PGI Release 2012 Technical Problem Reports
What problems have been fixed in PGI 2012?
Each release of the PGI compilers brings a number of technical corrections to problems discovered by PGI and by users. Submit problems found using our online technical support request form. PGI refers to all such reports as Technical Problem Reports , or TPRs.
The following table is a list of problems that have been fixed in each release. Descriptions as provided may be incomplete due to proprietary nature of the sample code. In such cases simple replacement examples may not be available.
Note: ICE stands for Internal Compiler Error.
| TPR | Version | Affected Component |
Symptoms / Actions |
|---|---|---|---|
| 17192 | 12.10 | Fortran |
pgfortran reports the error "Non-constant expression where constant expression required" when using an ISO_C_BINDING kind parameter
Allow a USE-associated parameter to be used as a kind value in a FUNCTION declaration appearing before its USE statement |
| 18403 | 12.10 | Fortran |
pgfortran reports the error "Derived type has not been declared"
Corrected the accessibility of a TYPE definition with the BIND attribute which appears after a PRIVATE statement referring to other entities |
| 18420 | 12.10 | Fortran |
Add support for makedirqq to pgfortran on Windows
Added the routine makedirqq to the pgfortran dflib module |
| 18687 | 12.10 | C/C++ |
pgCC -g -O creates a debug entry which references a nonexistent label
Corrected the debug support for lexical blocks at higher optimization levels |
| 18778 | 12.10 | C/C++ |
pgCC fails to lookup member function in template
Corrected g++ compatibility of the lookup for dependent function calls in dependent base classes |
| 18779 | 12.10 | CUDA Fortran |
Correct the CUDA Fortran documentation for constant data
Clarified the definition of constant data in CUDA Fortran |
| 18801 | 12.10 | 64-bit Fortran |
OpenMP code using allocatable derived types fails
Corrected the handling of shared derived type variables containing allocatable array members in parallel regions |
| 18825 | 12.10 | MacOS |
On OSX, setting MANPATH is unnecessary for accessing the PGI man pages
Removed suggesting setting MANPATH from the OSX install instructions |
| 18879 | 12.10 | Fortran |
pgfortran example containing adjustable length character declarations fail to compile
Corrected the processing of length parameters in character declarations |
| 18880 | 12.10 | 64-bit C/C++ |
pgcc example using m128 sse intrinsics fails to compile
Corrected the handling of __m128 dummy arguments |
| 18913 | 12.10 | PGI Accelerator |
Incorrect results with OpenACC example containing nested reduction loops
Corrected the handling of a reduction scalar appearing in nested loops |
| 18919 | 12.10 | Fortran |
pgfortran -Mipa=inline -mp reports the error "Internal compiler error. ipa_ast:unexpected ast type 100"
Corrected inlining subroutines containing OpenMP critical sections |
| 18921 | 12.10 | Fortran |
Fortran MODULE containing a generic interface compiled -g causes an undefined reference when placed in a shared object
Corrected generating the debug interface information for user-defined generics |
| 18928 | 12.10 | Fortran |
CUDA fortran examples fails to compile
Corrected the handling of the builtin data objects and user common blocks |
| 18936 | 12.10 | Fortran |
Incorrect results when using RESHAPE
Corrected the optimization of reshape when its source argument is an assumed-shape array of rank 2 or greater |
| 18937 | 12.10 | Fortran |
Fortran example using the PACK intrinsic fails to execute
Corrected the handling of the PACK intrinsics when its arguments contain sections of allocatable arrays |
| 18942 | 12.10 | Fortran |
OpenACC examples fails to compile
Corrected the handling of loops where the loop count cannot be computed |
| 18946 | 12.10 | Fortran |
Add emulation support for threadfence_system and threadfence_block
Added support for threadfence_system and threadfence support to the compiler and run-time |
| 17969 | 12.9 | Fortran |
pgfortran reports an error for a zero-size array parameters
Corrected the handling of an array initializer composed of a zero-length iterator |
| 18364 | 12.9 | C/C++ | pgcc reports the error, "Sizeof dimensionless array"
Corrected the handling of VLA declarations in abstract array types with more than one dimension |
| 18752 | 12.9 | PGI Acclerator Fortran | "failed to translate accelerator region Kernel argument list is > 256 bytes"
Improved the handling of large kernels with a large number of input values |
| 18785 | 12.9 | Fortran |
pgfortran reports the error, "operation illegal to external procedure kmod", when compiling -i8
Corrected the processing of a generic appearing in the INTRINSIC statement and as a variable |
| 18848 | 12.9 | PGI Acclerator Fortran |
Incorrect answers due to an optimization error
Corrected copy propagating the right-hand sides of multiple assignments |
| 18857 | 12.9 | PGI Accelerator C |
Error reports, "Internal compiler error. max_ili_ili argument error"
