Single Instruction Multiple Thread Research Articles

A GPU-Based, Real-Time Dealiasing Framework for High-Frame-Rate Vector Doppler Imaging.

Vector Doppler is well regarded as a potential way of deriving flow vectors to intuitively visualize complex flow profiles, especially when it is implemented at high frame rates. However, this technique's performance is known to suffer from aliasing artifacts. There is a dire need to devise real-time dealiasing solutions for vector Doppler. In this paper, we present a new methodological framework for achieving aliasing-resistant flow vector estimation at real-time throughput from precalculated Doppler frequencies. Our framework comprises a series of compute kernels that have synergized: 1) an extended least squares vector Doppler (ELS-VD) algorithm, 2) single-instruction, multiple-thread (SIMT) processing principles, and 3) implementation on a graphical processing unit (GPU). Results show that this new framework, when executed on an RTX-2080 GPU, can effectively generate aliasing-free flow vector maps using high-frame-rate imaging datasets acquired from multiple transmit-receive angle pairs in a carotid phantom imaging scenario. Over the entire cardiac cycle, the frame processing time for aliasing-resistant vector estimation was measured to be less than 16 ms, which corresponds to a minimum processing throughput of 62.5 fps. In a human femoral bifurcation imaging trial with fast flow (150 cm/s), our framework was found to be effective in resolving two-cycle aliasing artifacts at a minimum throughput of 53 fps. The framework's processing throughput was generally in the real-time range for practical combinations of ELS-VD algorithmic parameters. Overall, this work represents the first demonstration of real-time, GPU-based aliasing-resistant vector flow imaging using vector Doppler estimation principles.

The FastLanes Compression Layout: Decoding > 100 Billion Integers per Second with Scalar Code

The open-source FastLanes project aims to improve big data formats, such as Parquet, ORC and columnar database formats, in multiple ways. In this paper, we significantly accelerate decoding of all common Light-Weight Compression (LWC) schemes: DICT, FOR, DELTA and RLE through better data-parallelism. We do so by re-designing the compression layout using two main ideas: (i) generalizing the value interleaving technique in the basic operation of bit-(un)packing by targeting a virtual 1024-bits SIMD register, (ii) reordering the tuples in all columns of a table in the same Unified Transposed Layout that puts tuple chunks in a common "04261537" order (explained in the paper); allowing for maximum independent work for all possible basic SIMD lane widths: 8, 16, 32, and 64 bits. We address the software development, maintenance and future-proofness challenges of increasing hardware diversity, by defining a virtual 1024-bits instruction set that consists of simple operators supported by all SIMD dialects; and also, importantly, by scalar code. The interleaved and tuple-reordered layout actually makes scalar decoding faster, extracting more data-parallelism from today's wide-issue CPUs. Importantly, the scalar version can be fully auto-vectorized by modern compilers, eliminating technical debt in software caused by platform-specific SIMD intrinsics. Micro-benchmarks on Intel, AMD, Apple and AWS CPUs show that FastLanes accelerates decoding by factors (decoding >40 values per CPU cycle). FastLanes can make queries faster, as compressing the data reduces bandwidth needs, while decoding is almost free.

Offloading electromagnetic shower transport to GPUs

Making general particle transport simulation for high-energy physics (HEP) single-instruction-multiple-thread (SIMT) friendly, to take advantage of accelerator hardware, is an important alternative for boosting the throughput of simulation applications. To date, this challenge is not yet resolved, due to difficulties in mapping the complexity of Geant4 components and workflow to the massive parallelism features exposed by graphics processing units (GPU). The AdePT project is one of the R&D initiatives tackling this limitation and exploring GPUs as potential accelerators for offloading some part of the CPU simulation workload. Our main target is to implement a complete electromagnetic shower demonstrator working on the GPU. The project is the first to create a full prototype of a realistic electron, positron, and gamma electromagnetic shower simulation on GPU, implemented as either a standalone application or as an extension of the standard Geant4 CPU workflow. Our prototype currently provides a platform to explore many optimisations and different approaches. We present the most recent results and initial conclusions of our work, using both a standalone GPU performance analysis and a first implementation of a hybrid workflow based on Geant4 on the CPU and AdePT on the GPU.

