PGI Release 2010 Technical Problem Reports
What problems have been fixed in PGI 2010?
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 | Release | Compiler/Other | Symptoms/Description |
|---|---|---|---|
| 16194 | 10.9 | Fortran |
USE ONLY does not exclude other module entities when rename is present
Corrected the selection of public modules in a nested USE context with 'only' and '=>'. |
| 16195 | 10.9 | Fortran |
Renamed USE variable cannot be locally declared
Corrected the compiler to allow a renamed use-associated variable to be locally declared. |
| 16197 | 10.9 | Fortran |
USE rename hides variables from subsequent USE statements
Corrected the selection of public modules in a nested USE context with 'only' and '=>'. |
| 16741 | 10.9 | 64-bit C/C++ |
pgcc reports the error, "Use of undeclared variable", for valid C source
Corrected the processing of array formal arguments using VLA syntax. |
| 16834 | 10.9 | Fortran |
pgfortran fails to compile f2003 code
Implemented the f2003 features for type bound procedures, source allocation, and ASSOCIATE. |
| 16926 | 10.9 | 64-bit Fortran |
pgfortran reports the error, "Internal compiler error. Subscript for array is out-of-bounds"
Corrected the use of reshape in an initialzation expression when its argument is an implied-do. |
| 16990 | 10.9 | C/C++ |
pgcpp failures with "omp parallel for schedule(static, 1) collapse(2)"
Corrected the compiler's handling of schedule(static, 1). |
| 17051 | 10.9 | Fortran |
pgfortran reports the error, "Internal compiler error. no matching ref for symbol"
Corrected the handling of various comparisons within accelerator regions. |
| 17165 | 10.9 | 32-bit Fortran |
Compiler reports an error for BIND(C) applied to a derived type
Fixed the compiler to allow a derived type to have the BIND(C) attribute. |
| 17173 | 10.9 | C/C++ |
pgcpp reports the error "Internal compiler error. flowgraph: node is zero"
Corrected the handling of nested sizeof expressions containing ?: expressions as arguments. |
| 17174 | 10.9 | Fortran |
pgfortran reports the error "Could not resolve generic procedure"
Corrected the handling of extended class types when resolving generic functions. |
| 17180 | 10.9 | Fortran |
pgfortran fails to compile example -g
Corrected the debug generation of extended type variables. |
| 17185 | 10.9 | C/C++ |
pgcc generates spurious warnings for comparisons
Fixed the type checks of arrays and strings as an operand in "?:" where the other operand is the integer value 0. |
| 17188 | 10.9 | Debugger |
pgfortran debug information missing for arrays
Corrected the debug generation for use-associated arrays. |
| 17191 | 10.9 | Fortran |
pgfortran reports the error, "CUDALIB is not properly set", when -ta=nvidia:cc20 is used
Corrected which CUDA toolkit to use when compiling for compute capability 2.0. |
| 17193 | 10.9 | 32-bit Fortran |
pgfortran compilation fails
Corrected calling a procedure component procedure without arguments. |
| 17195 | 10.9 | Fortran |
Fortran non-advancing read fails on Windows
Corrected run-time's handling of redirected stdin. |
| 17196 | 10.9 | Fortran |
pgfortran -Mcuda reports the error, "Internal compiler error. unsupported procedure"
Inhibit collapsing array assignments into memset/memmove calls in device code. |
| 17198 | 10.9 | Fortran |
pgfortran -mp reports the error, "Internal compiler error. stack error while lowering"
Corrected the handling of allocatable array assignments in the presence of the workshare construct. |
| 17201 | 10.9 | 64-bit Fortran |
Possible memory leak in pgfortran-compiled code
Corrected the handling of intent(out) arguments which contain allocatable components. |
| 17205 | 10.9 | Fortran |
!dir$ ignore_tkr not recognized when ! begins in a column other than 1
Fixed the compiler to allow !dir$ ignore_tkr to begin in a column other than 1 |
| 17206 | 10.9 | Debugger |
pgdbg gui does not pass command line arguments
Corrected pgdbg to pass the command-line arguments to the program being executed. |
| 17213 | 10.9 | Fortran |
pgfortran -mp reports the error, "Internal compiler error. size_of:bad dtype 0"
Corrected the use of use-associated threadprivate allocatable arrays in an OpenMP copyin clause. |
| 17214 | 10.9 | Fortran |
pgfortran OpenMP missing implicit barrier at the end of a sections construct
Corrected the compiler to generate a barrier at the end of a sections region when the nowait clause is absent. |
| 3799 | 10.8 | All |
Same name is used for scratch files created on multiple MPI nodes
Corrected the naming scheme for scratch files. |
| 16605 | 10.8 | Accelerator |
CUDA - illegal use of the constant attribute
The compiler reports an error when the constant attribute appears in a host subprogram. |
| 16628 | 10.8 | Accelerator |
CUDA - compiler crash with complex declarations
Corrected the computation of 2**n for nvcc |
| 16733 | 10.8 | Accelerator |
CUDA:compiler crashes when using derived types in device code
Corrected the processing of user generic procedures in device code. |
