PGI Release 2015 Technical Problem Reports
What problems have been fixed in PGI 2015?
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 |
---|---|---|---|
22011 | 15.10 | Fortran | LBOUND with a kind argument results in an undefined reference Corrected the run-time support for LBOUND and UBOUND when the kind argument is present |
21992 | 15.10 | Fortran | pgfortran reports the error, "Lowering Error: bad ast optype in expression", for ASSOCIATE Corrected the handling of an ASSOCIATE expression when accessing an array section of a derived type component |
21973 | 15.10 | Fortran | pgfortran compilation fails for long source file names (122 chars) Increased the limit of the length of source file names |
21965 | 15.10 | Fortran | pgfortran compilation fails with "TERMINATED by signal 11" Corrected the processing of certain arguments whose corresponding dummy arguments have the POINTER attribute |
21958 | 15.10 | Fortran | pgfortran incorrectly reports the error, "must be a POINTER variable" for a procedure pointer assignment Corrected the handling of a pointer assignment when the target is a use-associated external procedure |
21944 | 15.10 | Fortran | Execution fails for use of parameterized derived type and character arrays Corrected the use of a type parameter as the extent of a character array |
21939 | 15.10 | Fortran | pgfortran is_contiguous fails for an assumed-shape array Corrected the use of assumed-shape array arguments passed to the is_contiguous intrinsic |
21907 | 15.10 | Fortran | pgfortran incorrectly reports the error, "Attempt to use private component", for a procedure pointer component Corrected the handling of a procedure pointer component declared PUBLIC in a type declared PRIVATE |
21906 | 15.10 | Fortran | pgfortran reports the error, "Keyword form of an argument illegal in this context for …" Corrected the handling of the keyword form a CALL when the interface is visible |
21902 | 15.10 | 64-bit Fortran | pgfortran auto-reallocation of an allocatable character fails Corrected the assignment to an allocatable scalar character of deferred length |
21890 | 15.10 | Fortran | pgfortran example with CONTAINed procedures and module type bound procedures fails Corrected referencing a procedure component with the same name as a CONTAINed procedure |
21876 | 15.10 | Fortran | pgfortran example containing the sum intrinsic produces incorrect results when compiled -mcmodel=medium Corrected the handling of the sum intrinsic appearing in a do loop whose index variable is integer(kind=4) in the presence of -mcmodel=medium or -i8 |
21806 | 15.10 | Fortran | pgfortran reports the error, "Internal compiler error. interf:new_symbol, symbol not found" Corrected the handling of a module procedure which appears in multiple interface blocks, one of which has the same name as the module procedure |
21663 | 15.10 | Fortran | pgfortran reports the error, "Internal compiler error. linearize: bad ili …", for code referencing type bound procedures Corrected the use of NOPASS in certain type bound procedure definitions |
21299 | 15.10 | C/C++ | pgc++ reports, "internal error: assertion failed …", on cv++11 code Corrected the handling of certain c++11 constructors |
19218 | 15.10 | PGI Accelerator | OpenACC: struct assignment in an acc kernel fails to parallelize Corrected the handling of certain structs containing multiple members |
21897 | 15.9 | C/C++ | pgc++ reports the error, "Internal compiler error. invalid region range 79" Corrected the processing of the compiler's internal region numberings |
21843 | 15.9 | 64-bit C/C++ | Incorrect results returned from an m128 function Corrected how the result of an m128 function is passed as an actual argument |
21842 | 15.9 | 64-bit C/C++ | pgcc fails to compile code containing the use of an m128 sse intrinsic Corrected how the value of an m128 sse intrinsic is stored into an element of an m128 array |
21841 | 15.9 | C/C++ | pgcc -O2 error occurs for an m128 sse intrinsic, _mm_movemask_ps Corrected the definition of _mm_movemask_ps in xmmintrin.h |
21840 | 15.9 | 64-bit C/C++ | pgc++ -Mipa reports the error, "Assert failed: ../src/ipanew.c:1552 - vfr != NULL" Corrected the handling of variable references which begin with '&' |
21836 | 15.9 | C/C++ | pgc++ --c++11 exception error Corrected the destructors generated for exception handling regions |
21831 | 15.9 | CUDA Fortran | Inconsistency in the documentation of atomicadd in the CUDA Fortran Programming Guide Corrected the documentation to indicate double precision atomcadd is supported |
21830 | 15.9 | PGI Accelerator | OpenACC program gives incorrect results for ta=tesla,host Corrected the processing of code blocks targeted for both the device and host |
