notebooks/real_time_nrx.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "8bf69742",
   "metadata": {},
   "source": [
    "# Toward Real-Time Inference of Neural Receivers with TensorRT\n",
    "\n",
    "This tutorial demonstrates a proof-of-concept implementation for real-time inference of the proposed neural receiver on a GPU. The input/output shapes are optimized for compatibility with the [NVIDIA Aerial platform](https://developer.nvidia.com/aerial), enabling seamless deployment in a commercial-grade 5G NR ecosystem.\n",
    "\n",
    "Please note that the exact inference latency depends on various factors, including the hardware platform, the specific neural receiver configuration, and the 5G NR parameters. We have simplified the neural receiver architecture to achieve an inference latency of approximately 1 ms on an NVIDIA A100 GPU. This provides a trade-off between BLER performance and latency.\n",
    "\n",
    "To minimize latency, we set the batch size to 1, as buffering slots would significantly increase processing latency and is therefore not an option.\n",
    "\n",
    "This notebook uses [TensorRT](https://developer.nvidia.com/tensorrt) as the inference engine. Refer to the [TensorRT Best Practices](https://docs.nvidia.com/deeplearning/tensorrt/pdf/TensorRT-Best-Practices.pdf) document for more details on the inference performance evaluation.\n",
    "\n",
    "Running the provided code assumes that the training of the NRX is already\n",
    "completed. For more details on the training and evaluation pipeline, see the\n",
    "notebook [Jumpstart into Multi-User MIMO Neural Receiver for 5G NR PUSCH](jumpstart_tutorial.ipynb).\n",
    "\n",
    "Additional packages required for this notebook include:\n",
    "- `tf2onnx` and `onnx==1.14` for `export_onnx.py` (available on PyPI)\n",
    "- `polygraphy` for running the inference pipeline (available on PyPI)\n",
    "- Full installation of [TensorRT](https://developer.nvidia.com/tensorrt-getting-started) must be installed locally (including GPU drivers). For installation instructions, refer to the [TensorRT Installation Guide](https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html). *Hint:* If you are using the Sionna Docker container, you can simply install TensorRT by `sudo apt-get install tensorrt`. This still might require to run `pip install tensorrt` afterwards and by updating the path variable ``export PATH=\"/usr/src/tensorrt/bin/:$PATH``.\n",
    "\n",
    "Remark: the Python package `onnx==1.14` is required (`onnx==1.15` does not work due to a known bug).\n",
    "\n",
    "Known limitations:\n",
    "- The number of allocated time/frequency resources is static due to current TensorRT limitations. However, padding can be used to keep input/output dimensions constant.\n",
    "- The DMRS configuration is *baked* into the exported ONNX model (can be made dynamic if required). This approach eliminates the need for pre-computed indices for LS channel estimation.\n",
    "- Variable MCS schemes are currently not supported in the TensorRT engine for a\n",
    "  neural receiver with the  *Var-IO* scheme (see [Neural Receiver with Support for Variable Modulation and Coding Schemes](variable_mcs_nrx.ipynb)).\n",
    "- After the ONNX export, the model cannot be trained anymore. This would currently require to re-run the export scripts to re-generate the engine with new weights.\n",
    "- The TensorRT engine does not include the FEC decoder and rate-matching. However, this can be done by using components from the Aerial platform.\n",
    "\n",
    "\n",
    "Please consider this code as a proof-of-concept rather than production-ready. Further, we aim to give researchers an idea of what network dimensions are feasible for real-time deployment."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc3232f2",
   "metadata": {},
   "source": [
    "![NRX_Flow](data:image/png;base64,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)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4d3b57e4",
   "metadata": {},
   "source": [
    "## Table of Contents\n",
    "\n",
    "* [Exporting the Receiver to TensorRT](#Exporting-the-Receiver-to-TensorRT)\n",
    "* [Running the TensorRT Engine](#Running-the-TensorRT-Engine)\n",
    "* [Inference Using TensorRT](#Inference-Using-TensorRT)\n",
    "* [BLER Performance Validation](#BLER-Performance-Validation)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4f61d459",
   "metadata": {},
   "source": [
    "## GPU Configuration and Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2692aa51",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Only GPU number 0 used.\n"
     ]
    }
   ],
   "source": [
    "gpu_num = 0 # Number of the GPU to be used\n",
    "\n",
    "# Suppress Tensorflow warnings\n",
    "import os\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = f\"{gpu_num}\"\n",
    "os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "import tensorflow as tf\n",
    "tf.get_logger().setLevel('ERROR')\n",
    "\n",
    "gpus = tf.config.list_physical_devices('GPU')\n",
    "if gpus:\n",
    "    try:\n",
    "        print('Only GPU number', gpu_num, 'used.')\n",
    "        tf.config.experimental.set_memory_growth(gpus[0], True)\n",
    "    except RuntimeError as e:\n",
    "        print(e)\n",
    "\n",
    "# Import libraries\n",
    "import sys\n",
    "sys.path.append('../') # to import utils\n",
    "\n",
    "import sionna as sn\n",
    "from utils import DataGeneratorAerial, DataEvaluator, Parameters\n",
    "\n",
    "from polygraphy.backend.common import bytes_from_path\n",
    "from polygraphy.backend.trt import engine_from_bytes, Profile\n",
    "from polygraphy.backend.trt.runner import TrtRunner\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "729442c1",
   "metadata": {},
   "source": [
    "## Exporting the Receiver to TensorRT\n",
    "\n",
    "After training, the neural receiver can be exported to the [ONNX](https://onnx.ai/) file format. Note that the exported model uses the weights from the previous training procedure ( `scripts/train_neural_rx.py`), as specified in the configuration file `nrx_rt.cfg`.\n",
    "\n",
    "The model export can be done by running `scripts/export_onnx.py` which automatically exports\n",
    "the receiver model to the ONNX file format and builds a TensorRT engine.\n",
    "Note that the engine is hardware and software platform specific and must be build on your target platform."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "20f8b4ab",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tf2onnx.tf_loader:Could not search for non-variable resources. Concrete function internal representation may have changed.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "&&&& RUNNING TensorRT.trtexec [TensorRT v8601] # trtexec --fp16 --onnx=../onnx_models/nrx_rt.onnx --saveEngine=../onnx_models/nrx_rt.plan --minShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --optShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --maxShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4\n",
      "[08/07/2024-17:24:35] [I] === Model Options ===\n",
      "[08/07/2024-17:24:35] [I] Format: ONNX\n",
      "[08/07/2024-17:24:35] [I] Model: ../onnx_models/nrx_rt.onnx\n",
      "[08/07/2024-17:24:35] [I] Output:\n",
      "[08/07/2024-17:24:35] [I] === Build Options ===\n",
      "[08/07/2024-17:24:35] [I] Max batch: explicit batch\n",
      "[08/07/2024-17:24:35] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default\n",
      "[08/07/2024-17:24:35] [I] minTiming: 1\n",
      "[08/07/2024-17:24:35] [I] avgTiming: 8\n",
      "[08/07/2024-17:24:35] [I] Precision: FP32+FP16\n",
      "[08/07/2024-17:24:35] [I] LayerPrecisions: \n",
      "[08/07/2024-17:24:35] [I] Layer Device Types: \n",
      "[08/07/2024-17:24:35] [I] Calibration: \n",
