Repository navigation
399 lines (365 loc) · 14.6 KB
/
Copy pathbench.yml
File metadata and controls
399 lines (365 loc) · 14.6 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
name: Benchmark Compare
defaults:
run:
shell: bash --noprofile --norc -euo pipefail {0}
on:
workflow_dispatch:
inputs:
gpu:
description: "GPU selector. Ex: 'rtxa6000'. See ci/bench.yaml for all options."
required: true
default: "rtxa6000"
type: string
launch_args:
description: "'.devcontainer/launch.sh -d' additional args. Ex: '--cuda 13.4 --host gcc14'"
required: false
default: "--cuda 13.4 --host gcc14"
type: string
arch:
description: "CMAKE_CUDA_ARCHITECTURES value passed as ci/bench/bench.sh --arch"
required: false
default: "native"
type: string
base_ref:
description: "Base Ref/SHA"
required: false
default: "origin/main"
type: string
test_ref:
description: "Test Ref/SHA"
required: false
default: ""
type: string
cpp_filters:
description: "C++ filters, space-separated (ex: '^cub.bench.copy.memcpy.base$')"
required: false
default: ""
type: string
python_filters:
description: "Python filters, space-separated (ex: 'compute/reduce/sum\\.py')"
required: false
default: ""
type: string
nvbench_args:
description: "NVBench extra args"
required: false
default: ""
type: string
nvbench_compare_args:
description: "nvbench-compare-robust extra args"
required: false
default: ""
type: string
nvbench_compare_legacy_args:
description: "nvbench-compare-legacy extra args"
required: false
default: ""
type: string
raw_args:
description: "Raw Args (advanced, bypasses all other inputs)"
required: false
default: ""
type: string
workflow_call:
inputs:
gpu:
description: "GPU selector. Ex: 'rtxa6000'. See ci/bench.yaml for all options."
required: true
type: string
launch_args:
description: "'.devcontainer/launch.sh -d' additional args. Ex: '--cuda 13.4 --host gcc14'"
required: false
default: "--cuda 13.4 --host gcc14"
type: string
arch:
description: "CMAKE_CUDA_ARCHITECTURES value passed as ci/bench/bench.sh --arch"
required: false
default: "native"
type: string
base_ref:
description: "Base Ref/SHA"
required: false
default: "origin/main"
type: string
test_ref:
description: "Test Ref/SHA"
required: false
default: ""
type: string
cpp_filters:
description: "C++ filters, space-separated (ex: '^cub.bench.copy.memcpy.base$')"
required: false
default: ""
type: string
python_filters:
description: "Python filters, space-separated (ex: 'compute/reduce/sum\\.py')"
required: false
default: ""
type: string
nvbench_args:
description: "NVBench extra args"
required: false
default: ""
type: string
nvbench_compare_args:
description: "nvbench-compare-robust extra args"
required: false
default: ""
type: string
nvbench_compare_legacy_args:
description: "nvbench-compare-legacy extra args"
required: false
default: ""
type: string
raw_args:
description: "Raw Args (advanced, bypasses all other inputs)"
required: false
default: ""
type: string
jobs:
run:
name: "Bench: ${{inputs.gpu}} ${{inputs.arch}} ${{inputs.launch_args}}"
if: ${{ contains(fromJSON('["NVIDIA", "NVIDIA-dev"]'), github.repository_owner) }}
runs-on: linux-amd64-gpu-${{ inputs.gpu }}-latest-1
permissions:
id-token: write
contents: read
steps:
- name: Checkout repo
uses: actions/checkout@v6
with:
fetch-depth: 0
persist-credentials: false
- name: Resolve NVBench SHA
run: |
nvbench_repo="https://github.com/NVIDIA/nvbench.git"
nvbench_ref="refs/heads/main"
nvbench_remote_ref="$(git ls-remote "${nvbench_repo}" "${nvbench_ref}")"
read -r nvbench_sha resolved_ref <<< "${nvbench_remote_ref}"
if [[ "${resolved_ref:-}" != "${nvbench_ref}" || ! "${nvbench_sha}" =~ ^[0-9a-fA-F]{40}$ ]]; then
echo "::error::Failed to resolve ${nvbench_repo} ${nvbench_ref}; got '${nvbench_remote_ref}'"
exit 1
fi
echo "Resolved NVBench ${nvbench_ref} to ${nvbench_sha}"
echo "CCCL_BENCH_NVBENCH_SHA=${nvbench_sha}" >> "${GITHUB_ENV}"
- name: Get AWS credentials for sccache bucket
if: ${{ contains(fromJSON('["NVIDIA", "NVIDIA-dev"]'), github.repository_owner) }}
uses: aws-actions/configure-aws-credentials@254c19bd240aabef8777f48595e9d2d7b972184b # v6
with:
