Rapsodia

log

age author description
6 months ago utke on macbook yeardefault tip
7 months ago utke on macbook force double precision
11 months ago Jean Utke extract module search settings from IPATH to not mess up things for the mixed language compiles
12 months ago utke on macbook move this to public for regression testing
12 months ago utke on macbook for regression testing
12 months ago utke on macbook typedef this and return a const ref instead
12 months ago utke on macbook need this dependency
13 months ago wang970 Solved the bug described in /RapsodiaExample/F90ReverseStack. By "de-finalize" (not to) the active variables as left values.
13 months ago wang970 A minor change in Regression/revF90. Add some comments about why we don't set the adjoint as 0 when the finalizer in F90 is called. (Because in F90, the finalizer will be called for left value of an assignment before the assign happens. And this will erase the adjoints.)
13 months ago wang970 In reverse mode, tan(x) should directly call r.v=tan(x.v) instead of tan()=sin()/cos(). The floating-point arthimic will cause slitely difference between them. (tan=sin/cos) is still in use in Forward mode. Which may suffer a same problem.
13 months ago wang970 For Cpp, 2 classes have to be introduced for the return expression issue. A base type is for OO and temporary, the derived type is for explicitly usage. So we can force the copy constructor from base type to derived type is invoked at the end of a function.
13 months ago wang970 Solved a bug in reverse/location.cpp. When the location array is expanded, loc_ad is not reset correctly.
13 months ago Jean Utke mixed language settings for nagfor/g++
13 months ago Jean Utke make the output more comprehensive
13 months ago wang970 Change the name and description of mixed language configurations. Enable the Regression/revF90 test.
13 months ago wang970 Delete temp directory if fortran finalization test fails in configure.py
13 months ago wang970 New mixed language linking configurations.
13 months ago wang970 Change configure.py so mixed language linking is set up for pgcpp and pgf95.
13 months ago wang970 Force explicit type conversion for every partial values in Fortran. (pgf95 requires). Set Mixed Language Linker in Regression/revF90. (Still not working for pgcpp/pgf95)
13 months ago wang970 Temporarily remove Regression/revF90 from test. (IPATH should not be exported and implicit type conversion is not default in pgf95)
13 months ago wang970 get rid of 'inline' in location.cpp and RA_adacc_d.c
13 months ago utke on macbook branch merge
13 months ago utke on macbook when everything is inlined we don't compile anything into the library but the rest is transparent, i.e. the library may still contain queue code
13 months ago utke on macbook for inlined definitions it only occurs once during the declaration
13 months ago utke on macbook add it here to be compatible even though it is not uesd (yet)
13 months ago utke on macbook set the inlineDefs here too
13 months ago wang970 Remove the comment line in common.h if we are using F90 printer.
13 months ago wang970 Removing unnecessary files
13 months ago wang970 Several changes: Remove implicit type convert which is forbidden in Cpp. Try to make Regression/rev functional.
13 months ago utke on macbook update F90 reference
13 months ago utke on macbook better qualify the change introduced with changeset 8dcb3fa3f796 such that we still avoid creating definitons for (presumably faster) generated assignment operators when NOT in reverse mode - also reintroduce the generated comment pointing to Rapsodia
13 months ago utke on macbook update CPP reference
13 months ago utke on macbook temporarily undo this change introduced with changeset 8dcb3fa3f796 until it is clear why changing struct to class is needed
13 months ago utke on macbook spacing typo
13 months ago utke on macbook There is no point in creating a separate rule for files with a .c extension if all we do is compile them with the C++ compiler. Rather we should - if C linkage is required - have the code in .cpp files like the rest and specify extern "C" linkage. Also - this injection messed up the compile flags that follow up and were expected to extend the .cpp rule
13 months ago utke on macbook generated code should not require fpermissive
13 months ago utke on macbook temporarily disabling tests for non-standard C++ generated for reverse mode
13 months ago utke on macbook refactor configure to remove nearly identical duplicate, rename make variable, temporarily disable reverse regression tests because they don't work yet and make the whole regression test fail
14 months ago wang970 extern "C" for all C code. Add the finalization test for Fortran compilers. Change Makefile so that no error message even if compiler does not support
14 months ago wang970 The scalar operation of *= should be performed after the push in queue mode and /= should be performed before the push.
14 months ago wang970 compound operator for queue in forward mode. Pending for tests.
14 months ago wang970 icpc does not recognize the expression "1.0d+0" (g++ does). Explicit double type is not necessary any more.
14 months ago wang970 initial the adjoint in default constructor
14 months ago Mu Wang rev 0.0.2
15 months ago Jean Utke strict gfortran syntax checking fails on empty contains block. vertex.defs can be an empty StatementGroup
15 months ago utke on macbook branch merge
15 months ago azamat adding nvcc
15 months ago utke on macbook do more checking in the debug version
15 months ago utke on macbook branch merge
15 months ago utke on macbook dump all the tensors
15 months ago utke on macbook introduce a reset so we can have a helper tensor instance that can be reused for different settings of #indeps and deriv degree
15 months ago utke on macbook make a short cut for n=1 which is a corner case that doesn't work with the generic logic
16 months ago Jean Utke slightly improve the configure test - there is a problem with the new version (13.4) of pgcpp in connection with openpa.
17 months ago utke on macbook update compiler versions
18 months ago utke on macbook intel compiler version
20 months ago utke on macbook rename CFLAGS to CXXFLAGS (should have been like that all along)
20 months ago utke on macbook turn the dependency relations into actual make rules; the old dependency output was largely nonsensical anyway
20 months ago utke on macbook newline
20 months ago utke on macbook reference update for reordered members after generator changes
20 months ago utke on macbook change the content of typeList and precDict for the interoperable case;