Squashed feature branch:
* Added CUDA support * Performance improvements * API improvements * Added "depth" parameter to GEMM * Android cross-compilation
This commit is contained in:
Philippe Tillet
@@ -2,7 +2,7 @@ from __future__ import division
|
||||
|
||||
import argparse, itertools, os, sys, json
|
||||
import misc_tools, optimize, dataset
|
||||
import pyatidlas as atd
|
||||
import pyisaac as atd
|
||||
import numpy as np
|
||||
|
||||
from numpy import random
|
||||
@@ -71,27 +71,20 @@ def do_tuning(args):
|
||||
return optimize.genetic(symbolic, Template, lambda t: TYPES[operation]['perf-index']([datatype(0).size, sizes, t]),
|
||||
TYPES[operation]['perf-measure'], archive)
|
||||
|
||||
def log_uniform_sample(a,b):
|
||||
return np.exp(np.random.uniform(low=np.log(a), high=np.log(b), size=1)).astype(int)
|
||||
|
||||
def space_gen_product(a,b,N,dim,method):
|
||||
N = int(N**(1.0/dim))
|
||||
def space_gen(a,b,method):
|
||||
for i in range(N):
|
||||
if method == 'linear':
|
||||
v = int(a + (b-a)*i/N)
|
||||
if method == 'log':
|
||||
v = int(np.exp(np.log(a) + (np.log(b) - np.log(a))*i/N))
|
||||
yield (v//64 + 1)*64
|
||||
return tuple(itertools.product(*[space_gen(a,b,method) for i in range(dim)]))
|
||||
def log_spaced_points(a,b,N,r=128):
|
||||
t = np.ceil(np.exp(np.linspace(np.log(a), np.log(b), N))/r)*r
|
||||
return t.reshape(t.size,1).astype(int)
|
||||
|
||||
|
||||
#Helper for tuning
|
||||
def tune(execution_handler, a, b, dimsample, layouts, sample_method_profiles, sample_method_dataset):
|
||||
def tune(execution_handler, layouts, tuning_sizes, training_sizes):
|
||||
print('-----')
|
||||
print(' '.join(map(str, ("Now tuning:", dtypestr, '-', operation, '-'.join(layouts), '[' + device.name, '(' + device.platform.name + ')]'))))
|
||||
#Update JSON
|
||||
full_operation = operation + ''.join(layouts)
|
||||
prefix = os.path.join('data',os.path.join(full_operation,dtypestr))
|
||||
if not os.path.exists(prefix):
|
||||
os.makedirs(prefix)
|
||||
if full_operation not in json_out:
|
||||
json_out[full_operation] = {}
|
||||
json_out[full_operation][dtypestr] = {}
|
||||
@@ -103,15 +96,13 @@ def do_tuning(args):
|
||||
else:
|
||||
def compute_perf(x, t):
|
||||
return TYPES[operation]['perf-index']([datatype(0).size, x, t])
|
||||
profiles_generator = space_gen_product(a, b, args.sample_size, dimsample, sample_method_profiles)
|
||||
profiles = dataset.sample_profiles(execution_handler, profiles_generator)
|
||||
#profiles = dataset.sample_profiles(execution_handler, tuning_sizes)
|
||||
if args.build_model:
|
||||
dataset_generator = space_gen_product(a, b, 1000, dimsample, sample_method_dataset)
|
||||
X, Y, profiles = dataset.sample_dataset(os.path.join(full_operation,dtypestr), profiles, execution_handler, dataset_generator)
|
||||
# profiles = np.loadtxt('data/'+full_operation+'/'+datatype+'/profiles.csv')
|
||||
# X = np.loadtxt('data/'+full_operation+'/'+datatype+'/X.csv',ndmin=2)
|
||||
# Y = np.loadtxt('data/'+full_operation+'/'+datatype+'/Y.csv',ndmin=2)
|
||||
clf = train_model(X, Y, profiles, TYPES[operation]['perf-measure'])
|
||||
#X, Y, profiles = dataset.sample_dataset(prefix, profiles, execution_handler, training_sizes)
|
||||
profiles = np.loadtxt(prefix+'/profiles.csv')
|
||||
X = np.loadtxt(prefix+'/X.csv',ndmin=2)
|
||||
Y = np.loadtxt(prefix+'/Y.csv',ndmin=2)
|
||||
clf = train_model(X, Y, profiles, compute_perf, TYPES[operation]['perf-measure'])
|
||||
