PGI Release 2014 Technical Problem Reports
What problems have been fixed in PGI 2014?
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 |
|---|---|---|---|
| 20400 | 14.10 | PVF | MPI debugging in PVF may hang Corrected the availability of mpiexec.exe |
| 20772 | 14.10 | PGI CDK | The CDK MVAPICH does not support InfiniBand (OFED) Updated the supported versions of MVAPICH2 and OFED in the CDK |
| 20779 | 14.10 | C/C++ | pgc++ compiled code fails to catch exception Adding the missing exception region for a lone function pointer |
| 20811 | 14.10 | PGI Accelerator | pgcpp faults compiling incorrect OpenACC code Improved the processing of erroneous OpenACC c++ data clauses |
| 20822 | 14.10 | Fortran | pgfortran faults compiling code -O2 Corrected the optimization of subprograms containing numerous pointer assignments |
| 20903 | 14.10 | Fortran | pgfortran faults compiling code containing derived types with type bound procedures from multiple modules Corrected the processing of a type bound procedure binding name used in different contexts |
| 20796 | 14.9 | Fortran | pgfortran "Internal Compiler Error" occurs with code containing move_alloc
Corrected move_alloc when transferring an allocatable component between types |
| 20784 | 14.9 | C/C++ | Add the g++ push_macro/pop_macro pragmas to pgc++
Implemented the g++ push_macro/pop_macro pragmas |
| 20758 | 14.9 | Fortran | pgfortran compilation error occurs with -Mallocatable=03
Corrected the assignment to an allocatable component when the parent of the component is subscripted |
| 20748 | 14.9 | Fortran | pgfortran intrinsic IBITS(I,POS,LEN) fails when LEN is 0
Corrected the computation of IBITS when the length argument is 0 |
| 20742 | 14.9 | C/C++ | "File with unknown suffix" warnings occur for shared objects named libname.so.0.0.0
Updated the driver to delete the warning before passing the file to the linker |
| 20734 | 14.9 | CUDA Fortran | CUDA Fortran fails to compile code containing many GOTO statements
Corrected the detection of loops containing blocks with multiple entries |
| 20727 | 14.9 | Fortran | Execution error occurs in pgfortran code containing ASSOCIATE
Corrected the processing of ASSOCIATE when the selector is an array expression |
| 20720 | 14.9 | Fortran | pgfortran reports the error, "Internal compiler error. Errors in ILM file 1"
Corrected the handling of ASSOCIATE when the selector is a subscripted component |
| 20719 | 14.9 | PGI Accelerator | Correct the acc loop default in the presence of routine
Corrected the default for routine to be acc routine sequential |
| 20716 | 14.9 | Fortran | pgfortran fails to compile code containing a generic whose specific is a module procedure with the same name
Corrected the declaration of a generic defined with multiple interfaces blocks |
| 20710 | 14.9 | PGI Accelerator | Eliminate the cuCtxAttach error when using cuda-gdb
Added the option noattach to -ta=tesla |
| 20708 | 14.9 | PGI Accelerator | OpenACC fails to compile code containing the __ldg routine accessing "bool" member variables
Extended the types allowed for the arguments of the __ldg routine |
| 20693 | 14.9 | PGI Accelerator | -Minfo output does not occur for classes declared in header files
Improved the collection of file and line information for code in header files |
| 20688 | 14.9 | PGI Accelerator | OpenACC -Minfo output is not the same when a C++ class is changed to a struct
Corrected the association of line and file information with code from include files |
| 20681 | 14.9 | 64-bit Fortran | Windows linking conflicts occur with the floating point environment routines
Removed the fp environment routines from the libpgc library |
| 20680 | 14.9 | 64-bit Fortran | pgfortran fails to compile code containing FORALLs
Corrected the handling of array-valued functions appearing as arguments to MATMUL in FORALLs |
| 20661 | 14.9 | PGI Accelerator | OpenACC c++ emits the error, "multiple definition of local value named 'li61_t0"
Corrected the naming of certain compiler temporaries |
| 20658 | 14.9 | C/C++ | Add the g++ __attribute__ ((section ("name"))) feature to pgc++
Implemented the section naming __attribute__ |