| 16895 | 10.8 | Accelerator |
CUDA: using cudaSetDevice results in the error "setting the device when a process is active is not allowed"
Corrected the processing of cudaSetDevice in module procedures. |
| 16912 | 10.8 | Accelerator |
pgfortran reports the error "unrec dtype in cp_iregarg "
Corrected the compiler's handling of derived type members in reductions. |
| 16948 | 10.8 | All |
Using -ta=analysis will give undefined reference errors
Corrected the processing of the -ta=analysis feature. |
| 16949 | 10.8 | Fortran |
pgfortran reports the error, "make_sec_from_ast: shape null"
Corrected the processing of dot_product within a block forall. |
| 16959 | 10.8 | Accelerator |
CUDA: errors cudaMalloc3D
Corrected the implementation of the routine. |
| 16991 | 10.8 | Debugger |
Request for better pgdbg access of Fortran PARAMETERs
Correct the debug information for USE-associated PARAMETERs. |
| 16995 | 10.8 | Accelerator |
pgfortran -ta=nivida reports the error, "could not find use for ili"
Corrected the handling of the bounds of USE-associated allocatable arrays within an accelerator region. |
| 17004 | 10.8 | 64-bit Fortran |
>2GB unformatted files cannot be created on win64
Added support to the Fortran I/O run-time for large files. |
| 17049 | 10.8 | C/C++ |
Incorrect result of a reinterpret_cast expression
Corrected the code generation of certain reinterpret_cast expressions. |
| 17064 | 10.8 | Fortran |
pgfortran reports the error, "missing interface for type bound procedure"
Corrected the processing of certain type bound procedures without arguments. |
| 17073 | 10.8 | C/C++ |
Incorrect optimization of const char pointer
Corrected the target of the const attribute. |
| 17075 | 10.8 | 64-bit C/C++ |
pgcc asm statements fails
Corrected use of asm operands in 64-bit registers. |
| 17081 | 10.8 | Fortran |
pgfortran crashes on code containing a FORALL
Corrected certain internal data structures used when processing FORALLs. |
| 17084 | 10.8 | Accelerator |
Accelerator example fails
Corrected the evaluation of a loop count for certain accelerated loops. |
| 17095 | 10.8 | PVF |
PVF 2010 fails to install without warning
The installer now prevents installation of PVF 2010 on systems that do not support VS 2010. |
| 17096 | 10.8 | Accelerator |
Incorrect answers for user code
Corrected the collapsing of dimensions for copying arrays between the host and device. |
| 17098 | 10.8 | Accelerator |
CUDA - add shared version of the cudafor library
Added the shared cudafor library to the installation. |
| 17102 | 10.8 | Profiler |
pgcollect config file location confusion: 'etc' vs. 'cuda'
Corrected the search of the cudaprof config file. |
| 17104 | 10.8 | Fortran |
BTEST(I, POS) incorrect
Corrected BTEST of 64-bit integers when the default logical is logical(4). |
| 17105 | 10.8 | Accelerator |
Sum reductions in Monte Carlo example incorrect
The reduction code was missing a __syncthreads() call. |
| 17112 | 10.8 | 64-bit Fortran |
Honor memory binding and implement MP_BLIST wraparound
When MP_BIND is set to yes, do not override the effects of numactl --membind; wraparound the thread binding of MP_BIND. |
| 17114 | 10.8 | Accelerator |
pgfortran -ta=nvidia creates incorrect assembly
Corrected the evaluation of a 64-bit loop count. |
| 17116 | 10.8 | 32-bit Fortran |
pgfortran example causes 32-bit compilers to seg fault.
Corrected resolving a user generic procedure to certain function calls. |
| 17079 | 10.8 | 64-bit Fortran |
pgfortran reports the error, "bad ast optype in expression"
Corrected the allocatable semantics of an assignment to an allocatable component. |
| 17109 | 10.8 | Fortran |
CUDA Fortran request for the pinned attribute on a derived type component
Added the support of the pinned attribute on allocatable components. |
| 17113 | 10.8 | Fortran |
pgfortran reports the error, "bad ast optype in expression"
Corrected the allocatable semantics of an assignment to an allocatable component. |
| 17123 | 10.8 | Fortran |
-mp -Mipa=fast causes "IPA fatal error … multiply defined …"
Added OpenMP omp_ function names to the IPA library. |
| 16401 | 10.6 | Documentation |
Add PVF example for calling Fortran from C/C++
Added a PVF sample and updated the documentation. |
| 16533 | 10.6 | 64-bit Fortran |
OpenMP ordered fails when DO index is integer*8
Corrected the compiler and run-time to handle 64-bit integer index expressions. |
| 16649 | 10.6 | Fortran |
Component of derived type has PRIVATE attribute
Implemented the f2003 feature of mixed component accessibility. |
| 16692 | 10.6 | Fortran |
Predefine a preprocessor variable for CUDA Fortran
Define the preprocessor variable, _CUDA, when -Mcuda is present or when the suffix of the source input file is .cuf or .CUF. |
| 16810 | 10.6 | Fortran |
OpenMP task causes compiler to abort
Corrected the compiler's processing of nested tasks. |
| 16815 | 10.6 | Fortran |
OpenMP DO with COLLAPSE fails
Updated the compiler to handle OpenMP DO COLLAPSE when CYCLE is present. |
