Opened 16 years ago
Closed 15 years ago
#150 closed enhancement (fixed)
PGF and NAG behaviour in ropp_io_thin.f90
| Reported by: | Huw Lewis | Owned by: | Huw Lewis |
|---|---|---|---|
| Priority: | normal | Milestone: | 4.0 |
| Component: | ropp_io | Version: | 2.0beta |
| Keywords: | thinner, compiler bug | Cc: | dave.offiler@… |
Description (last modified by )
Bugs found with PGF and NAG compilers in ropp_io_thin.f90.
- these compilers require an explicit format statement in conversion from string to real for determining number of thinned levels and thinned heights (see 'FIXME' comments in code). Note previously found problems in using CASE with these compilers in this routine.
The original code works fine with all other compilers, and the bug cannot be replicated in simple test programs run with nag.
Need to investigate what in particular is distinct about the ROPP code (cf test routines) which causes this performance, and find more satisfactory long-term solutions, i.e. input thinning information need not be of particular format.
Change history (3)
comment:1 by , 16 years ago
| Type: | defect → enhancement |
|---|
comment:2 by , 15 years ago
| Description: | modified (diff) |
|---|---|
| Milestone: | 3.0 → 4.0 |
comment:3 by , 15 years ago
| Resolution: | → fixed |
|---|---|
| Status: | new → closed |
See [2257].
Updated ropp_io_thin to tidy up compiler-specific 'quirks'.
Use of CASE statements was complicated by variations in final terminating character in the method string. It is more appropriate to use 'INDEX()' function for determining the different thinner variations in any case.
The required formatting fix applied at ROPP-2 should be maintained in future versions.
Fix tested in Test Folder (MT-IO-03) and ./t_roppthin routine with PGF and NAG passing successfully.
Ticket closed as fixed.
Minor issue, not addressed in ROPP-3. Move milestone to ROPP-4 when more opportunity time to review thinner code in more detail.
Ticket remains open.