benchdnn: inputs: graph: skip partition num check for some cases

Author: TaoLv

tests/benchdnn/inputs/graph/complex_fusion/harness_mha_all

@@ -2,13 +2,13 @@
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/JAX-MHA-inf-fp32.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/JAX-MQA-inf-fp32.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-GPT-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-bert_large-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-distill_bert-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-stable_diffusion-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA_backward-Bert_large-train-fp32-bs4.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA_forward-Bert_large-train-fp32-bs4.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA_backward-Bert_large-train-fp32-bs4.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA_forward-Bert_large-train-fp32-bs4.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/sdpa-plain-simplified-f16.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/sdpa-plain-wo-scale-f16-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/GQA-fp16.json
@@ -17,11 +17,11 @@
 
 # int8 graphs
 --reset --case=complex_fusion/mha/MHA-GPT-inf-int8-bs1.json
---reset --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
 --reset --case=complex_fusion/mha/MHA-bert_large-inf-int8-bs1.json
 --reset --case=complex_fusion/mha/MHA-distill_bert-inf-int8-bs1.json
---reset --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
---reset --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
 --reset --case=complex_fusion/mha/sdpa-plain-wo-scale-int8-bs1.json
 --reset --case=complex_fusion/mha/sdpa-compressed-kv-int4-gs32.json
 --reset --case=complex_fusion/mha/sdpa-compressed-kv-int8-gs128.json
@@ -30,14 +30,14 @@
 
 # Re-written graphs
 --reset --dt=f32,bf16,f16 --in-shapes=4:4x16x32x256+5:4x16x256x33+0:4x16x33x256+1:4x1x1x33+3:4x1x32x33 --case=complex_fusion/mha/MHA-GPT-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --in-shapes=3:4x32x32x128+4:4x32x128x33+0:4x32x33x128+1:4x1x32x33 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --in-shapes=3:4x32x32x128+4:4x32x128x33+0:4x32x33x128+1:4x1x32x33 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --in-shapes=3:20x16x384x64+4:20x16x64x384+0:20x16x384x64+1:20x1x1x384 --case=complex_fusion/mha/MHA-bert_large-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --in-shapes=3:10x16x384x64+4:10x1x64x384+0:10x1x384x64+1:10x1x1x384 --case=complex_fusion/mha/MHA-bert_large-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --in-shapes=4:56x12x128x64+5:56x12x64x128+0:56x12x128x64+1:56x1x1x128 --case=complex_fusion/mha/MHA-distill_bert-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --in-shapes=0:56x8x1024x80+1:56x8x77x80+2:56x8x77x80 --case=complex_fusion/mha/MHA-stable_diffusion-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --in-shapes=5:20x117x48x128+6:20x1x128x117+19:20x1x117x128 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --in-shapes=2514:32x16x512x64+2518:32x16x512x64+2543:32x1x512x512+2547:32x16x512x512+2525:32x16x512x64 --op-attrs=4837:shape:16384x1024 --case=complex_fusion/mha/MHA_forward-Bert_large-train-fp32-bs4.json
---reset --dt=f32,bf16,f16 --in-shapes=2514:32x16x512x64+2518:32x16x512x64+2591:32x16x512x512+2545:32x16x512x512+2547:32x16x512x512+2525:32x16x512x64+2548:32x16x512x512+5178:16384x1024 --op-attrs=7392:shape:32x512x16x64 --case=complex_fusion/mha/MHA_backward-Bert_large-train-fp32-bs4.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --in-shapes=5:20x117x48x128+6:20x1x128x117+19:20x1x117x128 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --in-shapes=2514:32x16x512x64+2518:32x16x512x64+2543:32x1x512x512+2547:32x16x512x512+2525:32x16x512x64 --op-attrs=4837:shape:16384x1024 --case=complex_fusion/mha/MHA_forward-Bert_large-train-fp32-bs4.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --in-shapes=2514:32x16x512x64+2518:32x16x512x64+2591:32x16x512x512+2545:32x16x512x512+2547:32x16x512x512+2525:32x16x512x64+2548:32x16x512x512+5178:16384x1024 --op-attrs=7392:shape:32x512x16x64 --case=complex_fusion/mha/MHA_backward-Bert_large-train-fp32-bs4.json
 --reset --dt=f32,bf16,f16 --in-shapes=0:20x16x384x64+1:20x16x384x64+8:20x16x384x64+5:20x1x1x384 --case=complex_fusion/mha/sdpa-plain-wo-scale-f16-bs1.json
 --reset --dt=f32,bf16,f16 --in-shapes=5:1x1x384x384,5:1x16x384x384 --case=complex_fusion/mha/sdpa-plain-simplified-f16.json
 --reset --dt=f32,bf16,f16 --in-shapes=0:2x16x384x64+1:2x16x384x64+5:2x1x1x384+8:2x16x384x64  --case=complex_fusion/mha/sdpa-plain-simplified-f16.json
@@ -47,9 +47,9 @@
 