Corrected the handling of data pragmas with a large number of variables |
| 18858 | 12.9 | PGI Accelerator |
Confusing duplicate copy messages
Corrected the generation of the copyin messages |
| 18873 | 12.9 | Fortran |
pgfortran reports the error, "Internal compiler error. select_rtemp: bad ili"
Corrected the processing of functions appearing as arguments when the corresponding argument is declared CLASS |
| 18874 | 12.9 | Fortran |
f2003 code containing a type-bound operator (*) and a type-bound assignment (=) fails to execute
Corrected the handling of an assignment which has a TYPE(scalar) input but a CLASS(scalar) output |
| 18880 | 12.9 | 64-bit C/C++ |
Code using m128 sse intrinsics fails to execute
Corrected the referencing misaligned 128-bit variables |
| 18882 | 12.9 | Fortran |
pgfortran reports the error, "Illegal use of symbol mpi_wtime - not public entity of module"
Corrected the handling of accessibility in nested modules |
| 18883 | 12.9 | CUDA Fortran |
Add support for more Fortran elemental functions
Added support for the bit intrinsics, btest, iand, ibclr, ibset, ieor, iand |
| 18885 | 12.9 | 64-bit C/C++ |
Large c++ code fails to compile on Windows -g
Corrected the generation of debug information for function parameter entries |
| 18906 | 12.9 | PGI Acclerator Fortran |
User code fails to compile
Corrected the processing of an acc kernel containing a large number of private variables |
| 18907 | 12.9 | C/C++ |
pgcc reports the error, "Illegal type conversion required", for a VLA
Corrected processing of a VLA declaration when its dimension specifier is a const int variable |
| 18913 | 12.9 | PGI Accelerator |
Incorrect answers for code containing nested reduction loops
Corrected handling of a loop nest which uses the same reduction variable |
| 18915 | 12.9 | Windows Fortran |
Driver does not pick up new MSMPI locations
Corrected the driver to allow the new install location of the MSMPI components |
| 18917 | 12.9 | 32-bit Fortran |
pgfortran reports the error, "Internal compiler error: check_member: member arrived with wrong derived type"
Corrected the handling of a module which has the same name as a member of use-associated derived type |
| 17192 | 12.8 | Fortran | pgfortran reports the error, "Non-constant expression where constant expression required" Corrected the use of a USE-associated PARAMETER which appears in a FUNCTION statement |
| 17702 | 12.8 | Fortran | pgfortran reports the error, "Could not resolve generic procedure" Corrected the visibility of a generic interface which is extended in nested modules |
| 17943 | 12.8 | Fortran | OpenMP task's copy of shared variable not updated Corrected the handling of shared variables within a task which are private to an enclosing parallel region |
| 18475 | 12.8 | Fortran | pgfortran example containing CHARACTER VALUE arguments fails Corrected the length handling of CHARACTER dummy arguments which have the VALUE attribute |
| 18714 | 12.8 | Fortran | pgfortran reports the error "Cannot invoke Abstract Interface get_name" Corrected the handling of a procedure member named the same as a module-contained subprogram |
| 18746 | 12.8 | Fortran | pgfortran reports the error, "Internal compiler error. ipa_ast:unexpected ast" Corrected the processing of OpenMP constructs when -Mipa=fast is used |
| 18751 | 12.8 | Fortran | MPI pgfortran program fails when one routine is compiled -fast Corrected an optimization error wih a derived type containing many array members |
| 18785 | 12.8 | Fortran | pgfortran reports the error, "operation illegal to external procedure kmod", when compiling -i8 Corrected naming conflicts with various integer intrinsics when -i8 is used |
| 18788 | 12.8 | Fortran | pgfortran reports the error, "Non-constant expression where constant expression required" Corrected the handling of use-renamed parameters |
| 18819 | 12.8 | PVF | PVF project added to Team Foundation Server (TFS) does not load Corrected how PVF interacts with TFS |
| 18823 | 12.8 | PGI Accelerator Fortran | ACC: pgfortran reports the warning, "Inconsistent vector0/worker0 setting" Corrected the handling of a scalar variable which is used as a reduction variable and a copy-out variable |
| 18829 | 12.8 | C/C++ | pgCC faults on code when compiled with -Mbounds Corrected the subscript checking of certain arrays |
| 18832 | 12.8 | CUDA Fotran | Request support for double precision atomic aadd in CUDA FORTRAN Added support for the double precision atomic add |
| 18847 | 12.8 | Fortran | pgfortran non-advancing I/O fails Corrected the handling of certain variables being read when non-advancing I/O is in effect |
| 18851 | 12.8 | PGI Accelerator | OpenACC problem with multiple sum reductions Corrected the handling of a float sum reduction in an OpenACC worker loop |
| 18855 | 12.8 | PGI Accelerator | OpenACC reports the error, "float *" cannot be assigned to an entity of type "float" Corrected the handling of a float sum reduction in an OpenACC worker loop |
| 16211 | 12.6 | Fortran | bind(c) function returning derived type fails during execution Corrected calling the C function returning struct |