| 20647 | 14.9 | PGI Accelerator | OpenACC reports the error, "Illegal address during kernel execution"
Corrected referencing large data structures in a kernel |
| 20642 | 14.9 | Fortran | pgfortran compilation error occurs for code containing type bound procedures
Corrected the error checking of the declarations of type-bound procedures |
| 20587 | 14.9 | PGI Accelerator | OpenACC reports the error, "Internal compiler error. Unsupported CMP"
Corrected the handling of 64-bit integer comparisons |
| 20579 | 14.9 | PGI Accelerator | Incorrect results when calling cuFFT from OpenACC
Corrected synchronizing certain operations when -Mcuda and -acc are used |
| 20539 | 14.9 | PGI Accelerator | OpenACC emits the message, "Compiler failed to translate accelerator region (see -Minfo messages): Load of NULL symbol"
Corrected allocating device-resident module allocatables |
| 20525 | 14.9 | PGI Accelerator | OpenACC fails to compile kernel code accessing derived types containing allocatable components
Improved the processing of data clauses containing derived type component references |
| 20305 | 14.9 | CUDA Fortran | CUDA Fortran performance regression of kernel code referencing a module allocatable array
Improved the memory placement of the descriptors for module allocatable array objects |
| 20173 | 14.9 | PGI Accelerator | OpenACC fails to detect certain max reductions
Corrected processing of a reduction clause in collapsed loops |
| 20066 | 14.9 | Fortran | pgfortran reports the error, "Subscript, substring, or argument illegal in this context"
Corrected the handling of ASSOCIATE when a subscripted reference appears as a term in the selector expression |
| 19975 | 14.9 | PGI Accelerator | OpenACC code with ROUTINE copying data despite data region
Improved the handling of ROUTINEs with assumed-shape array arguments |
| 19921 | 14.9 | C/C++ | pgcc #pragma weak conflicts with initialized external variables
Inhibit generating "weak" declarations for data-initialized global variables |
| 19448 | 14.9 | Fortran | Undefined references to IEEE and ISO_C related objects occur with pgfortran -Mcuda -Bdynamic on Windows
Corrected the generation of MODULE DLL export information on Windows |
| 19115 | 14.9 | CUDA Fortran | Execution errors occur for CUDA Fortran compiled code containing device data assignments
Corrected the generated code for performing a device to device copy for non-allocatable module variables |
| 18726 | 14.9 | PGI Accelerator | OpenACC should report an error when an update directive appears in a compute region
Updated the compiler to detect the presence of an update clause in a compute region and report an error |
| 16726 | 14.9 | CUDA Fortran | CUDA Fortran the appearance of DEVICE variables in an EQUIVALENCE statement
Updated the compiler to detect the use of a device variable as an equivalence object and report an error |
| 20663 | 14.7 | Fortran | Use of CONTIGUOUS with INTENT(IN) reported as an error Corrected the set of attributes which may appear with CONTIGUOUS |
| 20650 | 14.7 | PGI Accelerator | pgcpp reports the error, "main.cpp", line 30: internal error: coalesce_and_lookup_generalized_identifier: not identifier" Corrected the declaration of an array of structs appearing in a data clause |
| 20632 | 14.7 | Fortran | Illegal FMA instruction generated Corrected the handling of certain expressions during the FMA generation |
| 20613 | 14.7 | 64 bit C/C++ | Request to add more g++ features to pgc++ Added the definition of __gnu_linux__ and suppressed the -mp and pthreads warning |
| 20596 | 14.7 | PGI Accelerator | pgc++ fails to compile code containing cache directive Corrected the processing of subarray expression appearing in the cache directive |
| 20593 | 14.7 | Documentation | Specify the supported OpenMP version in the documentation Updated the documentation and FAQs to include the supported OpenMP version |
| 20586 | 14.7 | C/C++ | Incorrect value returned by a function containing a VLA Corrected the processing of a return statement within a lexical block containing a VLA declaration |
| 20585 | 14.7 | C/C++ | Report an error for a VLA declared static Disallow the appearance of extern and static on a VLA declaration |