21825 | 15.9 | PGI Accelerator | OpenACC: missing support for modf and fmod Added fmod, fmodf, modf, and modff to the set of allowed math.h functions |
21797 | 15.9 | PGI Accelerator | OpenACC: pgfortran -O2 -Mcuda=...cc35 reports the error, "no instance of overloaded function "__ldg" matches the argument list" Corrected the how a derived type argument is loaded into global memory |
21795 | 15.9 | CUDA Fortran | Error in CUDA Fortran with real(8) textures Corrected how a double is loaded from the texture cache |
21778 | 15.9 | 32/64-bit Fortran | pgfortran error with a C_LOC of a whole section of a pointer array Corrected the address computation of the whole section of a pointer by subscripting the pointer with its lower bounds |
21776 | 15.9 | All | Add the echo flag to the drivers Added the command flag -echo which directs the driver to echo the actual command line to stdout and stop |
21772 | 15.9 | PGI Accelerator | OpenACC: pgcc compiling code contain reduction emits a "parse error" Corrected the handling of a reduction formed from a struct argument |
21770 | 15.9 | Fortran | pgfortran date_and_time fails with -i8 Corrected how Integer(8) values are returned from the date_and_time intrinsic |
21754 | 15.9 | C/C++ | pgcc program fails to print -0.00000 Corrected the handling of the unary negate operator when its operand is +0.0 |
21741 | 15.9 | PGI Accelerator | OpenACC: pgfortran reports the error, "Warning: Linking two modules of different target triples!" Corrected the processing of certain derived type component references |
21736 | 15.9 | Fortran | pgfortran -Mallocatable=03 incorrectly reports the error, "Argument missing for formal argument …" Corrected the inheritance of a non-overloaded type bound procedure |
21721 | 15.9 | Fortran | pgfortran incorrectly reports type error " Illegal number or type of arguments to unpack - keyword argument field" Corrected the type checking of the derived type arguments of the UNPACK and PACK intrinsics |
21693 | 15.9 | Fortran | pgfortran compiled code fails to execute during finalization Corrected the processing in the final procedure of derived types containing device pointer members |
21667 | 15.9 | PGI Accelerator | OpenACC: pgc++ reports the error, "parse error: '%li72' defined with type 'i8''" Corrected the processing of narrow ints in acc routines |
21531 | 15.9 | PGI Accelerator | OpenACC: execution errors occur when using async with multiple OpenMP threads Corrected a race condition accessing and updating the OpenACC-generated 'present table' |
21428 | 15.9 | All | -Bstatic_pgi should include the static version of libcudadevice Added -lcudadevice to the set of "Bstatic_pgi" libraries |
21358 | 15.9 | PGI Accelerator | OpenACC: Allow user to pass the address of local scalar variables to device routines Improved the data analysis to auto-privatize certain local variables whose addresses are passed to acc routines |
21000 | 15.9 | PGI Accelerator | OpenACC: Use of % (mod) operator on 64-bit unsigned values results in the error, "unsupported operation: NUKMOD" Added the support of % on unsigned long data for the device |
19344 | 15.9 | PGI Accelerator | OpenACC: pgfortran errors occur when using c_f_pointer on device allocatable variables Corrected the use of device allocatable components arrays as arguments to the c_f_pointer procedure |
18610 | 15.9 | C/C++ | Link errors occur with pgcc -pgcpplibs on pgcpp-compiled object files Corrected the support of the option -pgcpplibs when presented to pgcc |
18386 | 15.9 | PGI Accelerator | OpenACC: Improve the performance of the nbody2 integration code Added the global memory optimization which uses wide (126 and 256 bit) loads and stores |
14756 | 15.9 | C/C++ | pgcpp is missing vfwprintf Added the declaration of vfwprintf to the PGI-supplied wchar.h header files |
21740 | 15.7 | Fortran | Incorrect results occur with conditional vectorization Corrected the vectorization of multiple guarded assignments to the same array element |
21737 | 15.7 | Documentation | man pages missing -tp haswell Added the description of haswell to the -tp switch |
21698 | 15.7 | All | pgrepro fails to execute Corrected the build of pgrepro |
21693 | 15.7 | Fortran | program execution faults during finalization Corrected the finalization of certain pointer types which are nullified |
21683 | 15.7 | Documentation | man pages missing description of cuda 7.0 Added 7.0 to the set of cuda versions |
21660 | 15.7 | Fortran | pgfortran reports the error, "-Illegal attempt to redefine symbol procslaveprocedure" Corrected the handling of a subprogram declared in an interface within a module-contained subprogram |
21658 | 15.7 | Fortran | pgfortran incorrectly reports the error, "Array objects are not conformable" Corrected the assignment of certain scalar expressions to array sections |
