NOTE: The Jupyter Notebook below is included in the Chimera SDK and can be run interactively by running the following CLI command:
$ quadric sdk notebook
From the Jupyter Notebook window in your browser, select the notebook named /quadric/sdk-cli/examples/custom_op/prototype_custom_op_perf.ipynb.
import tvm.contrib.epu.graphutils as graphutils
import onnx
import os, pathlib
import logging
import onnxruntime
from tvm.contrib.epu.chimera_job.chimera_job import ChimeraJob
onnx_file = "./ultraface_quant_v2.onnx"
onnx_session = onnxruntime.InferenceSession(onnx_file)
model = onnx.load(onnx_file)
start
util = graphutils.CustomOpReplacer(model)
_, custom_model = util.replace_subgraph(
"concat_output_1",
[
"Transpose_48",
"Transpose_70",
"Transpose_88",
"Transpose_102",
"Transpose_53",
"Transpose_75",
"Transpose_93",
"Transpose_105",
],
"ultrafaceBoundingBox",
False,
keep_constants=[
"317_scale",
"363_scale",
"403_scale",
"435_scale",
"331_scale",
"377_scale",
"417_scale",
"447_scale",
"467_quantized",
"467_scale",
"465_quantized",
"465_scale",
],
)
onnx_file_custom = os.path.basename(onnx_file.replace(".onnx", "-prototype_custom_op.onnx"))
print(onnx_file_custom)
onnx.save(custom_model, onnx_file_custom)
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: concat_output_0
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: scores_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 459_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 328_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 317_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 316
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 315_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 314
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 313_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 309_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 308_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 289_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 288
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 286_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 284_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 283
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 281_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 280
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 278_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 277
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 275_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 274
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 272_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 271
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 269_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 268
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 266_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 265
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 263_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 262
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 260_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 259
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 257_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 256
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 254_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 253
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 251_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 250
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 248_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 247
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 245_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 296_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 295
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 293_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 291_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 306_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 305
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 303_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 302
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 300_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 298_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 311_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 374_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 363_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 362
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 361_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 360
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 358_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 357
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 355_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 354
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 352_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 351
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 349_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 348
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 346_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 345
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 343_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 414_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 403_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 402
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 401_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 400
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 398_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 397
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 395_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 394
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 392_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 391
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 389_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 446_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 435_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 434
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 433_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 432
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 431_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 430
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 429_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: boxes_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 480
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 481
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 475_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 468_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 466_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 464_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 462
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 460_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 342_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 331_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 330
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 329_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 388_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 377_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 376
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 375_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 428_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 417_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 416
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 415_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 458_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 447_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 474_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 472
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 471_quantized
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 469
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 484
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 482
WARNING:root:ONNX node does not have a name, deriving temp name from its outputs. Node was named to: 485_quantized
ultraface_quant_v2-prototype_custom_op.onnx
end
Automatically generating the required CCL (C++) prototypes for Custom Op's in header files
cgc_job = ChimeraJob(model_p=onnx_file_custom)
custom_file = cgc_job.analyze_network()
2026-06-19 03:59 - INFO - epu - chimera_job - START==================================onnx_ingest
2026-06-19 03:59 - INFO - epu - codegen - START===============================optimize_relay
2026-06-19 03:59 - INFO - epu - codegen - START====================quantize_to_cpu_runnable_fx
2026-06-19 03:59 - INFO - epu - fx -
Source name Op Output 0 Range Output 0 Frac Bits
------------------------------ ----------------------------- ----------------------------- --------------------
CustomOp/ultrafaceBoundingBox0 contrib.epu.quadric_custom_op (-32768.0, 32767.99998474121) 16
Analysis of ultraface_quant_v2-prototype_custom_op.onnx
2026-06-19 03:59 - INFO - epu - chimera_job - Function Prototypes for Custom OPs found
2026-06-19 03:59 - INFO - epu - chimera_job - Please see the file: ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1_analyze_custom_op_template.hpp for a template example of custom op implementation
List of custom ops found in the network:
ultrafaceBoundingBox
2026-06-19 03:59 - INFO - epu - chimera_job - Wrote custom op seed file to ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1_analyze_custom_op_template.hpp.