| 16824 | 10.6 | C/C++ |
Application fails compiled -O2 and above
Corrected the compiler handling of guarded pointer dereferences within loops containing switch statements. |
| 16831 | 10.6 | Fortran |
Compilation fails for the use the log and exp intrinsics in an initializer
Implemented the f2003 feature where certain elemental numerical intrinsics in initialization expressions. |
| 16841 | 10.6 | 64-bit Fortran |
User code fails at -O2 and above
Corrected the compiler's merging of array assignments when one array was used as a subscripting in an ensuing array statement. |
| 16874 | 10.6 | Fortran |
INT(z'000000ffffffffff',i8) produces wrong answers
Corrected the computation of an INT intrinsic when the argument is a large BOZ literal and a kind argument indicates KIND=8. |
| 16889 | 10.6 | Fortran |
No error for allocatable in derived type without deferred shape
Report an error if an array component has deferred shape and does not the allocatable or pointer attribute. |
| 16890 | 10.6 | Fortran |
CUDA Fortran device scalar allocation not hoisted out of loop
Corrected the compiler to properly allocate device scalars. |
| 16892 | 10.6 | Fortran |
Add support for "__threadfence" to CUDA Fortran
Implemented. |
| 16896 | 10.6 | Accelerator |
CUDA Fortran program fails with "ConfigureCall FAILED: 9"
Corrected the compiler to report an error for a PARAMETER which is typed differently in a later statement. |
| 16899 | 10.6 | C/C++ |
pgcpp doesn't detect repeated default(shared) clauses
Corrected the compiler to report an error. |
| 16910 | 10.6 | 64-bit Fortran |
Accelerator compiler reports the error "pgnvd job exited with nonzero status code"
Corrected the compiler. |
| 16940 | 10.6 | Fortran |
pgfortran reports the error "stack error while lowering"
Corrected the compiler's processing of nested tasks. |
| 16941 | 10.6 | Fortran |
pgfortran reports the error "Illegal attribute - conflict with volatile"
Corrected the compiler to allow bind(c) and volatile as a legal combination of attributes. |
| 16942 | 10.6 | Fortran |
PUBLIC attribute on an extended type not working
Allow PUBLIC/PRIVATE to appear after the EXTENDS clause. |
| 16946 | 10.6 | Fortran |
Module defining two extensions of a type fails to compile
Corrected the compiler. |
| 16950 | 10.6 | 64-bit Fortran |
pgfortran reports the error "Constant expression of wrong data type" for legal code
Corrected the compiler's processing for an initialization expression for an array PARAMETER of derived type. |
| 16954 | 10.6 | Installation |
Install script deletes user defined TMPDIR
Corrected the installation process. |
| 16955 | 10.6 | Installation |
2010 symbolic links not updated to latest release
Corrected the installation process. |
| 16957 | 10.6 | Fortran |
Illegal context error reported for an OpenMP PARALLEL DO
Corrected the compiler to allow combined workshare directives within MASTER, CRITICAL, and ORDERED regions. |
| 16996 | 10.6 | C/C++ |
pgcc -Mbounds reports the error "illegal use of void type"
Corrected the -Mbounds processing for an extern array with an empty extent. |
| 16997 | 10.6 | C/C++ |
pgcc code fails -O2 and above
Corrected the compiler's use of a register in a loop which is not preserved across function calls |
| 17003 | 10.6 | 64-bit Fortran |
User code causes pgf902 to seg fault
Corrected the compiler's management of internal data structures when processing calls with several (100+) arguments. |
| 17005 | 10.6 | Fortran |
-pgcpplibs is missing -lcpp on OSX
Updated the driver to add -lcpp when -pgcpplibs is present on OSX. |
| 16274 | 10.5 | Fortran |
Name clash between subroutine and variable causes the error "descriptor does not exist"
Corrected the processing of internal procedures with dummy arguments which have the same names as components in the host routine. |
| 16402 | 10.5 | 64-bit Fortran |
ACC data region causes compiler to seg fault
Corrected the handling of variables appearing in copyin/copyout lists multiple times. |
| 16453 | 10.5 | Fortran |
OpenMP code gets the error, "flowgraph: node is zero"
Improved the compiler's handling of a critical section appearing as deadcode. |
| 16508 | 10.5 | Fortran |
Format string with missing ')' not detected.
Corrrected the I/O runtime to report an error if a format string is missing the terminating right parenthesis. |
| 16569 | 10.5 | C/C++ |
Threadprivate class being handled as shared
Enforce the OpenMP rule requiring threadprivate pragmas to lexically precede all references to the variables in the list. |
| 16574 | 10.5 | 32-bit Fortran |
pgfortran MATMUL seg faults on 32-bit windows
Corrected the win32 version of the matmul routine. |
| 16668 | 10.5 | Fortran |
User example produces wrong answers with -fast (-Mvect=sse)
Corrected the vectorizer's detection of reductions. |
| 16689 | 10.5 | C/C++ |
Preprocessing of file by pgcpp fails to define __GNU_SOURCE
Updated pthread.h to enable defining __GNU_SOURCE. |
| 16711 | 10.5 | 64-bit C/C++ |
OpenMP C hangs -reduction involving large array & OMP_STACKSIZE set