| 18191 | 12.6 | Fortran | Incorrect error message for an identifier which is too long Corrected the error message to indicate truncating to 63 characters instead of 31 |
| 18210 | 12.6 | Documentation | Documentation error in the ieee_arithmetic section Corrected the name of the type: ieee_round_type |
| 18364 | 12.6 | C/C++ | Error "PGC-S-0061-Sizeof dimensionless array required" Corrected the use of the sizeof operator on a VLA declarator |
| 18454 | 12.6 | Fortran | Error "Lowering Error: symbol pol$sd is an inconsistent array descriptor" Corrected the handling of an allocatable member of a derived type |
| 18481 | 12.6 | Fortran | WinApp program containing a list-directed write fails Corrected the run-time to detect a stdio error when in a WinApp |
| 18508 | 12.6 | Fortran | Error "PRIVATE type must not be used for PUBLIC variable" Corrected the accessibility of public variables declared with a private derived type |
| 18554 | 12.6 | PGI Accelerator | Add FAQ for NVIDIA error message "unsupported GNU version! gcc 4.5 and up are not supported!" Added message to FAQs |
| 18590 | 12.6 | PVF | Crash in Visual Studio when opening Fortran files from Windows Explorer Corrected the detection of fixed-form files |
| 18618 | 12.6 | Fortran | Error "illegal use of symbol my_var - not public entity of module" Corrected use-association of nested USE statements mixing the PUBLIC and PRIVATE statements |
| 18635 | 12.6 | PGI Accelerator Fortran | Kernel fails to be generated due to error, "induction variable live-out from loop" Corrected the handling of kernels appearing in subprograms containing internal procedures |
| 18642 | 12.6 | Fortran | Error "Cannot use variable or subprogram name in USE statement" Corrected the scoping of a module name and a local name |
| 18667 | 12.6 | Fortran | Failure with the use of the pack intrinsic Corrected the handling of the pack intrinsic when the array and mask arguments are array sections |
| 18672 | 12.6 | PGI Accelerator Fortran | pgfortran without -Mcuda reports an internal error for CUDA DEVICE variables Added an error check for using DEVICE variables without the -Mcuda option |
| 18677 | 12.6 | Fortran | Errors with the POS, POSITION and IOSTAT I/O specifiers Report an error for a POS value of 0 and allow POSITION and IOSTAT for STREAM files |
| 18686 | 12.6 | Fortran | pgfortran -F creates a file with the wrong suffix Corrected pgfortran to produce a .f90 file if the suffix of the input file is .F90 |
| 18687 | 12.6 | C/C++ | pgCC -O debug information for lexical blocks is incorrect Corrected the label references in the debug information for lexical blocks |
| 18688 | 12.6 | 64-bit C/C++ | Error "-Internal compiler error. linearize: bad ili" when using -mp Corrected the handling of an expression appearing in an OpenMP if clause containing the && or || operators |
| 18690 | 12.6 | Tools | xrefresh required for pgdbg and pgprof in text mode on Linux systems. Removed the requirement of xrefresh being present when using pgdbg/pgprof in text mode |
| 18694 | 12.6 | PGI Accelerator Fortran | Error "expression must have arithmetic or enum type" Corrected the handling of a scalar variable appearing in a kernel's copy clause |
| 18695 | 12.6 | Fortran | pgfortran intrinsic reshape fails with -i8 Corrected the handling of reshape of an array constructor when -i8 is used |
| 18699 | 12.6 | OSX | Improve the OSX license generation by using localhost as hostname Added information regarding localhost to the installation instructions for OSX |
| 18706 | 12.6 | Fortran | Error "Internal compiler error. more RDFO nodes than DFN nodes" Corrected the assignment of allocatables when the right-hand side must first be fully evaluated |
| 18711 | 12.6 | Fortran | Error "Internal compiler error. Missing region ILI" when compiled with -acc -Mipa Corrected the removal of scalar regions from the host code |
| 18713 | 12.6 | Fortran | Error with a READ from a direct access unformatted file with a trailing record that is too short. Corrected the error handling of a read of a record which is either non-existent or too short |
| 18715 | 12.6 | Fortran | Error "Syntax error at or near real constant" Corrected the recognition of generic type procedures |
| 18717 | 12.6 | PGI Accelerator Fortran | Error "expression must have arithmetic or enum type" Corrected the handling of a scalar variable appearing in a kernel's copy clause |
| 18723 | 12.6 | PGI Accelerator Fortran | cuModuleGetGlobal error "500 Not Found" Corrected the handling of the asynchronous clause |
| 18727 | 12.6 | Fortran | pgfortran failure with type-bound procedure Corrected the processing of typed allocation |
| 18760 | 12.6 | Fortran/C/C++ | -fast with -Mprof=lines fails on AVX systems Corrected calling the profiling support routines |
| 18767 | 12.6 | CUDA Fortran | Error "identifier "atomicaddl" is undefined" when compiling with -i8 Added the support for the integer(8) version of atomicadd |
