Skip to content

Commit 294cf9f

Browse files
anmyachevanmyachev
committed
completely remove build PTI step for benchmarks
Signed-off-by: Anatoly Myachev <[email protected]>
1 parent fbf75bd commit 294cf9f

File tree

1 file changed

+0
-30
lines changed

1 file changed

+0
-30
lines changed

‎.github/workflows/triton-benchmarks.yml‎

Lines changed: 0 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -117,17 +117,9 @@ jobs:
117117
cd benchmarks
118118
pip install .
119119
120-
- name: Build PTI
121-
run: |
122-
./scripts/install-pti.sh --build-level-zero
123-
PTI_LIBS_DIR=$(python ./scripts/pti_lib.py)
124-
ls $PTI_LIBS_DIR
125-
echo "PTI_LIBS_DIR=$PTI_LIBS_DIR" >> $GITHUB_ENV
126-
127120
- name: Run Triton Softmax kernel benchmark
128121
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'fused_softmax.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'fused_softmax.py') }}
129122
run: |
130-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
131123
cd benchmarks/triton_kernels_benchmark
132124
python fused_softmax.py --reports $REPORTS --n_runs $N_RUNS
133125
source ../../scripts/capture-hw-details.sh
@@ -138,15 +130,13 @@ jobs:
138130
- name: Run Triton Softmax kernel benchmark with Proton
139131
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'fused_softmax.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'fused_softmax.py') }}
140132
run: |
141-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
142133
cd benchmarks/triton_kernels_benchmark
143134
BENCHMARKING_METHOD=PROTON_PROFILER python fused_softmax.py
144135
source ../../scripts/capture-hw-details.sh
145136
146137
- name: Run Triton GEMM kernel benchmark
147138
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py') }}
148139
run: |
149-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
150140
cd benchmarks/triton_kernels_benchmark
151141
python gemm_benchmark.py --reports $REPORTS --n_runs $N_RUNS
152142
mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-base.csv
@@ -160,7 +150,6 @@ jobs:
160150
- name: Run Triton GEMM kernel benchmark - with tensor of pointer
161151
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_tensor_of_ptr_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_tensor_of_ptr_benchmark.py') }}
162152
run: |
163-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
164153
cd benchmarks/triton_kernels_benchmark
165154
python gemm_tensor_of_ptr_benchmark.py --reports $REPORTS --n_runs $N_RUNS
166155
source ../../scripts/capture-hw-details.sh
@@ -173,7 +162,6 @@ jobs:
173162
- name: Run Triton GEMM kernel benchmark - with tensor descriptor
174163
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_tensor_desc_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_tensor_desc_benchmark.py') }}
175164
run: |
176-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
177165
cd benchmarks/triton_kernels_benchmark
178166
python gemm_tensor_desc_benchmark.py --reports $REPORTS --n_runs $N_RUNS
179167
source ../../scripts/capture-hw-details.sh
@@ -186,7 +174,6 @@ jobs:
186174
- name: Run Triton GEMM (A@B^t) kernel benchmark
187175
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_benchmark.py_abt')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_abt') }}
188176
run: |
189-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
190177
cd benchmarks/triton_kernels_benchmark
191178
TRANSPOSE_B=1 python gemm_benchmark.py --reports $REPORTS --n_runs $N_RUNS
192179
mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-bt.csv
@@ -198,7 +185,6 @@ jobs:
198185
- name: Run Triton GEMM (A^t@B) kernel benchmark
199186
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_benchmark.py_atb')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_benchmark.py_atb') }}
200187
run: |
201-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
202188
cd benchmarks/triton_kernels_benchmark
203189
TRANSPOSE_A=1 python gemm_benchmark.py --reports $REPORTS --n_runs $N_RUNS
204190
mv $REPORTS/matmul-performance.csv $REPORTS/matmul-performance-at.csv
@@ -210,7 +196,6 @@ jobs:
210196
- name: Run Triton GEMM (stream-k) kernel benchmark
211197
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_streamk_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_streamk_benchmark.py') }}
212198
run: |
213-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
214199
cd benchmarks/triton_kernels_benchmark
215200
python gemm_streamk_benchmark.py --reports $REPORTS --n_runs $N_RUNS
216201
source ../../scripts/capture-hw-details.sh
@@ -220,7 +205,6 @@ jobs:
220205
- name: Run Triton GEMM (split-k) kernel benchmark
221206
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_splitk_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_splitk_benchmark.py') }}
222207
run: |
223-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
224208
cd benchmarks/triton_kernels_benchmark
225209
python gemm_splitk_benchmark.py --reports $REPORTS --n_runs $N_RUNS
226210
source ../../scripts/capture-hw-details.sh
@@ -230,7 +214,6 @@ jobs:
230214
- name: Run Triton GEMM + PreOp (exp) kernel benchmark
231215
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_preop_exp_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_preop_exp_benchmark.py') }}
232216
run: |
233-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
234217
cd benchmarks/triton_kernels_benchmark
235218
python gemm_preop_exp_benchmark.py --reports $REPORTS --n_runs $N_RUNS
236219
source ../../scripts/capture-hw-details.sh
@@ -239,7 +222,6 @@ jobs:
239222
- name: Run Triton GEMM + PostOp (Gelu) kernel benchmark
240223
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_postop_gelu_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_postop_gelu_benchmark.py') }}
241224
run: |
242-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
243225
cd benchmarks/triton_kernels_benchmark
244226
python gemm_postop_gelu_benchmark.py --reports $REPORTS --n_runs $N_RUNS
245227
source ../../scripts/capture-hw-details.sh
@@ -248,7 +230,6 @@ jobs:
248230