21652 | 15.7 | Fortran | Incorrect behavior of get_environment_variable Corrected get_environment_variable when the length of the value argument is less that the length of the environment variable's value |
21641 | 15.7 | PGI Accelerator | cuEventRecord returns error 700: Illegal address during kernel execution Corrected the allocation of private arrays in a collapsed gang loop |
21631 | 15.7 | PGI Accelerator | Fortran aggregate data types are not allowed in worker/vector transitions Corrected the processing of accelerator regions contains references to derived type objects |
21612 | 15.7 | Fortran | pgfortran compilation of code containing assignments to sections of pointers fails at -O2 Corrected the analysis of loops constructed from array pointer assignments |
21608 | 15.7 | 64-bit Fortran | pgfortran compiler faults on an array pointer reshape of a derived type component Corrected the handling of certain derived type pointer targets |
21605 | 15.7 | 64-bit Fortran | pgfortran -mcmodel=medium reports the error, "Internal compiler error. contiguous_array_section: unexpected dimension type 0" Corrected checking for contiguous array sections |
21600 | 15.7 | Fortran | pgfortran reports the error, "-Internal compiler error. string_expr_length: ast not string op" Corrected the processing of functions returning an allocatable character |
21598 | 15.7 | Fortran | pgfortran compilation faults on a declaration of a zero-length character variable Corrected the processing of zero-length character variables when declared within a module |
21583 | 15.7 | Fortran | pgfortran detects the error, "Could not resolve generic procedure genericfsum" Corrected the resolution of generic procedures when an intrinsic appears an actual argument |
21567 | 15.7 | Fortran | pgfortran reports the error, "Constant expression of wrong data type" Corrected the handling of a derived type containing multiple pointer components as well as default initializers |
21539 | 15.7 | 64-bit C/C++ | -Bstatic does not produce static executables with --export-dynamic --eh-frame-hdr Removed --export_dynamic from the list of default link options |
21522 | 15.7 | Fortran | Add an option to prevent Fortran formatted output from producing -0.0 Allow the environment variable, FORTRANOPT, to have the value "no_minus_zero" to inhibit emitting the '-' for IEEE -0.0 |
21450 | 15.7 | Fortran | pgfortran reports the error, "Non-constant expression where constant expression required" Corrected the handling of the integer kind values from the iso_c_binding module |
21434 | 15.7 | PGI Accelerator | Program execution faults when a loop index variable appears in the wait clause of an acc region Corrected the translation of the wait directive to a wait call |
21428 | 15.7 | All | libcudadive.so is still used with -Bstatic_pgi Add the static version of libcudadevice when -Bstatic_pgi is present |
21388 | 15.7 | Fortran | pgfortran reports the error, "No default initialization in structure constructor- member …" Corrected the processing of structure constructors of derived types whose names begin with "operator" |
21309 | 15.7 | Fortran | ISE-associated error not detected Corrected the checking for visible USE-associated derived types in a typed-ALLOCATE statement |
21304 | 15.7 | Fortran | Incorrect length and size of a character array component of a parameterized derived type Corrected applying the parameter value to the length and size of a character array component |
21161 | 15.7 | Fortran | pgfortran reports the error, "unknown intrinsic; unknown operator for LHS or argument" Corrected computing LEN of a TRIM call |
21067 | 15.7 | C/C++ | Incorrect placement of -lpgmp with -pgf90libs Corrected the driver to expand -pgf90libs after the occurrence of -lpgmp |
20600 | 15.7 | PGI Accelerator | OpenACC: Allow calling cuBlas device routines from compute regions Added the openacc_cublas module and the openacc_cublas header file |
19844 | 15.7 | Installation | makelocalrc fails on systems where LANG is not English Corrected makelocalrc when the default LANG is not n_US.UTF-8 |
19024 | 15.7 | PGI Accelerator | Issue a message when duplicate variable names appear in a reduction clause Corrected the compiler to detect duplicate names in a reduction clause and issue a warning |
18668 | 15.7 | PGI Accelerator | Add Climate Data Operators (CDO) to the Performance and Tuning Guide Updated the documentation |
17874 | 15.7 | C/C++ | Improve pgc++ to indicate that a typename is missing Enhanced the compiler's error checking to detect situations which are missing the typename keyword |
21539 | 15.5 | 64-bit C/C++ | Link errors occur with -Bstatic Removed --export-dynamic from the default set of linker options when -Bstatic is present |