| 18769 | 12.6 | OSX Fortran | Undefined symbols _pgf90io_nmlw_init_i8 with -Mlarge_arrays and a namelist write Added the support of large arrays for namelist I/O |
| 18779 | 12.6 | CUDA Fortran Documentation | CUDA Fortran doc needs updated section on constant data Updated the section on constant data in the CUDA Fortran Programming Guide to reflect actual functionality |
| 18794 | 12.6 | Fortran | Wrong answers when "transpose(derc)" is passed to matmul Corrected the handling of matmul when the first argument is a transpose of a non-square matrix |
| 18097 | 12.5 | Fortran | pgfortran user code exhibits poor performance Optimized MATMUL of small matrices and RESHAPE of contiguous array arguments |
| 18446 | 12.5 | Fortran | pgfortran reports the error, "Can't get array shape for PACK's mask arg' and 'mkshape: bad numdim 0" Corrected passing a compound MASK argument to the PACK intrinsic |
| 18532 | 12.5 | Fortran | pgfortran reports the error, "logical result for arithmetic operation" Allow the unary + or - operator on logical values as an extension |
| 18533 | 12.5 | PGI Accelerator Fortran | Mixed ACC/CUF program reports the error, "symbol `.STATICS3' is already defined" Corrected the handling of certain local variables when both -Mcuda and -ta=nvidia are present |
| 18542 | 12.5 | 64-bit Fortran | pgfortran does not report a type mismatch between function and character*(*) Detect type mismatch errors when the arguments are functions |
| 18592 | 12.5 | PGI Accelerator | OpenACC: program reports the error, "cuMemcpy2DUnaligned", when copying subarrays Corrected the handling of module allocatables in certain data clauses |
| 18593 | 12.5 | Fortran | pgfortran reports the error, "size_of: bad dtype" when compiling a module. Corrected the processing of pointer members of a type which also contains type bound procedures |
| 18598 | 12.5 | Fortran | Calling the lib3f "ftell" function fails to link with mpif90 Corrected the search order of libraries used by mpif90 |
| 18605 | 12.5 | PGI Accelerator | Perform real-to-complex transforms with CUFFT Provided a prototype of a 'cufft' Fortran interface module for CUFFT routines that allow real/complex or double/complex(8) arguments |
| 18608 | 12.5 | Fortran | pgfortran ALLOCATE reports the error, "not enough memory" Corrected the length computation of a substring when the start position is larger than the end position |
| 18610 | 12.5 | PGI Accelerator Windows C | Windows pgcc fails to link C++ and C objects containing accelerator code with ‑pgcpplibs Implemented the option -acclibs to add the required accelerator libraries |
| 18615 | 12.5 | 32-bit Windows Fortran | pgfortran read/write fails on 32-bit WinXP Removed the large file support from pgfortran on 32-bit windows since large files are not supported on XP |
| 18616 | 12.5 | Fortran | pgfortran reports the error, "Lowering Error: array zero-base is not a symbol for data type" Corrected the implementation of character initialization expressions consisting of array and scalar concatenations |
| 18617 | 12.5 | Fortran | pgfortran reports the error, "PGF90-S-0084-Illegal use of symbol proc - not public entity of module" Corrected the case where a module defines a subprogram with the same name as a renamed use-associated subprogram |
| 18622 | 12.5 | Fortran | pgfortran reports a syntax error for a generic operator declaration Corrected the lexical analyzer for the keyword "operator" and the ensuing "(" |
| 18624 | 12.5 | Fortran | f2003 get_environment_variable intermittently fails Corrected the run-time support routine for get_environment_variable |
| 18631 | 12.5 | C/C++ | pgcc -Mdse reports the error, "Sizeof dimensionless array x required" Corrected the analysis of VLAs |
| 18636 | 12.5 | Fortran | pgfortran example fails during execution with "glibc detected *** double free or corruption" Corrected applying the allocatable assignment semantics when the lhs is an array section of a derived type containing allocatable members |
| 18637 | 12.5 | CUDA Fortran | CUDA Fortran reports the error, "State space incorrect for instruction 'st'" Detect the user error of assigning to a variable with CONSTANT attribute in a global or device subprogram |
| 18639 | 12.5 | 64-bit CUDA Fortran | CUDA Fortran reports the error, "cf_data_init: incorrect offset 0" Corrected the initialization of derived type PARAMETERs with -Mcuda |
| 18643 | 12.5 | Fortran | pgfortran -mp=trace fails to link properly Added the missing symbols to the mp run-time library |
| 18647 | 12.5 | Fortran | pgfortran program incorrectly reports the error, "PGFIO-F-219/unformatted read/unit=11/attempt to read/write past end of record" Corrected the error detection for unformatted read/writes and rewind |
| 18650 | 12.5 | PGI Accelerator Fortran | ACC program with mirror clause fails to compile Corrected the handling of a use-associated allocatable variable in a mirror clause |
| 18660 | 12.5 | 64-bit CUDA C/C++ | CUDA C++ instantiation of a _global_ function for its address fails Allow templates instantiated for their address to have CUDA attributes |