      "[08/07/2024-17:24:35] [I] Refit: Disabled\n",
      "[08/07/2024-17:24:35] [I] Version Compatible: Disabled\n",
      "[08/07/2024-17:24:35] [I] TensorRT runtime: full\n",
      "[08/07/2024-17:24:35] [I] Lean DLL Path: \n",
      "[08/07/2024-17:24:35] [I] Tempfile Controls: { in_memory: allow, temporary: allow }\n",
      "[08/07/2024-17:24:35] [I] Exclude Lean Runtime: Disabled\n",
      "[08/07/2024-17:24:35] [I] Sparsity: Disabled\n",
      "[08/07/2024-17:24:35] [I] Safe mode: Disabled\n",
      "[08/07/2024-17:24:35] [I] Build DLA standalone loadable: Disabled\n",
      "[08/07/2024-17:24:35] [I] Allow GPU fallback for DLA: Disabled\n",
      "[08/07/2024-17:24:35] [I] DirectIO mode: Disabled\n",
      "[08/07/2024-17:24:35] [I] Restricted mode: Disabled\n",
      "[08/07/2024-17:24:35] [I] Skip inference: Disabled\n",
      "[08/07/2024-17:24:35] [I] Save engine: ../onnx_models/nrx_rt.plan\n",
      "[08/07/2024-17:24:35] [I] Load engine: \n",
      "[08/07/2024-17:24:35] [I] Profiling verbosity: 0\n",
      "[08/07/2024-17:24:35] [I] Tactic sources: Using default tactic sources\n",
      "[08/07/2024-17:24:35] [I] timingCacheMode: local\n",
      "[08/07/2024-17:24:35] [I] timingCacheFile: \n",
      "[08/07/2024-17:24:35] [I] Heuristic: Disabled\n",
      "[08/07/2024-17:24:35] [I] Preview Features: Use default preview flags.\n",
      "[08/07/2024-17:24:35] [I] MaxAuxStreams: -1\n",
      "[08/07/2024-17:24:35] [I] BuilderOptimizationLevel: -1\n",
      "[08/07/2024-17:24:35] [I] Input(s)s format: fp32:CHW\n",
      "[08/07/2024-17:24:35] [I] Output(s)s format: fp32:CHW\n",
      "[08/07/2024-17:24:35] [I] Input build shape: rx_slot_real=1x1584x14x4+1x1584x14x4+1x1584x14x4\n",
      "[08/07/2024-17:24:35] [I] Input build shape: rx_slot_imag=1x1584x14x4+1x1584x14x4+1x1584x14x4\n",
      "[08/07/2024-17:24:35] [I] Input build shape: h_hat_real=1x1584x2x4+1x1584x2x4+1x1584x2x4\n",
      "[08/07/2024-17:24:35] [I] Input build shape: h_hat_imag=1x1584x2x4+1x1584x2x4+1x1584x2x4\n",
      "[08/07/2024-17:24:35] [I] Input calibration shapes: model\n",
      "[08/07/2024-17:24:35] [I] === System Options ===\n",
      "[08/07/2024-17:24:35] [I] Device: 0\n",
      "[08/07/2024-17:24:35] [I] DLACore: \n",
      "[08/07/2024-17:24:35] [I] Plugins:\n",
      "[08/07/2024-17:24:35] [I] setPluginsToSerialize:\n",
      "[08/07/2024-17:24:35] [I] dynamicPlugins:\n",
      "[08/07/2024-17:24:35] [I] ignoreParsedPluginLibs: 0\n",
      "[08/07/2024-17:24:35] [I] \n",
      "[08/07/2024-17:24:35] [I] === Inference Options ===\n",
      "[08/07/2024-17:24:35] [I] Batch: Explicit\n",
      "[08/07/2024-17:24:35] [I] Input inference shape: h_hat_imag=1x1584x2x4\n",
      "[08/07/2024-17:24:35] [I] Input inference shape: h_hat_real=1x1584x2x4\n",
      "[08/07/2024-17:24:35] [I] Input inference shape: rx_slot_imag=1x1584x14x4\n",
      "[08/07/2024-17:24:35] [I] Input inference shape: rx_slot_real=1x1584x14x4\n",
      "[08/07/2024-17:24:35] [I] Iterations: 10\n",
      "[08/07/2024-17:24:35] [I] Duration: 3s (+ 200ms warm up)\n",
      "[08/07/2024-17:24:35] [I] Sleep time: 0ms\n",
      "[08/07/2024-17:24:35] [I] Idle time: 0ms\n",
      "[08/07/2024-17:24:35] [I] Inference Streams: 1\n",
      "[08/07/2024-17:24:35] [I] ExposeDMA: Disabled\n",
      "[08/07/2024-17:24:35] [I] Data transfers: Enabled\n",
      "[08/07/2024-17:24:35] [I] Spin-wait: Disabled\n",
      "[08/07/2024-17:24:35] [I] Multithreading: Disabled\n",
      "[08/07/2024-17:24:35] [I] CUDA Graph: Disabled\n",
      "[08/07/2024-17:24:35] [I] Separate profiling: Disabled\n",
      "[08/07/2024-17:24:35] [I] Time Deserialize: Disabled\n",
      "[08/07/2024-17:24:35] [I] Time Refit: Disabled\n",
      "[08/07/2024-17:24:35] [I] NVTX verbosity: 0\n",
      "[08/07/2024-17:24:35] [I] Persistent Cache Ratio: 0\n",
      "[08/07/2024-17:24:35] [I] Inputs:\n",
      "[08/07/2024-17:24:35] [I] === Reporting Options ===\n",
      "[08/07/2024-17:24:35] [I] Verbose: Disabled\n",
      "[08/07/2024-17:24:35] [I] Averages: 10 inferences\n",
      "[08/07/2024-17:24:35] [I] Percentiles: 90,95,99\n",
      "[08/07/2024-17:24:35] [I] Dump refittable layers:Disabled\n",
      "[08/07/2024-17:24:35] [I] Dump output: Disabled\n",
      "[08/07/2024-17:24:35] [I] Profile: Disabled\n",
      "[08/07/2024-17:24:35] [I] Export timing to JSON file: \n",
      "[08/07/2024-17:24:35] [I] Export output to JSON file: \n",
      "[08/07/2024-17:24:35] [I] Export profile to JSON file: \n",
      "[08/07/2024-17:24:35] [I] \n",
      "[08/07/2024-17:24:36] [I] === Device Information ===\n",
      "[08/07/2024-17:24:36] [I] Selected Device: NVIDIA GeForce RTX 3090\n",
      "[08/07/2024-17:24:36] [I] Compute Capability: 8.6\n",
      "[08/07/2024-17:24:36] [I] SMs: 82\n",
      "[08/07/2024-17:24:36] [I] Device Global Memory: 24257 MiB\n",
      "[08/07/2024-17:24:36] [I] Shared Memory per SM: 100 KiB\n",
      "[08/07/2024-17:24:36] [I] Memory Bus Width: 384 bits (ECC disabled)\n",
      "[08/07/2024-17:24:36] [I] Application Compute Clock Rate: 1.695 GHz\n",
      "[08/07/2024-17:24:36] [I] Application Memory Clock Rate: 9.751 GHz\n",
      "[08/07/2024-17:24:36] [I] \n",
      "[08/07/2024-17:24:36] [I] Note: The application clock rates do not reflect the actual clock rates that the GPU is currently running at.\n",
      "[08/07/2024-17:24:36] [I] \n",
      "[08/07/2024-17:24:36] [I] TensorRT version: 8.6.1\n",
      "[08/07/2024-17:24:36] [I] Loading standard plugins\n",
      "[08/07/2024-17:24:36] [I] [TRT] [MemUsageChange] Init CUDA: CPU +2, GPU +0, now: CPU 19, GPU 1420 (MiB)\n",
      "[08/07/2024-17:24:40] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +1444, GPU +268, now: CPU 1540, GPU 1688 (MiB)\n",
      "[08/07/2024-17:24:40] [I] Start parsing network model.\n",
      "[08/07/2024-17:24:40] [I] [TRT] ----------------------------------------------------------------\n",
      "[08/07/2024-17:24:40] [I] [TRT] Input filename:   ../onnx_models/nrx_rt.onnx\n",
      "[08/07/2024-17:24:40] [I] [TRT] ONNX IR version:  0.0.8\n",
      "[08/07/2024-17:24:40] [I] [TRT] Opset version:    15\n",
      "[08/07/2024-17:24:40] [I] [TRT] Producer name:    tf2onnx\n",
      "[08/07/2024-17:24:40] [I] [TRT] Producer version: 1.15.1 37820d\n",
      "[08/07/2024-17:24:40] [I] [TRT] Domain:           \n",
      "[08/07/2024-17:24:40] [I] [TRT] Model version:    0\n",
      "[08/07/2024-17:24:40] [I] [TRT] Doc string:       \n",
      "[08/07/2024-17:24:40] [I] [TRT] ----------------------------------------------------------------\n",
      "[08/07/2024-17:24:40] [I] Finished parsing network model. Parse time: 0.0127732\n",
      "[08/07/2024-17:24:40] [I] [TRT] Graph optimization time: 0.0343629 seconds.\n",
      "[08/07/2024-17:24:40] [I] [TRT] Local timing cache in use. Profiling results in this builder pass will not be stored.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[08/07/2024-17:24:40] [W] [TRT] onnx2trt_utils.cpp:374: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.\n",
      "[08/07/2024-17:24:40] [W] [TRT] onnx2trt_utils.cpp:514: Your ONNX model has been generated with double-typed weights, while TensorRT does not natively support double. Attempting to cast down to float.\n",
      "[08/07/2024-17:24:40] [W] [TRT] onnx2trt_utils.cpp:70: TensorRT is using FLOAT32 precision to run an INT32 ArgMax / ArgMin. Rounding errors may occur for large integer values\n",
      "[08/07/2024-17:24:40] [W] [TRT] Tensor DataType is determined at build time for tensors not marked as input or output.\n",
      "[08/07/2024-17:24:40] [W] [TRT] onnx2trt_utils.cpp:70: TensorRT is using FLOAT32 precision to run an INT32 ArgMax / ArgMin. Rounding errors may occur for large integer values\n",
