mirror of
https://github.com/amd/blis.git
synced 2026-05-14 03:02:08 +00:00
b37634540f
Details: - Updated the test/sup source file (test_gemm.c) and Makefile to support building matrices with small or large leading dimensions, and updated runme.sh to support executing both kinds of test drivers. - Updated runme.sh to allow for executing sup drivers with unpacked (the default) or packed matrices (via setting BLIS_PACK_A, BLIS_PACK_B environment variables), and for capturing output to files that encode both the leading dimension (small or large) and packing status into the filenames. - Consolidated octave scripts in test/sup/octave_st, test/sup/octave_mt into test/sup/octave and updated the octave code in that consolidated directory to read the new output filename format (encoding ldim and packing). Also added comments and streamlined code, particularly in plot_panel_trxsh.m. Tested the octave scripts with octave 5.2.0. - Moved old octave_st, octave_mt directories to test/sup/old.
293 lines
8.0 KiB
Bash
Executable File
293 lines
8.0 KiB
Bash
Executable File
#!/bin/bash
|
|
|
|
# File pefixes.
|
|
exec_root="test"
|
|
out_root="output"
|
|
|
|
sys="blis"
|
|
#sys="lonestar5"
|
|
#sys="ul252"
|
|
#sys="ul264"
|
|
|
|
if [ ${sys} = "blis" ]; then
|
|
|
|
export GOMP_CPU_AFFINITY="0-3"
|
|
nt=4
|
|
|
|
elif [ ${sys} = "lonestar5" ]; then
|
|
|
|
export GOMP_CPU_AFFINITY="0-23"
|
|
nt=12
|
|
|
|
elif [ ${sys} = "ul252" ]; then
|
|
|
|
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/home/field/intel/mkl/lib/intel64"
|
|
export GOMP_CPU_AFFINITY="0-51"
|
|
nt=26
|
|
|
|
elif [ ${sys} = "ul264" ]; then
|
|
|
|
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/home/field/intel/mkl/lib/intel64"
|
|
export GOMP_CPU_AFFINITY="0-63"
|
|
nt=32
|
|
|
|
fi
|
|
|
|
# Delay between test cases.
|
|
delay=0.02
|
|
|
|
# Threadedness to test.
|
|
#threads="st mt"
|
|
threads="st"
|
|
|
|
# Datatypes to test.
|
|
#dts="d s"
|
|
dts="d"
|
|
|
|
# Operations to test.
|
|
ops="gemm"
|
|
|
|
# Transpose combintions to test.
|
|
trans="nn nt tn tt"
|
|
|
|
# Storage combinations to test.
|
|
# NOTE: mixed storage cases are not yet implemented in test_gemm.c.
|
|
#stors="rrr rrc rcr rcc crr crc ccr ccc"
|
|
stors="rrr ccc"
|
|
|
|
# Problem shapes to test.
|
|
shapes="sll lsl lls lss sls ssl lll"
|
|
|
|
# Small problem dimensions to use.
|
|
# FGVZ: figure out how to probe what's in the directory and
|
|
# execute everything that's there?
|
|
sms="6"
|
|
sns="8"
|
|
sks="4"
|
|
#sks="10"
|
|
|
|
# Leading dimensions to use (small or large).
|
|
# When a leading dimension is large, it is constant and set to the largest
|
|
# problem size that will be run.
|
|
#ldims="s l"
|
|
ldims="s"
|
|
|
|
# Packing combinations for blissup. The first char encodes the packing status
|
|
# of matrix A and the second char encodes the packing status of matrix B.
|
|
# NOTE: This string must always contain 'uu' if other implementations are also
|
|
# being tested at the same time.
|
|
#pcombos="uu up pu pp"
|
|
pcombos="uu"