D['predictor'] = [{'children_left': e.tree_.children_left.tolist(),
|
||||
'children_right': e.tree_.children_right.tolist(),
|
||||
'threshold': e.tree_.threshold.astype('float64').tolist(),
|
||||
@@ -128,15 +119,15 @@ def do_tuning(args):
|
||||
x = atd.empty(sizes[0], datatype, context=context)
|
||||
y = atd.empty(sizes[0], datatype, context=context)
|
||||
return execute(x + y, sizes, Template, parameters, fname)
|
||||
tune(execution_handler, 1e3, 2e7, 1, (),'log', 'log')
|
||||
#dot
|
||||
tune(execution_handler, (), log_spaced_points(1e4, 1e7, 20), log_spaced_points(1e4, 1e7, 1000))
|
||||
#Dot
|
||||
if operation=='dot':
|
||||
def execution_handler(sizes, fname=os.devnull, parameters=None):
|
||||
x = atd.empty(sizes[0], datatype, context=context)
|
||||
y = atd.empty(sizes[0], datatype, context=context)
|
||||
s = atd.scalar(datatype)
|
||||
return execute(atd.dot(x, y), sizes, Template, parameters, fname)
|
||||
tune(execution_handler, 1e3, 2e7, 1, (),'log', 'log')
|
||||
tune(execution_handler, (), log_spaced_points(1e4, 1e7, 50), log_spaced_points(1e4, 1e7, 1000))
|
||||
#Matrix AXPY
|
||||
if operation=='maxpy':
|
||||
def execution_handler(sizes, fname=os.devnull, parameters=None):
|
||||
@@ -152,7 +143,14 @@ def do_tuning(args):
|
||||
x = atd.empty(sizes[1], datatype, context=context)
|
||||
LHS = A if A_trans=='N' else A.T
|
||||
return execute(atd.dot(LHS, x), sizes, Template[A_trans], parameters, fname)
|
||||
tune(execution_handler, 64, 6000, 2, (A_trans,),'log', 'log')
|
||||
tuning_sizes = itertools.chain( itertools.product([128, 512, 2048, 8192], [128, 512, 2048, 8192]),
|
||||
itertools.product([128, 512, 2048, 8192], [16384, 32768, 65536]),
|
||||
itertools.product([16384, 32768, 65536], [128, 512, 2048, 8192]))
|
||||
|
||||
training_sizes = itertools.chain( itertools.product([2**k for k in range(4, 13)], [2**k for k in range(4, 13)]),
|
||||
itertools.product([2**k for k in range(4, 13)], [2**k for k in range(13, 17)]),
|
||||
itertools.product([2**k for k in range(13, 17)], [2**k for k in range(4, 13)]))
|
||||
tune(execution_handler, (A_trans,), tuning_sizes, training_sizes)
|
||||
#Matrix Product
|
||||
if operation=='gemm':
|
||||
for L in args.gemm_layouts:
|
||||
@@ -164,8 +162,11 @@ def do_tuning(args):
|
||||
LHS = A if A_trans=='N' else A.T
|
||||
RHS = B if B_trans=='N' else B.T
|
||||
return execute(atd.dot(LHS, RHS), sizes, Template[(A_trans, B_trans)], parameters, fname)
|
||||
tune(execution_handler, 100, 2000, 3,(A_trans,B_trans), 'linear', 'linear')
|
||||
|
||||
|
||||
tuning_sizes = itertools.product([64, 256, 1024, 2560], [64, 256, 1024, 2560], [256, 2560, 32768, 65536])
|
||||
training_sizes = itertools.product([2**k for k in range(6, 13)], [2**k for k in range(6, 13)], [2**k for k in range(6, 17)])
|
||||
tune(execution_handler,(A_trans,B_trans), tuning_sizes, training_sizes)
|
||||
|
||||
json.dump(json_out, open(args.out,'w'))
|
||||
|
||||
|
||||
@@ -194,7 +195,7 @@ class ArgumentsHandler:
|
||||
|
||||
full_parser = tune_subparsers.add_parser('full', help = 'Tune each operation for randomly chosen sizes')
|
||||
full_parser.add_argument("--build-model", default=True, type=bool)
|
||||
full_parser.add_argument("--sample-size", default=60, type=int)
|
||||
full_parser.add_argument("--sample-size", default=64, type=int)
|
||||
|
||||
args = parser.parse_args()
|
||||
self.__dict__ = args.__dict__.copy()
|
||||
|
||||
Reference in New Issue
Block a user