Details:
- SYRK for small matrix was implemented by reusing small GEMM routine. This was
resulting in output written to the full C matrix, and C being symmetric the
lower and upper triangles of C matrix contained same results. BLAS SYRK API
spec demands either lower or upper triangle of C matrix to be written with
results. So, this was resulting in BLAS test failures, even though testsuite
of BLIS was passing small SYRK operation.
- To fix BLAS test failures of small matrix SYRK, separate kernel routines are
implemented for small SYRK for both single and double precision. The newly
added small SYRK routines are in file kernels/zen/3/bli_syrk_small.c.
Now the intermediate results of matrix C are written to a scratch buffer.
Final results are written from scratch buffer to matrix C using SIMD
copy to either lower or upper traingle part of matrix C.
- Source and header files frame/3/syrk/bli_syrk_front.c and
frame/3/syrk/bli_syrk_front.h are changed to invoke new small SYRK routines.
Change-Id: I9cfb1116c93d150aefac673fca033952ecac97cb
Issue: For the default values of mc, kc and nc with multi instance mode the performance across the cores dip drastically.
Fix: After experimentation found different set of values (mc, kc and nc) which fits in the cache size, and performance across the remains same across all the cores.
Change-Id: I98265e3b7e61cd7602a0cc5596240e86c08c03fe
BUG No: CPUPL-197 fixed by Thangaraj Santanu
The bli_clock_min_diff() function in BLIS assumed that if the time taken is greater than 1 hour then the reading must be wrong. However this is not the case in general, while the other checks such as time taken closer to zero or nsec is ofcourse valid.
gerrit review: http://git.amd.com:8080/#/c/118694/1/frame/base/bli_clock.c
Change-Id: I9dc313d7c5fdc20684f67a516bf3237de3e0694a
Details:
- Added a short section after the Intro of the README.md file titled
"Education and Learning" that directs interested readers to the
"LAFF-On Programming for High-Performance" massive open online course
(MOOC) hosted via edX.
Details:
- Added a new standalone test driver directory named '1m4m' that can
build and run performance experiments for BLIS 1m, 4m1a, assembly,
OpenBLAS, and the vendor library (MKL). This new driver directory
was used to regenerate performance results for the 1m paper.
- Added alternate (commented-out) cache blocksizes to
config/haswell/bli_cntx_init_haswell.c. These blocksizes tend to
work well on an a 12-core Intel Xeon E5-2650 v3.
Details:
- Updated the default MC cache blocksizes used by the haswell subconfig
for both row-preferential (the default) and column-preferential
microkernels.
Details:
- Updated the '-march=corei7-avx' flag in the sandybridge subconfig
to '-march=sandybridge' and the '-march=core-avx2' flag in the
haswell subconfig to '-march=haswell'. The older flags were used
by older versions of gcc and should have been updated to the newer
forms a long time ago. (The older flags were clearly working, even
though they are no longer documented in the gcc man page.)
Details:
- Updated most comments in testsuite modules that describe how the
correctness test is performed so that it is clear whether the vector
(normfv) or matrix (normfm) form of Frobenius norm is used.
Details:
- Defined two new static functions in bli_cntx.h:
bli_cntx_set_blksz_def_dt()
bli_cntx_set_blksz_max_dt()
which developers may find convenient when experimenting with different
values of cache blocksizes.
- Updated one- and two-socket multithreaded problem size range and
increment values in test/3/Makefile.
- Changed default to column storage in test/3/test_gemm.c.
- Fixed typo in comment in testsuite/src/test_subm.c.
Details:
- Fixed a bug in bli_l3_thrinfo_print_gemm_paths() and
bli_l3_thrinfo_print_trsm_paths(), defined in bli_l3_thrinfo.c,
whereby subnodes of the thrinfo_t tree are "dereferenced" near the
beginning of the functions, which may lead to segfaults in certain
situations where the thread tree was not fully formed because the
matrix problem was too small for the level of parallelism specified.
(That is, too small because some problems were assigned no work due
to the smallest units in the m and n dimensions being defined by the
register blocksizes mr and nr.) The fix requires several nested levels
of if statements, and this is one of those few instances where use of
goto statements results in (mostly) prettier code, especially in the
case of _gemm_paths(). And while it wasn't necessary, I ported this
goto usage to the loop body that prints the thrinfo_t work_id and
comm_id values for each thread. Thanks to Nicholai Tukanov for helping
to find this bug.