Corrected the mp run-time to properly allocate large stacks. |
| 16723 | 10.5 | C/C++ |
pgcpp OpenMP failure for "#pragma omp parallel for collapse(3)"
Detect 'for inits' of parallel fors. |
| 16731 | 10.5 | C/C++ |
sched.h errors with gcc 4.3
Corrected sched.h for systems where gcc 4.3 is installed. |
| 16738 | 10.5 | All |
-ta=nvidia or -Mcuda with Japanese locale creates bad assembler source
Corrected the generated code to accommodate locales using multi-byte characters. |
| 16740 | 10.5 | Fortran |
Erroneous SAVEd automatic array error "assumed size array with dummy arg""
Corrected the error message when applying the SAVE attribute to automatic arrays. |
| 16741 | 10.5 | 64-bit C/C++ |
pgcc reports the error, "Use of undeclared variable", for valid C source
Corrected the processing of array formal arguments using VLA syntax. |
| 16745 | 10.5 | 64-bit Fortran |
BTEST in code compiled -Mstandard fails with "Expression has wrong data type"
Corrected the handling of the BTEST intrinsic when the arguments are integer(kind=1). |
| 16747 | 10.5 | C/C++ |
OpenMP pgcpp error for the sequence for ..., #pragma omp task, one-line assignment
Accounted for multiple OpenMP pragmas when processing the private declarations of a block statement. |
| 16750 | 10.5 | C/C++ |
pgcpp rejects "#pragma omp task if (0)"
Corrected the handling of a constant if expression. |
| 16752 | 10.5 | C/C++ |
-zc_eh needs proper cleanup code outside of try blocks
Corrected zc_eh processing of cleanup code outside of try blocks. |
| 16757 | 10.5 | 64-bit C/C++ |
pgcpp incorrectly uses the movapd instruction non-16-byte aligned data
Corrected accessing a by-value struct argument consisting of 2 long double members. |
| 16758 | 10.5 | C/C++ |
OpenMP pgcpp not rejecting barrier within a task
Report an error if a barrier appears within a task. |
| 16760 | 10.5 | C/C++ |
OpenMP pgcpp does not report errors for non-conforming constructs
Improved the detection of errors for illlegal uses of flush, barrier, and taskwait. |
| 16762 | 10.5 | C/C++ |
OpenMP pgcpp does not report an error for a taskwait following a label
Report an error if a taskwait is preceded by a label. |
| 16769 | 10.5 | C/C++ |
pgcc syntax error occurs for a very long pragma line
Corrected the preprocessor so that a pragma is not split into multiple lines. |
| 16776 | 10.5 | 64-bit Fortran |
OpenMP example hangs when stacksize > 2GB
Corrected the mp run-time to properly allocate large stacks. |
| 16799 | 10.5 | Fortran |
CUDA fortran pow raised to integer not supported
Added support for raising a floating value to an integer power. |
| 16809 | 10.5 | Fortran |
False -mp error "assumed size array must appear in a SHARED or PRIVATE clause"
Apply the rule that an assumed-size array is always shared. |
| 16811 | 10.5 | Fortran |
Namelist END problem
Corrected the detection of "end" or "error" for namelist I/O. |
| 16812 | 10.5 | Fortran |
Legal fixed format code fails with pgfortran
Corrected the use of derived type object whose name begins with "parameter". |
| 16817 | 10.5 | C/C++ |
Compiler faults with -Mpre
Improved analysis for the PRE optimization. |
| 16819 | 10.5 | 64-bit Fortran |
Allocatable array descriptor not being set properly
Corrected the access of an allocatable which has the TARGET attribute. |
| 16821 | 10.5 | Fortran |
ieee_other not supported
Added support of the ieee_other rounding mode. |
| 16825 | 10.5 | 64-bit Fortran |
Compile time too long when using -Mpre
Improved analysis for the PRE optimization. |
| 16829 | 10.5 | 64-bit Fortran |
OpenMP code which changes the number of threads fails
Corrected allocation of the data structures needed for dynamic scheduling. |
| 16830 | 10.5 | Fortran |
pgfortran should accept the extension .f03 for source
Added f03 as a suffix for fortran files. |
| 16863 | 10.5 | Accelerator |
Request to add "acc_get_device_num"
Added the acc_get_device_num function to return the device number on which the code is executing. |
| 16887 | 10.5 | 32-bit Fortran |
Linker reports the warning "alignment 8 of symbol ... is smaller than 16 in ..."
Corrected the alignment of data initialized variables in a module. |
| 16732 | 10.4 | All |
Threads do not inherit floating point control status word on OSX and Windows
Corrected the run-time to propagate the master thread's value of the MXCSR to the slave threads. |
| 16720 | 10.4 | 64-bit Fortran |
Win64: incorrect value for INTEGER*8 **
Corrected the computation of ** for 64-bit integers. |
| 16691 | 10.4 | 64-bit Fortran |
CUDA Fortran argument mismatch errors
Corrected the processing of character assignments. |
| 16687 | 10.4 | Fortran |
CUDA Fortran shared scalar variables are not working
Added support of the shared attribute for scalars |
| 16676 | 10.4 | Fortran |
Add -Mcuda=nofma flag
Added the new option. |
| 16674 | 10.4 | C/C++ |
pgCC file does not compile -mp
Corrected the process of locally scoped for index variables within a parallel for. |
| 16670 | 10.4 | 64-bit Fortran |
