gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Bench: re-order GEMM-bench order (Putting DeepBench on top as it's most relevant)

Browse Source
This commit is contained in:
Philippe Tillet
2016-09-30 01:20:05 -04:00
parent 284fb5e109
commit b514638d86
2 changed files with 3 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
bench/blas.cpp
View File

@@ -163,9 +163,6 @@ void bench(sc::numeric_type dtype, std::string operation)
if(operation.substr(0,4)=="gemm") if(operation.substr(0,4)=="gemm")
{ {
std::vector<std::tuple<std::string, int_t, int_t, int_t, char, char> > MNKs; std::vector<std::tuple<std::string, int_t, int_t, int_t, char, char> > MNKs;
//Square
MNKs.push_back(make_tuple("Square",896,896,896,'N','T'));
MNKs.push_back(make_tuple("Square",2560,2560,2560,'N','T'));
//DeepBench //DeepBench
for(size_t MK: std::vector<size_t>{1760, 2048, 2560}) for(size_t MK: std::vector<size_t>{1760, 2048, 2560})
for(size_t N: std::vector<size_t>{16, 32, 64, 128, MK}) for(size_t N: std::vector<size_t>{16, 32, 64, 128, MK})
@@ -182,6 +179,9 @@ void bench(sc::numeric_type dtype, std::string operation)
MNKs.push_back(make_tuple("Lapack",4096,4096,32,'N','T')); MNKs.push_back(make_tuple("Lapack",4096,4096,32,'N','T'));
MNKs.push_back(make_tuple("Lapack",3456,3456,32,'N','T')); MNKs.push_back(make_tuple("Lapack",3456,3456,32,'N','T'));
MNKs.push_back(make_tuple("Lapack",896,896,32,'N','T')); MNKs.push_back(make_tuple("Lapack",896,896,32,'N','T'));
//Square
MNKs.push_back(make_tuple("Square",896,896,896,'N','T'));
MNKs.push_back(make_tuple("Square",2560,2560,2560,'N','T'));
std::cout << color_stream(ITALIC) << color_stream(BOLD) ; std::cout << color_stream(ITALIC) << color_stream(BOLD) ;
std::cout << "BENCH\tM\tN\tK\tAT\tBT\tISAAC"; std::cout << "BENCH\tM\tN\tK\tAT\tBT\tISAAC";

