Pulsar2 SIN operator failed to convert.

[nnn112358]

Issue Overview :

In pulsar2 version 3.3, when target_hardware is AX620E, SIN operator could not be converted.
It proceeds to generate quant_axmodel.onnx, but fails to generate axmodel.

According to “2. NPU Operators support list(AX620E)”, Sin is Supported to Unlimited.
https://pulsar2-docs.readthedocs.io/en/latest/appendix/op_support_list_ax620e.html

When target_hardware was AX650,converting the SIN operator passed.
I hope that target_hardware is AX620E can also convert SIN operators.

Enviroment

• pulsar2 version 3.3
• target_hardware AX620E

reproduction procedure:

I have placed the file to be reproduced here.
https://github.com/nnn112358/pulsar2_onnxcheck_sin/tree/main

# pulsar2 version
version: 3.3
commit: 3cdead5e

# python sin_export_model.py
Sin Operation:
Input shape: (1, 255)
Output shape: (1, 255)
Output range: [-1.0000, 0.9999]
Match: True

Tar file creation complete: input_data.tar
Temporary files deleted


root#  pulsar2 build --config sin_model_AX620E.json
~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~
===>export io data to folder: sin_model_AX620E/quant/debug/io
Building native ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% 0:00:00
2025-02-04 00:59:31.819 | INFO     | yamain.command.build:compile_ptq_model:980 - group 0 compiler transformation
2025-02-04 00:59:31.862 | WARNING  | yamain.command.load_model:pre_process:616 - preprocess tensor [x]
2025-02-04 00:59:31.862 | WARNING  | yamain.command.load_model:post_process:638 - postprocess tensor [z]
2025-02-04 00:59:31.863 | INFO     | yamain.command.load_model:ir_compiler_transformation:824 - use quant data as gt input: x, float32, (1, 255)
2025-02-04 00:59:31.891 | INFO     | yamain.command.build:compile_ptq_model:999 - group 0 QuantAxModel macs: 0
2025-02-04 00:59:31.892 | INFO     | yamain.command.build:compile_ptq_model:1139 - subgraph [0], group: 0, type: GraphType.NPU
2025-02-04 00:59:31.892 | INFO     | yamain.command.npu_backend_compiler:compile:157 - compile npu subgraph [0]
tiling op... ⠋ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 1/3 0:00:00
Traceback (most recent call last):
  File "<frozen yasched.graph_proc.graph_tiling>", line 109, in run
  File "<frozen backend>", line 158, in f
  File "<frozen backend.ax620e.backend_impl>", line 121, in build
AssertionError: AxQuantizedSin dont support

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<frozen yamain.common.error>", line 59, in guard_context
  File "<frozen yamain.command.build>", line 1169, in compile_ptq_model
  File "<frozen yamain.command.npu_backend_compiler>", line 199, in compile
  File "<frozen yasched.test_onepass>", line 3259, in test_onepass_ir
  File "<frozen yasched.test_onepass>", line 3243, in build
  File "<frozen yasched.test_onepass>", line 3383, in graph2models
  File "<frozen yasched.test_onepass>", line 670, in graph_sched
  File "<frozen yasched.graph_proc.graph_tiling>", line 134, in run
yasched.exceptions.TileFailException: AxQuantizedSin, AxQuantizedSin dont support
    op: /Sin
    attrs: {'const_inputs': {'x_scale': array(0.99607843, dtype=float32), 'r_scale': array(0.00784275, dtype=float32), 'x_zeropoint': array(0, dtype=int32), 'r_zeropoint': array(128, dtype=int32)}, 'output_dtype': 'U8', 'quant_method': 0}
    inputs: {'x': Tensor(U8, name=x_QuantizeLinear_out_link_var, shape=(1, 255), offset=0, bit_strides=(2048, 8))}
    mem_limit: MemLimit(workspace=524288, max_mem_size=720640)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "</opt/pulsar2/yamain/main.py>", line 4, in <module>
  File "<frozen yamain.main>", line 329, in <module>
  File "<frozen yamain.main>", line 257, in pulsar2
  File "<frozen yamain.main>", line 152, in wrapper
  File "<frozen yamain.common.error>", line 23, in wrapper
  File "<frozen yamain.command.build>", line 105, in build_error
  File "<frozen yamain.common.error>", line 14, in wrapper
  File "<frozen yamain.command.build>", line 507, in build
  File "<frozen yamain.command.build>", line 1172, in compile_ptq_model
  File "/usr/local/lib/python3.9/contextlib.py", line 137, in __exit__
    self.gen.throw(typ, value, traceback)
  File "<frozen yamain.common.error>", line 61, in guard_context
  File "<frozen yamain.common.error>", line 73, in error_func
yamain.common.error.CodeException: (<ErrorCode.NPUBackendError: 8>, TileFailException("AxQuantizedSin, AxQuantizedSin dont support\n    op: /Sin\n    attrs: {'const_inputs': {'x_scale': array(0.99607843, dtype=float32), 'r_scale': array(0.00784275, dtype=float32), 'x_zeropoint': array(0, dtype=int32), 'r_zeropoint': array(128, dtype=int32)}, 'output_dtype': 'U8', 'quant_method': 0}\n    inputs: {'x': Tensor(U8, name=x_QuantizeLinear_out_link_var, shape=(1, 255), offset=0, bit_strides=(2048, 8))}\n    mem_limit: MemLimit(workspace=524288, max_mem_size=720640)"))