| 20584 | 14.7 | PGI Accelerator | Incorrect array bounds in an -Minfo message with -Mlarge_arrays Corrected the presentation of certain bounds of an array with -Mlarge_arrays |
| 20576 | 14.7 | Documentation | Correct the reference to "float" in the CUDA Fortran manual Changed the reference to "float" to "REAL" |
| 20565 | 14.7 | Fortran | pgfortran reports the error, "Could not resolve generic procedure" Corrected the resolution of a generic which has an external procedure specific with the same name |
| 20546 | 14.7 | Installation | Create a silent install option for the Linux installers Added the silent install option to the Linux installation process |
| 20543 | 14.7 | Fortran | pgfortran-compiled OpenMP code does not execute correctly Corrected the handling of DEFAULT(PRIVATE) for certain variables appearing multiple times in a statement |
| 20537 | 14.7 | Fortran | Fortran function returning a procedure pointer fails to compile Corrected checking the type bound procedure semantics on a function returning procedure pointer |
| 20526 | 14.7 | Fortran | pgfortran fails to allow a legal empty structure constructor Corrected the handling of an item-less structure constructor when the all of the derived type components have default initializers |
| 20523 | 14.7 | PGI Accelerator | OpenACC code executes incorrectly using a loop with fixed iteration count in routine seq Corrected the handling of loops which can be expressed with constant loop counts |
| 20503 | 14.7 | PGI Accelerator | OpenACC reduction loops with very large trip counts execute incorrectly Corrected the computation of large loop counts whose values are greater larger than maximum value of a 32-bit signed int |
| 20490 | 14.7 | C/C++ | pgc++ --c++11 fails to compile certain gnu 4.8 header files Upgrade the pgc++ header files to be compatible with gnu 4.8 |
| 20482 | 14.7 | PGI Accelerator | OpenACC: routine vector width defaulting to 1 Corrected the default vector width |
| 20472 | 14.7 | 64-bit Fortran | pgfortran code containing method call fails at execution Corrected the handling of a scalar pointer and class type |
| 20409 | 14.7 | PGI Accelerator | Kernel launch failure when using "-Mcuda=nordc" Corrected referencing module data in other modules |
| 20408 | 14.7 | PGI Accelerator | Link errors occurs when using CUDA Fortran in a library Corrected the naming of the CUDA-specific section in the object file produced by the compile |
| 20370 | 14.7 | PGI Accelerator | MVAPICH fails to build when -acc is used Corrected the MPVAPICH build configuration |
| 20235 | 14.7 | Fortran | Array assignment of a structure constructor fails Corrected the processing of certain structure constructors containing multiple allocatable components |
| 20172 | 14.7 | CUDA Fortran | CUDA Fortran fails to compile code containing OpenMP threadprivate Corrected the handling of threadprivate declarations in CUDA global and device routines |
| 20077 | 14.7 | All | Undefined reference occurs on OS X when linking in certain archives Corrected the referencing of USE-associated modules placed in archives |
| 20072 | 14.7 | CUDA Fortran | CUDA Fortran fails to compile a module initializing a constant complex variable Corrected the initialization of complex values |
| 20067 | 14.7 | PGI Accelerator | pgfortran code does not execute correctly Corrected the handling of a kernel containing an acc reduction loop and an acc loop which uses the value of the reduction variable |
| 19975 | 14.7 | PGI Accelerator | Excessive data movement for ROUTINE with data region Improved calling a routine with assumed-shape array arguments |
| 17771 | 14.7 | PGI Accelerator | CUDA: Add support for device allocatables within user defined types A derived type object cannot be device if the type contains device allocatables; however, the object can be declared managed |
| 20418 | 14.6 | Fortran | Windows drivers do not handle wildcards
Corrected the drivers to expand wildcards on the command line |
| 20406 | 14.6 | Fortran | pgfortran reports the error, "attempt to redefine symbol" with conflated class entities.
