saiph: simplify granularity and ss experiments
This commit is contained in:
parent
0e0f1b265f
commit
10b1ff8f7a
@ -32,8 +32,8 @@
|
||||
};
|
||||
|
||||
saiph = {
|
||||
granularity-saiph = callPackage ./saiph/granularity-saiph.nix { };
|
||||
scalability-saiph = callPackage ./saiph/scalability-saiph.nix { };
|
||||
granularity = callPackage ./saiph/granularity.nix { };
|
||||
ss = callPackage ./saiph/ss.nix { };
|
||||
};
|
||||
|
||||
creams = rec {
|
||||
|
||||
@ -1,4 +1,4 @@
|
||||
######################################################################################
|
||||
#######################################################################
|
||||
# Saiph, granularity experiment:
|
||||
#
|
||||
# App:Heat 3D - garlic/tampi+isend+oss+task+simd branch
|
||||
@ -15,12 +15,13 @@
|
||||
# Granularity experiment configuration:
|
||||
# Single-core run
|
||||
# MPI binded to sockets: MPI procs = 2
|
||||
# Mesh distributed across third dimension to ensure contiguous communications
|
||||
# Mesh distributed across third dimension to ensure contiguous
|
||||
# communications
|
||||
# --> nbgx = 1, nbgy = 1
|
||||
# First dimension cannot be locally blocked (simd reasons)
|
||||
# Second and third dimension local blocking limited by local mesh size
|
||||
#
|
||||
######################################################################################
|
||||
#######################################################################
|
||||
|
||||
# Common packages, tools and options
|
||||
{
|
||||
@ -29,81 +30,102 @@
|
||||
, bsc
|
||||
, targetMachine
|
||||
, stages
|
||||
, garlicTools
|
||||
}:
|
||||
|
||||
with stdenv.lib;
|
||||
with garlicTools;
|
||||
|
||||
let
|
||||
|
||||
#*** Variable configurations ***
|
||||
varConf = with bsc; {
|
||||
# Variable configurations
|
||||
varConf = with targetMachine.config; {
|
||||
# Local blocks per dimension
|
||||
nbl1 = [ 1 2 3 4 6 12 24 48 96 ];
|
||||
nbl2 = [ 1 2 3 4 6 12 24 48 96 ];
|
||||
nblx = [ 1 ]; # SIMD
|
||||
nbly = range2 1 (hw.cpusPerNode * 8);
|
||||
nblz = [ 8 ];
|
||||
sizex = [ 3 ];
|
||||
gitBranch = [ "garlic/tampi+isend+oss+task+simd" ];
|
||||
};
|
||||
|
||||
#*** Generate the complete configuration for each unit ***
|
||||
genConf = with bsc; c: targetMachine.config // rec {
|
||||
# Generate the complete configuration for each unit
|
||||
genConf = c: targetMachine.config // rec {
|
||||
|
||||
# Experiment, units and job names
|
||||
expName = "saiph-granularity";
|
||||
unitName = "${expName}-N${toString nodes}" + "nbg_${toString nbgx}-${toString nbgy}-${toString nbgz}" + "nbl_1-${toString nbly}-${toString nblz}";
|
||||
jobName = "${unitName}";
|
||||
unitName = "${expName}"
|
||||
+ "-N${toString nodes}"
|
||||
+ "-nbg.x${toString nbgx}.y${toString nbgy}.z${toString nbgz}"
|
||||
+ "-nbl.x${toString nblx}.y${toString nbly}.z${toString nblz}";
|
||||
|
||||
jobName = unitName;
|
||||
|
||||
# saiph options
|
||||
totalTasks = ntasksPerNode * nodes;
|
||||
nodes = 1;
|
||||
enableManualDist = true; # allows to manually set nbg{x-y-z}
|
||||
nbgx = 1;
|
||||
nbgy = 1;
|
||||
nbgz = nodes*2; # forcing distribution by last dim
|
||||
nblx = 1; # simd reasons
|
||||
nbly = c.nbl1; # takes values from varConf