Compiler reports the error, 'support members', for an accelerator region code
Corrected the handling of complex operations within a region. |
| 16667 | 10.4 | Fortran |
SYSTEM_CLOCK does not reset to 0 when it overflows
Reset the clock if the value of count exceeds the maximum. |
| 16659 | 10.4 | Fortran |
f90 OpenMP example reports the error, 'redundss:mark unexpected AST type 4'
Corrected passing a private allocatable to an internal procedure. |
| 16654 | 10.4 | Fortran |
f90 code reports the error 'symbol has not been explicitly declared'
Corrected declaring a variable which also appears as a component and a dummy argument in an interface. |
| 16653 | 10.4 | Fortran |
CUDA Fortran does not report an error when using automatic arrays in device code
Added the check for automatic arrays in a device routine. |
| 16648 | 10.4 | Fortran |
f90 code reports the error 'sptr_of_subscript: not a subscript '
Corrected the processing of elemental functions whose arguments are arrays of derived types containing allocatable components. |
| 16643 | 10.4 | Fortran |
F2003 procedure pointer component with explicit interface fails
Corrected the TKR processing of routines with procedure pointers appearing as dummy arguments. |
| 16642 | 10.4 | Fortran |
CUDA Fortran global subroutine should not be inlined
Corrected the inliner to only allow device routines to be inlined into device/global routines. |
| 16632 | 10.4 | Fortran |
real(ktype), parameter:: lil0=(mu*1836.0/gma2)**0.5 causes error
Added the support of ** in initialization expressions. |
| 16607 | 10.4 | Fortran |
OpenMP lastprivate does not return expected value
Corrected the last value in the presence of a chunk expression in the schedule clause. |
| 16593 | 10.4 | Fortran |
OpenMP copyin of threadprivate allocatable fails
Corrected copyin/copyprivate for threadprivate variables with the allocatable attribute. |
| 16591 | 10.4 | 64-bit Fortran |
User module naming clash
Corrected the naming of module variables to prevent conflicts with the names of user global symbols. |
| 16586 | 10.4 | Fortran |
idamax prefetch incorrect
Corrected the address being prefetched in the main loop of idamax. |
| 16560 | 10.4 | Fortran |
CUDA Fortran code reports the error I'unsupported operation 311' due to missing VALUE attribute
Corrected pass-by-value for device routines. |
| 16544 | 10.4 | C/C++ |
C++ --zc_eh wrong processing for throwing functions outside of try
Account for throwing functions appearing outside of a try block. |
| 16507 | 10.4 | Fortran |
OpenMP program gives wrong answers
Corrected the declaration of private allocatables. |
| 16450 | 10.4 | 64-bit Fortran |
Vectorization being prevented by call to sqrtf()
Added the builtin recognition of float libm functions. |
| 16438 | 10.4 | Fortran |
User requests the '.ftn'/'.FTN' suffix
Corrected the driver to allow .ftn and .FTN. |
| 16406 | 10.4 | Fortran |
f90 code reports the error 'sptr_of_subscript: not a subscript'
Corrected the processing of elemental functions whose arguments are arrays of derived types containing allocatable components. |
| 14928 | 10.3 | Fortran |
Enhancement Request - Mod Files should convey presence of data initialization
Enhanced the mod files to require their object files at link-time when initializations are present |
| 15989 | 10.3 | Fortran |
Fortran Write statements are extremely slow to compile
Greatly reduced the compile-time of a subprogram containing a large number of WRITE statements |
| 16276 | 10.3 | C/C++ |
pgcc -Xa option, the "empty macro arguments" warning should not apply
pgcc no longer reports the warning for system header files |
| 16380 | 10.3 | 64-bit Fortran |
pgfortran reports the error "Errors in Lowering 1"
Added a compile-time check for static arrays which are too large |
| 16396 | 10.3 | Fortran |
CUDA Fortran reports an internal error for a global subroutine without value attribute
Corrected the default argument passing for global subroutines |
| 16417 | 10.3 | C/C++ |
--instantiation_dir, --prelink_objects' switches need to be deprecated
Corrected pgcpp to produce the "deprecated" warning for --instantiation_dir, --one_instantiation_per_object, and --template_info_file |
| 16450 | 10.3 | 64-bit Fortran |
Vectorization being prevented by call to sqrtf()
Enabled vectorization of loops containing calls to the libm float functions |
| 16487 | 10.3 | Fortran |
Need to force generation of external reference for module data
Enhanced the mod files to require their object files at link-time when initializations are present |
| 16493 | 10.3 | Fortran |
pgfortran reports the error "dwo_emit_threadprivate_location'" with -mp -g
Corrected the compiler to report an error for a THREADPRIVATE variable missing the SAVE attribute |
| 16504 | 10.3 | Fortran |
OpenMP 3.0 violation not detected as an error.
Corrected the compiler to report an error for a THREADPRIVATE variable missing the SAVE attribute |
| 16506 | 10.3 | C/C++ |
Add the g++ atomic update intrinsics to pgcpp.