Corrected the handling of an overloaded type parameter |
| 20405 | 14.6 | C/C++ | pgcc reports the error, "Illegal type conversion required" , for a designated initializer
Corrected the processing of a designated initializer of a struct member |
| 20403 | 14.6 | PGI Accelerator | OpenACC linking fails with "multiply defined symbols 'acc_fortran_constructor_0"
Corrected the generation of accelerator constructors for multiple files |
| 20378 | 14.6 | PGI Accelerator | OpenACC: logical scalar value not updated each time a kernel is launched
Corrected passing a scalar logical variable as an argument to a kernel |
| 20375 | 14.6 | PGI Accelerator | OpenACC: using a reduction variable of type bool fails
Corrected the implementation of reductions when the reduction variable is type bool and other 1-byte integer types |
| 20337 | 14.6 | 64-bit C++ | OSX: FreePGI emits the warning, "directory not found", when linking with -fpic
Added to the installation the necessary directory which is searched for -fpic |
| 20313 | 14.6 | 32-bit Fortran | pgfortran reports an error for a type bound procedure with same name as a procedure in another module
Corrected the processing of a dummy passed argument which does not match the data type of its type bound procedure |
| 20305 | 14.6 | PGI Accelerator | CUDA Fortran performance issue with allocating module allocatables from within device code
Optimized the placement of the descriptors for module allocatable arrays |
| 20212 | 14.6 | Fortran | Linking error for containing a recursive type specified in a module that has the SAVE attribute
Corrected the processing of a POINTER with the SAVE attribute whose type is a recursive derived type |
| 20211 | 14.6 | Fortran | pgfortran compilation error for a generic binding to an abstract type
Corrected the processing of a generic binding in a module which does not contain any module procedures |
| 20207 | 14.6 | Fortran | pgfortran reports an error for a valid structure constructor
Corrected the processing of a procedure pointer appearing in a structure constructor |
| 20203 | 14.6 | Fortran | pgfortran reports an error for a valid structure constructor
Corrected the processing of a procedure pointer appearing in a structure constructor |
| 20202 | 14.6 | Fortran | pgfortran faults when compiling an illegal module
Added a check for an argument-less type bound procedure that is not declared NOPASS |
| 20200 | 14.6 | Fortran | pgfortran faults after reporting an error for an illegal object in an ALLOCATE statement
Corrected the compiler to inhibit checking the source expression when an error was detected for the object |
| 20190 | 14.6 | Fortran | pgfortran reports the error, "Lowering Error: array upper bound is not a symbol for datatype 176"
Corrected the use of the SIZE intrinsic in the bounds of certain array dummy arguments |
| 20128 | 14.6 | Fortran | pgfortran code containing allocatable class types fails to compile
Corrected the processing of move_alloc whose arguments are allocatable class types |
| 20066 | 14.6 | Fortran | pgfortran reports the error, "Subscript, substring, or argument illegal in this context for theta"
Corrected the processing of certain associate selectors containing subscripted references |
| 19997 | 14.6 | PGI Accelerator | Using kernels region over sequential code fails
Corrected the processing of certain assignments when the right-hand side uses variables which were defined on the device |
| 19990 | 14.6 | Fortran | pgfortran reports the error, "Internal compiler error. Mkexpr1:bad id 14", for uses of type bound operators
Corrected the handling of generic operators when the passed argument is the first argument |
| 19878 | 14.6 | Fortran | pgfortran code containing pointer assignments does not execute correctly
Corrected certain pointer assignments when both the left hand and right hand sides contain strided sections |
| 19604 | 14.6 | PGI Accelerator | CUDA Fortran device allocation fails on Windows
Corrected the run-time support of allocating device data on Windows |
| 17296 | 14.6 | C/C++ | Misleading pgCC --remark output for OpenMP
Corrected the handling of --remark in the presence of OpenMP constructs |
| 17023 | 14.6 | C/C++ | pgCC code using sstream fails
Corrected the STL support of sstream |
| 14947 | 14.6 | 32-bit Fortran | Windows drivers do not handle wildcards
Corrected the drivers to expand wildcards on the command line |
