Macros perform similar purpose as phase driver (see phase.c), the
main difference is to decouple it away from just running phases
cyclically. Still uses counters, but the user controls what these
are.

User for minimize_transfers.c and algebraic_wlfi.c.

Some testing showed that the implementation was not ideal, leading to
problems in later CUDA traversals. I'm removing this, in favour of a
better implementation.

Completely forgot that EMRL is one by default, therefore whenever we
entered IWLMEM, we entered EMRTU, which in some very rare cases did
cause problems. We now check that the EMRCI and EMRTU flag is on as well!

Have moved out a function used in both traversals to the cuda_utils
module.

Also, I've removed the ISEMRLIFTED flag from EMRL/EMRTU, as it was
preventing certain cases from being identified (such as cond functions).

This optimisation works together with the CUDA IWLMEM optimisation to introduce
CUDA memory transfers into the AST. We need this traversal when using the EMR
optimisation as IWLMEM would otherwise lead to inoptimal code generation.

IWLMEM, when encountering a lift ERC, would within the loop function create a
host2device and cause a memory allocation and free on the CUDA device. The
EMRTU (EMR Type Update) traversal identifies such cases, and correclty
transforms the lifted ERCs into CUDA device_types *and* fixes the arguments of
the loop function applications (initial and recursi

We will use this later for EMRTU (relating to CUDA IWLMEM
optimisation).

The insert_memory_transfers module was split into several different
files (insert_withloop_memtran.c & insert_cudast_memtran.c) by Jing
starting from the 26-03-2010, and this was left in the git history.

This module is not used anywhere, and as far as I can tell, provides
nothing in addition to the other modules. As such its safe to remove the
files.

Additionally, I've identified the following libsac2c modules which are
also stale:

icm_betest.c
icm_trace.c
icm_comment.c
BEtest.c
wl_empty_result_handling.c
create_lac_fun.c
wl_descalarization.c
cuda_create_dfg.c
idag.c
elemqueue.c
elemlist.c
cygcompat.c
cygwinhelpers.c

One needs to take a closer look at this and see if they are still
useful.

Extend modulo primitive to support all "whole number" types

See merge request !107

Us `integral type` instead of `integer types`.

Additionally,
* we add an error when trying to use _mod_SxV_, _mod_VxS_, and _mod_VxV_
  which are not actually implemented.
* add a test to make sure _mod_SxS_ functions correctly.

[HOTFIX] fix incorrect assert check

See merge request !104

Remove Package Build to Seperate Project

See merge request !103

* also changed to the special `merge_request` specifier to build on
  MR creation and update.

fix incorrect symlinking in archive dist... we now use the version
manager script instead.

add manual requires for the moment

Extended Memory Optimisations

See merge request !80

We no longer print RC/ERC via print.c, but use the pre/post traversal
function injection fascility to _nicely_ associate this with the
relevent traversals (i.e. wrci and emrci). Additionally we update
documentation and include some more tests (and improve some existing
ones as well).

Main changes:
* moved some repeated code into header file (used in three places)
* fixed up various comments throughout
* resolved some inconsistencies in memory profiling code (had
incorrectly counted allocation/free sizes)
* some minor refactoring throughout
* added some further tests for memory profiler

Artem Shinkarov authored Dec 02, 2018 and Hans-Nikolai Viessmann committed Dec 31, 2018

Signed-off-by: Hans-Nikolai Viessmann <hv15@hw.ac.uk>

We now correctly allocate the lifted N_avis from EMRL when using the
CUDA backend - this is critically as we're replace a host2device with
a plain alloc + fill. Additionally we fixup/correclty handle the the
seq/mt_pth cases as well.

Support for dynamic pre/post traversal function inject is not
being handled *yet*, so lets remove these print functions for the
moment. They can be trivially restored later.

For the EMR tests we need to use the mini-stdlib, but because we have
placed them in their own directory, we get missing include errors. This
happens in run.sh when we first generate the test makefile. We extend
the flags to find the missing includes correctly.

We also activate EMRCF and EMRL per-default. One needs only specify
`-doEMRCI` to activate the whole spiel.