Interruptible Nodes: Reducing Queueing Costs in Irregular Streaming Dataflow Applications on Wide-SIMD Architectures

Streaming dataflow applications are an attractive target to parallelize on wide-SIMD processors such as GPUs. These applications can be expressed as a pipeline of compute nodes connected by edges, which feed outputs from one node to the next. Streaming applications often exhibit irregular dataflow, where the amount of output produced for one input is unknown a priori. Inserting finite queues between pipeline nodes can ameliorate the impact of irregularity and improve SIMD lane occupancy. The sizing of these queues is driven by both performance and safety considerations- relative queue sizes should be chosen to reduce runtime overhead and maximize throughput, but each node’s output queue must be large enough to accommodate the maximum number of outputs produced by one SIMD vector of inputs to the node. When safety and performance considerations conflict, the application may incur excessive memory usage and runtime overhead. In this work, we identify properties of applications that lead to such undesirable behaviors, with examples from applications implemented in our MERCATOR framework for irregular streaming on GPUs. To address these issues, we propose extensions to support interruptible nodes that can be suspended mid-execution if their output queues fill. We illustrate the impacts of adding interruptible nodes to the MERCATOR framework on representative irregular streaming applications from the domains of branching search and bioinformatics.

Adaptive Contention Management for Fine-Grained Synchronization on Commodity GPUs

As more emerging applications are moving to GPUs, fine-grained synchronization has become imperative. However, their performance can be severely impaired in case of frequent synchronization failures caused by high data contention. Differently from CPUs, GPUs own thousands of hardware threads and adopt single instruction multiple threads paradigm, making it impractical to deploy the CPU contention management mechanisms directly on GPUs. In this article, we design a Software Warp Controlling Framework (SWCF), which employs producer-consumer execution model and leverages GPU hardware barriers to dynamically control the execution of warps at runtime. On the basis of SWCF, we propose a contention management strategy to decrease frequent synchronization failures while avoiding the over-reducing of parallelism. We evaluate SWCF and the proposed strategy on commodity GPUs using a set of applications with fine-grained synchronization. The results show that on V100 GPU our contention management achieves a 4.7X speedup and outperforms the conventional GPU software backoff solution by 42% on average.

A GPGPU microarchitecture supports multi-path execution and branch compaction

GPGPU supports high-performance execution due to its powerful parallel computing capability. However, when meets branch divergent, only a subset of its SIMD lanes could issue and execute thread, which leads to the degradation of performance. Current branch compaction mechanisms such as thread block compaction (TBC) and SIMD lane permutation (SLP) can effectively alleviate this phenomenon, but none of these mechanisms support interleave execution of multiple control flows generated by branch divergent, which limits the improvement of thread-level parallelism (TLP) and performance.In this paper we propose a new microarchitecture that supports multi-path interleave execution and branch compaction at the same time. Our microarchitecture records the information of both two control flows into the same entry. Then two execution masks are used to provide executable information of interleaved control flows. We use a set of workloads with massive branch divergent to evaluate our microarchitecture. The result demonstrates that our design outperforms baseline, TBC and SLP, with a harmonic mean of 4.7%, 3.4% and 2.3%.

GPUTreeShap: massively parallel exact calculation of SHAP scores for tree ensembles

SHapley Additive exPlanation (SHAP) values (Lundberg & Lee, 2017) provide a game theoretic interpretation of the predictions of machine learning models based on Shapley values (Shapley, 1953). While exact calculation of SHAP values is computationally intractable in general, a recursive polynomial-time algorithm called TreeShap (Lundberg et al., 2020) is available for decision tree models. However, despite its polynomial time complexity, TreeShap can become a significant bottleneck in practical machine learning pipelines when applied to large decision tree ensembles. Unfortunately, the complicated TreeShap algorithm is difficult to map to hardware accelerators such as GPUs. In this work, we present GPUTreeShap, a reformulated TreeShap algorithm suitable for massively parallel computation on graphics processing units. Our approach first preprocesses each decision tree to isolate variable sized sub-problems from the original recursive algorithm, then solves a bin packing problem, and finally maps sub-problems to single-instruction, multiple-thread (SIMT) tasks for parallel execution with specialised hardware instructions. With a single NVIDIA Tesla V100-32 GPU, we achieve speedups of up to 19× for SHAP values, and speedups of up to 340× for SHAP interaction values, over a state-of-the-art multi-core CPU implementation executed on two 20-core Xeon E5-2698 v4 2.2 GHz CPUs. We also experiment with multi-GPU computing using eight V100 GPUs, demonstrating throughput of 1.2 M rows per second—equivalent CPU-based performance is estimated to require 6850 CPU cores.