| 20210 | 14.4 | Fortran | Compilation errors occur for TYPEIS Corrected the processing of certain TYPEIS expressions |
| 20173 | 14.4 | PGI Accelerator | Kernel not generated for a max reduction Corrected the handling of reduction clauses in collapsed loops |
| 20170 | 14.4 | Fortran | Optimization error occurs for a target of a pointer assignment Corrected the analysis of pointers and their targets |
| 20167 | 14.4 | 64-bit Fortran | pgfortran reports the error, "Internal compiler error. compute_initial_aryaddr: addr not found" Corrected the optimization of certain conditionals within loops which may be vectorized |
| 20141 | 14.4 | Fortran | Optimization error occurs for an assignment of two arrays with the pointer attribute Corrected the optimization of an array copy when the arrays have the pointer attribute |
| 20108 | 14.4 | Fortran | Debug information is missing for a Fortran dummy argument with -gopt Corrected the processing of dummy arguments for -gopt when the optimization level is less than two |
| 20065 | 14.4 | Fortran | pgfortran reports the error, "Ambiguous type bound procedure output in generic set for .write" Corrected the processing of generic type bound procedures for defined I/O |
| 20057 | 14.4 | Fortran | pgfortran reports the error, "Interface is not compatible with parent's interface for type bound procedure" Corrected the handling of a type bound procedure when the child's type bound procedure is not overloaded |
| 19996 | 14.4 | Fortran | pgfortran reports the error, "Result is not compatible with parent's result for type bound procedure name" Corrected the processing of a child type bound procedure when its result is an adjustable-length character |
| 19989 | 14.4 | PGI Visual Fotran | PVF fails to install when VC++ redistributables are present Corrected the installer |
| 19981 | 14.4 | PGI Accelerator | PGI_ACC_NOTIFY and PGI_ACC_TIME give confusing information for large arrays Corrected the data size in PGI_ACC_NOTIFY and corrected PGI_ACC_TIME to present the number of data transfers |
| 19968 | 14.4 | Fortran | Invalid instruction, vpcmpneq_oqq, generated by the vectorizer Corrected the handling of certain integer comparisons within loops which are vectorized |
| 19965 | 14.4 | C/C++ | OpenMP code with C99 for fails to compile Corrected the handling of an OpenMP C99 for when it appears in an orphaned function |
| 19932 | 14.4 | PGI Accelerator | Zeroing a complex array fails on the GPU Corrected cudamemset when the object is a complex array |
| 19915 | 14.4 | Fortran | Invalid hoist of a reference to an array element Corrected the handling of guarded references to an allocatable array |
| 19823 | 14.4 | PGI Accelerator | max reduction fails to execute correctly Corrected the handling of unsigned int index expressions |
| 19797 | 14.4 | PGI Accelerator | Add examples of OpenACC routine Added examples of the OpenACC routine directive to the installation |
| 19792 | 14.4 | 32-bit Fortran | pgcc IPA unreferenced externals on 32-bit Windows, Added entry points for the Windows' Interlock routines to the libc IPA library |
| 19639 | 14.4 | PGI Accelerator | pgfortran -Mcuda reports the error, 'a value of type "void" cannot be assigned to an entity of type "double"' Corrected the processing of certain return types of Fortran functions |
| 19283 | 14.4 | C/C++ | pgc++ --c++11 constexpr and function templates fail to compile Corrected the processing of c++11 lambda STL template definitions |
| 19140 | 14.4 | Fortran | pgfortran INQUIRE by SIZE fail. Corrected the processing of the SIZE specifier of an INQUIRE statement |
| 18992 | 14.4 | Packaging | .so libraries in the static library directory Moved the shared libraries to the libso directory |
| 18026 | 14.4 | Fortran | 3F fstat64 fails on win32 & win64 Corrected the 3F fstat64 routine to retrieve the proper size of the system's file stat structure |
| 14267 | 14.4 | Fortran | pgf90 -mp fails with 'unknown intrinsic function' and ICE 'Errors in Lowering' Corrected the |
| 17814 | 14.3 | C/C++ | Alignment of a struct/union unaffected when followed by__declspec(aligned(16)) Corrected the processing of __declspec appearing after the struct/union |
| 19761 | 14.3 | Installation | Licensing fails on systems using the Consistent Network Device Naming convention Added support for "biosdevname" to the licensing utilities |
