GitLab

Remove support for direct pointers in a DMA request, with the goal to
remove va_args in the entire DMA API soon. As scratchs still work with
pointers, also modify scratchs to convert tiles in layouts.

Various fixes related to dma API, missing validation on arguments,
pthread management and so on.

Unit tests were missing a few corner cases on dense_create. In
particular, there was a bug in the handling of the NULL pitch.

The bitmap tests were slowing down the entire pipeline, while at the
same time missing important checks on the inputs. This version is a bit
more exhaustive, while also speeding up the test by only checking the
corner-cases (first bits, last bits, partial ranges, overlapping
ranges).

Since vector resize can cause pointers into the vector to become
invalid, this patch introduces a level of indirection for DMAs and
Scratchs, using uuid on the user-side of a request to index into the
request vector.

Refactor the DMA API to work on layouts internally. The resulting code
ends up simpler for most cases, as long as we add a generic copy
function.

One things missing:
- a good API to expose the collection of copy functions we have.

One question not solved:
- should the DMA ask for a copy function on each operation ?

Use the newly added INNER_MALLOC for all the relevant creators.

Add a set of macros to help us deal with several mallocs back-to-back in
creators, while ensuring that pointers stay aligned properly.

See #45 for details. Related to the !68 cleanup.

Nicolas Denoyelle authored Aug 06, 2019 and Swann Perarnau committed Aug 06, 2019

Add new building block describing how data
is organized in memory.

Removes static allocations of all objects from the library.
While this will make our code less compact in theory,
this part of the API is too confusing for new users.

One more silent error. ax_pthreads will fail on -Werror, so we still
need to fix that.

Still fails, some issues in configure.ac/Makefile.am

Note that this still fails on make check, some problem with
CPPFLAGS/LDFLAGS

Last bit of API lacking consistency.

Implement #31 for vector.
Improve error management.
Checkpatch vector.

Implement #31 for tiling.
Improve error management.
Checkpatch tiling.

Implement #31 for scratch.
Improve error management.
Checkpatch scratch.

Implement #31 for dma.
Improve error management.
Checkpatch dma

Convert area/area_linux to new error handler.
Implement #31 for area.
Checkpatch area

Note that area_linux is still using a dynamic data structure no matter
what: a bitmask to the set of authorized nodes. The conversion to
aml_bitmap wouldn't help as there's no default value available for all
nodes on that type (similar to numa_all_nodes_ptr).

The source repo has been reorganized, but the repoquality checks were
not updated. Also make sure that style checks happen early, to fail the
CI quickly.

Same as the rest.

We're starting to agree on one main header for all the generic APIs of
building blocks, struct definitions included, and one header per
specific implementation.

Improve the spelling while we're at it.

Since we removed `move` from the dma API, bindings are now useless.

Poorly supported feature that doesn't play well with the rest of the
library and limits what we can do in the future.

Use m4 to define autoconf-level version variables, following the naming
scheme of semver.org

To make use of these variables in the headers and sources, a
generated-header is added in aml/utils/version.h

Also add a simple test for that part of the lib.

The overall strategy for now is to split the implementations into
distinct headers, but keep the generic APIs inside the main aml.h

Related to #27

The overall strategy for now is to split the implementations into
distinct headers, but keep the generic APIs inside the main aml.h

Related to #27

Nicolas Denoyelle authored Mar 14, 2019 and Swann Perarnau committed Mar 14, 2019

While the current bitmask management is heavily inspired by libnuma, it is not
as easy to use has the libnuma API. This patch is an attempt to refactor
to code towards something cleaner.

Nicolas Denoyelle authored Mar 12, 2019 and Swann Perarnau committed Mar 13, 2019

- create one directory per building block in src and include
- keep one directory for tests, \
  otherwise automake make them "test suites"
- move to AC_OPENMP, which is from autoconf 2.62 (2008)

Force libtool to static link the PIC version of our jemalloc import into
libaml, making libaml standalone. This requires us to test some
additional libraries in our own configure (pthread, and dlopen).

This also solves the long-standing issue of `make check` only working after
`make install`, while removing our custom jemalloc from the installed
libraries.

Fixes #26.

Mbind is giving us trouble again, will need to spend time looking at it
carefully.

Instead of asking the user to provide the offsets into a tiling, add a
function providing a tileid. This tileid corresponds to the in-memory
order of tiles, making the tilestart functions a lot simpler.

We still need to split the tileid for tilestart because scratchpads
create requests based on tileids.

Also add a unit test for tiling_2d, to make sure we're not doing
anything crazy.

This patch provides aligned allocations for all areas. Simple tests
included.

Note that I haven't tested if it conflicts with arena-wide alignements.

We are starting to work on benchmarks to evaluate the usefulness of this
library. Instead of integrating them into the testing infrastructure, it
makes more sense for them to have their own directory and a different
way of handling them.

This patch:
 - creates a benchmark directory for actual codes that we want to use as
   benchmarks of our library.
 - moves functional tests into it.
 - add proper OpenMP detection for these codes

Mempolicy seems to be inconsistant depending on the distro. Need a more
complete investigation.