21533 | 15.5 | Fortran | Error reported: "Internal compiler error. make_unrolled_cvect_pred_list" Corrected vectorizing a loop containing nested conditionals |
21524 | 15.5 | PGI Accelerator | pgfortran faults on code containing "# lines " when compiled -acc -Minfo Corrected the processing of the file information presented in "# lines" |
21520 | 15.5 | CUDA Fortran | Presence of SAVE leads to the error, "parse error: global variable reference must have pointer type" Corrected the handling of variables with the SAVE attribute within device routines |
21519 | 15.5 | PGI Accelerator | pgfortran -acc emits the error, "Compiler failed to translate accelerator region: Unknown device array size" of ACC ROUTINE Corrected the handling of arrays whose leading dimension is not a constant |
21514 | 15.5 | CUDA Fortran | cublasInit undefined at link time Corrected the declaration of cublasinit in the cublas module |
21465 | 15.5 | Fortran | Error reported: "Symbol, ddot, has not been explicitly declared" Corrected the processing of renamed use-associated procedure interfaces |
21464 | 15.5 | CUDA Fortran | Error reported: "Basic LLVM base data type required" Corrected the processing of COMPLEX variables with the SHARED attribute |
21431 | 15.5 | Fortran | Error reported: "Illegal use of derived type" Corrected the handling of a component reference whose parent is an array |
21356 | 15.5 | Fortran | A '#' marking a Fortran continuation is reported as a preprocessing error Corrected the detection of preprocessing directives when the source form is fixed-form |
21341 | 15.5 | Fortran | Error reported: "Could not resolve generic procedure circe2_generate" Corrected the processing of use-associated generics |
21310 | 15.5 | Fortran | Errors occur for POINTER variables in NAMELIST Corrected the NAMELIST processing of assumed length character POINTER variables |
21305 | 15.5 | Fortran | Error reported: "Too many data constants in initialization statement" Corrected the handling of array constructors for certain arrays declared with non-default lower bounds |
21264 | 15.5 | Fortran | Error reported: "Internal compiler error. ast_visit sees ast of 0", for an incorrect use of a user operator Corrected the error message when the operands to a user-defined operator have an incorrect type |
21067 | 15.5 | C/C++ | The linker reports undefined reference errors with the option -pgf90libs Corrected the placement of the "-pgf90libs" libraries on the link line |
20990 | 15.5 | Fortran | OpenMP DO with SCHEDULE(DYNAMIC) does not execute correctly when appearing in a nested parallel region Corrected the allocation and initialization of the scheduling data structures created by the compiler |
20024 | 15.5 | PGI Accelerator | Support basic fortran WRITE and STOP statements on the device Added the support for the STOP statement and list-directed WRITE and PRINT to unit * |
19732 | 15.5 | Windows Fortran | Writing a large Fortran file faults during execution on Windows Corrected writing large unformatted records on Windows |
19616 | 15.5 | Fortran | Error reported: "Illegal use of symbol str - not public entity of module" Corrected the processing of use-associated generics |
19407 | 15.5 | PGI Accelerator | pgcc -acc reports the error, "Could not find allocated-variable index for symbol" Corrected the processing of firstprivate on an acc loop |
18203 | 15.5 | CUDA Fortran | CUF kernel does not execute correctly when the loop bound variables have the CONSTANT attribute Corrected the setup of a kernel loop when the bounds require access to variables in constant memory |
21455 | 15.4 | 64-bit Fortran | Undefined references to omp_ routines are reported by the linker when -nomp Is used Added the necessary libraries to the link process when -nomp Is used |
21449 | 15.4 | Fortran | Error reported for an array section appearing in an associate construct Corrected the handling of an associate expression consisting of an array section when the rank of the array is 2 or greater and the section is rank one |
21444 | 15.4 | Fortran | Compilation error is reported when invoking a function whose argument is an INTENT(OUT) allocatable array Corrected the automatic deallocation of certain allocatable arrays |
21443 | 15.4 | Fortran | Compilation errors are reported for an array constructor of derived types which are type extension Corrected the data initialization of structure constructors where the derived type is a type extension of a type without data components |
21436 | 15.4 | 64-bit Fortran | An "Illegal context" error is reported for an OpenMP DO -mp in a nested parallel region Corrected the binding of the OpenMP DO in a nested parallel region where the outermost construct is a combined PARALLEL DO |