Added support of _gnu_cxx::__atomic_add and __exchange_and_add to pgcpp's omp.h |
| 16507 | 10.3 | Fortran |
Incorrect behavior for OpenMP private allocatables
Corrected the default allocation for allocatables declared private |
| 16509 | 10.3 | Fortran |
CUDA Fortran doesn't support amod
Implemented the support for the amod intrinsic |
| 16513 | 10.3 | 64-bit Fortran |
pgfortran reports the error "Errors in Lowering 44"
Corrected then deallocate of a nested derived type containing allocatables |
| 16515 | 10.3 | C/C++ |
pgcpp fstream failure
Corrected fstream open() to reset is error status |
| 16527 | 10.3 | Fortran |
pgfortran reports the error "array upper bound is not a symbol for datatype"
Corrected then deallocate of a nested derived type containing allocatables |
| 16528 | 10.3 | Fortran |
pgfortran reports the error "Illegal use of symbol wp - KIND parameter"
Corrected a scoping issue with a KIND parameter which is also the name of a component |
| 16535 | 10.3 | Fortran |
pgfortran seg faults on ACC code
Corrected the processing of accelerator reductions |
| 16538 | 10.3 | Fortran |
Compiling CUDA program -Msave causes internal compiler error
Do not apply the -Msave option to device variables |
| 16547 | 10.3 | Fortran |
system_clock() count_max incorrect when compiled -i8
Corrected the maximum count with the count_max is integer(8) |
| 16549 | 10.3 | Fortran |
Errors with the Fortran 2003 iso_fortran_env module routines, iostat_end and iostat_eor
Corrected the values returned by iostat_end() and iostat_eor() |
| 16560 | 10.3 | Fortran |
CUDA pgfortran reports the error "unsupported operation 311" due to missing VALUE attribute
Corrected the default argument passing for global subroutines |
| 16566 | 10.3 | 64-bit Fortran |
Errors with arrays of large derived types
Corrected the compilers when subscripting arrays of large derived types under -mcmodel=medium |
| 16567 | 10.3 | 64-bit Fortran |
Fortran PAUSE on win64 does not pause
Corrected the run-time to report: enter <return> or <ctrl> d to continue |
| 16585 | 10.3 | Fortran |
ACC pgfortran reports the error "semsmp: bad rednum"
Corrected the compiler's processing of ACCUPDATE |
| 16589 | 10.3 | C/C++ |
The link error "undefined reference to `.LB2498'" occurs for code compiled pgcc -O3 -g
Corrected the debug generation for certain optimized loops |
| 16594 | 10.3 | C/C++ |
pgcc -mp compilation errors with sched.h
Updated sched.h to be usable on SLES 11 systems |
| 16598 | 10.3 | Accelerator |
Error with the copy clause for ACC code
Corrected the loop count used for the cache-filling assignments |
| 16602 | 10.3 | Fortran |
OpenMP completion of tasks fails
Corrected the run-time to ensure tasks are completed by the time the program exits |
| 16618 | 10.3 | Fortran |
CUDA pgfortran reports the error "Const space overflowed"
Added a check for a 64KB limit on static arrays |
| 16619 | 10.3 | Fortran |
move_alloc of allocatable character causes severe error
Corrected the compiler's processing of the move_alloc subroutine |
| 16620 | 10.3 | 64-bit C/C++ |
asm test fails with pgcc
Corrected the processing of asm unsigned long arguments |
| 16626 | 10.3 | Fortran |
-Bstatic_pgi should apply to -Mmpi=mpich1,mppich2,mvapich1
Updated the drivers to apply static linking to the supplied mpi libraries |
| 16638 | 10.3 | All |
Implement -Mmpi=mpich2, -Mmpi=hpmpi, -Mmpi=mvapich1
Added the options to the drivers |
| 16640 | 10.3 | Fortran |
Wrong answers for ACC code
Corrected the processing of cached array copies |
| 16641 | 10.3 | Fortran |
Request for "-Mcuda=keepgpu"
Implemented -ta=nvidia,keepgpu and -Mcuda=keepgpu |
| 16655 | 10.3 | C/C++ |
OpenMP pgcpp example reports the error a "pragma omp barrier may only be contained in a block statement"
Corrected the detection of where a barrier may occur |
| 16656 | 10.3 | All |
Problems with MPI shared libraries
Corrected the installation of the supplied MPI shared libraries |
| 15899 | 10.2 | 64-bit Fortran |
pgfortran reports the error "unsupported operation 74" with -i8 -ta nvidia
pgfortran now supports 64-bit integers for the accelerator. |
| 16200 | 10.2 | Documentation |
Add 'Accelerator Installation' information to the CDK install instructions
Updated the documentation. |
| 16381 | 10.2 | 64-bit Fortran |
pgfortran reports the error "expression must be a modifiable lvalue"
Corrected the handling of objects manged in the accelerator's shared memory. |
| 16382 | 10.2 | 64-bit Fortran |
pgfortran reports the error, "Unaligned memory accesses not supported"
Corrected the handling of objects manged in the accelerator's shared memory |
| 16397 | 10.2 | Fortran |
CUDA Fortran missing intrinsics
Added support for the ceiling, dim, floor, and modulo intrinsics |
| 16415 | 10.2 | Fortran |
pgfortran example fails with "Signalled SIGABRT at 0x3918830215"
Corrected the processing of maxval of an array section in a specified dimension. |
| 16435 | 10.2 | Fortran |
CUDA Fortran modulo function doesn't allow gridDim%x as argument
Corrected the compiler's handling of the modulo function. |
| 16441 | 10.2 | 64-bit Fortran |
pgfortran reports the error "unsupported datatype"
Corrected the accelerator compiler's handling of index arrays. |
| 16442 | 10.2 | Fortran |
CUDA Fortran simple code using modulo fails at -O2 and higher
Corrected the handling of variables appearing as arguments to modulo and on the left-hand side of assignments. |
| 16446 | 10.2 | Fortran |
incorrect kind for the result of bit intrinsics
Corrected the kind for the bit intrinsics, such as iand, when the arguments are integer(1) or integer(2). |
| 16460 | 10.2 | C/C++ |
C++ --zc_eh error during nested template processing
Corrected the processing of constructors/destructors which contain throwing functions as arguments. |
| 16466 | 10.2 | ALL |
Update CDK product with mvapich built on 1.4 OFED
Corrected the install of mvapich with the 1.4 version of OFED. |
| 16479 | 10.2 | 64-bit Fortran |
pgfortran reports cuMemcpy2D and cuMemcpyDtoH errors
Corrected the accelerator compiler's handling of copyin and copyout. |
| 16483 | 10.2 | Accelerator |
pgaccelinfo feature requests
Enhanced pgaccelinfo with -short and -quiet flags and corrected the command's return status. |
| 16484 | 10.2 | Fortran |
pgfortran reports "unsupported operation 224" for ishft
Corrected the accelerator compiler's processing of ishft when the shift count is negative. |
| 16494 | 10.2 | Installation |
install script performs 'chmod' on more directories than it is installing
Updated the install script to chmod the directory just installed. |
| 16500 | 10.2 | Fortran |
pgfortran reports "launch" error due to bad value in cache initialization
Corrected the initialization of cache copies for the accelerator. |
| 16512 | 10.2 | Fortran |
CUDA Fortran should check for missing chevron
The compiler issues an error message if the name of a global subroutine is used in a call without chevrons. |
| 16514 | 10.2 | Fortran |
pgfortran -r8 -i8 does not type amod correctly
Corrected the result type of the amod intrinsic when -r8 and -i8 are present. |
| 16517 | 10.2 | 32-bit Fortran |
pgfortran accelerator code for summing integers is incorrect
Corrected the acclerator code for of an integer summation. |
| 16537 | 10.2 | Accelerator |
CUDA program gives wrong answers with -fast
Corrected the handling of variables appearing as arguments to modulo and on the left-hand side of assignments. |
| 4354 | 10.1 | C++ |
C++ gnu compatibility with template containing an unnamed enum
Corrected the name mangling of certain templates |
| 15550 | 10.1 | C++ |
Compiler option to support gcc's -fexceptions.