Details:
- Updated the BLASFEO performance graphs shown in PerformanceSmall.md
using a new commit of BLASFEO (2c9f312); updated PerformanceSmall.md
accordingly.
- Updated test/sup/octave/plot_l3sup_perf.m so that the .m files
containing the mpnpkp results do not need to be preprocessed in order
to plot half the problem size range (ie: up to 400 instead of the
800 range of the other shape cases).
- Trivial updates to runme.m.
Details:
- Use -qopenmp-simd instead of -fopenmp-simd when compiling with Intel
icc. Recall that this option is used for SIMD auto-vectorization in
reference kernels only. Support for the -f option has been completely
deprecated and removed in newer versions of icc in favor of -q. Thanks
to Victor Eijkhout for reporting this issue and suggesting the fix.
Details:
- Added a cpp-conditional directive block to bli_arch_config.h that
#includes "bli_family_thunderx2.h". The code has been missing since
adf5c17f. However, this never manifested as an error because the file
is virtually empty and not needed for thunderx2 (or most subconfigs).
Thanks to Jeff Diamond for helping to spot this.
Details:
- Updated the graphs linked in PerformanceSmall.md with BLASFEO results,
and added documenting language accordingly.
- Updated scripts in test/sup/octave to plot BLASFEO data.
- Minor tweak to language re: how OpenBLAS was configured for
docs/Performance.md.
Details:
- Changed starting problem and increment from 16 to 4.
- Added 'lll' (square problems) to list of problem size shapes to
compile and run with.
- Define BLASFEO location and added BLASFEO-related definitions.
Details:
- Added a new markdown document, docs/PerformanceSmall.md, which
publishes new performance graphs for Kaby Lake and Epyc showcasing
the new BLIS sup (small/skinny/unpacked) framework logic and kernels.
For now, only single-threaded dgemm performance is shown.
- Reorganized graphs in docs/graphs into docs/graphs/large, with new
graphs being placed in docs/graphs/sup.
- Updates to scripts in test/sup/octave, mostly to allow decent output
in both GNU octave and Matlab.
- Updated README.md to mention and refer to the new PerformanceSmall.md
document.
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:
- Attempted a fix to issue #313, which reports that when building only
a shared library (ie: static library build is disabled), running the
BLAS test drivers can fail because those drivers provide their own
local version of xerbla_() as a clever (albeit still rather hackish)
way of checking the error codes that result from the individual tests.
This local xerbla_() function is never found at link-time because the
BLAS test drivers' Makefile imports BLIS compilation flags via the
get-user-cflags-for() function, which currently conveys the
-fvisibility=hidden flag, which hides symbols unless they are
explicitly annotated for export. The -fvisibility=hidden flag was
only ever intended for use when building BLIS (not for applications),
and so the attempted solution here is to omit the symbol export
flag(s) from get-user-cflags-for() by storing the symbol export
flag(s) to a new BULID_SYMFLAGS variable instead of appending it
to the subconfigurations' CMISCFLAGS variable (which is returned by
every get-*-cflags-for() function). Thanks to M. Zhou for reporting
this issue and also to Isuru Fernando for suggesting the fix.
- Renamed BUILD_FLAGS to BUILD_CPPFLAGS to harmonize with the newly
created BUILD_SYMFLAGS.
- Fixed typo in entry for --export-shared flag in 'configure --help'
text.
Details:
- Documented the BLIS environment variables that were set
(e.g. BLIS_JC_NT, BLIS_IC_NT, BLIS_JR_NT) for each machine and
threading configuration in order to achieve the parallelism reported
on in docs/Performance.md.
Details:
- Increased the double-precision real MT threshold (which controls
whether the sup implementation kicks for smaller m dimension values)
from 80 to 180, and this change was made for both haswell and zen
subconfigurations. This is less about the m dimension in particular
and more about facilitating a smoother performance transition when
m = n = k.
Details:
- Commented out redundant setting of LIBBLIS_LINK within all driver-
level Makefiles. This variable is already set within common.mk, and
so the only time it should be overridden is if the user wants to link
to a different copy of libblis.
- Very minor changes to build/gen-make-frags/gen-make-frag.sh.
- Whitespace and inconsequential quoting change to configure.
- Moved top-level 'windows' directory into a new 'attic' directory.