role-to-assume: arn:aws:iam::279114543810:role/gha-oidc-NVIDIA
aws-region: us-east-2
role-duration-seconds: 43200 # 12 hours
- name: Prepare AWS config for devcontainer
run: |
# The devcontainer mounts this path to /home/coder/.aws.
aws_dir="${{ github.workspace }}/.aws"
mkdir -p "${aws_dir}"
cat > "${aws_dir}/config" <<EOF
[default]
bucket=rapids-sccache-devs
region=us-east-2
EOF
cat > "${aws_dir}/credentials" <<EOF
[default]
aws_access_key_id=${AWS_ACCESS_KEY_ID:-}
aws_session_token=${AWS_SESSION_TOKEN:-}
aws_secret_access_key=${AWS_SECRET_ACCESS_KEY:-}
EOF
chmod 0600 "${aws_dir}/credentials"
chmod 0664 "${aws_dir}/config"
- name: Resolve benchmark args
id: resolve-args
env:
INPUT_ARCH: ${{ inputs.arch }}
INPUT_BASE_REF: ${{ inputs.base_ref }}
INPUT_TEST_REF: ${{ inputs.test_ref }}
INPUT_CPP_FILTERS: ${{ inputs.cpp_filters }}
INPUT_PYTHON_FILTERS: ${{ inputs.python_filters }}
INPUT_RAW_ARGS: ${{ inputs.raw_args }}
INPUT_NVBENCH_ARGS: ${{ inputs.nvbench_args }}
INPUT_NVBENCH_COMPARE_ARGS: ${{ inputs.nvbench_compare_args }}
INPUT_NVBENCH_COMPARE_LEGACY_ARGS: ${{ inputs.nvbench_compare_legacy_args }}
run: |
parse_quoted_args() {
local quoted_args="$1"
# xargs handles shell-style quoting without eval.
# Ignore empty / whitespace-only inputs to avoid emitting a spurious empty arg.
if [[ -z "${quoted_args}" || "${quoted_args}" =~ ^[[:space:]]*$ ]]; then
return 0
fi
if ! printf '%s' "${quoted_args}" | xargs printf '%s\0' \
2> >(while IFS= read -r line; do
echo "::error::Failed to parse quoted args: ${line}" >&2
done); then
return 2
fi
}
declare -a bench_args
bench_args=()
if [[ -n "${INPUT_RAW_ARGS}" ]]; then
parsed_raw_args_file="$(mktemp "${RUNNER_TEMP}/bench-raw-args-XXXXXX")"
if ! parse_quoted_args "${INPUT_RAW_ARGS}" > "${parsed_raw_args_file}"; then
rm -f "${parsed_raw_args_file}"
exit 2
fi
mapfile -d '' -t bench_args < "${parsed_raw_args_file}"
rm -f "${parsed_raw_args_file}"
else
if [[ -z "${INPUT_BASE_REF}" || -z "${INPUT_TEST_REF}" ]]; then
echo "::error::When Raw Args is empty, Base Ref and Test Ref must be set."