|
|
|
|
# Implementations to test.
|
|
impls="vendor blissup blisconv openblas eigen blasfeo libxsmm"
|
|
#impls="vendor blisconv openblas eigen blasfeo libxsmm"
|
|
#impls="vendor"
|
|
#impls="blissup"
|
|
#impls="blisconv"
|
|
#impls="openblas"
|
|
#impls="eigen"
|
|
|
|
# Save a copy of GOMP_CPU_AFFINITY so that if we have to unset it, we can
|
|
# restore the value.
|
|
GOMP_CPU_AFFINITYsave=${GOMP_CPU_AFFINITY}
|
|
|
|
# Example: test_dgemm_nn_rrc_m6npkp_blissup_st.x
|
|
|
|
for th in ${threads}; do
|
|
|
|
for dt in ${dts}; do
|
|
|
|
for op in ${ops}; do
|
|
|
|
for tr in ${trans}; do
|
|
|
|
for st in ${stors}; do
|
|
|
|
for sh in ${shapes}; do
|
|
|
|
for sm in ${sms}; do
|
|
|
|
for sn in ${sns}; do
|
|
|
|
for sk in ${sks}; do
|
|
|
|
for ld in ${ldims}; do
|
|
|
|
for im in ${impls}; do
|
|
|
|
for pc in ${pcombos}; do
|
|
|
|
if [ "${th}" = "mt" ]; then
|
|
|
|
# Prohibit attempts to run blasfeo or libxsmm as
|
|
# multithreaded.
|
|
if [ "${im}" = "blasfeo" ] || \
|
|
[ "${im}" = "libxsmm" ]; then
|
|
continue;
|
|
fi
|
|
|
|
# Specify the multithreading depending on which
|
|
# implementation is about to be tested.
|
|
if [ "${im:0:4}" = "blis" ]; then
|
|
unset OMP_NUM_THREADS
|
|
export BLIS_NUM_THREADS=${nt}
|
|
elif [ "${im}" = "openblas" ]; then
|
|
unset OMP_NUM_THREADS
|
|
export OPENBLAS_NUM_THREADS=${nt}
|
|
elif [ "${im}" = "eigen" ]; then
|
|
export OMP_NUM_THREADS=${nt}
|
|
elif [ "${im}" = "vendor" ]; then
|
|
unset OMP_NUM_THREADS
|
|
export MKL_NUM_THREADS=${nt}
|
|
fi
|
|
export nt_use=${nt}
|
|
|
|
else # if [ "${th}" = "st" ];
