What are the tactics in quantizing a model

[alsozatch]

TensorRT (with command trtexec) has the following output. What are the 3 tactics being compared here?

[12/12/2023-12:31:07] [V] [TRT] --------------- Timing Runner: Optimizer Reformat( → /model.12/m.0/cv2/act/Mul_output_0) (Reformat)
[12/12/2023-12:31:07] [V] [TRT] Setting a default quantization params because quantization data is missing for
[12/12/2023-12:31:08] [V] [TRT] Tactic: 0x00000000000003e8 Time: 0.289504
[12/12/2023-12:31:08] [V] [TRT] Setting a default quantization params because quantization data is missing for
[12/12/2023-12:31:08] [V] [TRT] Tactic: 0x00000000000003ea Time: 0.605257
[12/12/2023-12:31:08] [V] [TRT] Setting a default quantization params because quantization data is missing for
[12/12/2023-12:31:08] [V] [TRT] Tactic: 0x0000000000000000 Time: 0.291301
[12/12/2023-12:31:08] [V] [TRT] Fastest Tactic: 0x00000000000003e8 Time: 0.289504

[AastaLLL]

Hi,

TensorRT has several algorithms for a layer.
The log indicates that TensorRT has tried the possible algorithm and picked the fastest one.

Thanks.