      "[08/07/2024-17:24:40] [W] [TRT] Tensor DataType is determined at build time for tensors not marked as input or output.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[08/07/2024-17:25:56] [I] [TRT] Detected 7 inputs and 2 output network tensors.\n",
      "[08/07/2024-17:25:56] [I] [TRT] Total Host Persistent Memory: 109632\n",
      "[08/07/2024-17:25:56] [I] [TRT] Total Device Persistent Memory: 5120\n",
      "[08/07/2024-17:25:56] [I] [TRT] Total Scratch Memory: 50383872\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MemUsageStats] Peak memory usage of TRT CPU/GPU memory allocators: CPU 0 MiB, GPU 128 MiB\n",
      "[08/07/2024-17:25:56] [I] [TRT] [BlockAssignment] Started assigning block shifts. This will take 75 steps to complete.\n",
      "[08/07/2024-17:25:56] [I] [TRT] [BlockAssignment] Algorithm ShiftNTopDown took 2.25591ms to assign 8 blocks to 75 nodes requiring 72029184 bytes.\n",
      "[08/07/2024-17:25:56] [I] [TRT] Total Activation Memory: 72028672\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in building engine: CPU +0, GPU +4, now: CPU 0, GPU 4 (MiB)\n",
      "[08/07/2024-17:25:56] [I] Engine built in 80.2636 sec.\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[08/07/2024-17:25:56] [W] [TRT] TensorRT encountered issues when converting weights between types and that could affect accuracy.\n",
      "[08/07/2024-17:25:56] [W] [TRT] If this is not the desired behavior, please modify the weights or retrain with regularization to adjust the magnitude of the weights.\n",
      "[08/07/2024-17:25:56] [W] [TRT] Check verbose logs for the list of affected weights.\n",
      "[08/07/2024-17:25:56] [W] [TRT] - 26 weights are affected by this issue: Detected subnormal FP16 values.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[08/07/2024-17:25:56] [I] [TRT] Loaded engine size: 1 MiB\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in engine deserialization: CPU +0, GPU +0, now: CPU 0, GPU 0 (MiB)\n",
      "[08/07/2024-17:25:56] [I] Engine deserialized in 0.0071929 sec.\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MS] Running engine with multi stream info\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MS] Number of aux streams is 1\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MS] Number of total worker streams is 2\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MS] The main stream provided by execute/enqueue calls is the first worker stream\n",
      "[08/07/2024-17:25:56] [I] [TRT] [MemUsageChange] TensorRT-managed allocation in IExecutionContext creation: CPU +0, GPU +69, now: CPU 0, GPU 69 (MiB)\n",
      "[08/07/2024-17:25:56] [I] Setting persistentCacheLimit to 0 bytes.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input rx_slot_real\n",
      "[08/07/2024-17:25:56] [I] Input binding for rx_slot_real with dimensions 1x1584x14x4 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input rx_slot_imag\n",
      "[08/07/2024-17:25:56] [I] Input binding for rx_slot_imag with dimensions 1x1584x14x4 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input h_hat_real\n",
      "[08/07/2024-17:25:56] [I] Input binding for h_hat_real with dimensions 1x1584x2x4 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input h_hat_imag\n",
      "[08/07/2024-17:25:56] [I] Input binding for h_hat_imag with dimensions 1x1584x2x4 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input active_dmrs_ports\n",
      "[08/07/2024-17:25:56] [I] Input binding for active_dmrs_ports with dimensions 1x2 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input dmrs_ofdm_pos\n",
      "[08/07/2024-17:25:56] [I] Input binding for dmrs_ofdm_pos with dimensions 2x2 is created.\n",
      "[08/07/2024-17:25:56] [I] Using random values for input dmrs_subcarrier_pos\n",
      "[08/07/2024-17:25:56] [I] Input binding for dmrs_subcarrier_pos with dimensions 2x6 is created.\n",
      "[08/07/2024-17:25:56] [I] Output binding for output_1 with dimensions 1x4x2x1584x14 is created.\n",
      "[08/07/2024-17:25:56] [I] Output binding for output_2 with dimensions 1x2x1584x14x8 is created.\n",
      "[08/07/2024-17:25:56] [I] Starting inference\n",
      "[08/07/2024-17:25:59] [I] Warmup completed 158 queries over 200 ms\n",
      "[08/07/2024-17:25:59] [I] Timing trace has 2351 queries over 3.00373 s\n",
      "[08/07/2024-17:25:59] [I] \n",
      "[08/07/2024-17:25:59] [I] === Trace details ===\n",
      "[08/07/2024-17:25:59] [I] Trace averages of 10 runs:\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26996 ms - Host latency: 1.40822 ms (enqueue 0.430486 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26883 ms - Host latency: 1.4067 ms (enqueue 0.427335 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26945 ms - Host latency: 1.4072 ms (enqueue 0.430724 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26873 ms - Host latency: 1.40631 ms (enqueue 0.425497 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26925 ms - Host latency: 1.40787 ms (enqueue 0.434364 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27017 ms - Host latency: 1.40903 ms (enqueue 0.426923 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26986 ms - Host latency: 1.40815 ms (enqueue 0.4285 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26935 ms - Host latency: 1.40746 ms (enqueue 0.425076 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2676 ms - Host latency: 1.40535 ms (enqueue 0.42605 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26884 ms - Host latency: 1.40686 ms (enqueue 0.425275 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26873 ms - Host latency: 1.40748 ms (enqueue 0.431015 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26771 ms - Host latency: 1.4065 ms (enqueue 0.434619 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26843 ms - Host latency: 1.40621 ms (enqueue 0.43493 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26863 ms - Host latency: 1.40663 ms (enqueue 0.430106 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26852 ms - Host latency: 1.40688 ms (enqueue 0.429507 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26833 ms - Host latency: 1.40687 ms (enqueue 0.42702 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26823 ms - Host latency: 1.40663 ms (enqueue 0.426074 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26925 ms - Host latency: 1.40775 ms (enqueue 0.427698 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26894 ms - Host latency: 1.40688 ms (enqueue 0.425415 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26914 ms - Host latency: 1.40761 ms (enqueue 0.429254 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26853 ms - Host latency: 1.40659 ms (enqueue 0.427679 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26802 ms - Host latency: 1.40577 ms (enqueue 0.430289 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26894 ms - Host latency: 1.40666 ms (enqueue 0.426096 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26884 ms - Host latency: 1.4067 ms (enqueue 0.430838 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26792 ms - Host latency: 1.40609 ms (enqueue 0.427261 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26947 ms - Host latency: 1.4076 ms (enqueue 0.431995 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26884 ms - Host latency: 1.40667 ms (enqueue 0.427081 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26955 ms - Host latency: 1.41003 ms (enqueue 0.428888 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26882 ms - Host latency: 1.40736 ms (enqueue 0.431812 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26811 ms - Host latency: 1.40641 ms (enqueue 0.429181 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26903 ms - Host latency: 1.40707 ms (enqueue 0.431592 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26791 ms - Host latency: 1.40591 ms (enqueue 