## The analyze command will generate a custom op template file that can be used for compliation!!!
!cat {custom_file}
template <typename OcmAllocatorType> void ultrafaceBoundingBox (DdrTensor <std::int8_t, 1, 6, 30, 40>& ext_tensor_in0 /* 317_quantized */ ,
DdrTensor <std::int8_t, 1, 4, 15, 20>& ext_tensor_in1 /* 363_quantized */ ,
DdrTensor <std::int8_t, 1, 4, 8, 10>& ext_tensor_in2 /* 403_quantized */ ,
DdrTensor <std::int8_t, 1, 6, 4, 5>& ext_tensor_in3 /* 435_quantized */ ,
DdrTensor <std::int8_t, 1, 12, 30, 40>& ext_tensor_in4 /* 331_quantized */ ,
DdrTensor <std::int8_t, 1, 8, 15, 20>& ext_tensor_in5 /* 377_quantized */ ,
DdrTensor <std::int8_t, 1, 8, 8, 10>& ext_tensor_in6 /* 417_quantized */ ,
DdrTensor <std::int8_t, 1, 12, 4, 5>& ext_tensor_in7 /* 447_quantized */ ,
FixedPoint32<31> scalar_in8 /* 317_quantized.317_scale */ ,
FixedPoint32<31> scalar_in9 /* 363_quantized.363_scale */ ,
FixedPoint32<31> scalar_in10 /* 403_quantized.403_scale */ ,
FixedPoint32<31> scalar_in11 /* 435_quantized.435_scale */ ,
FixedPoint32<31> scalar_in12 /* 331_quantized.331_scale */ ,
FixedPoint32<31> scalar_in13 /* 377_quantized.377_scale */ ,
FixedPoint32<31> scalar_in14 /* 417_quantized.417_scale */ ,
FixedPoint32<31> scalar_in15 /* 447_quantized.447_scale */ ,
DdrTensor <std::int8_t, 1, 4420, 2>& const_in16 /* CustomOp/ultrafaceBoundingBox0.467_quantized */ ,
FixedPoint32<31> scalar_in17 /* CustomOp/ultrafaceBoundingBox0.467_scale */ ,
DdrTensor <std::int8_t, 1, 4420, 2>& const_in18 /* CustomOp/ultrafaceBoundingBox0.465_quantized */ ,
FixedPoint32<31> scalar_in19 /* CustomOp/ultrafaceBoundingBox0.465_scale */ ,
DdrTensor <FixedPoint32<16>, 1, 6, 4420>& ext_tensor_in20 /* CustomOp/ultrafaceBoundingBox0 */ ,
OcmAllocatorType& ocmMem) {
/* User implementation of custom op */
}
Compile network with custom op prototype stub
cgc_job = ChimeraJob(
model_p=onnx_file_custom,
custom_ops=custom_file,
)
cgc_job.compile()
2026-06-19 03:59 - INFO - epu - chimera_job - START==================================onnx_ingest
2026-06-19 03:59 - INFO - epu - codegen - START===============================optimize_relay
2026-06-19 03:59 - INFO - epu - codegen - START====================quantize_to_cpu_runnable_fx
2026-06-19 03:59 - INFO - epu - fx -
Source name Op Output 0 Range Output 0 Frac Bits
------------------------------ ----------------------------- ----------------------------- --------------------
CustomOp/ultrafaceBoundingBox0 contrib.epu.quadric_custom_op (-32768.0, 32767.99998474121) 16
2026-06-19 03:59 - INFO - epu - codegen - START====================build_cpu_runnable_fx_relay
2026-06-19 03:59 - INFO - epu - codegen - START=======================quantize_to_chimera_fx
2026-06-19 03:59 - INFO - epu - codegen - START=================================relay_to_tir
2026-06-19 03:59 - INFO - epu - codegen - START===========================relay_to_epu_relay
2026-06-19 03:59 - INFO - epu - codegen - START==============================adapt_and_order
2026-06-19 03:59 - INFO - epu - mac_counter -
2026-06-19 03:59 - INFO - epu - mac_counter - ============================================================
2026-06-19 03:59 - INFO - epu - mac_counter - MAC Operation Count Summary
2026-06-19 03:59 - INFO - epu - mac_counter - ============================================================
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 16,588,800 ops (8,294,400 MACs) - 245_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 5,529,600 ops (2,764,800 MACs) - 248_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 19,660,800 ops (9,830,400 MACs) - 251_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,764,800 ops (1,382,400 MACs) - 254_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 257_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,764,800 ops (1,382,400 MACs) - 260_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 263_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 691,200 ops (345,600 MACs) - 266_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 4,915,200 ops (2,457,600 MACs) - 269_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,382,400 ops (691,200 MACs) - 272_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 275_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,382,400 ops (691,200 MACs) - 278_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 281_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,228,800 ops (614,400 MACs) - 284_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,764,800 ops (1,382,400 MACs) - 286_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 5,529,600 ops (2,764,800 MACs) - 289_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,228,800 ops (614,400 MACs) - 291_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,764,800 ops (1,382,400 MACs) - 293_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 5,529,600 ops (2,764,800 MACs) - 296_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,228,800 ops (614,400 MACs) - 298_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,073,600 ops (1,036,800 MACs) - 300_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 4,147,200 ops (2,073,600 MACs) - 303_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 5,529,600 ops (2,764,800 MACs) - 306_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 7,372,800 ops (3,686,400 MACs) - 309_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 311_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,382,400 