr70017.c  (test for warnings, line 21)
PASS: gcc.dg/pr70017.c (test for excess errors)
...

The test-case looks like this:
...
/* { dg-do compile } */
/* { dg-require-stack-check "generic" } */
/* { dg-options "-fstack-check=generic" } */

/* Check that the expected warning is issued for large frames.  */

#define ONE(s) char a##s[32];
#define TEN(s) ONE(s##0) ONE(s##1) ONE(s##2) ONE(s##3) ONE(s##4) \
               ONE(s##5) ONE(s##6) ONE(s##7) ONE(s##8) ONE(s##9)
#define HUNDRED(s) TEN(s##0) TEN(s##1) TEN(s##2) TEN(s##3) TEN(s##4) \
                   TEN(s##5) TEN(s##6) TEN(s##7) TEN(s##8) TEN(s##9)

void foo(void)
{
  HUNDRED(a)
  HUNDRED(b)
#undef ONE
#define ONE(s) a##s[0] = 0;
  HUNDRED(a)
  HUNDRED(b)
} /* { dg-warning "frame size too large for reliable stack checking" "" {
target { ! alpha*-*-* } } } */
...

The ptx specification abstracts away from the stack, so I'm not sure
fstack-check makes sense for the target (although it might for msoft-stack).

It's probably a good idea to error out in the target, which should make all
dg-require-stack-check fail.