add support for TensorRT 10.10

Context

This MR aims to add support for newer version of TensorRT 10.10.

Modified files

Added tensorrt_10.10 folder which is a copy of tensorrt_8.6 folder in static/ directory. The following modifications have been applied to match with the 10.10 API version of TensorRT :

• Modified tools/tensorrt10.10_compiler.Dockerfile & Makefile to use correct version of tensorrt
• Updated unsupported dims32 → dims in src/Graph.cpp to match TRT 10 API
• Updated deprecated PlatformHasFast*() → custom cudaHasFast*() for datamode() mapping in include/Utils.cpp and src/Graph.cpp
• Updated trt_version = 10.10 by default in export() in __init__.py

Detailed major modifications

Added a static bool cudaSupportsDatatype(nvinfer1::Datatype datatype) to confirm that a datatype is supported on the CUDA SM. The support matrix is available at Support Hardware & Precision Matrix

Acknowledged deprecations

• The implicit-quantization API (IInt8Calibrator) is deprecated in TensorRT 10.10; while aidge integrates ONNX Q/DQ layer support for the recommended explicit-quantization workflow, we’ll keep implicit quantization now and migrate in a future refactor. More informations are available at Quantized types documentation

changed target branch from main to dev

mentioned in merge request !7 (closed)

changed target branch from dev to main

changed title from Draft: feat: add support for TensorRT 10.10 to Draft: add support for TensorRT 10.10

changed the description

marked the checklist item Resolve any remaining deprecated API warnings as completed

marked the checklist item Resolve any remaining deprecated API warnings as incomplete

assigned to @mmarchal

requested review from @cmoineau

added StatusWork in Progress label

added Enhancement ⭐ label

marked this merge request as ready

@cmoineau the MR is ready

changed the description

merged