21400 | 15.4 | PGI Accelerator | Program using PGI_ACC_TIME fails to complete on Windows Corrected how the application terminates in the absence of a call to acc_shutdown() |
21385 | 15.4 | All | MPI profile information does not include a profile for each MPI rank Added the Fortran bindings for the MPI routines in the MPICH profiling libraries |
21374 | 15.4 | Fortran | Lengths of records larger than 2GB are incorrect in a BIG_ENDIAN unformatted file Corrected the presentation of the trailing record length of the first sub-record representing a large unformatted record |
21344 | 15.4 | Fortran | pgfortran reports the error, "Internal compiler error. module:new_dtype, illegal type 4" Corrected the processing of interfaces for type bound procedures |
21283 | 15.4 | Fortran | Incorrect warning reported for an INTENT(IN) actual argument corresponding to an INTENT(OUT) POINTER argument Corrected the check of the argument to reflect the association status of the POINTER rather than the definition of its target |
21222 | 15.4 | PGI Accelerator | The ACC_DEVICE_NUM environment variable is ignored when -Mcuda is used with -acc Corrected the CUDA run-time to first check if the device context is in ACC_DEVICE_NUM |
21010 | 15.4 | 64-bit Fortran | Performance of pgfortran TRIM Improved the performance of TRIM for relatively short arguments |
20979 | 15.4 | Fortran | pgfortran fails with "pgf902 TERMINATED by signal 11" Corrected the processing of certain array constructors containing both REAL(4) and REAL(8) items |
20697 | 15.4 | Fortran | Missing f2008 intrinsic, STORAGE_SIZE Added the STORAGE_SIZE intrinsic to pgfortran |
19916 | 15.4 | Fortran | Options -r8 and -Mr8 have different behavior Improved the documentation of the options |
19657 | 15.4 | PGI Accelerator | OpenACC: incorrect copyin behavior if the variable is already present Corrected the processing of copyin to first check if the variable is already present on the device |
19435 | 15.4 | Debugger | pgdbg reports the error, "Could not open linux-vdso.so.1", when debugging an MPICH-1 program Corrected the search of shared objects which are not present on the system |
18870 | 15.4 | Debugger | Add the breakpoint hit count feature to pgdbg Extended the implementation of breakpoints to include hit counts |
18745 | 15.4 | CUDA Fortran | Compiling CUDA fortran code compiled -mp reports the error, "Internal compiler error. Unsupported procedure" Improved the support for certain WRITE statements in device routines when compiling with and without -mp |
18277 | 15.4 | Fortran | Missing f2008 Bessel intrinsics Added the support for the elemental and transformation versions of the Bessel intrinsics |
21357 | 15.3 | Fortran | Fortran preprocessor reports the error, "EOF in comment" Corrected the preprocessor to inhibit certain preprocessing checks within Fortran comments |
21356 | 15.3 | Fortran | Fortran preprocessor reports the error, "Illegal directive name" Corrected the preprocessor to require preprocessor directives to begin in column 1 when fixed-form |
21348 | 15.3 | PGI Accelerator | pgcc -g -acc fails with the PTX assembly error, "Parsing error near '-', syntax error" Corrected the debug information for C arrays on the device |
21338 | 15.3 | Fortran | pgfortran program containing generic type bound procedures fails to compile Corrected the handling of a generic type bound procedure inherited from a parent type when there is an identical binding name associated with an unrelated type |
21328 | 15.3 | C/C++ | pgc++ -Bstatic fails with the error, " cannot find -lgcc_s" Removed -lgcc_s from the link line when -Bstatic is present |
21313 | 15.3 | Fortran | Memory leak occurs during the execution of a WRITE statement Corrected the processing of a format in the context of executing a nested/recursive WRITE |
21298 | 15.3 | PGI Accelerator | pgcc -g -acc fails with the PTX assembly error, "Parsing error near '-', syntax error" Corrected the debug information for C arrays on the device |
21212 | 15.3 | C/C++ | An alignment fault occurs during the execution of a C++ program Corrected the references of m128 variables in an extended asm statement |
21211 | 15.3 | C/C++ | pgc++ emits the errors, "'cc' clobber ignored bswap_16(i16)" and "undefined reference to `constant_p" Added the support for byteswap.h and __builtin_constant_p |
21112 | 15.3 | Fortran | pgfortran produces the error, "Lowering Error: array upper bound is not a symbol for datatype 65" Corrected the processing of abstract interfaces for array-valued functions |
20986 | 15.3 | PGI Accelerator | SDK examples should use cudadevicesynchronize() instead of cudathreadsynchronize() Corrected the examples containing the calls to cudathreadsynchronize() |
20722 | 15.3 | PGI Accelerator | CUDA Fortran dgemmdynamic example fails on 32-bit Windows Corrected the 32-bit compiler's handling of certain CUDA run-time routines |