Details:
- Added
#ifndef _POSIX_C_SOURCE
#define _POSIX_C_SOURCE 200809L
#endif
to bli_system.h so that an application that uses BLIS (specifically,
an application that #includes blis.h) does not need to remember to
#define the macro itself (either on the command line or in the code
that includes blis.h) in order to activate things like the pthreads.
Thanks to Christos Psarras for reporting this issue and suggesting
this fix.
- Commented out #include <sys/time.h> in bli_system.h, since I don't
think this header is used/needed anymore.
- Comment update to function macro for bli_?normiv_unb_var1() in
frame/util/bli_util_unb_var1.c.
Details:
- Fixed an obscure but in the bli_dgemmsup_rv_haswell_asm_5x8n() kernel
that only affected the beta == 0, column-storage output case. Thanks
to the BLAS test drivers for catching this bug.
- Previously, bli_gemmsup_ref_var1n() and _var2m() were returning if
k = 0, when the correct action would be to scale by beta (and then
return). Thanks to the BLAS test drivers to catching this bug.
- Changed the sup threshold behavior such that the sup implementation
only kicks in if a matrix dimension is strictly less than (rather than
less than or equal to) the threshold in question.
- Initialize all thresholds to zero (instead of 10) by default in
ref_kernels/bli_cntx_ref.c. This, combined with the above change to
threshold testing means that calls to BLIS or BLAS with one or more
matrix dimensions of zero will no longer trigger the sup
implementation.
- Added disabled debugging output to frame/3/bli_l3_sup.c (for future
use, perhaps).
Details:
- Removed already limited use of the BLIS_ENABLE_SUP_MR_EXT and
BLIS_ENABLE_SUP_NR_EXT macros in bli_gemmsup_ref_var1n() and
bli_gemmsup_ref_var2m(). Their purpose was merely to avoid a long
conditional that would determine whether to allow the last iteration
to be merged with the second-to-last iteration. Functionally, the
macros were not needed, and they ended up causing problems when
building configuration families such as intel64 and x86_64.
Details:
- Fixed an incorrectly-named macro guard that is intended to allow
disabling of the sup framework via the configure option
--disable-sup-handling. In this case, the preprocessor macro,
BLIS_DISABLE_SUP_HANDLING, was still named by its name from an older
uncommitted version of the code (BLIS_DISABLE_SM_HANDLING).
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:
- Somehow the variable name change (root_file_name -> root_inputname)
in flatten-headers.py mentioned in the commit log entry for 89a70cc
didn't make it into the actual commit. This commit applies that
change.
Details:
- Changed the default installation prefix from $HOME/lib to /usr/local.
- Modified the way configure internally handles the prefix, libdir,
includedir, and sharedir (and also added an --exec-prefix option).
The defaults to these variables are set as follows:
prefix: /usr/local
exec_prefix: ${prefix}
libdir: ${exec_prefix}/lib
includedir: ${prefix}/include
sharedir: ${prefix}/share
The key change, aside from the addition of exec_prefix and its use to
define the default to libdir, is that the variables are substituted
into config.mk with quoting that delays evaluation, meaning the
substituted values may contain unevaluated references to other
variables (namely, ${prefix} and ${exec_prefix}). This more closely
follows GNU conventions, including those used by GNU autoconf, and
also allows make to override any one of the variables *after*
configure has already been run (e.g. during 'make install').
- Updates to build/config.mk.in pursuant to above changes.
- Updates to output of 'configure --help' pursuant to above changes.
- Updated docs/BuildSystem.md to reflect the new default installation
prefix, as well as mention EXECPREFIX and SHAREDIR.
- Changed the definitions of the UNINSTALL_OLD_* variables in the
top-level Makefile to use $(wildcard ...) instead of 'find'. This
was motivated by the new way of handling prefix and friends, which
leads to the 'find' command being run on /usr/local (by default),
which can take a while almost never yielding any benefit (since the
user will very rarely use the uninstall-old targets).
- Removed periods from the end of descriptive output statements (i.e.,
non-verbose output) since those statements often end with file or
directory paths, which get confusing to read when puctuated by a
period.
- Trival change to 'make showconfig' output.
- Removed my name from 'configure --help'. (Many have contributed to it
over the years.)
- In configure script, changed the default state of threading_model
variable from 'no' to 'off' to match that of debug_type, where there
are similarly more than two valid states. ('no' is still accepted
if given via the --enable-debug= option, though it will be
standardized to 'off' prior to config.mk being written out.)
- Minor variable name change in flatten-headers.py that was intended for
32812ff.
- CREDITS file update.