With --zc_eh, exception code can unwind through C routines |
| 16041 | 10.1 | C++ |
C++ --zc_eh can't catch simple class type
Corrected the destruction of local variables in the context of copy constructors and throw |
| 16095 | 10.1 | Fortran |
pgf90 fails to evaluate an array extent as 'constant'
Corrected the compiler to evaluate a subscripted array parameter as a constant |
| 16170 | 10.1 | C++ |
pgcpp issues the error "Illegal context for ordered"
Corrected the use of the ordered clause in an omp for |
| 16171 | 10.1 | C++ |
C++ omp for static schedule has no effect ion C++ iterators
Applied the static schedule to iterators |
| 16173 | 10.1 | C++ |
Floating point exception for code compiled -mp
Corrected the use of pointers as omp for variables |
| 16176 | 10.1 | Fortran |
pgf90 reports the error "size_of:bad dtype"
Corrected the use of a vector subscript for an array or component array of derived type |
| 16220 | 10.1 | Documentation |
Example wrong for CUDA Fortran
Updated the example. |
| 16221 | 10.1 | 64-bit C/C++ |
GPU example fails
Corrected the processing of global scalar live-out variables |
| 16369 | 10.1 | 32-bit Fortran |
Feature request - f2003 POS specifier for data transfers
Added the support for POS in INQUIRE and READ/WRITE statements |
| 16377 | 10.1 | C++ |
Preprocessor error occurs when compiling Scipy
Corrected an error with preprocessing embedded concatenations |
| 16378 | 10.1 | 64-bit Fortran |
-Msmartalloc=hugebss not working
Corrected the implementation of hugebss |
| 16379 | 10.1 | Fortran |
pgf90 reports the error "k may not appear in a PRIVATE clause"
Allow variables, which are also dummy arguments to statement functions, to appear in a PRIVATE clause |
| 16384 | 10.1 | 32-bit Fortran |
CUDA Fortran doc not included in linux tar file
Updated the tar file |
| 16391 | 10.1 | Fortran |
CUDA Fortran reports the error ".i0002, has not been explicitly declared"
Updated the compiler's processing of loop temporaries |
| 16393 | 10.1 | 64-bit C/C++ |
Scipy code gets relocation errors when compiled -fpic
Corrected the detection of static const functions which were still referenced |
| 16394 | 10.1 | Fortran |
pgf90 reports the error "process_forall: not a FORALL"
Corrected the processing of a function call in the right-hand side of a forall assignment |
| 16397 | 10.1 | Fortran |
CUDA Fortran numeric intrinsics are missing some implementations
Added the support for ceiling, dim, floor, and modulo |
| 16398 | 10.1 | Documentation |
CUDA Fortran doc has kernel invocation wrong
Corrected the example. |
| 16403 | 10.1 | Fortran |
pgf90 ignored or unrecognized import statement
Corrected IMPORT of a renamed use-associated object |
| 16404 | 10.1 | Fortran |
pgf90 reports the error "attempt to CALL a FUNCTION" for procedure pointers
Updated the compiler's processing of CALL to subroutines via procedure pointer components |
| 16405 | 10.1 | Fortran |
c_bool in iso_c_binding not implemented
Added the definition of c_bool |
| 16407 | 10.1 | Fortran |
pgf90 reports the error "transform_call:Array Expression can't be here"
Corrected the processing of a unary operation of the result of an array-valued elemental function |
| 16411 | 10.1 | 32-bit Fortran |
accel code gets wrong answers
Corrected the compiler's processing of if statements within an acc region |
| 16412 | 10.1 | Fortran |
STOP statement in a parallel region leads to an infinite loop
Updated the run-time when multiple threads exit a parallel region |
| 16425 | 10.1 | Installation |
Linux install script fails to link the mpi directory to 2010/mpi
Updated the script |
| 16426 | 10.1 | Fortran |
accel compiler removes 'if' from acc region
Corrected the compiler's processing of if statements within an acc region |
| 16449 | 10.1 | C++ |
for init statement processing incorrect with --zc_eh
Corrected the processing of for inits |
| 16451 | 10.1 | Fortran |
Compiling OpenMPI code fails with "I may not appear in a PRIVATE clause"
Allow variables, which are also dummy arguments to statement functions, to appear in a PRIVATE clause |
| 3425 | 10.0 | 64-bit Fortran |
Request for directives to ignore possible dependences in array assignments
The IVDEP directive can now be applied to array assignments involving pointers. |
| 4161 | 10.0 | Fortran |
Option -Mipa=fast module procedure import error
Module procedures are now imported properly. |
| 15948 | 10.0 | All |
Request to support higher counts for OMP_NUM_THREADS and MP_BLIST
