Details:
- Fine-tuned the double-precision real MT threshold (which controls
whether the sup implementation kicks for smaller m dimension values)
from 180 to 201 for haswell and 180 to 256 for zen.
- Updated octave scripts in test/sup/octave to include a seventh column
to display performance for m = n = k.
Details:
- Implemented a new sub-framework within BLIS to support the management
of code and kernels that specifically target matrix problems for which
at least one dimension is deemed to be small, which can result in long
and skinny matrix operands that are ill-suited for the conventional
level-3 implementations in BLIS. The new framework tackles the problem
in two ways. First the stripped-down algorithmic loops forgo the
packing that is famously performed in the classic code path. That is,
the computation is performed by a new family of kernels tailored
specifically for operating on the source matrices as-is (unpacked).
Second, these new kernels will typically (and in the case of haswell
and zen, do in fact) include separate assembly sub-kernels for
handling of edge cases, which helps smooth performance when performing
problems whose m and n dimension are not naturally multiples of the
register blocksizes. In a reference to the sub-framework's purpose of
supporting skinny/unpacked level-3 operations, the "sup" operation
suffix (e.g. gemmsup) is typically used to denote a separate namespace
for related code and kernels. NOTE: Since the sup framework does not
perform any packing, it targets row- and column-stored matrices A, B,
and C. For now, if any matrix has non-unit strides in both dimensions,
the problem is computed by the conventional implementation.
- Implemented the default sup handler as a front-end to two variants.
bli_gemmsup_ref_var2() provides a block-panel variant (in which the
2nd loop around the microkernel iterates over n and the 1st loop
iterates over m), while bli_gemmsup_ref_var1() provides a panel-block
variant (2nd loop over m and 1st loop over n). However, these variants
are not used by default and provided for reference only. Instead, the
default sup handler calls _var2m() and _var1n(), which are similar
to _var2() and _var1(), respectively, except that they defer to the
sup kernel itself to iterate over the m and n dimension, respectively.
In other words, these variants rely not on microkernels, but on
so-called "millikernels" that iterate along m and k, or n and k.
The benefit of using millikernels is a reduction of function call
and related (local integer typecast) overhead as well as the ability
for the kernel to know which micropanel (A or B) will change during
the next iteration of the 1st loop, which allows it to focus its
prefetching on that micropanel. (In _var2m()'s millikernel, the upanel
of A changes while the same upanel of B is reused. In _var1n()'s, the
upanel of B changes while the upanel of A is reused.)
- Added a new configure option, --[en|dis]able-sup-handling, which is
enabled by default. However, the default thresholds at which the
default sup handler is activated are set to zero for each of the m, n,
and k dimensions, which effectively disables the implementation. (The
default sup handler only accepts the problem if at least one dimension
is smaller than or equal to its corresponding threshold. If all
dimensions are larger than their thresholds, the problem is rejected
by the sup front-end and control is passed back to the conventional
implementation, which proceeds normally.)
- Added support to the cntx_t structure to track new fields related to
the sup framework, most notably:
- sup thresholds: the thresholds at which the sup handler is called.
- sup handlers: the address of the function to call to implement
the level-3 skinny/unpacked matrix implementation.
- sup blocksizes: the register and cache blocksizes used by the sup
implementation (which may be the same or different from those used
by the conventional packm-based approach).
- sup kernels: the kernels that the handler will use in implementing
the sup functionality.
- sup kernel prefs: the IO preference of the sup kernels, which may
differ from the preferences of the conventional gemm microkernels'
IO preferences.
- Added a bool_t to the rntm_t structure that indicates whether sup
handling should be enabled/disabled. This allows per-call control
of whether the sup implementation is used, which is useful for test
drivers that wish to switch between the conventional and sup codes
without having to link to different copies of BLIS. The corresponding
accessor functions for this new bool_t are defined in bli_rntm.h.
- Implemented several row-preferential gemmsup kernels in a new
directory, kernels/haswell/3/sup. These kernels include two general
implementation types--'rd' and 'rv'--for the 6x8 base shape, with
two specialized millikernels that embed the 1st loop within the kernel
itself.
- Added ref_kernels/3/bli_gemmsup_ref.c, which provides reference
gemmsup microkernels. NOTE: These microkernels, unlike the current
crop of conventional (pack-based) microkernels, do not use constant
loop bounds. Additionally, their inner loop iterates over the k
dimension.
