Skip to content

Graph API benchmarks added #2560

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
pbalcer merged 1 commit into from
Jan 20, 2025
Merged

Graph API benchmarks added #2560

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 56 additions & 1 deletion scripts/benchmarks/benches/compute.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,7 +22,7 @@ def setup(self):
if options.sycl is None:
return

repo_path = git_clone(self.directory, "compute-benchmarks-repo", "https://github.com/intel/compute-benchmarks.git", "d13e5b4d8dd3d28926a74ab7f67f78c10f708a01")
repo_path = git_clone(self.directory, "compute-benchmarks-repo", "https://github.com/intel/compute-benchmarks.git", "578a7ac6f9bc48f6c2b408ef64a19a2ef9a216e7")
build_path = create_build_path(self.directory, 'compute-benchmarks-build')

configure_command = [
Expand Down Expand Up @@ -74,6 +74,18 @@ def benchmarks(self) -> list[Benchmark]:
MemcpyExecute(self, 10, 16, 1024, 10000, 0, 1, 1),
MemcpyExecute(self, 4096, 1, 1024, 10, 0, 1, 0),
MemcpyExecute(self, 4096, 4, 1024, 10, 0, 1, 0),
GraphApiSinKernelGraphSYCL(self, 0, 10),
Copy link
Contributor

@pbalcer pbalcer Jan 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please pick max 2/3 scenarios per benchmark. We need to keep the runtime of the whole job reasonable (I'm aiming for <30 minutes).

GraphApiSinKernelGraphSYCL(self, 1, 10),
GraphApiSinKernelGraphSYCL(self, 0, 100),
GraphApiSinKernelGraphSYCL(self, 1, 100),
# Submit
GraphApiSubmitExecGraph(self, 0, 1, 10),
GraphApiSubmitExecGraph(self, 1, 1, 10),
GraphApiSubmitExecGraph(self, 1, 1, 100),
# Exec
GraphApiSubmitExecGraph(self, 0, 0, 10),
GraphApiSubmitExecGraph(self, 1, 0, 10),
GraphApiSubmitExecGraph(self, 1, 0, 100),
]

if options.ur is not None:
Expand Down Expand Up @@ -357,3 +369,46 @@ def bin_args(self) -> list[str]:
f"--SrcUSM={self.srcUSM}",
f"--DstUSM={self.dstUSM}",
]

class GraphApiSinKernelGraphSYCL(ComputeBenchmark):
def __init__(self, bench, withGraphs, numKernels):
self.withGraphs = withGraphs
self.numKernels = numKernels
super().__init__(bench, "graph_api_benchmark_sycl", "SinKernelGraph")

def explicit_group(self):
return "SinKernelGraph"

def name(self):
return f"graph_api_benchmark_sycl SinKernelGraph graphs:{self.withGraphs}, numKernels:{self.numKernels}"

def bin_args(self) -> list[str]:
return [
"--iterations=100",
Copy link
Contributor

@pbalcer pbalcer Jan 16, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this enough iterations for the benchmarks to have reproducible results?
We aim for the stddev between runs (or, rather, the coefficient of variation of all the runs) to be smaller than 2%.

f"--numKernels={self.numKernels}",
f"--withGraphs={self.withGraphs}",
]

class GraphApiSubmitExecGraph(ComputeBenchmark):
def __init__(self, bench, ioq, submit, numKernels):
self.ioq = ioq
self.submit = submit
self.numKernels = numKernels
super().__init__(bench, "graph_api_benchmark_sycl", "SubmitExecGraph")

def name(self):
return f"graph_api_benchmark_sycl SubmitExecGraph ioq:{self.ioq}, submit:{self.submit}, numKernels:{self.numKernels}"

def explicit_group(self):
if self.submit:
return "SubmitGraph"
else:
return "ExecGraph"

def bin_args(self) -> list[str]:
return [
"--iterations=100",
f"--measureSubmit={self.submit}",
f"--ioq={self.ioq}",
f"--numKernels={self.numKernels}",
]
Loading