| 19775 | 14.3 | Fortran | pgfortran - Minline reports the error, "Internal compiler error, readin_func:can't scan 1st line" Corrected the inliner when source paths contain spaces |
| 19782 | 14.3 | Fortran | Comparing NaNs fails at -O2 or higher optimizations with -Kieee Corrected the constant evaluation of floating comparisons when -Kieee is present |
| 19795 | 14.3 | Fortran | pgfortran fails to compile type-bound procedure named "size" Corrected the appearance of intrinsic names as binding names |
| 19796 | 14.3 | Fortran | pgfortran reports the error, "Lowering Error: array upper bound is not a symbol for datatype" Corrected the use of certain intrinsics in the bounds of an array declaration when -i8 is present |
| 19812 | 14.3 | PGI Accelerator | Certain complex arithmetic operations in accelerator regions are incorrect Corrected the handling of complex scalar arguments |
| 19860 | 14.3 | PGI Accelerator | acc parallel loop reduction produces incorrect results Corrected the handling of multiple variables appearing in a reduction clause |
| 19863 | 14.3 | CUDA Fortran | pgfortran -Mcuda=llvm reports the error, "parse error: multiple definition of local value ..." Corrected the processing of a kernel containing expressions of type complex |
| 19877 | 14.3 | Fortran | pgfortran -fast -i8 reports the error, "Internal compiler error. make_cvect_cmp_vili(): bad predicate" Corrected the handling of 64-bit integer compares in if statements when analyzing loops for vectorization |
| 19834 | 14.2 | Windows | awk and other Cygwin commands fail to execute Corrected the installation of Cygwin on all Window platforms |
| 19821 | 14.2 | 64-bit Fortran | pgfortran test fails to compile with '-mcmodel=medium -O2' Corrected the flow analysis of certain allocatable and assumed-shape arrays |
| 19810 | 14.2 | PGI Accelerator | OpenACC: use of a user defined derived type in an accelerator region fails Corrected handling of derived types when appearing in multiple kernels |
| 19792 | 14.2 | 32-bit Fortran | Use of IPA on Win32 generates references to undefined functions Added IPA information for various Interlocked Windows functions |
| 19779 | 14.2 | 64-bit C/C++ | pgcpp --use_pch and --pch_dir report 'Catastrophic error: cannot open precompiled header file "dir"' Corrected the handling of the directory path for the options use_pch and pch_dir |
| 19778 | 14.2 | Fortran | pgfortran derived type assignment fails when there is a pointer and an allocatable component Corrected the processing of the allocatable assignment semantics for a derived type containing a pointer component |
| 19773 | 14.2 | PGI Accelerator | "OpenACC: pgfortran reports the error, "Internal compiler error, import_const: unknown constant datatype" Corrected the processing of module variables having the POINTER and TEXTURE attributes" |
| 19756 | 14.2 | Fortran | "pgfortran reports the error, "Internal compiler error. mk_mem_ptr_shape: no static desc for pointer" Corrected the processing of certain derived types containing allocatable scalar components" |
| 19690 | 14.2 | Fortran | pgfortran compilation error on code containing parameterized derived types Corrected the handling of type parameters which have the same name as entities outside of the type |
| 19654 | 14.2 | Fortran | Memory leaks in I/O statements Updated the run-time's exit processing to cleanup various data structures allocated by the I/O run-time |
| 19371 | 14.2 | CUDA Fortran | CUDA Fortran: compiler needs to report an error when host data is used in device code. Added error checks for host (MODULE) date being used in a device or global subprogram |
| 18741 | 14.2 | CUDA Fortran | CUDA Fortran: compiler needs to report an error when host data is used in device code. Added error checks for host (MODULE) date being used in a device or global subprogram |
| 4281 | 14.1 | Fortran | RECL specified in OPEN has no effect on list-directed writes Corrected the handling of list-directed writes if the RECL specifier was present in the corresponding OPEN |
| 17397 | 14.1 | 64-bit Fortran | The presence of internal procedures may cause a performance degradation Improved the analysis of a host subprogram containing internal procedures |
| 17879 | 14.1 | Fortran | Add F2008 kind values to the iso_fortran_env module Implemented the f2008 iso_fortran_env kind constant such as int8, int16, int32, int64, real32, real64 |