0.425934 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26844 ms - Host latency: 1.40663 ms (enqueue 0.42583 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26934 ms - Host latency: 1.40714 ms (enqueue 0.426263 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26852 ms - Host latency: 1.40694 ms (enqueue 0.430902 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26813 ms - Host latency: 1.40618 ms (enqueue 0.432904 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26843 ms - Host latency: 1.4071 ms (enqueue 0.428845 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26823 ms - Host latency: 1.40591 ms (enqueue 0.428845 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26793 ms - Host latency: 1.40544 ms (enqueue 0.437073 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26852 ms - Host latency: 1.4064 ms (enqueue 0.429181 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26946 ms - Host latency: 1.40816 ms (enqueue 0.435205 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26843 ms - Host latency: 1.40699 ms (enqueue 0.431232 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26904 ms - Host latency: 1.40762 ms (enqueue 0.429034 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26884 ms - Host latency: 1.40733 ms (enqueue 0.42738 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26957 ms - Host latency: 1.40836 ms (enqueue 0.433343 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26906 ms - Host latency: 1.4074 ms (enqueue 0.429065 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26882 ms - Host latency: 1.40636 ms (enqueue 0.427277 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26935 ms - Host latency: 1.40765 ms (enqueue 0.427423 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26904 ms - Host latency: 1.40757 ms (enqueue 0.42442 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27109 ms - Host latency: 1.40914 ms (enqueue 0.427112 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27292 ms - Host latency: 1.41113 ms (enqueue 0.426172 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.29166 ms - Host latency: 1.43103 ms (enqueue 0.427399 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.30181 ms - Host latency: 1.4399 ms (enqueue 0.42193 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.31974 ms - Host latency: 1.45819 ms (enqueue 0.42547 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.31473 ms - Host latency: 1.45314 ms (enqueue 0.425085 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.31002 ms - Host latency: 1.44811 ms (enqueue 0.422308 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.29772 ms - Host latency: 1.43881 ms (enqueue 0.422729 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.29555 ms - Host latency: 1.43358 ms (enqueue 0.426892 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.29127 ms - Host latency: 1.42838 ms (enqueue 0.432397 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2843 ms - Host latency: 1.41908 ms (enqueue 0.424304 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28399 ms - Host latency: 1.41992 ms (enqueue 0.428809 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28052 ms - Host latency: 1.41535 ms (enqueue 0.424683 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28163 ms - Host latency: 1.41675 ms (enqueue 0.426117 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28113 ms - Host latency: 1.41565 ms (enqueue 0.426501 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28008 ms - Host latency: 1.41517 ms (enqueue 0.424133 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28102 ms - Host latency: 1.41597 ms (enqueue 0.430872 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28101 ms - Host latency: 1.41584 ms (enqueue 0.427148 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27825 ms - Host latency: 1.41339 ms (enqueue 0.427234 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27867 ms - Host latency: 1.41333 ms (enqueue 0.428381 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27908 ms - Host latency: 1.41425 ms (enqueue 0.424573 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27877 ms - Host latency: 1.41317 ms (enqueue 0.427832 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28041 ms - Host latency: 1.41534 ms (enqueue 0.424414 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27971 ms - Host latency: 1.41503 ms (enqueue 0.425757 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27454 ms - Host latency: 1.40912 ms (enqueue 0.427551 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27527 ms - Host latency: 1.40992 ms (enqueue 0.422864 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27528 ms - Host latency: 1.41029 ms (enqueue 0.428516 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27528 ms - Host latency: 1.41049 ms (enqueue 0.425598 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27498 ms - Host latency: 1.41063 ms (enqueue 0.423547 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27459 ms - Host latency: 1.40955 ms (enqueue 0.425574 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2759 ms - Host latency: 1.41067 ms (enqueue 0.423486 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27572 ms - Host latency: 1.41014 ms (enqueue 0.428687 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27509 ms - Host latency: 1.41034 ms (enqueue 0.428845 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27459 ms - Host latency: 1.40964 ms (enqueue 0.428235 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2752 ms - Host latency: 1.41011 ms (enqueue 0.429395 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27509 ms - Host latency: 1.41013 ms (enqueue 0.424805 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2761 ms - Host latency: 1.41075 ms (enqueue 0.42959 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27449 ms - Host latency: 1.40941 ms (enqueue 0.427405 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27537 ms - Host latency: 1.4098 ms (enqueue 0.425049 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27417 ms - Host latency: 1.40909 ms (enqueue 0.424463 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2752 ms - Host latency: 1.40959 ms (enqueue 0.422083 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27446 ms - Host latency: 1.40902 ms (enqueue 0.427332 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27283 ms - Host latency: 1.40737 ms (enqueue 0.423535 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27316 ms - Host latency: 1.40815 ms (enqueue 0.428381 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27026 ms - Host latency: 1.40502 ms (enqueue 0.423181 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2675 ms - Host latency: 1.40195 ms (enqueue 0.426343 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26895 ms - Host latency: 1.40378 ms (enqueue 0.426196 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26832 ms - Host latency: 1.40273 ms (enqueue 0.429211 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26876 ms - Host latency: 1.40398 ms (enqueue 0.428772 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26794 ms - Host latency: 1.40239 ms (enqueue 0.433105 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27017 ms - Host latency: 1.40547 ms (enqueue 0.426733 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26852 ms - Host latency: 1.40311 ms (enqueue 0.424744 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26783 ms - Host latency: 1.40286 ms (enqueue 0.424854 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26937 ms - Host latency: 1.40406 ms (enqueue 0.424194 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2686 ms - Host latency: 1.40317 ms (enqueue 0.424707 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26914 ms - Host latency: 1.40353 ms (enqueue 0.42677 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27 