4
python/setup.py
View File

@@ -73,11 +73,7 @@ def main():
libraries += ['gnustl_shared'] libraries += ['gnustl_shared']
#Source files #Source files
<<<<<<< Updated upstream
src = 'src/lib/exception/api.cpp src/lib/exception/driver.cpp src/lib/value_scalar.cpp src/lib/random/rand.cpp src/lib/driver/check.cpp src/lib/driver/ndrange.cpp src/lib/driver/platform.cpp src/lib/driver/backend.cpp src/lib/driver/program.cpp src/lib/driver/command_queue.cpp src/lib/driver/event.cpp src/lib/driver/kernel.cpp src/lib/driver/handle.cpp src/lib/driver/device.cpp src/lib/driver/program_cache.cpp src/lib/driver/buffer.cpp src/lib/driver/context.cpp src/lib/driver/dispatch.cpp src/lib/jit/generation/engine/stream.cpp src/lib/jit/generation/engine/keywords.cpp src/lib/jit/generation/reduce_1d.cpp src/lib/jit/generation/elementwise_1d.cpp src/lib/jit/generation/base.cpp src/lib/jit/generation/elementwise_2d.cpp src/lib/jit/generation/matrix_product.cpp src/lib/jit/generation/reduce_2d.cpp src/lib/jit/syntax/engine/object.cpp src/lib/jit/syntax/engine/macro.cpp src/lib/jit/syntax/engine/process.cpp src/lib/jit/syntax/engine/binder.cpp src/lib/jit/syntax/expression/operations.cpp src/lib/jit/syntax/expression/expression.cpp src/lib/jit/syntax/expression/preset.cpp src/lib/api/blas/clBLAS.cpp src/lib/api/blas/cublas.cpp src/lib/runtime/execute.cpp src/lib/runtime/scheduler/dag.cpp src/lib/runtime/scheduler/strategies/heft.cpp src/lib/runtime/inference/predictors/random_forest.cpp src/lib/runtime/inference/profiles.cpp src/lib/runtime/inference/database.cpp src/lib/array.cpp '.split() + [os.path.join('src', 'bind', sf) for sf in ['_isaac.cpp', 'core.cpp', 'driver.cpp', 'kernels.cpp', 'exceptions.cpp']]
=======
src = 'src/lib/random/rand.cpp src/lib/jit/syntax/expression/preset.cpp src/lib/jit/syntax/expression/expression.cpp src/lib/jit/syntax/expression/operations.cpp src/lib/jit/syntax/engine/binder.cpp src/lib/jit/syntax/engine/macro.cpp src/lib/jit/syntax/engine/object.cpp src/lib/jit/syntax/engine/process.cpp src/lib/jit/generation/base.cpp src/lib/jit/generation/engine/keywords.cpp src/lib/jit/generation/engine/stream.cpp src/lib/jit/generation/reduce_1d.cpp src/lib/jit/generation/elementwise_1d.cpp src/lib/jit/generation/matrix_product.cpp src/lib/jit/generation/reduce_2d.cpp src/lib/jit/generation/elementwise_2d.cpp src/lib/runtime/execute.cpp src/lib/runtime/scheduler/dag.cpp src/lib/runtime/scheduler/strategies/heft.cpp src/lib/runtime/inference/predictors/random_forest.cpp src/lib/runtime/inference/database.cpp src/lib/runtime/inference/profiles.cpp src/lib/value_scalar.cpp src/lib/exception/api.cpp src/lib/exception/driver.cpp src/lib/driver/kernel.cpp src/lib/driver/program.cpp src/lib/driver/check.cpp src/lib/driver/backend.cpp src/lib/driver/context.cpp src/lib/driver/command_queue.cpp src/lib/driver/device.cpp src/lib/driver/platform.cpp src/lib/driver/handle.cpp src/lib/driver/dispatch.cpp src/lib/driver/buffer.cpp src/lib/driver/program_cache.cpp src/lib/driver/event.cpp src/lib/driver/ndrange.cpp src/lib/array.cpp src/lib/api/blas/cublas.cpp src/lib/api/blas/clBLAS.cpp '.split() + [os.path.join('src', 'bind', sf) for sf in ['_isaac.cpp', 'core.cpp', 'driver.cpp', 'kernels.cpp', 'exceptions.cpp']] src = 'src/lib/random/rand.cpp src/lib/jit/syntax/expression/preset.cpp src/lib/jit/syntax/expression/expression.cpp src/lib/jit/syntax/expression/operations.cpp src/lib/jit/syntax/engine/binder.cpp src/lib/jit/syntax/engine/macro.cpp src/lib/jit/syntax/engine/object.cpp src/lib/jit/syntax/engine/process.cpp src/lib/jit/generation/base.cpp src/lib/jit/generation/engine/keywords.cpp src/lib/jit/generation/engine/stream.cpp src/lib/jit/generation/reduce_1d.cpp src/lib/jit/generation/elementwise_1d.cpp src/lib/jit/generation/matrix_product.cpp src/lib/jit/generation/reduce_2d.cpp src/lib/jit/generation/elementwise_2d.cpp src/lib/runtime/execute.cpp src/lib/runtime/scheduler/dag.cpp src/lib/runtime/scheduler/strategies/heft.cpp src/lib/runtime/inference/predictors/random_forest.cpp src/lib/runtime/inference/database.cpp src/lib/runtime/inference/profiles.cpp src/lib/value_scalar.cpp src/lib/exception/api.cpp src/lib/exception/driver.cpp src/lib/driver/kernel.cpp src/lib/driver/program.cpp src/lib/driver/check.cpp src/lib/driver/backend.cpp src/lib/driver/context.cpp src/lib/driver/command_queue.cpp src/lib/driver/device.cpp src/lib/driver/platform.cpp src/lib/driver/handle.cpp src/lib/driver/dispatch.cpp src/lib/driver/buffer.cpp src/lib/driver/program_cache.cpp src/lib/driver/event.cpp src/lib/driver/ndrange.cpp src/lib/array.cpp src/lib/api/blas/cublas.cpp src/lib/api/blas/clBLAS.cpp '.split() + [os.path.join('src', 'bind', sf) for sf in ['_isaac.cpp', 'core.cpp', 'driver.cpp', 'kernels.cpp', 'exceptions.cpp']]
>>>>>>> Stashed changes
boostsrc = 'external/boost/libs/' boostsrc = 'external/boost/libs/'
for s in ['numpy','python','smart_ptr','system','thread']: for s in ['numpy','python','smart_ptr','system','thread']:
src = src + [x for x in recursive_glob('external/boost/libs/' + s + '/src/','.cpp') if 'win32' not in x and 'pthread' not in x] src = src + [x for x in recursive_glob('external/boost/libs/' + s + '/src/','.cpp') if 'win32' not in x and 'pthread' not in x]