ops (691,200 MACs) - 315_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 921,600 ops (460,800 MACs) - 317_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 345,600 ops (172,800 MACs) - 343_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 4,915,200 ops (2,457,600 MACs) - 346_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 691,200 ops (345,600 MACs) - 349_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 352_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 691,200 ops (345,600 MACs) - 355_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 9,830,400 ops (4,915,200 MACs) - 358_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 691,200 ops (345,600 MACs) - 361_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 307,200 ops (153,600 MACs) - 363_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 184,320 ops (92,160 MACs) - 389_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 5,242,880 ops (2,621,440 MACs) - 392_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 368,640 ops (184,320 MACs) - 395_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 10,485,760 ops (5,242,880 MACs) - 398_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 368,640 ops (184,320 MACs) - 401_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 163,840 ops (81,920 MACs) - 403_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 2,621,440 ops (1,310,720 MACs) - 429_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 23,040 ops (11,520 MACs) - 431_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 655,360 ops (327,680 MACs) - 433_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 552,960 ops (276,480 MACs) - 435_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,382,400 ops (691,200 MACs) - 329_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,843,200 ops (921,600 MACs) - 331_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 691,200 ops (345,600 MACs) - 375_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 614,400 ops (307,200 MACs) - 377_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 368,640 ops (184,320 MACs) - 415_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 327,680 ops (163,840 MACs) - 417_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - conv2d: 1,105,920 ops (552,960 MACs) - 447_quantized
2026-06-19 03:59 - INFO - epu - mac_counter - ------------------------------------------------------------
2026-06-19 03:59 - INFO - epu - mac_counter - Total: 200,837,120 ops (100,418,560 MACs)
2026-06-19 03:59 - INFO - epu - mac_counter - ============================================================
2026-06-19 03:59 - INFO - epu - mac_counter -
2026-06-19 03:59 - INFO - epu - codegen - START==============================amend_ctrl_flow
2026-06-19 03:59 - INFO - epu - codegen - START=============================plan_lrm_virtual
2026-06-19 03:59 - INFO - epu - codegen - START==============================amend_ctrl_flow
2026-06-19 03:59 - INFO - epu - codegen - START===============================lrm_alloc_loop
2026-06-19 03:59 - INFO - epu - codegen - START==============================amend_ctrl_flow
2026-06-19 03:59 - INFO - epu - codegen - START================================lrm_splitting
2026-06-19 04:00 - INFO - epu - fuse_weights - No optimization for node 306_quantized with properties { kernel_size: [3, 3], channels: 16, strides:[1, 1], padding:[5, 5, 5, 5], data_layout:NCHW, dilation:[5, 5], groups:1, kernel_layout:OIHW, out_dtype:int8, out_layout: }.Applying General Convolution algorithm. Performance for this node can be improved significantly. Please refer to the documentation or contact Quadric Support.
2026-06-19 04:00 - INFO - epu - fuse_weights - No optimization for node 289_quantized with properties { kernel_size: [3, 3], channels: 16, strides:[1, 1], padding:[2, 2, 2, 2], data_layout:NCHW, dilation:[2, 2], groups:1, kernel_layout:OIHW, out_dtype:int8, out_layout: }.Applying General Convolution algorithm. Performance for this node can be improved significantly. Please refer to the documentation or contact Quadric Support.
2026-06-19 04:00 - INFO - epu - fuse_weights - No optimization for node 296_quantized with properties { kernel_size: [3, 3], channels: 16, strides:[1, 1], padding:[3, 3, 3, 3], data_layout:NCHW, dilation:[3, 3], groups:1, kernel_layout:OIHW, out_dtype:int8, out_layout: }.Applying General Convolution algorithm. Performance for this node can be improved significantly. Please refer to the documentation or contact Quadric Support.
2026-06-19 04:00 - INFO - epu - codegen - START==============================ext_split_relay
2026-06-19 04:01 - INFO - epu - codegen - START====================================build_tir
2026-06-19 04:01 - INFO - epu - chimera_job - Compilation of ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1 successful
Evaluate performance of the network with an Empty Custom Operation
cgc_job.run_inference_harness()
print(cgc_job)
cgc_job.plot_run_statistics()
2026-06-19 04:01 - INFO - epu - iss_testing - No tranges found for input, use default float range: <tvm.contrib.epu.interval.Interval object at 0x7b2be76579d0>
FILM 15/15: 100%|███████████████████████████████████████████████████| 15/15 [00:04<00:00, 3.15it/s]
2026-06-19 04:01 - INFO - epu - iss_testing - No tranges found for input, use default float range: <tvm.contrib.epu.interval.Interval object at 0x7b2d24d8db40>
2026-06-19 04:01 - INFO - epu - iss_testing - Started Executing Onnxruntime...