ms - Host latency: 1.40425 ms (enqueue 0.430493 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26873 ms - Host latency: 1.40348 ms (enqueue 0.423572 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26832 ms - Host latency: 1.40291 ms (enqueue 0.430017 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26967 ms - Host latency: 1.40535 ms (enqueue 0.4302 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26843 ms - Host latency: 1.40358 ms (enqueue 0.429114 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26923 ms - Host latency: 1.40391 ms (enqueue 0.42439 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26833 ms - Host latency: 1.40344 ms (enqueue 0.426392 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26824 ms - Host latency: 1.40332 ms (enqueue 0.425598 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26815 ms - Host latency: 1.40249 ms (enqueue 0.425122 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26854 ms - Host latency: 1.40304 ms (enqueue 0.426208 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26945 ms - Host latency: 1.4047 ms (enqueue 0.42627 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2678 ms - Host latency: 1.40231 ms (enqueue 0.42793 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26842 ms - Host latency: 1.4037 ms (enqueue 0.42981 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26865 ms - Host latency: 1.40288 ms (enqueue 0.42699 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26852 ms - Host latency: 1.4027 ms (enqueue 0.423767 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26874 ms - Host latency: 1.40281 ms (enqueue 0.425281 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27006 ms - Host latency: 1.40555 ms (enqueue 0.422083 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26842 ms - Host latency: 1.40408 ms (enqueue 0.426379 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27252 ms - Host latency: 1.40726 ms (enqueue 0.428235 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27346 ms - Host latency: 1.41195 ms (enqueue 0.427307 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27407 ms - Host latency: 1.409 ms (enqueue 0.422778 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27891 ms - Host latency: 1.41423 ms (enqueue 0.429431 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28136 ms - Host latency: 1.41648 ms (enqueue 0.426672 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28049 ms - Host latency: 1.41559 ms (enqueue 0.426404 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27958 ms - Host latency: 1.41566 ms (enqueue 0.427246 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28018 ms - Host latency: 1.41483 ms (enqueue 0.422339 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28163 ms - Host latency: 1.41595 ms (enqueue 0.422314 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27949 ms - Host latency: 1.41409 ms (enqueue 0.427466 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28086 ms - Host latency: 1.41587 ms (enqueue 0.425879 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28134 ms - Host latency: 1.41611 ms (enqueue 0.425842 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27909 ms - Host latency: 1.41344 ms (enqueue 0.425427 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28082 ms - Host latency: 1.41621 ms (enqueue 0.422766 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2811 ms - Host latency: 1.41587 ms (enqueue 0.425806 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28031 ms - Host latency: 1.41549 ms (enqueue 0.427661 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28184 ms - Host latency: 1.41726 ms (enqueue 0.431909 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28102 ms - Host latency: 1.4161 ms (enqueue 0.429102 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28033 ms - Host latency: 1.41537 ms (enqueue 0.422766 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28091 ms - Host latency: 1.41588 ms (enqueue 0.427722 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28094 ms - Host latency: 1.41542 ms (enqueue 0.42688 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28099 ms - Host latency: 1.41531 ms (enqueue 0.423132 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28093 ms - Host latency: 1.41519 ms (enqueue 0.425977 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28083 ms - Host latency: 1.4155 ms (enqueue 0.4229 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2804 ms - Host latency: 1.41523 ms (enqueue 0.422974 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28066 ms - Host latency: 1.41631 ms (enqueue 0.430957 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28184 ms - Host latency: 1.41675 ms (enqueue 0.425146 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28083 ms - Host latency: 1.41589 ms (enqueue 0.426099 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28066 ms - Host latency: 1.41599 ms (enqueue 0.431079 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28101 ms - Host latency: 1.41587 ms (enqueue 0.429224 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27957 ms - Host latency: 1.4144 ms (enqueue 0.428198 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27888 ms - Host latency: 1.41338 ms (enqueue 0.423608 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27922 ms - Host latency: 1.41406 ms (enqueue 0.422974 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27944 ms - Host latency: 1.41389 ms (enqueue 0.427563 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27903 ms - Host latency: 1.41396 ms (enqueue 0.423853 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27852 ms - Host latency: 1.41426 ms (enqueue 0.430859 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27905 ms - Host latency: 1.41345 ms (enqueue 0.424097 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27849 ms - Host latency: 1.41433 ms (enqueue 0.425879 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27883 ms - Host latency: 1.41372 ms (enqueue 0.427954 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2782 ms - Host latency: 1.41267 ms (enqueue 0.42373 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27756 ms - Host latency: 1.41284 ms (enqueue 0.433203 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27263 ms - Host latency: 1.40718 ms (enqueue 0.424097 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27324 ms - Host latency: 1.40859 ms (enqueue 0.427271 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27253 ms - Host latency: 1.40649 ms (enqueue 0.428833 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27139 ms - Host latency: 1.40615 ms (enqueue 0.422705 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27266 ms - Host latency: 1.40732 ms (enqueue 0.424902 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27324 ms - Host latency: 1.40803 ms (enqueue 0.423511 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27207 ms - Host latency: 1.40637 ms (enqueue 0.429663 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27266 ms - Host latency: 1.40747 ms (enqueue 0.424097 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27168 ms - Host latency: 1.40547 ms (enqueue 0.425635 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27405 ms - Host latency: 1.40876 ms (enqueue 0.42373 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27512 ms - Host latency: 1.41062 ms (enqueue 0.429053 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27468 ms - Host latency: 1.4092 ms (enqueue 0.430835 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2739 ms - Host latency: 1.40903 ms (enqueue 0.424951 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2739 ms - Host latency: 1.40847 ms (enqueue 0.427173 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27434 ms - Host latency: 1.40891 ms (enqueue 