| 18670 | 12.5 | Fortran | pgfortran example produces wrong answers when transpose(b) is passed to matmul Corrected the small matmul optimization when its argument is a transpose |
| 18671 | 12.5 | PGI Accelerator Fortran | ACC example produces wrong answers Corrected the subscripting of complex arrays and loops which count down |
| 18682 | 12.5 | Fortran | pgfortran reports the error, "add_arg_ili :bad ili for BYVAL" Corrected passing certain arguments by value to an array-valued function |
| 18440 | 12.4 | Installation | License failure with 12.x on Linux A requirement is that LSB version 3.0 must be installed on the system. The command lsb_release can be used to determine the version number of LSB. |
18487 | 12.4 | Fortran | OpenMP PARALLEL fails to execute on more than one thread Corrected the handling of the IF clause on a PARALLEL construct so that it has no effect on any nested parallel regions |
| 18506 | 12.4 | Fortran | pgfortran reports the error, "FINAL subroutine must be a module procedure with one dummy argument" Corrected the processing of the declarations of module-contained final procedures. |
| 18507 | 12.4 | Fortran | pgfortran reports the error, "Internal compiler error. size_of: attempt to size assumed size character: Corrected the handling the LEN_TRIM intrinsic appearing as the CHARACTER length specifier |
| 18524 | 12.4 | CUDA Fortran | Errors with the ISO_C_BINDING c_int8_t kind Corrected the handling of c_int8_t variables in kernel regions |
| 18530 | 12.4 | CUDA Fortran | User code when using 'volatile' attribute gets wrong answers with -fast Corrected the compiler to use the volatile semantics |
| 18535 | 12.4 | Fortran | pgfortran fails with "Fatal-f901 completed with exit code -1073741819" on Windows Corrected the processing of a repeated constant within an array constructor |
| 18544 | 12.4 | CUDA Fortran | Program gets wrong answers in 12.x Corrected a data transfer optimization between the host and device |
| 18552 | 12.4 | CUDA Fortran | Using SUM on device array causes seg fault Corrected the size computation of the device array |
| 18553 | 12.4 | Installation | CDK not installing mpich2 libraries properly Corrected the installation of the mpich2 libraries on slave nodes |
| 18554 | 12.4 | PGI Accelerator | OpenACC produces the message, "error:#error -- unsupported GNU version! gcc 4.5 and up are not supported!" Added an entry to the FAQ which discusses this error |
| 18566 | 12.4 | Fortran | pgfortran example produces different answers at -O2 and higher Corrected the compiler to apply the semantics of the -Msave option on local allocatable arrays |
| 18571 | 12.4 | CUDA Fortran | Reports the error, "Internal compiler error, mk_offset: unexpected ast when using constant scalar in host expression Corrected the contiguous section optimization when copying between the host and device |
| 18581 | 12.4 | PGI Accelerator | FATAL ERROR: data in PRESENT clause was not found Corrected the handling of a nested "copy/in/out" inside a "create/copyin/copyout" |
| 18583 | 12.4 | C/C++ | Long compile-time for a routine from ESMF Corrected compilation problem of a module with many nested USEs |
| 18585 | 12.4 | C/C++ | pgcpp code produces "terminate called without an active exception" Corrected the handling of an exception which is thrown inside a 'jumped' context |
| 18586 | 12.4 | 32-bit Fortran | List-directed/formatted write error on 32-bit Windows XP Corrected the run-time when writing to a file whose OPEN status is 'unknown' |
| 18587 | 12.4 | C/C++ | FATAL ERROR: data in PRESENT clause was not found Corrected the handling of a nested "copy/in/out" inside a "create/copyin/copyout" |
| 18589 | 12.4 | PGI Accelerator | call to cuMemcpy2Dunaligned returns an error Corrected the upload of a 2D array section of a 4D array |
| 18596 | 12.4 | PGI Accelerator | CUDA 4.1 interfaces different than CUDA 4.0 Adjusted the cuda interface libraries to accommodate the interface changes in CUDA 4.1 |
| 18601 | 12.4 | Profiler | JAVA error, "java.lang.ArrayIndexOutOfBoundsException", occurs with the profiler Corrected an array index out of bounds error |
| 18503 | 12.3 | 64-bit C/C++ | Code including 12.2 versions of float.h and limits.h may fail to compile Updated the two .h files to properly guard with GNU_SOURCE |
| 18500 | 12.3 | Fortran | pgfortran reports the error, "Symbol, omp_get_thread_num, has not been explicitly declared" Corrected the handling of nested USE statements where the name of a procedure appears in an ONLY clause |
| 18495 | 12.3 | Fortran | pgfortran reports the error, "Illegal use of symbol close_file - not public entity of module" Corrected the handling of nested USE statements where the name of a procedure appears in an ONLY clause |
| 18483 | 12.3 | C/C++ | pgcpp reports the error, "Internal compiler error. union_find_last_lp_per_handler:empty throw_bih" Corrected the use of OpenMP private variables when processing a region for a throwing function or virtual table pointer |
