Added H100 and P100 benchmark scripts (#18)

Added benchmark scripts for P100 (based on the configuration of A100) and H100 GPUs
GridTools · Apr 8, 2024 · 33658cd · 33658cd
1 parent 8575e02
commit 33658cd
Show file tree

Hide file tree

Showing 3 changed files with 357 additions and 0 deletions.
diff --git a/setup.py b/setup.py
@@ -93,10 +93,14 @@ def pybind11_extension(m):
         'console_scripts': [
             'sbench=stencil_benchmarks.scripts.sbench:main',
             'sbench-analyze=stencil_benchmarks.scripts.sbench_analyze:main',
+            'sbench-h100-collection=stencil_benchmarks.scripts'
+            '.sbench_h100_collection:main',
             'sbench-a100-collection=stencil_benchmarks.scripts'
             '.sbench_a100_collection:main',
             'sbench-v100-collection=stencil_benchmarks.scripts'
             '.sbench_v100_collection:main',
+            'sbench-p100-collection=stencil_benchmarks.scripts'
+            '.sbench_p100_collection:main',
             'sbench-mi50-collection=stencil_benchmarks.scripts'
             '.sbench_mi50_collection:main',
             'sbench-mi100-collection=stencil_benchmarks.scripts'

diff --git a/stencil_benchmarks/scripts/sbench_h100_collection.py b/stencil_benchmarks/scripts/sbench_h100_collection.py
@@ -0,0 +1,176 @@
+# Stencil Benchmarks
+#
+# Copyright (c) 2017-2021, ETH Zurich and MeteoSwiss
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+import click
+
+from stencil_benchmarks.benchmarks_collection.stencils.cuda_hip import (
+    basic, horizontal_diffusion as hdiff, vertical_advection as vadv)
+from stencil_benchmarks.tools.multirun import (Configuration,
+                                               run_scaling_benchmark,
+                                               truncate_block_size_to_domain,
+                                               default_kwargs)
+
+
+@click.group()
+def main():
+    pass
+
+
+common_kwargs = default_kwargs(backend='cuda',
+                               compiler='nvcc',
+                               gpu_architecture='sm_90',
+                               verify=False,
+                               dry_runs=1,
+                               alignment=128,
+                               dtype='float32')
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def basic_bandwidth(output, executions, option):
+    kwargs = common_kwargs(
+        option,
+        loop='3D',
+        block_size=(128, 2, 1),
+        halo=(1, 1, 1),
+    )
+
+    stream_kwargs = kwargs.copy()
+    stream_kwargs.update(loop='1D', block_size=(1024, 1, 1), halo=(0, 0, 0))
+
+    configurations = [
+        Configuration(basic.Copy, name='stream', **stream_kwargs),
+        Configuration(basic.Empty, name='empty', **kwargs),
+        Configuration(basic.Copy, name='copy', **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-i', axis=0, **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-j', axis=1, **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-k', axis=2, **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-i',
+                      axis=0,
+                      **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-j',
+                      axis=1,
+                      **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-k',
+                      axis=2,
+                      **kwargs),
+        Configuration(basic.Laplacian,
+                      name='lap-ij',
+                      along_x=True,
+                      along_y=True,
+                      along_z=False,
+                      **kwargs)
+    ]
+
+    table = run_scaling_benchmark(configurations, executions)
+    table.to_csv(output)
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def horizontal_diffusion_bandwidth(output, executions, option):
+    kwargs = common_kwargs(option)
+
+    configurations = [
+        Configuration(hdiff.Classic, block_size=(32, 12, 1), **kwargs),
+        Configuration(hdiff.OnTheFly,
+                      block_size=(32, 16, 1),
+                      loop='3D',
+                      **kwargs),
+        Configuration(hdiff.OnTheFlyIncache, block_size=(32, 8, 1), **kwargs),
+        Configuration(hdiff.JScanSharedMem, block_size=(256, 32, 1), **kwargs),
+        Configuration(hdiff.JScanOtfIncache, block_size=(128, 4, 1), **kwargs),
+        Configuration(hdiff.JScanOtf, block_size=(128, 4, 1), **kwargs),
+        Configuration(hdiff.JScanShuffleIncache,
+                      block_size=(28, 8, 2),
+                      **kwargs),
+        Configuration(hdiff.JScanShuffle, block_size=(28, 8, 2), **kwargs),
+        Configuration(hdiff.JScanShuffleSystolic,
+                      block_size=(28, 4, 3),
+                      **kwargs)
+    ]
+
+    def truncate_block_size_to_domain_if_possible(**kwargs):
+        if kwargs['block_size'][0] != 28:
+            return truncate_block_size_to_domain(**kwargs)
+        return kwargs
+
+    table = run_scaling_benchmark(
+        configurations,
+        executions,
+        preprocess_args=truncate_block_size_to_domain_if_possible)
+    table.to_csv(output)
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def vertical_advection_bandwidth(output, executions, option):
+    kwargs = common_kwargs(option)
+
+    configurations = [
+        Configuration(vadv.Classic,
+                      block_size=(128, 1),
+                      unroll_factor=8,
+                      **kwargs),
+        Configuration(vadv.LocalMem,
+                      block_size=(128, 1),
+                      unroll_factor=28,
+                      **kwargs),
+        Configuration(vadv.SharedMem,
+                      block_size=(64, 1),
+                      unroll_factor=0,
+                      **kwargs),
+        Configuration(vadv.LocalMemMerged,
+                      block_size=(128, 1),
+                      unroll_factor=2,
+                      **kwargs)
+    ]
+
+    table = run_scaling_benchmark(
+        configurations,
+        executions,
+        preprocess_args=truncate_block_size_to_domain)
+    table.to_csv(output)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/stencil_benchmarks/scripts/sbench_p100_collection.py b/stencil_benchmarks/scripts/sbench_p100_collection.py