- Defined new typedef enums:
- stor3_t: captures the effective storage combination of the level-3
problem. Valid values are BLIS_RRR, BLIS_RRC, BLIS_RCR, etc. A
special value of BLIS_XXX is used to denote an arbitrary combination
which, in practice, means that at least one of the operands is
stored according to general stride.
- threshid_t: captures each of the three dimension thresholds.
- Changed bli_adjust_strides() in bli_obj.c so that bli_obj_create()
can be passed "-1, -1" as a lazy request for row storage. (Note that
"0, 0" is still accepted as a lazy request for column storage.)
- Added support for various instructions to bli_x86_asm_macros.h,
including imul, vhaddps/pd, and other instructions related to integer
vectors.
- Disabled the older small matrix handling code inserted by AMD in
bli_gemm_front.c, since the sup framework introduced in this commit
is intended to provide a more generalized solution.
- Added test/sup directory, which contains standalone performance test
drivers, a Makefile, a runme.sh script, and an 'octave' directory
containing scripts compatible with GNU Octave. (They also may work
with matlab, but if not, they are probably close to working.)
- Reinterpret the storage combination string (sc_str) in the various
level-3 testsuite modules (e.g. src/test_gemm.c) so that the order
of each matrix storage char is "cab" rather than "abc".
- Comment updates in level-3 BLAS API wrappers in frame/compat.
Details:
- Added preprocessor branches to test/3/test_gemm.c to explicitly
support row-stored matrices. Column-stored matrices are also still
supported (and is the default for now). (This is mainly residual work
leftover from initial integration of Eigen into the test drivers, so
if we ever want to test Eigen with row-stored matrices, the code will
be ready to use, even if it is not yet integrated into the Makefile
in test/3.)
Details:
- Updated the level-3 performance graphs in docs/graphs with new Eigen
results, this time using a development version cloned from their git
mirror on March 27, 2019 (version 3.3.90). Performance is improved
over 3.3.7, though still noticeably short of BLIS/MKL in most cases.
- Very minor updates to docs/Performance.md and matlab scripts in
test/3/matlab.
Details:
- Updated matlab scripts in test/3/matlab to optionally plot/display
Eigen performance curves. Whether Eigen is plotted is determined by
a new boolean function parameter, with_eigen.
- Updated runme.m scratchpad to reflect the latest invocations of the
plot_panel_4x5() function (with Eigen plotting enabled).
Details:
- Fixed the Makefile in test/3 so that it no longer incorrectly labels
the matlab output variables from Eigen-linked hemm, herk, trmm, and
trsm driver output as "vendor". (The gemm drivers were already
correctly outputing matlab variables containing the "eigen" label.)
Details:
- Adjusted test/3/Makefile so that the test drivers are linked against
Eigen's BLAS library for hemm, herk, trmm, and trsm. We have to do
this since Eigen's headers don't define implementations to the
standard BLAS APIs.
- Simplified #included headers in hemm, herk, trmm, and trsm source
driver files, since nothing specific to Eigen is needed at
compile-time for those operations.
Details:
- Use compile-time implementations of Eigen in test_gemm.c via new
EIGEN cpp macro, defined on command line. (Linking to Eigen's BLAS
library is not necessary.) However, as of Eigen 3.3.7, Eigen only
parallelizes the gemm operation and not hemm, herk, trmm, trsm, or
any other level-3 operation.
- Fixed a bug in trmm and trsm drivers whereby the wrong function
(bli_does_trans()) was being called to determine whether the object
for matrix A should be created for a left- or right-side case. This
was corrected by changing the function to bli_is_left(), as is done
in the hemm driver.
- Added support for running Eigen test drivers from runme.sh.
Details:
- Added targets to test/3/Makefile that link against a BLAS library
build by Eigen. It appears, however, that Eigen's BLAS library does
not support multithreading. (It may be that multithreading is only
available when using the native C++ APIs.)
- Updated runme.sh with a few Eigen-related tweaks.
- Minor tweaks to docs/Performance.md.
Details:
- Added a new markdown document, docs/Performance.md, which reports
performance of a representative set of level-3 operations across a
variety of hardware architectures, comparing BLIS to OpenBLAS and a
vendor library (MKL on Intel/AMD, ARMPL on ARM). Performance graphs,
in pdf and png formats, reside in docs/graphs.
- Updated README.md to link to new Performance.md document.