|
||||
nblz = c.nbl2; # takes values from varConf
|
||||
mpi = impi;
|
||||
gitBranch = "garlic/tampi+isend+oss+task+simd";
|
||||
gitCommit = "8052494d7dc62bef95ebaca9938e82fb029686f6"; # fix a specific commit
|
||||
rev = "0";
|
||||
nbgz = totalTasks; # forcing distribution by last dim
|
||||
|
||||
# Repeat the execution of each unit 30 times
|
||||
loops = 30;
|
||||
inherit (c) nblx nbly nblz gitBranch sizex;
|
||||
|
||||
blocksPerTask = nblx * nbly * nblz * 1.0;
|
||||
blocksPerCpu = blocksPerTask / cpusPerTask;
|
||||
|
||||
# fix a specific commit
|
||||
gitCommit = "8052494d7dc62bef95ebaca9938e82fb029686f6";
|
||||
|
||||
# Repeat the execution of each unit 10 times
|
||||
loops = 10;
|
||||
|
||||
# Resources
|
||||
inherit (targetMachine.config) hw;
|
||||
qos = "debug";
|
||||
ntasksPerNode = hw.socketsPerNode; # MPI binded to sockets
|
||||
cpusPerTask = hw.cpusPerSocket; # Using the 24 CPUs of each socket
|
||||
ntasksPerNode = hw.socketsPerNode; # MPI binded to sockets
|
||||
cpusPerTask = hw.cpusPerSocket; # Using the 24 CPUs of each socket
|
||||
};
|
||||
|
||||
#*** Compute the final set of configurations ***
|
||||
# Compute the array of configurations: cartesian product of all factors
|
||||
# Compute the array of configurations: cartesian product of all
|
||||
# factors
|
||||
allConfigs = stdexp.buildConfigs {
|
||||
inherit varConf genConf;
|
||||
};
|
||||
|
||||
# Filter to remove non-desired configurations:
|
||||
# --> tasks/proc < 0.5
|
||||
# --> nblz > 50
|
||||
configs = filter (el: if ((builtins.mul el.nbly el.nblz) < (builtins.mul 0.5 el.cpusPerTask) || el.nblz > 50) then false else true) allConfigs;
|
||||
isGoodConfig = c:
|
||||
let
|
||||
maxNblz = c.cpusPerTask * 2;
|
||||
in
|
||||
! (c.blocksPerCpu < 0.5 || c.nblz > maxNblz);
|
||||
|
||||
configs = filter (isGoodConfig) allConfigs;
|
||||
|
||||
#*** Sets the env/argv of the program ***
|
||||
exec = {nextStage, conf, ...}: with conf; stages.exec {
|
||||
exec = {nextStage, conf, ...}: stages.exec {
|
||||
inherit nextStage;
|
||||
env = ''
|
||||
export OMP_NUM_THREADS=${toString hw.cpusPerSocket}
|
||||
export OMP_NUM_THREADS=${toString conf.cpusPerTask}
|
||||
'';
|
||||
};
|
||||
|
||||
#*** Configure the program according to the app ***
|
||||
program = {nextStage, conf, ...}:
|
||||
let
|
||||
customPkgs = stdexp.replaceMpi conf.mpi;
|
||||
in
|
||||
customPkgs.apps.saiph.override {
|
||||
inherit (conf) enableManualDist nbgx nbgy nbgz nblx nbly nblz mpi gitBranch gitCommit;
|
||||
};
|
||||
program = {nextStage, conf, ...}: bsc.apps.saiph.override {
|
||||
inherit (conf) enableManualDist
|
||||
nbgx nbgy nbgz nblx nbly nblz
|
||||
sizex
|
||||
gitBranch gitCommit;
|
||||
|
||||
L3SizeKB = conf.hw.cacheSizeKB.L3;
|
||||
cachelineBytes = conf.hw.cachelineBytes;
|
||||
};
|
||||
|
||||
#*** Add stages to the pipeline ***
|
||||
pipeline = stdexp.stdPipeline ++ [ exec program ];
|
||||
|
||||
in
|
||||
stdexp.genExperiment { inherit configs pipeline; }
|
||||
|
||||
stdexp.genExperiment { inherit configs pipeline; }
|
||||
@ -31,62 +31,68 @@
|
||||
, bsc
|
||||
, targetMachine
|
||||
, stages
|
||||
, garlicTools
|
||||
}:
|
||||
|
||||
with stdenv.lib;
|
||||
with garlicTools;