@@ -0,0 +1,177 @@
+# Stencil Benchmarks
+#
+# Copyright (c) 2017-2021, ETH Zurich and MeteoSwiss
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are met:
+#
+# 1. Redistributions of source code must retain the above copyright notice,
+# this list of conditions and the following disclaimer.
+#
+# 2. Redistributions in binary form must reproduce the above copyright notice,
+# this list of conditions and the following disclaimer in the documentation
+# and/or other materials provided with the distribution.
+#
+# 3. Neither the name of the copyright holder nor the names of its contributors
+# may be used to endorse or promote products derived from this software without
+# specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+# AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+# ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+# LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+# CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+# SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+# INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+# CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+# ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+# POSSIBILITY OF SUCH DAMAGE.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+import click
+
+from stencil_benchmarks.benchmarks_collection.stencils.cuda_hip import (
+    basic, horizontal_diffusion as hdiff, vertical_advection as vadv)
+from stencil_benchmarks.tools.multirun import (Configuration,
+                                               run_scaling_benchmark,
+                                               truncate_block_size_to_domain,
+                                               default_kwargs)
+
+
+@click.group()
+def main():
+    pass
+
+
+common_kwargs = default_kwargs(backend='cuda',
+                               compiler='nvcc',
+                               gpu_architecture='sm_60',
+                               verify=False,
+                               dry_runs=1,
+                               alignment=128,
+                               dtype='float32',
+                               print_code=True)
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def basic_bandwidth(output, executions, option):
+    kwargs = common_kwargs(
+        option,
+        loop='3D',
+        block_size=(128, 2, 1),
+        halo=(1, 1, 1),
+    )
+
+    stream_kwargs = kwargs.copy()
+    stream_kwargs.update(loop='1D', block_size=(1024, 1, 1), halo=(0, 0, 0))
+
+    configurations = [
+        Configuration(basic.Copy, name='stream', **stream_kwargs),
+        Configuration(basic.Empty, name='empty', **kwargs),
+        Configuration(basic.Copy, name='copy', **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-i', axis=0, **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-j', axis=1, **kwargs),
+        Configuration(basic.OnesidedAverage, name='avg-k', axis=2, **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-i',
+                      axis=0,
+                      **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-j',
+                      axis=1,
+                      **kwargs),
+        Configuration(basic.SymmetricAverage,
+                      name='sym-avg-k',
+                      axis=2,
+                      **kwargs),
+        Configuration(basic.Laplacian,
+                      name='lap-ij',
+                      along_x=True,
+                      along_y=True,
+                      along_z=False,
+                      **kwargs)
+    ]
+
+    table = run_scaling_benchmark(configurations, executions)
+    table.to_csv(output)
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def horizontal_diffusion_bandwidth(output, executions, option):
+    kwargs = common_kwargs(option)
+
+    configurations = [
+        Configuration(hdiff.Classic, block_size=(32, 16, 1), **kwargs),
+        Configuration(hdiff.OnTheFly,
+                      block_size=(32, 16, 1),
+                      loop='3D',
+                      **kwargs),
+        Configuration(hdiff.OnTheFlyIncache, block_size=(32, 8, 1), **kwargs),
+        Configuration(hdiff.JScanSharedMem, block_size=(256, 32, 1), **kwargs),
+        Configuration(hdiff.JScanOtfIncache, block_size=(128, 4, 1), **kwargs),
+        Configuration(hdiff.JScanOtf, block_size=(128, 4, 1), **kwargs),
+        Configuration(hdiff.JScanShuffleIncache,
+                      block_size=(28, 8, 2),
+                      **kwargs),
+        Configuration(hdiff.JScanShuffle, block_size=(28, 8, 2), **kwargs),
+        Configuration(hdiff.JScanShuffleSystolic,
+                      block_size=(28, 4, 3),
+                      **kwargs)
+    ]
+
+    def truncate_block_size_to_domain_if_possible(**kwargs):
+        if kwargs['block_size'][0] != 28:
+            return truncate_block_size_to_domain(**kwargs)
+        return kwargs
+
+    table = run_scaling_benchmark(
+        configurations,
+        executions,
+        preprocess_args=truncate_block_size_to_domain_if_possible)
+    table.to_csv(output)
+
+
+@main.command()
+@click.argument('output', type=click.Path())
+@click.option('--executions', '-e', type=int, default=101)
+@click.option('--option', '-o', multiple=True)
+def vertical_advection_bandwidth(output, executions, option):
+    kwargs = common_kwargs(option)
+
+    configurations = [
+        Configuration(vadv.Classic,
+                      block_size=(512, 1),
+                      unroll_factor=8,
+                      **kwargs),
+        Configuration(vadv.LocalMem,
+                      block_size=(128, 1),
+                      unroll_factor=28,
+                      **kwargs),
+        Configuration(vadv.SharedMem,
+                      block_size=(64, 1),
+                      unroll_factor=0,
+                      **kwargs),
+        Configuration(vadv.LocalMemMerged,
+                      block_size=(512, 1),
+                      unroll_factor=2,
+                      **kwargs)
+    ]
+
+    table = run_scaling_benchmark(
+        configurations,
+        executions,
+        preprocess_args=truncate_block_size_to_domain)
+    table.to_csv(output)
+
+
+if __name__ == '__main__':
+    main()