| 18482 | 12.3 | Fortran | pgfortran reports the error, "Symbol, get_ncols_p, has not been explicitly declared" Corrected the handling of nested USE statements where the name of a module procedure appears in an ONLY clause |
| 18461 | 12.3 | CUDA Fortran | Errors with named array constants (PARAMETERs) Corrected the processing of array PARAMETERs when used on the GPU |
| 18458 | 12.3 | Fortran | pgfortran reports the error, "Symbol, iscsf has not been explicitly declared" Corrected the handling of nested USE statements where the name of a module procedure appears in an ONLY clause |
| 18449 | 12.3 | CUDA Fortran | Large code containing many reductions may cause failures in pgf902 Corrected the compiler's handling of a large number of reductions to be executed on the GPU |
| 17633 | 12.3 | Fortran | pgfortran reports the error, "Could not resolve generic procedure" Corrected extending a USE-associated generic |
| 16710 | 12.3 | C/C++ | OpenMP example compiled -mp -Mconcur/-Mvect fails Corrected an optimization problem with OpenMP parallel loops containing integer expressions which references INT_MAX |
| 16413 | 12.3 | 64-bit Fortran | OpenMP example compiled -mp -O2 fails Corrected an optimization problem with OpenMP parallel loops containing integer expressions which references the maximum 32-bit signed integer |
| 18450 | 12.2 | Linux Installer | Release 12.1 fails to install on Ubuntu 11.10 Adjusted the installer for Unbuntu 11.10. |
| 18433 | 12.2 | CUDA Fortran | Allow for stream in CUF kernel Added the ability to specify a stream in a kernel directive. |
| 18420 | 12.2 | Windows Fortran | Support the makedirqq routine in pgfortran for Windows Added the interface for makedirqq to the dflib module on windows. |
| 18418 | 12.2 | 64-bit Fortran | Length of file in OPEN is limited to 255 characters Removed the restriction on the length of the file value which may appear in an OPEN statement. |
| 18411 | 12.2 | Fortran | Error with calling a pointer-valued function from a OpenMP parallel region Corrected the processing of calling a pointer-valued function within the context of an OpenMP parallel region. |
| 18407 | 12.2 | Fortran | Link fails with pghpf -Mmpi Corrected the options passed to ld when -Mmpi is present. |
| 18371 | 12.2 | PGI Accelerator Fortran | Code compiled with "-ta=nvidia,time" has potential memory leak Corrected the handling of the CUDA events used for timing. |
| 18350 | 12.2 | PGI Accelerator | Copy should occur instead of copyout for variables which are conditionally updated Corrected the copy of arrays whose elements are conditionally updated. |
| 18348 | 12.2 | PGI Accelerator C | Allow a for loop index variable to be type char Extended the set of types of for loop index variables allowed within accelerator regions. |
| 18289 | 12.2 | PGI Accelerator | Allow a for loop index variable to be type unsigned long Extended the set of types of for loop index variables allowed within accelerator regions. |
| 18286 | 12.2 | Profiler | OpenMPI compiler wrapper instructions are obsolete Corrected the documentation of the compiler wrapper data files to reflect newer versions of OpenMPI. |
| 18230 | 12.2 | PGI Accelerator | Memory leak in multiple invocations of $acc region? Corrected the handling of the CUDA events used for timing. |
| 18153 | 12.2 | CUDA Fortran | Program with '!$cuf kernel do <<<*, 256>>>' causes many undefined reference linker Corrected the handling of a cuf kernel in a contained subprogram. |
| 18127 | 12.2 | PGI Accelerator | -mcmodel=medium causes undefined reference linker errors Corrected the handling of the option -mcmodel=medium when -ta is also present. |
| 18409 | 12.1 | C/C++ | pgcc reports the errors, "Unmatched double quote" and "Illegal character: '\0'" on Windows Corrected pgcc's handling of strings containing certain encodings of "Windows-1251" characters. |
| 18372 | 12.1 | 64-bit Fortran | OpenMP program fails Corrected a race condition when there are calls to omp_set_num_threads with a varying number of threads. |
| 18370 | 12.1 | 64-bit Fortran | For C interoperability, pass Fortran character lengths as size_t For x86_64, the lengths of character objects are now passed as 64-bit integers. |
| 18366 | 12.1 | Fortran | Fortran example containing a derived type with type-bound procedures seg fault. Corrected a deallocation error of a pointer CLASS member of a derived type which also contains type bound procedures. |
| 18363 | 12.1 | Fortran | Corrupted allocation of a member of derived type which also contains type-bound procedure Corrected an deallocation error of an allocatable CLASS member of a derived type which also contains type bound procedures. |
| 18357 | 12.1 | Fortran | Error message incorrectly reported for the ASSOCIATE statement. Corrected the error checking of ASSOCIATE expressions containing subscripted references. |
| 18355 | 12.1 | OSX Installer | OSX installation error of the licensing components Corrected the permissions of the OSX licensing components. |