| 17898 | 14.1 | Windows Fortran | Poor formatted read performance on Windows Minimized the use of the Windows file locking read routines |
| 18076 | 14.1 | PGI Accelerator | Host local variables with the REFLECTED attribute are copied in for an acc region in an internal procedure Corrected the processing of REFLECTED host-associated variables when in an internal procedure |
| 18174 | 14.1 | PGI Accelerator | Used error leads to pgfortran reporting "Internal compiler error: pgnvd job exited with nonzero status …" , "_data_ecc_gpu_0" is undefined Improved error checking of host symbols used in global subroutines |
| 18190 | 14.1 | Profiler | CCFF messages not presented with pgprof -feedbackonly Removed the Feedback Only Mode option from the File | Open dialog |
| 18250 | 14.1 | Fortran | Non-advancing reads fail on certain files Corrected the handling of the last record of a sequential formatted file which is not terminated with the newline character |
| 18448 | 14.1 | Fortran | pgfortran does not produce an error when allocating an already allocated object Added an allocated check for an ALLOCATE of an allocatable object |
| 18509 | 14.1 | Fortran | -gopt generates incorrect dwarf for Fortran arrays Improved the compiler and debugger to convey that debug information for arrays is not available |
| 18852 | 14.1 | 64-bit Fortran | pgfortran reports errors on legal array PARAMETER declarations Corrected the handling of array constructors containing derived type member references |
| 18894 | 14.1 | PGI Accelerator | Report when reduction is generated via the reduction clause Added the Minfo message for a user-specified reduction |
| 18901 | 14.1 | PGI Accelerator | Errors with the host_data clause Corrected the handling of the host_data clause |
| 19023 | 14.1 | PGI Accelerator | Add Compiler Feedback message when reduction clause is used. Added the Minfo message for a user-specified reduction |
| 19346 | 14.1 | OpenMP | Random failures occurred for device code appearing in OpenMP parallel regions Corrected the creation of certain device variables for code appearing in an OpenMP context |
| 19438 | 14.1 | PGI Accelerator | pgfortran reports the error, "error: expression must have class type" Corrected the handling of complex variables appearing in a present_or_copy clause |
| 19549 | 14.1 | 32-bit Fortran | Excessive compilation time of pgfortran -O2 on Windows. Corrected the handling of a large number of pointer assignments when optimizations are enabled |
| 19553 | 14.1 | Fortran | pgfortran reports the error, "Lowering Error: symbol city$p is a member reference", when -Mallocatable=03 is used Corrected the handling of allocatable deferred character components |
| 19560 | 14.1 | C/C++ | C code gets wrong answer with -fast Corrected data flow analysis of certain assignments for loops containing early exits |
| 19580 | 14.1 | Fortran | Compatibility error with the bits/types.h include file Corrected types.h to guard against using the type, __SWBLK_T_TYPE, not being defined |
| 19584 | 14.1 | PVF | PVF installs may fail silently Added a check to the installer to determine if the PVF compilers and tools were installed correctly |
| 19613 | 14.1 | 32-bit Fortran | pgfortran reports an error when an empty derived type is used to declare an allocatable or pointer object Corrected the compiler to allow zero-size derived types |
| 19623 | 14.1 | Fortran | pgfortran incorrectly reports argument type mismatch errors Corrected certain cases where actual arguments were composed of empty array constructors |
| 19624 | 14.1 | Fortran | pgfortran incorrectly reports the error, "illegal use of derived type" when processing generics Corrected the handling of arguments passed to a user-defined generic with the same name as a type bound procedure |
| 19633 | 14.1 | Fortran | pgfortran incorrectly reports the error, "Adjustable array can not have automatic bounds specifiers", with -i8 Corrected the checking of specification expressions in adjustable arrays |
| 19639 | 14.1 | PGI Accelerator | pgfortran reports the error, " error: a value of type "void" cannot be assigned to an entity of type "double" Corrected the processing of external functions |
| 19647 | 14.1 | Fortran | pgfortran code fails when compiled -O2 Corrected an error with loop fusion |
| 19658 | 14.1 | Fortran | pgfortran ASSOCIATE failing with array sections Corrected the subscripted use of an associated name when its selector is an array section |