 # Re-written int8 graphs
 --reset --in-shapes=5:4x16x32x256+4:4x16x256x33+0:4x16x33x256+1:4x1x1x33+3:4x1x32x33 --case=complex_fusion/mha/MHA-GPT-inf-int8-bs1.json
---reset --in-shapes=4:4x32x32x128+3:4x32x128x33+0:4x32x33x128+1:4x1x32x33 --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --in-shapes=4:4x32x32x128+3:4x32x128x33+0:4x32x33x128+1:4x1x32x33 --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
 --reset --in-shapes=4:20x16x384x64+3:20x16x64x384+0:20x16x384x64+1:20x1x1x384 --case=complex_fusion/mha/MHA-bert_large-inf-int8-bs1.json
 --reset --in-shapes=5:56x12x128x64+4:56x12x64x128+0:56x12x128x64+1:56x1x1x128 --case=complex_fusion/mha/MHA-distill_bert-inf-int8-bs1.json
---reset --in-shapes=4:20x117x48x128+3:20x1x128x117+0:20x1x117x128 --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
---reset --in-shapes=4:32x16x384x64+3:32x16x64x384+0:32x16x384x64+1:32x1x1x384 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
+--reset --expected-n-partitions=0 --in-shapes=4:20x117x48x128+3:20x1x128x117+0:20x1x117x128 --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --in-shapes=4:32x16x384x64+3:32x16x64x384+0:32x16x384x64+1:32x1x1x384 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
 --reset --in-shapes=4:20x16x384x64+3:20x16x64x384+0:20x16x384x64+1:20x1x1x384 --case=complex_fusion/mha/sdpa-plain-wo-scale-int8-bs1.json

tests/benchdnn/inputs/graph/complex_fusion/harness_mha_ci

@@ -2,11 +2,11 @@
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/JAX-MHA-inf-fp32.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/JAX-MQA-inf-fp32.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-GPT-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-LLaMa-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-bert_large-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-stable_diffusion-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-distill_bert-inf-fp32-bs1.json
---reset --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
+--reset --expected-n-partitions=0 --dt=f32,bf16,f16 --case=complex_fusion/mha/MHA-starcoder-inf-fp32-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/sdpa-plain-simplified-f16.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/sdpa-plain-wo-scale-f16-bs1.json
 --reset --dt=f32,bf16,f16 --case=complex_fusion/mha/GQA-fp16.json
@@ -15,11 +15,11 @@
 
 # int8 graphs
 --reset --case=complex_fusion/mha/MHA-GPT-inf-int8-bs1.json
---reset --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/MHA-LLaMa-inf-int8-bs1.json
 --reset --case=complex_fusion/mha/MHA-bert_large-inf-int8-bs1.json
 --reset --case=complex_fusion/mha/MHA-distill_bert-inf-int8-bs1.json
---reset --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
---reset --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/MHA-starcoder-inf-int8-bs1.json
+--reset --expected-n-partitions=0 --case=complex_fusion/mha/dynamic_quantized_mha-Bert_large-inf-int8-bs1-fake.json
 --reset --case=complex_fusion/mha/sdpa-plain-wo-scale-int8-bs1.json
 --reset --case=complex_fusion/mha/sdpa-compressed-kv-int8-gs128.json
 --reset --case=complex_fusion/mha/sdpa-compressed-v-int8-gs32.json

tests/benchdnn/inputs/graph/pattern/harness_bf16_all

@@ -3,8 +3,8 @@
 --reset --dt=bf16 --case=pattern/f32/bn_relu_fusion.json
 --reset --dt=bf16 --case=pattern/f32/conv_bias_post_ops_fusion.json
 # This fusion pattern is not support on GPU engine for now, will split into 2
-# partitions with GPU engine
---reset --dt=bf16 --case=pattern/f32/conv_depthwise_fusion_cpu.json
+# partitions with GPU engine. Skip the partition number check for it.
+--reset --dt=bf16  --expected-n-partitions=0 --case=pattern/f32/conv_depthwise_fusion_cpu.json
 --reset --dt=bf16 --case=pattern/f32/conv_post_ops_fusion.json
 --reset --dt=bf16 --case=pattern/f32/convtranspose_post_ops_fusion.json
 --reset --dt=bf16 --case=pattern/f32/matmul_bias_post_ops_chain_fusion.json

tests/benchdnn/inputs/graph/pattern/harness_f16_all

@@ -3,8 +3,8 @@
 --reset --dt=f16 --case=pattern/f32/bn_relu_fusion.json
 --reset --dt=f16 --case=pattern/f32/conv_bias_post_ops_fusion.json
 # This fusion pattern is not support on GPU engine for now, will split into 2
-# partitions with GPU engine
---reset --dt=f16 --case=pattern/f32/conv_depthwise_fusion_cpu.json
+# partitions with GPU engine. Skip the partition number check for it.
+--reset --dt=f16 --expected-n-partitions=0 --case=pattern/f32/conv_depthwise_fusion_cpu.json
 --reset --dt=f16 --case=pattern/f32/conv_post_ops_fusion.json
 --reset --dt=f16 --case=pattern/f32/convtranspose_post_ops_fusion.json
 --reset --dt=f16 --case=pattern/f32/matmul_bias_post_ops_chain_fusion.json