| 18353 | 12.1 | 64-bit Fortran | The .EQV. logical operator produces incorrect results with -i8 Corrected computing nested .EQV. operations in the presence of -i8. |
| 18332 | 12.1 | 64-bit Fortran | pgfortran reports the error, "Internal compiler error. getsname: bad stype" Corrected the processing of a derived type containing type-bound procedures with DEFERRED binding. |
| 18292 | 12.1 | Fortran | Name collision errors with LEN, ILEN, and NLEN Corrected an issue with the use of the name, LEN, ILEN, or NLEN, as the I/O SIZE specifier. |
| 18285 | 12.1 | Profiler | Java ExcecutionException error occurs with pgprof Corrected a profiler error parsing the pgprof.out file. |
| 18283 | 12.1 | Fortran | pgfortran reports the error, "type bound procedure must be a module procedure or an external procedure with an explicit interface" Corrected the processing of a PASS argument which appears in a subscripted reference which also contains multiple member references. |
| 18282 | 12.1 | 64-bit Fortran | A segmentation error occurs when compiling code containing a deferred length character member Corrected the compiler's handling of a derived type containing a CLASS member and a member which is a character of deferred length. |
| 18280 | 12.1 | Windows | Windows binding not working for more than 32 cores Corrected the binding of more than 32 cores on Windows. |
| 18274 | 12.1 | C/C++ | pgcc reports the error, "File write error occurred" Corrected the processing of #pragma ident when the string is empty. |
| 18270 | 12.1 | 64-bit Fortran | The assembler reports the message, ".size expression for MAIN_ does not evaluate to a constant" Corrected the .size computation for a program containing an internal procedure. |
| 18268 | 12.1 | 64-bit Fortran | Logical operation gives different results with -i8 Corrected computing nested .EQV. operations in the presence of -i8 |
| 18264 | 12.1 | PGI Accelerator Fortran | Error with conditional store Corrected an optimization involving a conditional store assigned to a register. |
| 18262 | 12.1 | Fortran | Incorrect results with -ta=nvidia,nofma Corrected the store of double**2 in the presence of nfma. |
| 18259 | 12.1 | Fortran | Program segfaults with pgfortran Corrected the computation of SIZE of a derivied type reference where the member is scalar and its parent is array |
| 18258 | 12.1 | Fortran | pgfortran incorrectly "Syntax error at or near WHERE" Allow a construct name to appear on the WHERE, ELSEWHERE, and ENDWHERE statements. |
| 18257 | 12.1 | Windows Fortran | Segmentation fault occurs in pgf901 whe comping WRF 3.3.1 on Win64 Increased the amount of stack space that can be used by pgf901. |
| 18247 | 12.1 | CUDA Fortran | Internal errors occur in CUDA Fortran when assigning literals representing infinity and NaN Use the binary representation for certain literals |
| 18246 | 12.1 | Fortran | Incorrect results with the POS specifier in a WRITE statement Corrected the handling of POS in a WRITE for unformatted files. |
| 18240 | 12.1 | Fortran | ReadConsoleOutputCharacter not working correctly Corrected the handling of the lpNumberOfCharsRead argument. |
| 18236 | 12.1 | PGI Accelerator | Function cudaMemset incompatible with -i8 Corrected the processing of the size/length arguments to routines such as cudaMemset. |
| 18234 | 12.1 | 64-bit Fortran | pgfortran reports the error, "Constant expression of wrong data type", for hexadecimal constants Corrected the handling of array constructors formed from hexadecimal constants. |
| 18222 | 12.1 | Fortran | pgfortran reports the error, "Too many data constants in initialization statement" Corrected the use of certain intrinsics, such as acos, as an initializer. |
| 18221 | 12.1 | 64-bit C/C++ | Optimization error with subscripting a large allocated array Corrected an optimization of subscript computation of an array whose number of elements is less than or equal to MAX_INT but its aggregate size is larger than 2GB. |
| 18211 | 12.1 | 32-bit Windows CUDA Fortran | Undefined symbols in Win32 libacc1cu.lib Corrected the build of the libacc1cu library on Win32. |
| 18210 | 12.1 | Documentation | Incorrect description of ieee_round in the Fortran reference manual Corrected the documentation. |
| 18200 | 12.1 | MPI | pgfortran -Mprof=mpich1 linking error Removed the use of the library, -lopa, from the link step of -Mprof=mpich1. |
| 18187 | 12.1 | Fortran | pgfortran aborts with -Mipa=fast Corrected the processing of certain derived type initializers when -Mipa is present. |
| 18179 | 12.1 | Fortran | pgfortran reports the error, "Illegal use of symbol i4 - not public entity of module" Corrected the handling of nested USE statements. |
| 18165 | 12.1 | Fortran | pgfortran reports the error, "cannot access PRIVATE type bound procedure xi" Corrected resolving a call to a generic type-bound procedure. |
| 18149 | 12.1 | Fortran | pgfortran example segfaults on the execution of an array assignment Corrected assigning allocatable arrays of derived type which contains pointer components. |