- name: Run Triton GEMM + PostOp (add matrix) kernel benchmark bfloat16
249231
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_postop_addmatrix_benchmark_bfloat16.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_postop_addmatrix_benchmark_bfloat16.py') }}
250232
run: |
251-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
252233
cd benchmarks/triton_kernels_benchmark
253234
python gemm_postop_addmatrix_benchmark.py --reports $REPORTS --n_runs $N_RUNS
254235
source ../../scripts/capture-hw-details.sh
@@ -258,7 +239,6 @@ jobs:
258239
- name: Run Triton GEMM + PostOp (add matrix) kernel benchmark int8
259240
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'gemm_postop_addmatrix_benchmark_int8.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'gemm_postop_addmatrix_benchmark_int8.py') }}
260241
run: |
261-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
262242
cd benchmarks/triton_kernels_benchmark
263243
INT8_ONLY=1 python gemm_postop_addmatrix_benchmark.py --reports $REPORTS --n_runs $N_RUNS
264244
source ../../scripts/capture-hw-details.sh
@@ -268,7 +248,6 @@ jobs:
268248
- name: Run Triton FA fwd kernel benchmark
269249
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flash_attention_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_benchmark.py') }}
270250
run: |
271-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
272251
cd benchmarks/triton_kernels_benchmark
273252
python flash_attention_benchmark.py --reports $REPORTS --n_runs $N_RUNS
274253
@@ -279,7 +258,6 @@ jobs:
279258
- name: Run Triton FA bwd kernel benchmark
280259
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flash_attention_bwd_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_bwd_benchmark.py') }}
281260
run: |
282-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
283261
cd benchmarks/triton_kernels_benchmark
284262
FA_KERNEL_MODE="bwd" \
285263
python flash_attention_benchmark.py --reports $REPORTS --n_runs $N_RUNS
@@ -292,7 +270,6 @@ jobs:
292270
- name: Run Triton FA fwd kernel benchmark - with tensor descriptors
293271
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flash_attention_tensor_desc_benchmark.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flash_attention_tensor_desc_benchmark.py') }}
294272
run: |
295-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
296273
cd benchmarks/triton_kernels_benchmark
297274
python flash_attention_tensor_desc_benchmark.py --reports $REPORTS --n_runs $N_RUNS
298275
mv $REPORTS/attn-performance.csv $REPORTS/attn-tensor-desc-performance.csv
@@ -304,7 +281,6 @@ jobs:
304281
- name: Run Triton FlexAttention Causal Mask fwd kernel benchmark
305282
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_causal_mask.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_causal_mask.py') }}
306283
run: |
307-
# export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
308284
cd benchmarks/triton_kernels_benchmark
309285
python flex_attention_benchmark_causal_mask.py --reports $REPORTS --n_runs $N_RUNS
310286
@@ -315,7 +291,6 @@ jobs:
315291
- name: Run Triton FlexAttention (batch_size=4) Causal Mask fwd kernel benchmark
316292
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_batch4-causal_mask.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_batch4-causal_mask.py') }}
317293
run: |
318-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
319294
cd benchmarks/triton_kernels_benchmark
320295
BATCH_SIZE=4 python flex_attention_benchmark_causal_mask.py --reports $REPORTS --n_runs $N_RUNS
321296
@@ -326,7 +301,6 @@ jobs:
326301
- name: Run Triton FlexAttention (batch_size=16) Causal Mask fwd kernel benchmark
327302
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_batch16-causal_mask.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_batch16-causal_mask.py') }}
328303
run: |
329-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
330304
cd benchmarks/triton_kernels_benchmark
331305
BATCH_SIZE=16 python flex_attention_benchmark_causal_mask.py --reports $REPORTS --n_runs $N_RUNS
332306
@@ -337,7 +311,6 @@ jobs:
337311
- name: Run Triton FlexAttention Causal Mask bwd kernel benchmark
338312
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_bwd_benchmark_causal_mask.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_bwd_benchmark_causal_mask.py') }}
339313
run: |
340-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
341314
cd benchmarks/triton_kernels_benchmark
342315
FA_KERNEL_MODE='bwd' \
343316
python flex_attention_benchmark_causal_mask.py --reports $REPORTS --n_runs $N_RUNS
@@ -349,7 +322,6 @@ jobs:
349322
- name: Run Triton FlexAttention Custom Masks fwd kernel benchmark
350323
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'flex_attention_benchmark_custom_masks.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'flex_attention_benchmark_custom_masks.py') }}
351324
run: |
352-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
353325
cd benchmarks/triton_kernels_benchmark
354326
python flex_attention_benchmark_custom_masks.py --reports $REPORTS --n_runs $N_RUNS
355327
@@ -363,7 +335,6 @@ jobs:
363335
- name: Run Prefix Sums kernel benchmark
364336
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'prefix_sums.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'prefix_sums.py') }}
365337
run: |
366-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
367338
cd benchmarks/triton_kernels_benchmark
368339
python prefix_sums.py --reports $REPORTS --n_runs $N_RUNS
369340
source ../../scripts/capture-hw-details.sh
@@ -372,7 +343,6 @@ jobs:
372343
- name: Run micro benchmark
373344
if: ${{ steps.install.outcome == 'success' && !cancelled() && (inputs.benchmarks == '' || contains(fromJson(inputs.benchmarks || '[]'), 'micro_benchmarks.py')) && !contains(fromJson(inputs.skip_benchmarks || '[]'), 'micro_benchmarks') }}
374345
run: |
375-
export LD_LIBRARY_PATH=$PTI_LIBS_DIR:$LD_LIBRARY_PATH
376346
cd benchmarks/micro_benchmarks
377347
python run_benchmarks.py --reports $REPORTS
378348

0 commit comments

Comments
 (0)