19652 | 15.3 | PGI Accelerator | pgfortran -Mcuda produces the error, " Unresolved extern function 'cublasCreate_v2'" Corrected the compiler to ensure that the cublas run-time is included |
21246 | 15.1 | PGI Accelerator | Execution fault occurs in the PGI Accelerator runtime when a zero-size array appears as an argument Added guards for zero-size arrays |
21203 | 15.1 | PGI Accelerator | OpenACC reports the error, "nvlink error Prototype doesn't match" Corrected the handling of sequential acc routine |
21192 | 15.1 | Fortran | pgfortran error occurs with an associate construct Corrected the processing of an associate of a whole array section |
21186 | 15.1 | 64-bit Fortran | pgfortran example causes a segmentation fault in the compiler Corrected the processing of array bounds in certain contexts |
21166 | 15.1 | Fortran | pgfortran incorrectly reports "has not been explicitly declared" errors Corrected the handling of a forward reference to an implementation of a type bound procedure |
21154 | 15.1 | Fortran | pgfortran reports the error, "Could not resolve generic procedure" Corrected resolving a generic to a module procedure with the same name as an abstract interface |
21153 | 15.1 | PVF | pgdbg execution is very slow after continuing from a removed breakpoint Corrected how execution resumes after a breakpoint is deleted |
21130 | 15.1 | Fortran | pgfortran sourced allocation produces incorrect results Corrected the allocation when the SOURCE is a function returning an allocatable class object |
21129 | 15.1 | Fortran | pgfortran code causes a segmentation fault in the compiler Corrected the processing of a structure constructor where the values for certain pointer components are missing |
21126 | 15.1 | Fortran | pgfortran reports multiple error messages when the floating point operation ** overflows Corrected the detection of overflow when evaluating ** at compile-time |
21101 | 15.1 | Fortran | References to undefined labels occur when compiling certain loops with -fast Corrected how certain loops containing if statements are vectorized |
21092 | 15.1 | Fortran | Need a compiler option to inhibit loop idiom recognition Added the option, -Mnoidiom |
21089 | 15.1 | PGI Accelerator | OpenACC: C++ does not detect illegal schedule Corrected the compiler to report an error if the number of gangs is specified with gang appearing within acc parallel |
21081 | 15.1 | Fortran | pgfortran incorrectly reports the warning, "INTENT(IN) argument cannot be defined" Corrected the compiler to not detect an error for an assignment to a dummy argument with the POINTER and INTENT(IN) attributes |
21080 | 15.1 | Fortran | pgfortran reports the error, "Internal compiler error. conv_deallocate: unexpected ast" with -Mallocatable=03 Corrected the handling of assignments when the left-hand side is a section of an allocatable component |
21079 | 15.1 | Fortran | pgfortran execution error of subprogram containing INTENT(OUT) allocatable arguments Corrected the deallocation of the allocatable components of the argument corresponding to an INTENT(OUT) allocatable argument |
21078 | 15.1 | Fortran | pgfortran OpenMP program containing a COPYPRIVATE of a local array fails Corrected the copying of the local arrays appearing in a COPYPRIVATE clause within an orphaned subprogram |
21077 | 15.1 | PGI Accelerator | pgfortran reports the error, "Internal compiler error. sem_err104:unk doif->ID", for a missing $ACC directive Corrected the checking for a missing closing construct of a $ACC loop directive |
21064 | 15.1 | PGI Accelerator | Code containing a call to a chain of sequential routines fails to executes Corrected how scalars are passed as arguments to sequential routines |
21063 | 15.1 | PGI Accelerator | CUDA Fortran example fails with "pgnvd-Fatal-Could not spawn nvlink.exe" on Windows Corrected nvlink's processing of certain object and library files |
21040 | 15.1 | PGI Accelerator | Compiler reports the error, "Internal compiler error. Create: Unexpected atomic store opcode" Corrected the handling of acc atomic capture of assignments of the form v=x++; |
21018 | 15.1 | Fortran | pgfortran code containing a call to a type bound procedure fails Corrected calling a type bound procedure whose name is the same as a type bound procedure in the base type |
21016 | 15.1 | Fortran | pgfortran code containing a call to a type bound procedure fails Corrected calling a type bound procedure whose name is the same as a type bound procedure in the base type |
21004 | 15.1 | PGI Accelerator | pgfortran reports the error, "Compiler failed to translate PGI Accelerator region" Added support for converting integer*1 and integer*2 to single precision float |