- Minor updates to CREDITS, docs/Multithreading.md.
- Minor updates to matlab scripts in test/3/matlab.
Details:
- Renamed '3m4m' directory to '3', which captures the directory nicely
since it builds test drivers to test level-3 operations.
- These test drivers ceased to be used to test the 3m and 4m (or even
1m) induced methods long ago, hence the name change.
Details:
- Further updates to matlab scripts, mostly for compatibility with
GNU Octave.
- More tweaks to runme.sh.
- Updates to runme.m that allow copy-paste into matlab interactive
session to generate graphs.
Details:
- Rewrote much of Makefile to generate executables for single- and dual-
socket multithreading as well as single-threaded. Each of the three
can also use a different problem size range/increment, as is often
appropriate when doubling/halving the number of threads.
- Rewrote runme.sh script to flexibly execute as many threading
parameter scenarios as is given in the input parameter string
(currently set within the script itself). The string also encodes
the maximum problem size for each threading scenario, which is used
to identify the executable to run. Also improved the "progress" output
of the script to reduce redundant info and improve readability in
terminals that are not especially wide.
- Minor updates to test_*.c source files.
- Updated matlab scripts according to changes made to the Makefile,
test drivers, and runme.sh script, and renamed 'plot_all.m' to
'runme.m'.
Details:
- Minor updates to matlab graph-generating scripts.
- Added a plot_all.m script that is more of a scratchpad for copying and
pasting function invocations into matlab to generate plots that are
presently of interest to us.
Details:
- Added a variant set of matlab scripts geared to producing plots that
reflect performance data gathered with and without extra memory
optimizations enabled. These scripts reside (for now) in
test/mixeddt/matlab/wawoxmem.
Details:
- Parameterized, reorganized, and added comments to matlab scripts in
test/mixeddt/matlab.
- Reordered some lines of code and added comments to plot_l3_perf.m in
test/3m4m/matlab.
Details:
- Updated 3m4m and mixeddt Makefiles and runme.sh scripts, mostly to
port recent changes to the former to the latter.
- Disabled (for now) code in 3m4m/test_*.c files that disables all
induced methods except for the one that is requested from the
Makefile via the IND macro. This is done because usually, we want to
test whatever method is enabled automatically for complex datatypes.
(That is, when native complex microkernels are missing, we usually
want to test performance of 1m.)
Details:
- Removed explicit reference to The University of Texas at Austin in the
third clause of the license comment blocks of all relevant files and
replaced it with a more all-encompassing "copyright holder(s)".
- Removed duplicate words ("derived") from a few kernels' license
comment blocks.
- Homogenized license comment block in kernels/zen/3/bli_gemm_small.c
with format of all other comment blocks.
Details:
- Added a new directory, test/3m4m/matlab, containing matlab scripts for
plotting 4x5 panels of performance graphs (using the subplot()
function) for gemm, hemm, herk, trmm, and trsm across all four
floating-point datatypes. I expect to further refine these scripts as
time goes on, but their current state constitutes a good start.
Details:
- Cleanups to Makefile to allow all test drivers to be built for
OpenBLAS and MKL in addition to BLIS.
- Fixed copy-paste typos in test_hemm in calls to ssymm_() and dsymm_().
- Fixed incorrect types for betap in BLAS cpp macro branch of
test_herk.c.
Details:
- Added a new test_hemm.c test driver to test/3m4m, which was modeled
after the driver by the similar name in test. Also updated Makefile
so that blis-nat-[sm]t would trigger builds for the new driver.
Details:
- Renamed scripts in test/mixeddt/matlab:
plot_case_all.m -> plot_dom_all.m
plot_case_md.m -> plot_dom_case.m
plot_all_md.m -> plot_dt_all.m
- Added plot_dt_select.m in order to plot select graphs for the main
body of the mixeddt paper, and added additional related legend
handling in plot_gemm_perf.m.
- Added test/mixeddt/matlab/output and a .gitkeep file within in order
to force git to recognize the directory.
Details:
- Cleaned up test/mixeddt Makefile in preparation for gathering new
data for mixeddt paper, including renaming implementations to
"internal" and "ad-hoc" to match the terminology to be used in the
paper.
- Added new matlab scripts for generating 8 figures, each covering all
mixed-precision cases for each mixed-domain case.
- Updated the runme.sh script according to changes to Makefile.