0.423291 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27344 ms - Host latency: 1.40796 ms (enqueue 0.425098 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27383 ms - Host latency: 1.4084 ms (enqueue 0.424048 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2729 ms - Host latency: 1.41033 ms (enqueue 0.42229 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27427 ms - Host latency: 1.40928 ms (enqueue 0.426904 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27449 ms - Host latency: 1.40891 ms (enqueue 0.425684 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26965 ms - Host latency: 1.40425 ms (enqueue 0.428369 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26938 ms - Host latency: 1.40393 ms (enqueue 0.432471 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26917 ms - Host latency: 1.40364 ms (enqueue 0.426904 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27004 ms - Host latency: 1.40457 ms (enqueue 0.425049 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26958 ms - Host latency: 1.40378 ms (enqueue 0.422974 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27065 ms - Host latency: 1.40593 ms (enqueue 0.430371 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27114 ms - Host latency: 1.40652 ms (enqueue 0.430347 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2709 ms - Host latency: 1.40605 ms (enqueue 0.426294 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27043 ms - Host latency: 1.40508 ms (enqueue 0.422925 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27007 ms - Host latency: 1.40425 ms (enqueue 0.432129 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27002 ms - Host latency: 1.4042 ms (enqueue 0.425562 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27029 ms - Host latency: 1.40554 ms (enqueue 0.4271 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26895 ms - Host latency: 1.40371 ms (enqueue 0.425293 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26836 ms - Host latency: 1.40291 ms (enqueue 0.428491 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26892 ms - Host latency: 1.40352 ms (enqueue 0.422705 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2686 ms - Host latency: 1.40396 ms (enqueue 0.428735 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26868 ms - Host latency: 1.40315 ms (enqueue 0.429004 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26982 ms - Host latency: 1.40486 ms (enqueue 0.426025 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.26863 ms - Host latency: 1.40317 ms (enqueue 0.427368 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27397 ms - Host latency: 1.40872 ms (enqueue 0.423218 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27432 ms - Host latency: 1.40852 ms (enqueue 0.424438 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27549 ms - Host latency: 1.41045 ms (enqueue 0.422485 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2759 ms - Host latency: 1.41064 ms (enqueue 0.427417 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27412 ms - Host latency: 1.40872 ms (enqueue 0.423608 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27524 ms - Host latency: 1.41008 ms (enqueue 0.428491 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27517 ms - Host latency: 1.41028 ms (enqueue 0.425439 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27605 ms - Host latency: 1.41128 ms (enqueue 0.422021 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27605 ms - Host latency: 1.41082 ms (enqueue 0.429395 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27764 ms - Host latency: 1.41299 ms (enqueue 0.429199 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28047 ms - Host latency: 1.41519 ms (enqueue 0.428882 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28059 ms - Host latency: 1.41536 ms (enqueue 0.424146 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28059 ms - Host latency: 1.41582 ms (enqueue 0.423242 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28081 ms - Host latency: 1.41577 ms (enqueue 0.425269 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28015 ms - Host latency: 1.4145 ms (enqueue 0.423755 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28237 ms - Host latency: 1.41724 ms (enqueue 0.423804 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28103 ms - Host latency: 1.4156 ms (enqueue 0.430078 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27949 ms - Host latency: 1.41448 ms (enqueue 0.424731 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27974 ms - Host latency: 1.41475 ms (enqueue 0.433569 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28052 ms - Host latency: 1.41519 ms (enqueue 0.42478 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28032 ms - Host latency: 1.41533 ms (enqueue 0.427905 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28069 ms - Host latency: 1.41528 ms (enqueue 0.432739 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.2814 ms - Host latency: 1.41646 ms (enqueue 0.422705 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28027 ms - Host latency: 1.41453 ms (enqueue 0.42688 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28125 ms - Host latency: 1.41599 ms (enqueue 0.425269 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28079 ms - Host latency: 1.41567 ms (enqueue 0.425757 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27993 ms - Host latency: 1.41487 ms (enqueue 0.432227 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28193 ms - Host latency: 1.41738 ms (enqueue 0.42749 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27998 ms - Host latency: 1.41514 ms (enqueue 0.429932 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28152 ms - Host latency: 1.41592 ms (enqueue 0.428125 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.28108 ms - Host latency: 1.41628 ms (enqueue 0.424243 ms)\n",
      "[08/07/2024-17:25:59] [I] Average on 10 runs - GPU latency: 1.27981 ms - Host latency: 1.41416 ms (enqueue 0.424292 ms)\n",
      "[08/07/2024-17:25:59] [I] \n",
      "[08/07/2024-17:25:59] [I] === Performance summary ===\n",
      "[08/07/2024-17:25:59] [I] Throughput: 782.693 qps\n",
      "[08/07/2024-17:25:59] [I] Latency: min = 1.39575 ms, max = 1.47571 ms, mean = 1.41066 ms, median = 1.40918 ms, percentile(90%) = 1.41748 ms, percentile(95%) = 1.41968 ms, percentile(99%) = 1.45068 ms\n",
      "[08/07/2024-17:25:59] [I] Enqueue Time: min = 0.417236 ms, max = 0.487579 ms, mean = 0.426978 ms, median = 0.424316 ms, percentile(90%) = 0.43689 ms, percentile(95%) = 0.443848 ms, percentile(99%) = 0.463135 ms\n",
      "[08/07/2024-17:25:59] [I] H2D Latency: min = 0.0466309 ms, max = 0.09198 ms, mean = 0.0504571 ms, median = 0.0498657 ms, percentile(90%) = 0.0531006 ms, percentile(95%) = 0.0537109 ms, percentile(99%) = 0.0549316 ms\n",
      "[08/07/2024-17:25:59] [I] GPU Compute Time: min = 1.26343 ms, max = 1.323 ms, mean = 1.27494 ms, median = 1.2738 ms, percentile(90%) = 1.28204 ms, percentile(95%) = 1.28412 ms, percentile(99%) = 1.31177 ms\n",
      "[08/07/2024-17:25:59] [I] D2H Latency: min = 0.0837402 ms, max = 0.0898438 ms, mean = 0.0852642 ms, median = 0.0852051 ms, percentile(90%) = 0.0863037 ms, percentile(95%) = 0.0866699 ms, percentile(99%) = 0.0871582 ms\n",
      "[08/07/2024-17:25:59] [I] Total Host Walltime: 3.00373 s\n",
      "[08/07/2024-17:25:59] [I] Total GPU Compute Time: 2.99738 s\n",
      "[08/07/2024-17:25:59] [I] Explanations of the performance metrics are printed in the verbose logs.\n",
      "[08/07/2024-17:25:59] [I] \n",
      "&&&& PASSED TensorRT.trtexec [TensorRT v8601] # trtexec --fp16 --onnx=../onnx_models/nrx_rt.onnx --saveEngine=../onnx_models/nrx_rt.plan --minShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --optShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --maxShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4\n",
