diff options
Diffstat (limited to 'test')
-rw-r--r-- | test/dt_arith.c | 46 |
1 files changed, 23 insertions, 23 deletions
diff --git a/test/dt_arith.c b/test/dt_arith.c index 9583b47..c90c524 100644 --- a/test/dt_arith.c +++ b/test/dt_arith.c @@ -87,7 +87,7 @@ static int skip_overflow_tests_g = 0; #endif /* OpenVMS doesn't have this feature. Make sure to disable it*/ -#ifdef VMS +#ifdef H5_VMS #undef HANDLE_SIGFPE #endif @@ -2679,7 +2679,7 @@ my_isnan(dtype_t type, void *val) retval = 1; } -#ifdef VMS +#ifdef H5_VMS /* For "float" and "double" on OpenVMS/Alpha, NaN is * actually a valid value of maximal value.*/ if(!retval) { @@ -2695,7 +2695,7 @@ my_isnan(dtype_t type, void *val) return 0; } } -#endif /*VMS*/ +#endif /*H5_VMS*/ return retval; } @@ -2726,7 +2726,7 @@ my_isinf(dtype_t type, int endian, unsigned char *val, size_t size, bits = (unsigned char*)calloc(1, size); -#ifdef VMS +#ifdef H5_VMS if(H5T_ORDER_VAX==endian) { for (i = 0; i < size; i += 4) { bits[i] = val[(size-2)-i]; @@ -2739,10 +2739,10 @@ my_isinf(dtype_t type, int endian, unsigned char *val, size_t size, for (i=0; i<size; i++) bits[size-(i+1)] = *(val + ENDIAN(size,i,endian)); } -#else /*VMS*/ +#else /*H5_VMS*/ for (i=0; i<size; i++) bits[size-(i+1)] = *(val + ENDIAN(size, i, endian)); -#endif /*VMS*/ +#endif /*H5_VMS*/ if((ret1=H5T_bit_find(bits, mpos, msize, H5T_BIT_LSB, 1))<0 && (ret2=H5T_bit_find(bits, epos, esize, H5T_BIT_LSB, 0))<0) @@ -2847,7 +2847,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) * The remainder of this function is executed only by the child if * HANDLE_SIGFPE is defined. */ -#ifndef VMS +#ifndef H5_VMS HDsignal(SIGFPE,fpe_handler); #endif @@ -2947,7 +2947,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) switch (run_test) { case TEST_NOOP: case TEST_NORMAL: -#ifdef VMS +#ifdef H5_VMS if(src_type == FLT_FLOAT) { INIT_FP_NORM(float, FLT_MAX, FLT_MIN, FLT_MAX_10_EXP, FLT_MIN_10_EXP, src_size, dst_size, buf, saved, nelmts); @@ -2973,7 +2973,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) #endif } else goto error; -#else /*VMS*/ +#else /*H5_VMS*/ if(src_type == FLT_FLOAT) { INIT_FP_NORM(float, FLT_MAX, FLT_MIN, FLT_MAX_10_EXP, FLT_MIN_10_EXP, src_size, dst_size, buf, saved, nelmts); @@ -2987,7 +2987,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) #endif } else goto error; -#endif /*VMS*/ +#endif /*H5_VMS*/ break; case TEST_DENORM: @@ -3063,7 +3063,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) hw = (unsigned char*)&hw_f; underflow = HDfabs(*((double*)aligned)) < FLT_MIN; overflow = HDfabs(*((double*)aligned)) > FLT_MAX; -#ifdef VMS +#ifdef H5_VMS maximal = HDfabs(*((double*)aligned)) == FLT_MAX; #endif } else if (FLT_DOUBLE==dst_type) { @@ -3083,7 +3083,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) hw = (unsigned char*)&hw_f; underflow = HDfabsl(*((long double*)aligned)) < FLT_MIN; overflow = HDfabsl(*((long