[flang] Improve warning message

[klausler]

Summary

When a floating-point exception occurs while folding an exponentiation
operation with a host library at compilation time, the warning message
should not say that it was an "intrinsic function".

[klausler]

Correct the test results for Evaluate/folding04.f90.

[jeanPerier]

Evaluate/folding04.f90 still seems to fail after your patch update. It seems llvm buildbots run yield "overflow" errors and not "division by zero" errors in clog((0., 0.)) and others. Could it be a difference in what the underlying runtime math library is signaling when folding ?

[klausler]

In D137030#3895792, @jeanPerier wrote:

Evaluate/folding04.f90 still seems to fail after your patch update. It seems llvm buildbots run yield "overflow" errors and not "division by zero" errors in clog((0., 0.)) and others. Could it be a difference in what the underlying runtime math library is signaling when folding ?

I can make the test specific to x86, or I can make the error messages more generic ("floating-point exception") just to make the test pass.

[klausler]

Adjust test for portability.

[jeanPerier]

LGTM

flang/lib/Evaluate/host.cpp

if (flags_.empty()) {
     }
     if (errnoCapture == ERANGE) {
       // can't distinguish over/underflow from errno
       flags_.set(RealFlag::Overflow);
     }
   }
 
   if (!flags_.empty()) {
-    RealFlagWarnings(context, flags_, "intrinsic function");
+    RealFlagWarnings(
+        context, flags_, "evaluation of intrinsic function or operation");
   }
   errno = 0;
   if (fesetenv(&originalFenv_) != 0) {
     std::fprintf(
         stderr, "fesetenv() failed: %s\n", llvm::sys::StrError(errno).c_str());
     common::die(
         "Folding with host runtime: fesetenv() failed while restoring fenv: %s",
         llvm::sys::StrError(errno).c_str());
   }
 #if __x86_64__
   _mm_setcsr(originalMxcsr);
 #endif
 
   errno = 0;
 }
 } // namespace Fortran::evaluate::host

flang/test/Evaluate/folding04.f90

module real_tests
   real(4), parameter :: r4_nmax = -3.4028235E38
   !WARN: warning: invalid argument on division
   real(4), parameter :: r4_nan = 0._4/0._4
   !WARN: warning: division by zero
   real(4), parameter :: r4_pinf = 1._4/0._4
   !WARN: warning: division by zero
   real(4), parameter :: r4_ninf = -1._4/0._4
 
-  !WARN: warning: invalid argument on intrinsic function
+  !WARN: warning: invalid argument on evaluation of intrinsic function or operation
   real(4), parameter :: nan_r4_acos1 = acos(1.1)
   TEST_ISNAN(nan_r4_acos1)
-  !WARN: warning: invalid argument on intrinsic function
+  !WARN: warning: invalid argument on evaluation of intrinsic function or operation
   real(4), parameter :: nan_r4_acos2 = acos(r4_pmax)
   TEST_ISNAN(nan_r4_acos2)
-  !WARN: warning: invalid argument on intrinsic function
+  !WARN: warning: invalid argument on evaluation of intrinsic function or operation
   real(4), parameter :: nan_r4_acos3 = acos(r4_nmax)
   TEST_ISNAN(nan_r4_acos3)
-  !WARN: warning: invalid argument on intrinsic function
+  !WARN: warning: invalid argument on evaluation of intrinsic function or operation
   real(4), parameter :: nan_r4_acos4 = acos(r4_ninf)
   TEST_ISNAN(nan_r4_acos4)
-  !WARN: warning: invalid argument on intrinsic function
+  !WARN: warning: invalid argument on evaluation of intrinsic function or operation
   real(4), parameter :: nan_r4_acos5 = acos(r4_pinf)
   TEST_ISNAN(nan_r4_acos5)
   !WARN: warning: second argument to MOD must not be zero
   real(4), parameter :: nan_r4_mod = mod(3.5, 0.)
   TEST_ISNAN(nan_r4_mod)
-
-  !WARN: warning: overflow on intrinsic function
+  !WARN: warning: overflow on evaluation of intrinsic function or operation
   logical, parameter :: test_exp_overflow = exp(256._4).EQ.r4_pinf
 end module
 
 module parentheses
   ! Test parentheses in folding (they are kept around constants to keep the
   ! distinction between variable and expressions and require special care).
   real(4), parameter :: x_nop = 0.1_4
   real(4), parameter :: x_p = (x_nop)