- Fixed a minor bug in test_gemm.c that may have given incorrect
performance in complex, homogeneous storage datatype cases where
the computation precision was equal to the storage precisions.
(Examples: zzzd, cccs.)
Details:
- Expanded the bli_pthread_*() -> pthread_*() wrappers in
frame/thread/bli_pthread.c to include cases for Windows taken from
frame/base/bli_pthread_wrap.c. Now, bli_thread_*() is always defined
and always used by BLIS and the BLIS testsuite (in lieu of calling
pthreads directly, as before). The implementation used in this new
API depends on whether we are building for Windows, and to a lesser
extent, whether we are building on OS X. For the core API, Windows
uses Windows threads, non-Windows (Linux, OS X) uses pthreads.
OS X and Windows get barriers implemented in terms of other
bli_pthread_*() functions, and Linux gets barriers implemented in
terms of pthread_barrier*(). This commit addresses issue #273.
- Fixed a bug in the Linux definition of bli_pthread_mutex_unlock(),
which was erroneously calling pthread_mutex_lock().
- Minor changes to configure so that the auto-detection executable
can be built given the above changes (most notably, turning on
POSIX extensions via -D_GNU_SOURCE).
- Removed temporary play-test code for shiftd that accidentally got
committed into test/3m4m/test_gemm.c.
Details:
- Defined a new level-1d operation called 'shiftd', including object and
typed APIs. This operation adds a scalar value to every element along
an arbitrary diagonal of a matrix. Currently, shiftd is implemented in
terms of the addv kernel. (The scalar is passed in as the x vector
with an increment of zero.)
- Replaced ad-hoc usage of setd and addd (after creating a temporary
matrix object) with use of shiftd, which is much more concise, in
various test driver files in the testsuite. Similar changes were made
to the standalone test drivers and the example code.
- Added documentation entries in BLISObjectAPI.md and BLISTypedAPI.md
for bli_shiftd() and bli_?shiftd(), respectively.
- Added observed object properties to level-1d documentation in
BLISObjectAPI.md.
Details:
- Implemented support for gemm where A, B, and C may have different
storage datatypes, as well as a computational precision (and implied
computation domain) that may be different from the storage precision
of either A or B. This results in 128 different combinations, all
which are implemented within this commit. (For now, the mixed-datatype
functionality is only supported via the object API.) If desired, the
mixed-datatype support may be disabled at configure-time.
- Added a memory-intensive optimization to certain mixed-datatype cases
that requires a single m-by-n matrix be allocated (temporarily) per
call to gemm. This optimization aims to avoid the overhead involved in
repeatedly updating C with general stride, or updating C after a
typecast from the computation precision. This memory optimization may
be disabled at configure-time (provided that the mixed-datatype
support is enabled in the first place).
- Added support for testing mixed-datatype combinations to testsuite.
The user may test gemm with mixed domains, precisions, both, or
neither.
- Added a standalone test driver directory for building and running
mixed-datatype performance experiments.
- Defined a new variation of castm, castnzm, which operates like castm
except that imaginary values are not touched when casting a real
operand to a complex operand. (By contrast, in these situations castm
sets the imaginary components of the destination matrix to zero.)
- Defined bli_obj_imag_is_zero() and substituted calls in lieu of all
usages of bli_obj_imag_equals() that tested against BLIS_ZERO, and
also simplified the implementation of bli_obj_imag_equals().
- Fixed bad behavior from bli_obj_is_real() and bli_obj_is_complex()
when given BLIS_CONSTANT objects.
- Disabled dt_on_output field in auxinfo_t structure as well as all
accessor functions. Also commented out all usage of accessor
functions within macrokernels. (Typecasting in the microkernel is
still feasible, though probably unrealistic for now given the
additional complexity required.)
- Use void function pointer type (instead of void*) for storing function
pointers in bli_l0_fpa.c.
- Added documentation for using gemm with mixed datatypes in
docs/MixedDatatypes.md and example code in examples/oapi/11gemm_md.c.
- Defined level-1d operation xpbyd and level-1m operation xpbym.
- Added xpbym test module to testsuite.
- Updated frame/include/bli_x86_asm_macros.h with additional macros
(courtsey of Devin Matthews).
Details:
- Adjusted the method by which micropanels are assigned to threads in
the 2nd (jr) and 1st (ir) loops around the microkernel to (mostly)
employ contiguous "slab" partitioning rather than interleaved (round
robin) partitioning. The new partitioning schemes and related details
for specific families of operations are listed below:
- gemm: slab partitioning.