2026-06-19 04:01 - INFO - epu - iss_testing - Done 0:00:00.053078
╒═════════════════════╤══════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════╕
│ Module Name │ ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1 │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ ONNX File │ ultraface_quant_v2-prototype_custom_op.onnx │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Custom Ops │ /quadric/sdk-cli/examples/custom_op/ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1_analyze_custom_op_template.hpp │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Product Target │ QC-U │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Number of Cores │ 1 │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ ISS Clock Frequency │ 1.700 │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ L2M Size │ 16MB │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ LRM Size │ 4kB │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ External Read BW │ 128GBps │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ External Write BW │ 128GBps │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ MACS per PE │ 16 │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Max L2M │ 1.107MB │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Max LRM │ 0.506kB │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Max Temp Ext Bytes │ 0.127MB │
├─────────────────────┼──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────┤
│ Network GMACs │ 0.100 │
╘═════════════════════╧══════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════════╛
NOTE: CGC has used general convolution for some nodes, which may result in suboptimal performance. For performance-critical applications, please contact Quadric support to explore optimization strategies.
General convolution nodes: 306_quantized, 289_quantized, 296_quantized
For more details, see: https://app.quadric.ai/docs/latest/chimera-software-user-guide/chimera-graph-compiler-cgc/graph-optimizations-performed-by-cgc#general-convolution
╒════╤════════╤═════════════════╤══════════════════╤══════════════════════════╤═══════╕
│ │ Type │ Name │ shape │ type │ mse │
╞════╪════════╪═════════════════╪══════════════════╪══════════════════════════╪═══════╡
│ 0 │ Input │ input │ [1, 3, 240, 320] │ tensor[FixedPoint32<27>] │ n/a │
├────┼────────┼─────────────────┼──────────────────┼──────────────────────────┼───────┤
│ 1 │ Output │ concat_output_1 │ [1, 6, 4420] │ tensor[FixedPoint32<16>] │ 0.370 │
╘════╧════════╧═════════════════╧══════════════════╧══════════════════════════╧═══════╛
Post-ISS Report 1.7 GHz ***
Fully placed-and-routed gate simulation:
╒══════════════════════════════════╤═════════╕
│ Latency (ms) │ 0.19 │
├──────────────────────────────────┼─────────┤
│ FPS │ 5240.12 │
├──────────────────────────────────┼─────────┤
│ Average Power @ 3nm SSGNP (mW) │ 1375.14 │
├──────────────────────────────────┼─────────┤
│ FPS per Watt @ 3nm SSGNP (FPS/W) │ 3810.61 │
├──────────────────────────────────┼─────────┤
│ Ext Rd Bytes (MB) │ 1.25 │
├──────────────────────────────────┼─────────┤
│ Ext Wr Bytes (MB) │ 0.03 │
├──────────────────────────────────┼─────────┤
│ Avg Ext Rd BW (GBps) │ 6.42 │
├──────────────────────────────────┼─────────┤
│ Avg Ext Wr BW (GBps) │ 0.13 │
├──────────────────────────────────┼─────────┤
│ MAC Utilization │ 1.89% │
╘══════════════════════════════════╧═════════╛
*** Data generated using 7nm SSGNP gatesim and scaled to 3nm
[SDK-CLI] : TotalCycles: 324,420
[SDK-CLI] : Executions/second: 5,240.12
compute : ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 135.474K
data_array : ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 62.284K
mac : ▇▇▇▇▇▇▇▇▇▇▇▇▇ 36.557K
data_external: ▇▇▇ 10.185K
data_ocm : ▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 79.799K
for more information check run directory: /quadric/sdk-cli/examples/custom_op/ccl_build/ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1/run/20260619_040116_e35fc6
2026-06-19 04:01 - INFO - epu - chimera_job - Combined plots generated and saved to:
/quadric/sdk-cli/examples/custom_op/ccl_build/ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1/run/20260619_040116_e35fc6/data/ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1.combined.png
'/quadric/sdk-cli/examples/custom_op/ccl_build/ultraface_quant_v2_prototype_custom_op_QC_U_1d7_16MB_4kB_128GBps_128GBps_16_OFF_x1_x1/run/20260619_040116_e35fc6/data'