|
|
|
|
# Use single-threaded execution.
|
|
export OMP_NUM_THREADS=1
|
|
export BLIS_NUM_THREADS=1
|
|
export OPENBLAS_NUM_THREADS=1
|
|
export MKL_NUM_THREADS=1
|
|
export nt_use=1
|
|
fi
|
|
|
|
# Isolate the individual chars in the current pcombo
|
|
# string.
|
|
packa=${pc:0:1}
|
|
packb=${pc:1:1}
|
|
|
|
# For blissup implementations, set the BLIS_PACK_A and
|
|
# BLIS_PACK_B environment variables according to the
|
|
# chars in the current pcombo string.
|
|
if [ "${im:0:7}" = "blissup" ]; then
|
|
|
|
# Set BLIS_PACK_A if the pcombo char is 'p'; otherwise
|
|
# unset the variable altogether.
|
|
if [ ${packa} = "p" ]; then
|
|
export BLIS_PACK_A=1
|
|
else
|
|
unset BLIS_PACK_A
|
|
fi
|
|
|
|
# Set BLIS_PACK_B if the pcombo char is 'p'; otherwise
|
|
# unset the variable altogether.
|
|
if [ ${packb} = "p" ]; then
|
|
export BLIS_PACK_B=1
|
|
else
|
|
unset BLIS_PACK_B
|
|
fi
|
|
else
|
|
|
|
# Unset the variables for non-blissup implementations,
|
|
# just to be paranoid-safe.
|
|
unset BLIS_PACK_A
|
|
unset BLIS_PACK_B
|
|
fi
|
|
|
|
# Limit execution of non-blissup implementations to the
|
|
# 'uu' packing combination. (Those implementations don't
|
|
# use the pcombos string, but since we iterate over its
|
|
# words for all implementations, we have to designate one
|
|
# of them as a placeholder to allow those implementations
|
|
# to execute. The 'uu' string was chosen over the 'pp'
|
|
# string because it's more likely that this script will be
|
|
# used to run blissup on unpacked matrices, and so the
|
|
# sorting for the output files is nicer if the non-blissup
|
|
# implementations use the 'uu' string, even if it's more
|
|
# likely that those implementations use packing. Think of
|
|
# 'uu' as encoding the idea that explicit packing was not
|
|
# requested.)
|
|
if [ "${im:0:7}" != "blissup" ] && \
|
|
[ "${pc}" != "uu" ]; then
|
|
continue;
|
|
fi
|
|
|
|
# Multithreaded OpenBLAS seems to have a problem
|
|
# running properly if GOMP_CPU_AFFINITY is set.
|
|
# So we temporarily unset it here if we are about
|
|
# to execute OpenBLAS, but otherwise restore it.
|
|
if [ ${im} = "openblas" ]; then
|
|
unset GOMP_CPU_AFFINITY
|
|
else
|
|
export GOMP_CPU_AFFINITY="${GOMP_CPU_AFFINITYsave}"
|
|
fi
|
|
|
|
# Limit execution of non-BLIS implementations to
|
|
# rrr/ccc storage cases.
|
|
if [ "${im:0:4}" != "blis" ] && \
|
|
[ "${st}" != "rrr" ] && \
|
|
[ "${st}" != "ccc" ]; then
|
|
continue;
|
|
fi
|
|
|
|
# Further limit execution of libxsmm to
|
|
# ccc storage cases.
|
|
if [ "${im:0:7}" = "libxsmm" ] && \
|
|
[ "${st}" != "ccc" ]; then
|
|
continue;
|
|
fi
|
|
|
|
# Extract the shape chars for m, n, k.
|
|
chm=${sh:0:1}
|
|
chn=${sh:1:1}
|
|
chk=${sh:2:1}
|
|
|
|
# Construct the shape substring (e.g. m6npkp)
|
|
shstr=""
|
|
|
|
if [ ${chm} = "s" ]; then
|
|
shstr="${shstr}m${sm}"
|
|
else
|
|
shstr="${shstr}mp"
|
|
fi
|
|
|
|
if [ ${chn} = "s" ]; then
|
|
shstr="${shstr}n${sn}"
|
|
else
|
|
shstr="${shstr}np"
|
|
fi
|
|
|
|
if [ ${chk} = "s" ]; then
|
|
shstr="${shstr}k${sk}"
|
|
else
|
|
shstr="${shstr}kp"
|
|
fi
|
|
|
|
# Construct the ldim substring (e.g. lds or ldl)
|
|
ldstr="ld${ld}"
|
|
|
|
# Construct the pack substring (e.g. uaub, uapb, paub, or papb)
|
|
packstr="${packa}a${packb}b"
|
|
|
|
# Ex: test_dgemm_nn_rrc_m6npkp_blissup_st.x
|
|
# Ex: test_dgemm_nt_rrr_m6npkp_ldl_blissup_st.x
|
|
|
|
# Construct the name of the test executable.
|
|
exec_name="${exec_root}_${dt}${op}_${tr}_${st}_${shstr}_${ldstr}_${im}_${th}.x"
|
|
|
|
# Construct the name of the output file.
|
|
out_file="${out_root}_${th}_${dt}${op}_${tr}_${st}_${shstr}_${ldstr}_${packstr}_${im}.m"
|
|
|
|
echo "Running (nt = ${nt_use}) ./${exec_name} > ${out_file}"
|
|
|
|
# Run executable.
|
|
./${exec_name} > ${out_file}
|
|
|
|
sleep ${delay}
|
|
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
done
|
|
|