| 19660 | 14.1 | Fortran | pgfortran pointer remapping fails Corrected pointer assignments when both the pointer and target contain section expressions |
| 19692 | 14.1 | Fortran | pgfortran -Mbyteswapio fails Corrected the generation of the trailing length fields of an unformatted record |
| 19702 | 14.1 | Fortran | pgfortran -Mbyteswapio fails Corrected the generation of the trailing length fields of an unformatted record |
| 4138 | 14.1 | Fortran | pgfortran OpenMP code fails at -O2 Corrected the optimizer's processing of threadprivate local variables |
| 16835 | 14.1 | Windows Fortran | pgfortran mpi module is not present on Windows Added the MS_MPI Fortran module to the Windows installation |
| 17642 | 14.1 | Fortran | OpenMP atomic operations fail in the presence of recursive functions Corrected the handling of atomic operations which are recursive |
| 19216 | 14.1 | C/C++ | Correct the implementation of pow() as documented in ISO/IEC 9899:TC3 Corrected the implementation of pow to raise FE_INVALID as required by ISO/IEC 9899 |
| 19390 | 14.1 | Fortran | pghpf reports errors of the form "ILM file line 21: expecting …" Corrected the internal handling of certain constructs within pghpf |
| 19413 | 14.1 | C/C++ | Include file glob.h references missing symbol __REDIRECT_NTH Define __REDIRECT_NTH when the system gcc is version 4.7 or greater |
| 19492 | 14.1 | Fortran | pgfortran code fails during execution when performing a namelist read Corrected the handling of certain arrays in a namelist group |
| 19510 | 14.1 | Fortran | pgfortran source allocation fails when the source is a scalar Allow an ALLOCATE of an array when its source is scalar |
| 19544 | 14.1 | C/C++ | pgCC generates incorrect names for templated classes in the debug information Corrected the debug information to use the template class names rather than the mangled names |
| 19562 | 14.1 | Fortran | pgfortran code compiled -Mvect produces incorrect results Corrected the vectorization of loops containing multiple non-stride-1 array references |
| 19600 | 14.1 | All | Exiting an MVAPICH program may not terminate all of the slave processes Upgraded to a newer version of MVAPICH which corrects the exit processing |
| 19625 | 14.1 | C/C++ | pgCC reports multiple "no instance of function template" errors Added support for log2f and exp2f as built-ins |
| 19629 | 14.1 | Fortran | pgfortran incorrectly reports the error, "Illegal procedure interface - obs_unary_int" Corrected the handling of certain procedure pointer components |
| 19635 | 14.1 | PGI Accelerator | CUDA-x86-SDK example MonteCarloMultiGPU fails to compile Corrected the steps to compile the example |
| 19661 | 14.1 | Fortran | -gopt dwarf for allocatable is missing pointer type Corrected the debug generation of allocatables when -gopt is used |
| 19679 | 14.1 | Fortran | pgfortran reports the error, "ILM file: missing subtype for pointer datatype 74" Corrected the processing the type and subtype of procedure pointers |
| 19680 | 14.1 | Fortran | pgfortran reports errors of the form "symbol '…_td' is already defined" Corrected the processing of handling of host derived types within internal procedures |
| 19691 | 14.1 | Fortran | pgfortran reports the error, "illegal procedure interface - obs_unary_real" Corrected the handling of certain procedure pointer components |
| 19701 | 14.1 | CUDA Fortran | Error with passing an array element to a CUDA Fortran kernel within an PGI Accelerator region Corrected subscripting a device copy of an array |
| 19709 | 14.1 | PGI Accelerator | host_data of allocatable arrays fails Corrected the processing of allocatable arrays appearing in the host_data clause |
| 19710 | 14.1 | C/C++ | pgCC reports the error, "assertion failed: remap_ptr_to_entry_number: pointer in secondary "trans unit" Corrected the processing of the omp barrier pragma |
| 19741 | 14.1 | Fortran | pgfortran reports the error, "Internal compiler error. bld_lhs, ast nyd" with -acc Corrected the handling of a local allocatable passed to an internal procedure |
| 19742 | 14.1 | Fortran | pgfortran reports the error, "Internal compiler error. bld_lhs, ast nyd" at -O2 Corrected the handling of a local allocatable passed to an internal procedure |
| 19757 | 14.1 | Installation | Installation instructions for Network install need to be corrected Corrected the path information in the install script |