exit 2
fi
if [[ -z "${INPUT_CPP_FILTERS}" && -z "${INPUT_PYTHON_FILTERS}" ]]; then
echo "::error::At least one of C++ Filters or Python Filters must be set."
exit 2
fi
bench_args+=("${INPUT_BASE_REF}")
bench_args+=("${INPUT_TEST_REF}")
if [[ -n "${INPUT_ARCH}" ]]; then
bench_args+=(--arch "${INPUT_ARCH}")
fi
# Add C++ filters as --cpp-filter flags.
if [[ -n "${INPUT_CPP_FILTERS}" ]]; then
declare -a parsed_cpp_filters
parsed_cpp_filters=()
parsed_cpp_filters_file="$(mktemp "${RUNNER_TEMP}/bench-cpp-filters-XXXXXX")"
if ! parse_quoted_args "${INPUT_CPP_FILTERS}" > "${parsed_cpp_filters_file}"; then
rm -f "${parsed_cpp_filters_file}"
exit 2
fi
mapfile -d '' -t parsed_cpp_filters < "${parsed_cpp_filters_file}"
rm -f "${parsed_cpp_filters_file}"
for cpp_filter in "${parsed_cpp_filters[@]}"; do
bench_args+=(--cpp-filter "${cpp_filter}")
done
fi
# Add Python filters as --python-filter flags.
if [[ -n "${INPUT_PYTHON_FILTERS}" ]]; then
declare -a parsed_py_filters
parsed_py_filters=()
parsed_py_filters_file="$(mktemp "${RUNNER_TEMP}/bench-py-filters-XXXXXX")"
if ! parse_quoted_args "${INPUT_PYTHON_FILTERS}" > "${parsed_py_filters_file}"; then
rm -f "${parsed_py_filters_file}"
exit 2
fi
mapfile -d '' -t parsed_py_filters < "${parsed_py_filters_file}"
rm -f "${parsed_py_filters_file}"
for py_filter in "${parsed_py_filters[@]}"; do
bench_args+=(--python-filter "${py_filter}")
done
fi
if [[ -n "${INPUT_NVBENCH_ARGS}" ]]; then
bench_args+=(--nvbench-args "${INPUT_NVBENCH_ARGS}")
fi
if [[ -n "${INPUT_NVBENCH_COMPARE_ARGS}" ]]; then
bench_args+=(--nvbench-compare-args "${INPUT_NVBENCH_COMPARE_ARGS}")
fi
if [[ -n "${INPUT_NVBENCH_COMPARE_LEGACY_ARGS}" ]]; then
bench_args+=(--nvbench-compare-legacy-args "${INPUT_NVBENCH_COMPARE_LEGACY_ARGS}")
fi
fi
if [[ "${#bench_args[@]}" -lt 2 ]]; then
echo "::error::Resolved argument list must include at least <base-ref> and <test-ref>."