|
||||
|
||||
let
|
||||
|
||||
#*** Variable configurations ***
|
||||
varConf = with bsc; {
|
||||
varConf = with targetMachine.config; {
|
||||
# FIXME: None of those selected nbl* and problem size is able to give good
|
||||
# efficiency when testing strong scaling. We should find better values.
|
||||
# Local blocks per dimension
|
||||
nbl1 = [ 1 2 3 4 6 12 24 48 96 ];
|
||||
nbl2 = [ 1 2 3 4 6 12 24 48 96 ];
|
||||
# Number of nodes
|
||||
nodes = [ 1 2 4 8 ];
|
||||
nblx = [ 1 ]; # SIMD
|
||||
nbly = [ 32 ];
|
||||
nblz = [ 8 ];
|
||||
sizex = [ 3 6 ];
|
||||
gitBranch = [ "garlic/tampi+isend+oss+task+simd" ];
|
||||
nodes = range2 1 8;
|
||||
};
|
||||
|
||||
#*** Generate the complete configuration for each unit ***
|
||||
genConf = with bsc; c: targetMachine.config // rec {
|
||||
genConf = c: targetMachine.config // rec {
|
||||
|
||||
# Experiment, units and job names
|
||||
expName = "saiph-scalability";
|
||||
unitName = "${expName}-N${toString nodes}" + "nbg_${toString nbgx}-${toString nbgy}-${toString nbgz}" + "nbl_1-${toString nbly}-${toString nblz}";
|
||||
jobName = "${unitName}";
|
||||
expName = "saiph-ss";
|
||||
unitName = "${expName}"
|
||||
+ "-N${toString nodes}"
|
||||
+ "-nbg.x${toString nbgx}.y${toString nbgy}.z${toString nbgz}"
|
||||
+ "-nbl.x${toString nblx}.y${toString nbly}.z${toString nblz}";
|
||||
jobName = unitName;
|
||||
|
||||
# saiph options
|
||||
nodes = c.nodes; # takes values from varConf
|
||||
enableManualDist = true; # allows to manually set nbg{x-y-z}
|
||||
nbgx = 1;
|
||||
nbgy = 1;
|
||||
nbgz = nodes*2; # forcing distribution by last dim
|
||||
nblx = 1; # simd reasons
|
||||
nbly = c.nbl1; # takes values from varConf
|
||||
nblz = c.nbl2; # takes values from varConf
|
||||
mpi = impi;
|
||||
nbgz = nodes * ntasksPerNode; # forcing distribution by last dim
|
||||
|
||||
inherit (c) nblx nbly nblz nodes sizex;
|
||||
|
||||
gitBranch = "garlic/tampi+isend+oss+task+simd";
|
||||
gitCommit = "8052494d7dc62bef95ebaca9938e82fb029686f6"; # fix a specific commit
|
||||
rev = "0";
|
||||
# Repeat the execution of each unit 30 times
|
||||
loops = 30;
|
||||
|
||||
blocksPerTask = nblx * nbly * nblz * 1.0;
|
||||
blocksPerCpu = blocksPerTask / cpusPerTask;
|
||||
|
||||
# Repeat the execution of each unit 10 times
|
||||
loops = 10;
|
||||
|
||||
# Resources
|
||||
inherit (targetMachine.config) hw;
|
||||
qos = "bsc_cs";
|
||||
|
||||
qos = "debug";
|
||||
ntasksPerNode = hw.socketsPerNode; # MPI binded to sockets
|
||||
cpusPerTask = hw.cpusPerSocket; # Using the 24 CPUs of each socket
|
||||
};
|
||||
|
||||
#*** Compute the final set of configurations ***
|
||||
# Compute the array of configurations: cartesian product of all factors
|
||||
allConfigs = stdexp.buildConfigs {
|
||||
configs = stdexp.buildConfigs {
|
||||
inherit varConf genConf;
|
||||
};
|
||||
# Filter to remove non-desired configurations:
|
||||
# --> tasks/proc < 3
|
||||
# --> nblz > 25
|
||||
configs = filter (el: if ((builtins.mul el.nbly el.nblz) < (builtins.mul 3 el.cpusPerTask) || el.nblz > 25) then false else true) allConfigs;
|
||||
|
||||
|
||||
#*** Sets the env/argv of the program ***
|
||||
exec = {nextStage, conf, ...}: with conf; stages.exec {
|
||||
@ -97,17 +103,19 @@ let
|
||||
};
|
||||
|
||||
#*** Configure the program according to the app ***
|
||||
program = {nextStage, conf, ...}:
|
||||
let
|
||||
customPkgs = stdexp.replaceMpi conf.mpi;
|
||||
in
|
||||
customPkgs.apps.saiph.override {
|
||||
inherit (conf) enableManualDist nbgx nbgy nbgz nblx nbly nblz mpi gitBranch gitCommit;
|
||||
};
|
||||
program = {nextStage, conf, ...}: bsc.apps.saiph.override {
|
||||
inherit (conf) enableManualDist
|
||||
nbgx nbgy nbgz nblx nbly nblz
|
||||
sizex
|
||||
gitBranch gitCommit;
|
||||
|
||||
L3SizeKB = conf.hw.cacheSizeKB.L3;
|
||||
cachelineBytes = conf.hw.cachelineBytes;
|
||||
};
|
||||
|
||||
#*** Add stages to the pipeline ***
|
||||
pipeline = stdexp.stdPipeline ++ [ exec program ];
|
||||
|
||||
in
|
||||
stdexp.genExperiment { inherit configs pipeline; }
|
||||
|
||||
stdexp.genExperiment { inherit configs pipeline; }
|
||||
Loading…
Reference in New Issue
Block a user