21001 | 15.1 | Fortran | pgfortran positioning errors with STREAM files Corrected applying the POS specifier when the STREAM file is at end-of-file |
20956 | 15.1 | PGI Accelerator | CUDA Fortran code incorrect for multiple shared automatic arrays Corrected the computations of the bounds of multi-dimensioned automatic arrays appearing in a global subprograms |
20945 | 15.1 | Fortran | pgfortran reports the error, "Internal compiler error. Errors in Lowering", with -fast Corrected calling a subprogram via a procedure pointer component from an internal procedure |
20942 | 15.1 | C/C++ | Insufficient debug information for the pgc++ STL containers Corrected the compiler to emit the correct names for templated classes |
20940 | 15.1 | PGI Accelerator | pgcc OpenACC produces invalid debug information Corrected the debug information for the argument of a kernel which corresponds to a local array |
20936 | 15.1 | C/C++ | pgc++ should support the g++ option, -std=c++11 Added the option -std=c++11 in addition to the option, --c++11 |
20930 | 15.1 | Fortran | pgfortran code containing a call to a type bound procedure causes a fatal compilation Corrected calling a type bound procedure which has type deferred character |
20905 | 15.1 | Profiler | pgcollect fails on OpenSuse 13.1 with the error, "dwf_init: Incompatible Dwarf Version 4" Corrected pgcollect to accept object files containing Dwarf information of versions greater than 2 |
20904 | 15.1 | Profiler | pgprof does not display kernel region information Corrected the handling of unknown file/routine names |
20888 | 15.1 | PGI Accelerator | Compiling -Mcuda and -ta=nvidia causes the error, "nvlink fatal : Could not find fatbin" Corrected the processing of code containing only a data region |
20866 | 15.1 | PGI Accelerator | OpenACC C++ fails to translate PGI Accelerator region when using a static class as a member of another class member Corrected the evaluation of a loop trip count for an acc loop containing static class members |
20854 | 15.1 | PGI Accelerator | Add the cusparse pgfortran module Added the pgfortran cusparse .mod file to the installation |
20852 | 15.1 | Fortran | pgfortran code containing a call to a type bound procedure fails Corrected calling a type bound procedure whose name is the same as a type bound procedure in the base type |
20839 | 15.1 | Fortran | pgfortran reports the error, "Internal compiler error. expected derived type member in initialization assignment", on code containing a FORALL Corrected the processing of a structure constructor containing array constructors being assigned to a FORALL-subscripted array object |
20832 | 15.1 | All | Support for -Mipa when performing parallel makes Updated the documentation to state that IPA is not compatible with parallel make environments |
20816 | 15.1 | Fortran | pgfortran reports the error, "Cannot invoke Abstract Interface gather" Corrected adding the pass object when calling a type bound procedure |
20813 | 15.1 | All | Add the silent install feature to the network installs of PGI products Updated the makelocalrc and install commands to include the silent install capability |
20812 | 15.1 | PGI Accelerator | OpenACC PGI_ACC_TIME gives incorrect output on Windows Corrected the implementation of PGI_ACC_TIME |
20798 | 15.1 | C/C++ | pgcpp and pgc++ -E on a pgfortran program emits the output twice Corrected the pgcpp and pgc++ drivers to invoke the preprocessor just once on a fortran input file |
20786 | 15.1 | Fortran | The pgfortran compiler should not add the .f or .f90 suffix to the name of the fortran file specified by the user No longer have the compiler attempt to add a suffix to the name of the input file |
20774 | 15.1 | PGI Accelerator | OpenACC fails to parallelize loop without the presence of the pcopy clause Corrected handling how an array is passed to a routine within an acc parallel loop |
20769 | 15.1 | Fortran | pgfortran code containing a call to a type bound procedure fails with the error "Too many arguments specified" Corrected the processing of type bound procedures declared NOPASS |
20745 | 15.1 | PGI Accelerator | OpenACC code using COLLAPSE and KERNEL produces incorrect answers Corrected the processing of nested acc loop with the collapse clause |
20738 | 15.1 | 64-bit C/C++ | Preserve IEEE minus 0.0 in the PGI compilers Updated the c, c++, and fortran compilers and run-time to preserve the sign of IEEE minus 0.0 |
20732 | 15.1 | PGI Accelerator | OpenACC emits the error, "Unexpected flow graph" Corrected the construction of loops containing conditionals and calls to routines |
20694 | 15.1 | PGI Accelerator | Data assigned via a struct pointer is not copied back to the host Corrected the analysis of a struct assignment within a acc parallel loop |