double*)aligned)) > FLT_MAX; -#ifdef VMS +#ifdef H5_VMS maximal = HDfabs(*((long double*)aligned)) == FLT_MAX; #endif } else if (FLT_DOUBLE==dst_type) { @@ -3091,7 +3091,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) hw = (unsigned char*)&hw_d; underflow = HDfabsl(*((long double*)aligned)) < DBL_MIN; overflow = HDfabsl(*((long double*)aligned)) > DBL_MAX; -#ifdef VMS +#ifdef H5_VMS maximal = HDfabs(*((long double*)aligned)) == DBL_MAX; #endif } else { @@ -3127,7 +3127,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) if (k==dst_size) continue; /*no error*/ -#ifdef VMS +#ifdef H5_VMS /* For "float" and "double" on OpenVMS/Alpha, NaN is * a valid value of maximal value.*/ if (FLT_FLOAT==src_type && @@ -3137,7 +3137,7 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) my_isnan(src_type, saved+j*sizeof(double))) { continue; } -#endif /*VMS*/ +#endif /*H5_VMS*/ /* * Assume same if both results are NaN. There are many NaN bit @@ -3195,11 +3195,11 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) if (overflow && my_isinf(dst_type, dendian, buf+j*sizeof(float), dst_size, dst_mpos, dst_msize, dst_epos, dst_esize)) continue; /* all overflowed, no error */ -#ifdef VMS +#ifdef H5_VMS if (maximal && my_isinf(dst_type, dendian, buf+j*sizeof(float), dst_size, dst_mpos, dst_msize, dst_epos, dst_esize)) continue; /* maximal value, no error */ -#endif /*VMS*/ +#endif /*H5_VMS*/ check_mant[0] = HDfrexpf(x, check_expo+0); check_mant[1] = HDfrexpf(hw_f, check_expo+1); } else if (FLT_DOUBLE==dst_type) { @@ -3211,11 +3211,11 @@ test_conv_flt_1 (const char *name, int run_test, hid_t src, hid_t dst) if (overflow && my_isinf(dst_type, dendian, buf+j*sizeof(double), dst_size, dst_mpos, dst_msize, dst_epos, dst_esize)) continue; /* all overflowed, no error */ -#ifdef VMS +#ifdef H5_VMS if (maximal && my_isinf(dst_type, dendian, buf+j*sizeof(double), dst_size, dst_mpos, dst_msize, dst_epos, dst_esize)) continue; /* maximal value, no error */ -#endif /*VMS*/ +#endif /*H5_VMS*/ check_mant[0] = HDfrexp(x, check_expo+0); check_mant[1] = HDfrexp(hw_d, check_expo+1); #if H5_SIZEOF_LONG_DOUBLE !=0 && (H5_SIZEOF_LONG_DOUBLE!=H5_SIZEOF_DOUBLE) @@ -5009,12 +5009,12 @@ run_fp_int_conv(const char *name) run_test = FALSE; #endif -#ifdef VMS +#ifdef H5_VMS run_test = TRUE; #endif if(run_test) { -#ifdef VMS +#ifdef H5_VMS test_values = TEST_NORMAL; { #else @@ -5025,7 +5025,7 @@ run_fp_int_conv(const char *name) test_values = TEST_DENORM; else test_values = TEST_SPECIAL; -#endif /*VMS*/ +#endif /*H5_VMS*/ nerrors += test_conv_int_fp(name, test_values, H5T_NATIVE_FLOAT, H5T_NATIVE_SCHAR); nerrors += test_conv_int_fp(name, test_values, H5T_NATIVE_DOUBLE, H5T_NATIVE_SCHAR); @@ -5217,7 +5217,7 @@ main(void) * for user-defined integer types */ nerrors += test_derived_integer(); -#ifndef VMS +#ifndef H5_VMS /* Does floating point overflow generate a SIGFPE? */ generates_sigfpe(); #endif |