      "Only GPU number 0 used.\n",
      "Loading: nrx_rt.cfg.\n",
      "Evaluating the first 2 port sets.\n",
      "Loading DoubleTDL with low correlation.\n",
      "Loading pre-trained weights.\n",
      "---Converting ONNX model---\n",
      "---Saving ONNX model---\n",
      "trtexec --fp16 --onnx=../onnx_models/nrx_rt.onnx --saveEngine=../onnx_models/nrx_rt.plan  --minShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --optShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4 --maxShapes=rx_slot_real:1x1584x14x4,rx_slot_imag:1x1584x14x4,h_hat_real:1x1584x2x4,h_hat_imag:1x1584x2x4\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "# We recommend to run this command in the terminal outside of a jupyter notebook.\n",
    "#os.environ[\"PATH\"] += os.pathsep + \"/usr/src/tensorrt/bin/\"   # If this notebook's path variable doesn't include tensorrt\n",
    "os.system(f\"python ../scripts/export_onnx.py -config_name nrx_rt.cfg -num_tx 2 -gpu {gpu_num}\");\n",
    "# the gpu index is only required for multi-gpu systems\n",
    "\n",
    "# num_tx defines the maximum number of active DMRS ports; less users can be dynamically scheduled by setting the dmrs_port_mask during inference"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aa021dd2",
   "metadata": {},
   "source": [
    "Remark: if the above code throws an error related to \"Could not infer attribute `explicit_paddings` type from empty iterator\", please ensure that the `onnx==1.14. There is a known bug in version 1.15 of ONNX causing this issue.\n",
    "\n",
    "The latency numbers above are for an NVIDIA RTX 3090 GPU."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c4735ba2",
   "metadata": {},
   "source": [
    "The compilation script directly provides detailed latency measurements for the given architecture and system hardware.\n",
    "\n",
    "The reported numbers are obviously platform-specific and can be used to optimize the architecture such that real-time inference is possible.\n",
    "Further profiling could be done using the [NVIDIA Nsight Systems](https://developer.nvidia.com/nsight-systems)."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cd635768",
   "metadata": {},
   "source": [
    "## Running the TensorRT Engine"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7b569ad2",
   "metadata": {},
   "source": [
    "We can now run the TRT engine in standalone mode without the need for TensorFlow or Sionna. \n",
    "\n",
    "Alternatively, the [ONNX](https://onnx.ai/) model description could be used\n",
    "within any other neural network acceleration framework for deployment.\n",
    "\n",
    "Remark: the TRT engine is platform specific (hardware and software), i.e., must be rebuilt for the target platform.\n",
    "Thus, we do not provide any pre-compiled engines in this repository."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "1d939e8f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[W] 'colored' module is not installed, will not use colors when logging. To enable colors, please install the 'colored' module: python3 -m pip install colored\n",
      "[I] Loading bytes from ../onnx_models/nrx_rt.plan\n"
     ]
    }
   ],
   "source": [
    "# load pre-compiled TensorRT engine\n",
    "model_name = \"nrx_rt\" # equals label in config file\n",
    "engine = engine_from_bytes(bytes_from_path(f\"../onnx_models/{model_name}.plan\"))\n",
    "nrx_trt = TrtRunner(engine) # load the TRT engine\n",
    "nrx_trt.activate() # and activate TRT engine"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "beb2d9f7",
   "metadata": {},
   "source": [
    "### Input/Output NRX Dimensions\n",
    "\n",
    "The input/output dimensions are aligned with the [NVIDIA Aerial platform](https://developer.nvidia.com/6g-program) and, thus, differ from the definitions in [NVIDIA Sionna](https://nvlabs.github.io/sionna/api/nr.html#puschreceiver).\n",
    "\n",
    "Remark: Due to the limited support for complex-valued numbers in TensorRT, we define the inputs by their real and imaginary component.\n",
    "The detailed inputs are as follows:\n",
    "\n",
    "     rx_slot_real : [batch_size, num_subcarriers, num_ofdm_symbols, num_rx_ant], tf.float32\n",
    "        Real part of the received OFDM resource grid after cyclic prefix\n",
    "        removal and FFT.\n",
    "\n",
    "    rx_slot_imag : [batch_size, num_subcarriers, num_ofdm_symbols, num_rx_ant], tf.float32\n",
    "        Imaginary part of the received OFDM resource grid after cyclic prefix\n",
    "        removal and FFT.\n",
    "\n",
    "    h_hat_real : [batch_size, num_pilots, num_tx, num_rx_ant], tf.float32\n",
    "        Real part of the LS channel estimates at pilot positions.\n",
    "\n",
    "    h_hat_imag : [batch_size, num_pilots, num_tx, num_rx_ant], tf.float32\n",
    "        Imaginary part of the LS channel estimates at pilot positions.\n",
    "\n",
    "    dmrs_port_mask: [batch_size, max_num_tx], tf.float32\n",
    "        Mask of 0s and 1s to indicate that DMRS ports are active or not.\n",
    "\n",
    "    dmrs_ofdm_pos: [num_tx, num_dmrs_symbols], tf.int\n",
    "        DMRS symbol positions within slot.\n",
    "\n",
    "    dmrs_subcarrier_pos: [num_tx, num_pilots_per_prb/dmrs_ofdm_pos], tf.int\n",
    "        Pilot positions per PRB.\n",
    "\n",
    "And the output shapes are given as:\n",
    "\n",
    "    llr : [batch_size, num_bits_per_symbol, num_tx, num_effective_subcarriers, num_ofdm_symbols], tf.float\n",
    "        LLRs on bits. Remark: contrary to Sionna, we use the definition \n",
    "        LLR(X) = log[ P(X=0) / P(X=1)].\n",
    "\n",
    "    h_hat : [batch_size, num_tx, num_effective_subcarriers, num_ofdm_symbols, 2*num_rx_ant], tf.float\n",
    "        Refined channel estimates."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "15670951",
   "metadata": {},
   "source": [
    "The batch_size is fixed to 1 during the export.\n",
    "\n",
    "Remark: TensorRT supports dynamic input shapes during execution. However,\n",
    "this introduces a small kernel launch overhead whenever the input shapes are changing (e.g., the number of allocated PRBs).\n",
    "Thus, we currently use a fixed input dimension and consider zero-padding of the non-active PRBs.\n",
    "\n",
    "To support dynamic shapes, one needs to provide the desired range of input parameters at the end of the\n",
    "`scripts/export_onnx.py` script."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9b1f1657",
   "metadata": {},
   "source": [
    "Note that this implementation requires LS channel estimates at the DMRS pilot positions. \n",
    "The main reason for this is that the DMRS pilots depend on various parameters such as the `slot-id` and the `RNTI` which dynamically change.\n",
    "Thus, integrating an 5G NR LS channel estimator in the TensorRT engine would require a substantially amount of code just for the dynamic DMRS generation.\n",
    "For this reason, we have decoupled the LS channel estimation from the TensorRT engine and we refer to the [NVIDIA Aerial platform](https://developer.nvidia.com/aerial) which allows for standard compliant dynamic pilot generation.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3b249823",
   "metadata": {},
   "source": [
    "## Inference Using TensorRT\n",
    "\n",
    "We now verify the BLER of the system. As mentioned earlier, the input/output shapes of the TensorRT engine are not directly Sionna compatible."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "7a83f0c6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "----- Inputs -----\n",
      "rx_slot shape:  (1, 1584, 14, 4)\n",
      "h_hat shape:  (1, 1584, 2, 4)\n",
      "active_dmrs_ports shape:  (1, 2)\n",
      "dmrs_ofdm_pos shape:  (2, 2)\n",