20692 | 15.1 | PGI Accelerator | Operators from the C++ Complex class implemented as method functions returning complex fail Corrected how functions returning complex are presented |
20691 | 15.1 | PGI Accelerator | C/C++ complex operations prevents parallelization of loops Improved the data flow analysis of real and imaginary parts of complex variables |
20687 | 15.1 | PGI Accelerator | Data assigned via a struct pointer is not copied back to the host Corrected the analysis of a struct assignment within a acc parallel loop |
20550 | 15.1 | PGI Accelerator | Execution error occurs in OpenACC code using the declare directive Corrected the handling of declaring device_resident allocatable arrays |
20468 | 15.1 | PGI Accelerator | CUDA Fortran: data initialization of double precision variables is incorrect Corrected how double precision scalar variables are statically initialized |
20160 | 15.1 | PGI Accelerator | Add the cusparse pgfortran module Added the pgfortran cusparse .mod file to the installation |
20152 | 15.1 | PGI Accelerator | Automatic arrays not supported in CUDA Fortran and OpenACC acc routines Added the support for automatic arrays |
19964 | 15.1 | Fortran | Compiling a C++ program -fast cause the compiler to crash Corrected how the end condition of certain loops is computed by the vectorizer |
19790 | 15.1 | PGI Accelerator | ACC code causes pgcc to crash Corrected the flow-analysis when certain variables are missing from the present clause |
19789 | 15.1 | PGI Accelerator | ACC - request for method to allocate temporary arrays in subroutine kernels Added the support for automatic arrays |
19785 | 15.1 | PGI Accelerator | ACC parallel loop sum reduction is incorrect Corrected the computation of a sum reduction when the variable is integer |
19582 | 15.1 | PGI Accelerator | OpenACC: acc kernels with loop reduction does not produce correct results Corrected the processing of adjacent acc loops specifying a reduction of the same variable |
19493 | 15.1 | Fortran | DFPORT missing implementation of hostnm Removed hostnm from the documentation of DFPORT |
19491 | 15.1 | PGI Accelerator | Compilation error occurs when derived type component appears in a data clause Added the support of references to derived components to the data clause |
19451 | 15.1 | PGI Accelerator | OpenACC code causes pgCC to fault Corrected the handling of a acc loop containing a relatively large number of assignments |
19278 | 15.1 | PGI Accelerator | CUDA Fortran code causes the compiler to crash Corrected the handling of a name collision of a user variable and a CUDA fortran default module |
19252 | 15.1 | PGI Accelerator | OpenACC question on collapse Updated the documentation to clarify the usage and implications of the collapse clause |
19050 | 15.1 | PGI Accelerator | ACC: conflicting "loop is parallelizable message" Corrected the processing of a acc loop seq within a acc parallel |
19015 | 15.1 | All | -Bstatic should imply -mp=nonuma on Linux Added nonuma.o to the link if -mp and -Mconcur are not on the link with -Bstatic |
18684 | 15.1 | Fortran | pgfortran preprocessor needs to support c99 preprocessing features Upgraded the preprocessor used by pgfortran with c99 features |
18431 | 15.1 | Debugger | pgdbg does not allow program arguments to be passed that begin with a hyphen ('-') Updated the documentation to indicate that arguments beginning with '-' need to be passed via the '-program_args' option |
18267 | 15.1 | PGI Accelerator | ACC: struct in PGI Accelerator region causes the "Unsupported reference" error Added support for structs appearing in acc region |
18136 | 15.1 | PGI Accelerator | CUDA Fortran with large data requirements fails to execute Corrected the size computations of certain arrays to be allocated on the device |
18135 | 15.1 | PGI Accelerator | OpenACC emits the error, "Internal compiler error. ACC CG: could not find ALV index for symbol" Corrected the flow analysis of certain acc loops |
17802 | 15.1 | Manufacturing | Add OS-specific feature names for the tools Added the OS type to the feature names of the licensed PGI tools |
17295 | 15.1 | All | Installation errors occur if libnuma.so.1 is present Corrected the installation process to take into account the presence of libnuma.so or libnuma.so.1 |
17271 | 15.1 | 64-bit C/C++ | pgc++ and pgcpp reports an error when the c99 function, fwprintf, is invoked Added fwprintf to the c++ header files |
17199 | 15.1 | PGI Accelerator | OpenACC reports the error, " Internal compiler error. expecting array datatype" Added the support of derived type array components in data clauses |
3963 | 15.1 | Fortran | IEEE + and - 0.0 tests fail Updated the c, c++, and fortran compilers and run-time to preserve the sign of IEEE minus 0.0 |