exit 2
fi
args_file="${RUNNER_TEMP}/bench_args.txt"
printf '%s\0' "${bench_args[@]}" > "${args_file}"
{
echo "args_file=${args_file}"
echo "resolved_args<<EOF"
for arg in "${bench_args[@]}"; do
printf '%q\n' "${arg}"
done
echo "EOF"
} >> "${GITHUB_OUTPUT}"
- name: Show resolved benchmark args
run: |
echo "Resolved args passed to ci/bench/bench.sh:"
while IFS= read -r arg; do
echo " ${arg}"
done <<< "${{ steps.resolve-args.outputs.resolved_args }}"
- name: Run benchmark compare
id: run-benchmark-compare
env:
GH_TOKEN: ${{ github.token }}
AWS_ACCESS_KEY_ID: ${{ env.AWS_ACCESS_KEY_ID }}
AWS_REGION: ${{ env.AWS_REGION }}
AWS_SECRET_ACCESS_KEY: ${{ env.AWS_SECRET_ACCESS_KEY }}
AWS_SESSION_TOKEN: ${{ env.AWS_SESSION_TOKEN }}
GPU_NAME: ${{ inputs.gpu }}
LAUNCH_ARGS: ${{ inputs.launch_args }}
run: |
args_file="${{ steps.resolve-args.outputs.args_file }}"
ci_script="${RUNNER_TEMP}/ci.sh"
timestamp_utc="$(date -u +'%Y%m%dT%H%M%SZ')"
mapfile -d '' -t bench_args < "${args_file}"
base_ref="${bench_args[0]:-}"
test_ref="${bench_args[1]:-}"
base_sha_full="$(git rev-parse --verify "${base_ref}^{commit}")"
test_sha_full="$(git rev-parse --verify "${test_ref}^{commit}")"
base_sha_short="${base_sha_full:0:8}"
test_sha_short="${test_sha_full:0:8}"
artifact_name="bench-${GPU_NAME}-${timestamp_utc}-${base_sha_short}-${test_sha_short}"
mkdir -p "bench-artifacts"
GPU_ARGS=""
if [[ ! "${LAUNCH_ARGS}" =~ "--gpus" ]]; then
echo "Selected GPU '${GPU_NAME}'; enabling '--gpus all'"
GPU_ARGS="--gpus all"
else
echo "GPU selected ('${GPU_NAME}'), but '--gpus' already present in LAUNCH_ARGS"
fi
cat <<"EOF" > "${ci_script}"
#! /usr/bin/env bash
set -euo pipefail
# bench_args.txt contains NUL-delimited argv entries resolved by the workflow.
# This adds them to a bench_args array while preserving any quotes, whitespace, etc
mapfile -d '' -t bench_args < "/bench_args.txt"
# NVBench compare deps:
./ci/util/retry.sh 5 30 python3 -m pip install 'cuda-bench[compare]>=0.3.0'
./ci/bench/bench.sh "${bench_args[@]}"
EOF
chmod +x "${ci_script}"
rc=0
# These strings intentionally contain multiple launch arguments.
# shellcheck disable=SC2086
.devcontainer/launch.sh \
-d ${LAUNCH_ARGS} ${GPU_ARGS} \
--env "AWS_ROLE_ARN=" \
--env "AWS_REGION=${AWS_REGION:-}" \
--env "AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID:-}" \
--env "AWS_SESSION_TOKEN=${AWS_SESSION_TOKEN:-}" \
--env "AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY:-}" \
--env "CCCL_BENCH_ARTIFACT_ROOT=bench-artifacts" \
--env "CCCL_BENCH_GPU_NAME=${GPU_NAME}" \
--env "CCCL_BENCH_NVBENCH_SHA=${CCCL_BENCH_NVBENCH_SHA}" \
--env "GH_TOKEN=${GH_TOKEN:-}" \
--env "LAUNCH_ARGS=${LAUNCH_ARGS} ${GPU_ARGS}" \
--env "SCCACHE_REGION=${AWS_REGION:-}" \
--volume "${ci_script}:/ci.sh" \
--volume "${args_file}:/bench_args.txt" \
-- /ci.sh \
|| rc=$?
echo "artifact_name=${artifact_name}" >> "${GITHUB_OUTPUT}"
exit "${rc}"
- name: Append benchmark summaries
if: ${{ always() }}
run: |
shopt -s nullglob # in case no summary files are found, avoid a literal '*.md' entry
summaries=(bench-artifacts/*/summary.md)
if [[ "${#summaries[@]}" -eq 0 ]]; then
echo "No benchmark summary files found." >> "${GITHUB_STEP_SUMMARY}"
exit 0
fi
for summary in "${summaries[@]}"; do
echo "## ${summary}"
cat "${summary}"
echo
done >> "${GITHUB_STEP_SUMMARY}"
- name: Upload benchmark artifacts
if: ${{ always() }}
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
with:
name: ${{ steps.run-benchmark-compare.outputs.artifact_name }}
path: bench-artifacts/*
if-no-files-found: warn
retention-days: 14