      "dmrs_subcarrier_pos shape:  (2, 6)\n",
      "\n",
      "----- Outputs -----\n",
      "LLR shape:  (1, 4, 2, 1584, 14)\n",
      "h_hat_refined shape:  (1, 2, 1584, 14, 8)\n"
     ]
    }
   ],
   "source": [
    "# system parameters as defined in the config\n",
    "batch_size  = 1 # fixed to one to avoid buffering latency\n",
    "num_rx_ant = 4\n",
    "num_ofdm_symbols = 14 # entire grid (nrx slices 12 symbols)\n",
    "num_tx = 2 # transmitters can be dynamically deactivated via the dmrs_port_mask\n",
    "\n",
    "num_prbs = 132 # number of occupied PRBs\n",
    "num_pilots = num_prbs * 6 * 2 # 2 DMRS symbols each with 6 occupied REs\n",
    "\n",
    "# input specs for trt engine\n",
    "spec = {\n",
    "    'rx_slot_real': np.float32,\n",
    "    'rx_slot_imag': np.float32,\n",
    "    'h_hat_real': np.float32,\n",
    "    'h_hat_imag': np.float32,\n",
    "    'active_dmrs_ports': np.float32,\n",
    "    'dmrs_ofdm_pos': np.int32,\n",
    "    'dmrs_subcarrier_pos': np.int32}\n",
    "\n",
    "################################\n",
    "### Dummy input data for testing\n",
    "################################\n",
    "\n",
    "# received IQ samples\n",
    "rx_slot = np.ones([batch_size, num_prbs*12, num_ofdm_symbols, num_rx_ant],\n",
    "                   np.complex64)\n",
    "\n",
    "# DMRS pilots; only pilots with energy are considered, i.e.,\n",
    "# masked pilots are ignored\n",
    "h_hat = np.ones([batch_size, num_pilots, num_tx, num_rx_ant], np.complex64)\n",
    "\n",
    "# activate all DMRS ports\n",
    "active_dmrs_ports = np.ones([batch_size, num_tx], np.float32)\n",
    "\n",
    "dmrs_ofdm_pos = np.array([[2, 11],[2, 11]]) # which OFDM symbols are used for DMRS\n",
    "dmrs_subcarrier_pos = np.array([[0,2,4,6,8,10],[0,2,4,6,8,10]]) #subcarriers used for pilots (per PRB)\n",
    "\n",
    "print(\"----- Inputs -----\")\n",
    "print(\"rx_slot shape: \", rx_slot.shape)\n",
    "print(\"h_hat shape: \", h_hat.shape)\n",
    "print(\"active_dmrs_ports shape: \", active_dmrs_ports.shape)\n",
    "print(\"dmrs_ofdm_pos shape: \", dmrs_ofdm_pos.shape)\n",
    "print(\"dmrs_subcarrier_pos shape: \", dmrs_subcarrier_pos.shape)\n",
    "\n",
    "# prepare input\n",
    "feed_dict = {\n",
    "    'rx_slot_real': rx_slot.real,\n",
    "    'rx_slot_imag': rx_slot.imag,\n",
    "    'h_hat_real': h_hat.real,\n",
    "    'h_hat_imag': h_hat.imag,\n",
    "    'active_dmrs_ports': active_dmrs_ports,\n",
    "    'dmrs_ofdm_pos': dmrs_ofdm_pos,\n",
    "    'dmrs_subcarrier_pos': dmrs_subcarrier_pos}\n",
    "\n",
    "# and run TensorRT engine\n",
    "res = nrx_trt.infer({k: feed_dict[k].astype(dt) for k, dt in spec.items()})\n",
    "\n",
    "llr = res['output_1']\n",
    "h_hat_refined = res['output_2']\n",
    "\n",
    "print(\"\\n----- Outputs -----\")\n",
    "print(\"LLR shape: \", llr.shape)\n",
    "print(\"h_hat_refined shape: \", h_hat_refined.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4e16c4c5",
   "metadata": {},
   "source": [
    "## BLER Performance Validation\n",
    "\n",
    "We use the utility `DataGeneratorAerial` and `DataEvaluator` from `utils/onnx_utils.py` to generate compatible inputs for the NRX and to simulate the BLER of the generated TensorRT engine."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "ab964478",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating the first 2 port sets.\n",
      "Loading DoubleTDL with low correlation.\n"
     ]
    }
   ],
   "source": [
    "# initialize system parameters for given config\n",
    "config_name = \"nrx_rt.cfg\"\n",
    "sys_parameters = Parameters(config_name, system=\"nrx\", training=False)\n",
    "\n",
    "# allows to generate test vectors for the NRX via Sionna\n",
    "data_generator = DataGeneratorAerial(sys_parameters, training=False)\n",
    "\n",
    "# evaluates LLRs from NRX by applying FEC and BLER calculations\n",
    "data_evaluator = DataEvaluator(sys_parameters)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a15b9f4d",
   "metadata": {},
   "source": [
    "The DataGenerator generates input data (i.e., slots after transmission over the channel model) for the given system parameters.\n",
    "The DataEvaluator takes the output of the neural receiver and directly evaluates the BLER (i.e., demaps the resource grid of LLRs and applies FEC decoding)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "01929fb9",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Progress 90%; BER=0.0000,TBLER=0.0000\n",
      "Final Results: BER=0.0000, TBLER=0.0000\n"
     ]
    }
   ],
   "source": [
    "num_iter = 10 # batches to process\n",
    "# We have configured TRT to expect a batch size of 1\n",
    "batch_size = 10 # this is only for DataGeneration\n",
    "ebno_db = 6.\n",
    "num_tx = 2 # number of active transmitters\n",
    "\n",
    "# for BER statistic\n",
    "num_bit_errors = 0\n",
    "num_block_errors = 0\n",
    "num_bits = 0\n",
    "num_blocks = 0\n",
    "\n",
    "for i in range(num_iter):\n",
    "\n",
    "    # this loop runs for batch_size = 1; note that one could draw larger\n",
    "    # batches and uses an inner loop with batch_size=1 for the TRT engine.\n",
    "    nrx_inputs, c, b, _ = data_generator(batch_size, ebno_db, num_tx)\n",
    "\n",
    "    llr = []\n",
    "    for j in range(batch_size):\n",
    "\n",
    "        #and run the TRT engine\n",
    "        feed_dict = {\n",
    "            'rx_slot_real': nrx_inputs[0][j:j+1],\n",
    "            'rx_slot_imag': nrx_inputs[1][j:j+1],\n",
    "            'h_hat_real': nrx_inputs[2][j:j+1],\n",
    "            'h_hat_imag': nrx_inputs[3][j:j+1],\n",
    "            'active_dmrs_ports': nrx_inputs[4][j:j+1],\n",
    "            'dmrs_ofdm_pos': nrx_inputs[5],\n",
    "            'dmrs_subcarrier_pos': nrx_inputs[6],}\n",
    "\n",
    "        # and run TensorRT engine\n",
    "        res = nrx_trt.infer({k: feed_dict[k].astype(dt) for k, dt in spec.items()})\n",
    "\n",
    "        llr.append(np.copy(res['output_1']))\n",
    "    #h_hat_refined = res['output_2']\n",
    "\n",
    "    llr = np.concatenate(llr, axis=0)\n",
    "\n",
    "    _, ber_uncoded, b_hat = data_evaluator(llr, c)\n",
    "\n",
    "    # and track BER/BLER\n",
    "    num_block_errors += sn.utils.count_block_errors(b,b_hat).numpy()\n",
    "    num_bit_errors += sn.utils.count_errors(b, b_hat).numpy()\n",
    "    num_blocks += np.prod(b.shape[:-1])\n",
    "    num_bits += np.prod(b.shape)\n",
    "\n",
    "    ber = num_bit_errors/num_bits\n",
    "    bler = num_block_errors / num_blocks\n",
    "\n",
    "    print(f\"Progress {int(i/num_iter*100)}%; BER={ber:.4f},TBLER={bler:.4f}\",\n",
    "          end=\"\\r\")\n",
    "print(f\"\\nFinal Results: BER={ber:.4f}, TBLER={bler:.4f}\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9e539e03",
   "metadata": {},
   "source": [
    "Remark: most of the compute time in the above example is spend in the `data_generator` and `data_evaluator`.\n",
    "Throughput benchmarks are not part of this example."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "90dadb0d",
   "metadata": {},
   "source": [
    "## Outlook\n",
    "\n",
    "The TensorRT engine is now ready for deployment using the [NVIDIA Aerial platform](https://developer.nvidia.com/aerial).\n",
    "For further details including code examples of Aerial RAN, please consider joining the [NVIDIA 6G Developer Program](https://developer.nvidia.com/6g-program)."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.12"
  },
  "vscode": {
   "interpreter": {
    "hash": "916dbcbb3f70747c44a77c7bcd40155683ae19c65e1c03b4aa3499c5328201f1"
   }
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {},
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}