- herk: slab partitioning for region corresponding to non-triangular
region of C; round robin partitioning for triangular region.
- trmm: slab partitioning for region corresponding to non-triangular
region of B; round robin partitioning for triangular region.
(NOTE: This affects both left- and right-side macrokernels:
trmm_ll, trmm_lu, trmm_rl, trmm_ru.)
- trsm: slab partitioning.
(NOTE: This only affects only left-side macrokernels trsm_ll,
trsm_lu; right-side macrokernels were not touched.)
Also note that the previous macrokernels were preserved inside of
the 'other' directory of each operation family directory (e.g.
frame/3/gemm/other, frame/3/herk/other, etc).
- Updated gemm macrokernel in sandbox/ref99 in light of above changes
and fixed a stale function pointer type in blx_gemm_int.c
(gemm_voft -> gemm_var_oft).
- Added standalone test drivers in test/3m4m for herk, trmm, and trsm
and minor changes to test/3m4m/Makefile.
- Updated the arguments and definitions of bli_*_get_next_[ab]_upanel()
and bli_trmm_?_?r_my_iter() macros defined in bli_l3_thrinfo.h.
- Renamed bli_thread_get_range*() APIs to bli_thread_range*().
Details:
- Tweaked the cflags functions in common.mk so that a new preprocessor
macro, BLIS_IS_BUILDING_LIBRARY, is defined, but only when BLIS
itself is being built. This macro will not be defined when, for
example, the testsuite or example code compiles code local to those
applications. This was done in part by defining a new cflags function
get-user-cflags-for(), which is now the designated function for
application Makefiles if they wish to inherit a basic set of CFLAGS
from BLIS. (The compiler flags returned are identical to that of
get-frame-cflags-for() except that -DBLIS_IS_BUILDING_LIBRARY is
omitted.)
- Updated all test driver-like makefiles to call get-user-cflags-for()
instead of get-frame-cflags-for().
Details:
- Removed four trailing spaces after "BLIS" that occurs in most files'
commented-out license headers.
- Added UT copyright lines to some files. (These files previously had
only AMD copyright lines but were contributed to by both UT and AMD.)
- In some files' copyright lines, expanded 'The University of Texas' to
'The University of Texas at Austin'.
- Fixed various typos/misspellings in some license headers.
Details:
- Updated the build system so that "lesser" Makefiles, such as those in
belonging to example code or the testsuite, may be run even if the
directory is orphaned from the original build tree. This allows a
user to configure, compile, and install BLIS, delete the build tree
(that is, the source distribution, or the build directory for out-
of-tree builds) and then compile example or testsuite code and link
against the installed copy of BLIS (provided the example or testsuite
directory was preserved or obtained from another source). The only
requirement is that make be invoked while setting the
BLIS_INSTALL_PATH variable to the same installation prefix used when
BLIS was configured. The easiest syntax is:
make BLIS_INSTALL_PATH=/install/prefix
though it's also permissible to set BLIS_INSTALL_PATH as an
environment variable prior to running 'make'.
- Updated all lesser Makefiles to implement the new aforementioned build
behavior.
- Relocated check-blastest.sh and check-blistest.sh from build to
blastest and testsuite, respectively, so that if those directories are
copied elsewhere the user can still run 'make check' locally.
- Updated docs/Testsuite.md with language that mentions this new option
of building/linking against an installed copy of BLIS.
Details:
- Removed the dtime (delta time, or wallclock time) column from the
matlab output of all test drivers in test, test/3m4m, test/studies.
This value was rarely (if ever) really needed and usually only served
to take up screen space.
- Updated format specifier in test/studies/skx to use %7.2f instead of
%6.3f.
- For the test drivers in 'test' directory, added an initial line of
output that sets last entry of matlab matrix to zero in order to
induce a pre-allocation of the entire array of performance results.
Details:
- Changed the format specifier for the gflops column in the testsuite
output from %7.3f to %7.2f. This was done mainly to keep the output
aligned properly when the expected perfomance exceeded 1000 gflops.
Also, two decimal places still conveys plenty of precision for all
practical applications, including just eyeballing performance deltas
between two executions (let alone two implementations).
- Changed the format specifier for gflops in the test/3m4m drivers
from %6.3f to %7.2f (for the same reasons listed above).