The maximum number of OpenMP threads has been increased to 256. |
| 16012 | 10.0 | C/C++ |
C OpenMP example scales with pgcc but not with pgCC
Corrected the processing of nested OpenMP pragmas in pgCC. |
| 16024 | 10.0 | Fortran |
Large array example seg faults due to random_number()
Corrected the processing of large arrays in the run-time routine for random_number. |
| 16070 | 10.0 | Fortran |
Option -Mipa=fast module procedure import error
Module procedures are now imported properly. |
| 16085 | 10.0 | C/C++ |
GNU linkonce errors building c++ applications
Corrected the switch processing when compiling -fpic and using linkonce sections. |
| 16095 | 10.0 | Fortran |
pgf90 missed opportunity to evaluate an array extent as 'constant'
Corrected the handling of large arrays in the run-tiime routine routine for random_number. |
| 16096 | 10.0 | C/C++ |
Legal C code 'Illegal type conversion required'
Added support for the missing cast functions for the extended asm intrinsics. |
| 16117 | 10.0 | 64-bit Fortran |
pgf90 reports the error, "interf:new_symbol, symbol not found"
Corrected the processing of a MODULE appearing after a statement separator. |
| 16118 | 10.0 | C/C++ |
pgcc -Mbounds triggers false failure
Corrected array bounds checking of variable length arrays |
| 16119 | 10.0 | C/C++ |
pgCC openmp example produces wrong answers
Corrected failures with unsigned iterators in an ordered parallel for. |
| 16121 | 10.0 | C/C++ |
pgCC extended asm constraint list errors
Corrected the processing of constraints when clobbers list is empty. |
| 16144 | 10.0 | All |
omp_get_schedule() incorrect for 'auto' schedule
omp_get_schedule() returns the correct value when the sched type is auto. |
| 16145 | 10.0 | Fortran |
omp_lib.h lacks defines for omp_sched_* parameters
Added the appropriate PARAMETERs for the omp sched types. |
| 16161 | 10.0 | C/C++ |
pgCC gnu.linkonce error
Corrected the switch processing when compiling -fpic and using linkonce sections. |
| 16164 | 10.0 | Fortran |
pgf90 reports the error "PGF90-S-0310-CASE value must be a constant"
Corrected the evaluation of a user generic named ichar which also resolves to the ichar intrinsic. |
| 16165 | 10.0 | C/C++ |
pgcc -mp -Xa terminates with signal 11 for misspelled OpenMP routine.
Corrected the -Xa checking by the compiler when calling undeclared functions. |
| 16170 | 10.0 | C/C++ |
pgCC issues "PGCC-S-0155-Illegal context for ordered"
Corrected the processing of nested OpenMP pragmas in pgCC. |
| 16171 | 10.0 | C/C++ |
C++ omp for schedule(static,1) not being observed for C++ iterators
Corrected the processing of static schedule with a chunk value of 1. |
| 16172 | 10.0 | C/C++ |
false pgCC warning, "variable declared but never referenced", with -mp
The appearance of a local variable in private clause now counts as a reference. |
| 16173 | 10.0 | C/C++ |
Floating point exception for code compiled -mp
Corrected the compiler's handling of pointer iterators in a parallel for. |
| 16174 | 10.0 | Fortran |
implicit none triggers for procedure pointer with an interface
A procedure pointer with a procedure interface is now marked as being declared. |
| 16175 | 10.0 | Fortran |
user generic, byte_write, not resolved
A character substring argument now matches a fixed length character dummy when attempting to resolve generic procedures. |
| 16177 | 10.0 | Fortran |
pgf90 reports the error "PGF90-S-0038-Symbol, ptr, has not been explicitly declared"
A procedure pointer with a procedure interface is now marked as being declared. |
| 16188 | 10.0 | C/C++ |
Problems setting breakpoints in c++ code
Corrected the debug line information for multiple statements. |
| 16189 | 10.0 | Fortran |
pgf90 reports the error "PGF90-S-0146-Expression must be character type"
Corrected declaring derived type PARAMETERs which contain a pointer component. |
| 16190 | 10.0 | Fortran |
pgf90 example aborts at -O2 but not at -O1.
Corrected creating an array temp for a reshape intrinsic appearing as an acual argument. |
| 16219 | 10.0 | All |
Option -Minfo -ta=host,nvidia only provides host info
The nvidia information is now reported. |
| 16220 | 10.0 | 64-bit Fortran |
Matrix multiplicaton CUDA example wrong
Corrected the example in the documentation. |
| 16221 | 10.0 | 64-bit C/C++ |
GPU example has an error
Corrected the example in the documentation. |
| 16320 | 10.0 | All |
pgf90 -Mlarge_arrays reports "unknown switch" for 32-bit
Improved the message to indicate that the switch is only applicable for 64-bit compilers. |