| 18144 | 12.1 | Fortran | pgfortran reports the error, "unknown target type for conversion" Corrected the processing of array constructors containing references to c_loc. |
| 18139 | 12.1 | Fortran | pgfortran seg faults when using type bound procedures. Corrected the processing of a type bound procedure when its interface refers to a module-contained subprogram |
| 18138 | 12.1 | CUDA Fortran | pgf77 -Mcuda reports the error, "pgf77-Error-CUDA is not available in this installation" Corrected the error message to report that pgf77 does not support -Muda. |
| 18126 | 12.1 | Fortran | pgfortran reports type mismatch errors when using the DFWIN module Corrected DFWIN-specific errors related to using the CHAR intrinsic, the argument lpRect in function InvalidateRect, and the routine LOWORD. |
| 18124 | 12.1 | Fortran | Use of ISO_C_BINDING c_ptr type fails Corrected the call of a function whose return type is type(c_ptr) |
| 18083 | 12.1 | Fortran | pgfortran reports the error "not public entity of module" Corrected the handling of nested USE statements. |
| 18056 | 12.1 | Fortran | size() intrinsic an allocatable array with the SAVE attribute returns wrong result Corrected an optimization involving the use of the bounds of an allocatable array with the SAVE attribute. |
| 18052 | 12.1 | 64-bit Windows Fortran | Passing float complex by value from Fortran to C on Win64 is incorrect Corrected the handling of complex value argrument on Win64. |
| 18051 | 12.1 | 64-bit Windows Fortran | Passing double complex by value from Fortran to C on Win64 is incorrect Corrected the handling of complex value argrument on Win64. |
| 18040 | 12.1 | Fortran | pgfortran reports the error, " Constant expression of wrong data type" Corrected the handling of an array constructor item consisting of the negate of a complex constant. |
| 18026 | 12.1 | Windows Fortran | 3F fstat64 fails on Windows Corrected the Win32 and Win64 versions of fstat64 and stat64. |
| 18010 | 12.1 | Fortran | pgfortran should give an error when intent(out) is used with the value attribute Updated the compiler to report an error for the combination of VALUE and INTENT(OUT). |
| 17983 | 12.1 | PGI Accelerator Fortran | Program reports the error, "call to cuMemFree returned error 1: Invalid value" Corrected the intialization of device pointer for array in mirrored regions. |
| 17969 | 12.1 | Fortran | pgfortran reports errors for named array constants of zero size Corrected the processing of zero-sized array constructors. |
| 17917 | 12.1 | MPI | mpicc -Mprof=mpich1 linking error Removed the use of the library, -lopa, from the link step of -Mprof=mpich1 . |
| 17907 | 12.1 | MPI | mpichversion produces misleading information Corrected mpichversion to produce the correct location of the mpich installation. |
| 17806 | 12.1 | CUDA Fortran | CUDA: double precision constants not properly initialized. Corrected the initialization of double precision objects on the device |
| 17776 | 12.1 | Fortran | Use of ISO_C_BINDING c_loc fails Corrected the processing of nested USE statements. |
| 17739 | 12.1 | Fortran | pgfortran reports the error, "Illegal use of symbol i1 - not public entity of module" Corrected the processing of nested USE statements. |
| 17733 | 12.1 | Fortran | pgfortran reports the error, "Argument missing for formal argument x" Corrected the processing of a procedure pointer member when its interface refers to a module-contained subprogram. |
| 17501 | 12.1 | Fortran | pgfortran reports the error, "class is use associated from module ftl_pgi_5e and ftl_pgi_5c and cannot be accessed" Corrected the processing of nested USE statements. |
| 17192 | 12.1 | Fortran | pgfortran reports the error, "Non-constant expression where constant expression required", for a function Corrected the processing of USE-associated named integer constants referenced as a kind value in a function declaration. |
| 17122 | 12.1 | Debugger | Errors reported with the use of pgdbg -s -mpi Corrected the behavior of pgdbg -s -mpi when locale not supported by Xlib. |
| 17059 | 12.1 | Fortran | Missed opportunity for optimizing loops containing use of integer and floating point operations Improved the optimizer to vectorize converting integer to floating point. |
| 16058 | 12.1 | 64-bit Fortran | ALLOCATE/DEALLOCATE does not scale with array size Improved the performance of the run-time routines for ALLOCATE/DEALLOCATE. |
| 15105 | 12.1 | Debugger | pgdbg attach command reports the error, "pgserv (vm-suse103-64: Corrected the attach behavior after a sequence of attach/detach commands. |
| 14601 | 12.1 | 64-bit Fortran | Add a feature to align allocatables on page boundaries Added the ALIGN extension to the ALLOCATE statement. |
| 4345 | 12.1 | Debugger | When using pgdbg on an mpich2 program, the program's output does not appear in the I/O window Corrected handling of the output of an mpich2 program while debugging. |
| 4126 | 12.1 | Fortran | pgfortran reports the error, "must appear in a SHARED or PRIVATE", for an implied-do variable when "IMPLICIT NONE" is present Corrected the scoping of implied-do index variables. |