GSoFa: Scalable Sparse Symbolic LU Factorization on GPUs

Decomposing a matrix <inline-formula><tex-math notation="LaTeX">$\mathbf {A}$</tex-math></inline-formula> into a lower matrix <inline-formula><tex-math notation="LaTeX">$\mathbf {L}$</tex-math></inline-formula> and an upper matrix <inline-formula><tex-math notation="LaTeX">$\mathbf {U}$</tex-math></inline-formula> , which is also known as LU decomposition, is an essential operation in numerical linear algebra. For a sparse matrix, LU decomposition often introduces more nonzero entries in the <inline-formula><tex-math notation="LaTeX">$\mathbf {L}$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$\mathbf {U}$</tex-math></inline-formula> factors than in the original matrix. A <i>symbolic factorization</i> step is needed to identify the nonzero structures of <inline-formula><tex-math notation="LaTeX">$\mathbf {L}$</tex-math></inline-formula> and <inline-formula><tex-math notation="LaTeX">$\mathbf {U}$</tex-math></inline-formula> matrices. Attracted by the enormous potentials of the Graphics Processing Units (GPUs), an array of efforts have surged to deploy various LU factorization steps except for the symbolic factorization, to the best of our knowledge, on GPUs. This article introduces <small>gSoFa</small> , the first <u>G</u> PU-based <u>s</u> ymb <u>o</u> lic <u>fa</u> ctorization design with the following three optimizations to enable scalable LU symbolic factorization for <i>nonsymmetric pattern</i> sparse matrices on GPUs. First, we introduce a novel fine-grained parallel symbolic factorization algorithm that is well suited for the <i>Single Instruction Multiple Thread</i> (SIMT) architecture of GPUs. Second, we tailor supernode detection into a SIMT friendly process and strive to balance the workload, minimize the communication and saturate the GPU computing resources during supernode detection. Third, we introduce a three-pronged optimization to reduce the excessive space consumption problem faced by multi-source concurrent symbolic factorization. Taken together, <small>gSoFa</small> achieves up to 31× speedup from 1 to 44 Summit nodes (6 to 264 GPUs) and outperforms the state-of-the-art CPU project, on average, by 5×. Notably, <small>gSoFa</small> also achieves up to 47 percent of the peak memory throughput of a V100 GPU in the Summit Supercomputer.

A Survey of Network-Based Hardware Accelerators

Many practical data-processing algorithms fail to execute efficiently on general-purpose CPUs (Central Processing Units) due to the sequential matter of their operations and memory bandwidth limitations. To achieve desired performance levels, reconfigurable (FPGA (Field-Programmable Gate Array)-based) hardware accelerators are frequently explored that permit the processing units’ architectures to be better adapted to the specific problem/algorithm requirements. In particular, network-based data-processing algorithms are very well suited to implementation in reconfigurable hardware because several data-independent operations can easily and naturally be executed in parallel over as many processing blocks as actually required and technically possible. GPUs (Graphics Processing Units) have also demonstrated good results in this area but they tend to use significantly more power than FPGA, which could be a limiting factor in embedded applications. Moreover, GPUs employ a Single Instruction, Multiple Threads (SIMT) execution model and are therefore optimized to SIMD (Single Instruction, Multiple Data) operations, while in FPGAs fully custom datapaths can be built, eliminating much of the control overhead. This review paper aims to analyze, compare, and discuss different approaches to implementing network-based hardware accelerators in FPGA and programmable SoC (Systems-on-Chip). The performed analysis and the derived recommendations would be useful to hardware designers of future network-based hardware accelerators.

Hardware Accelerators for Real-Time Face Recognition: A Survey

Real-time face recognition has been of great interest in the last decade due to its wide and variant critical applications which include biometrics, security in public places, and identification in login systems. This has encouraged researchers to design fast and accurate embedded and portable systems that are capable to detect and recognize a large number of faces at almost video frame rate. Due to the increasing volume of reference faces, traditional general purpose computing engines such as the ones based on Intel&#x2019;s Pentium processors have shown not to be adequate and various dedicated hardware accelerators based on either Graphical Processing Unit (GPU), Field Programmable Gate Arrays (FPGA), Application Specific Integrated Circuit (ASIC), or even multi-core Central Processing Units (CPU) have emerged. Earlier published review papers on face detection/recognition have discussed face detection and face recognition algorithms enhancement that improve the detection accuracy. Nevertheless, none of them has reviewed the hardware accelerators used for this application. Accordingly, this paper aims to provide a comprehensive review of the most recent face recognition algorithms and associated embedded hardware systems targeting real-time performance. A detailed comparison between neural network and non-neural network-based algorithms in terms of accuracy and processing time is provided. Discussions on their suitability to be implemented into parallel hardware architectures such as Single Instruction Multiple Thread (SIMT) or Single Instruction Multiple Data (SIMD) is also discussed.

Computation analysis of regularization methods and parameter selection for acoustics radiation modes source reconstruction of vibrating plates

Source localization and power estimation is a topic of great interest in acoustics and vibration. Acoustic source radiation modes reconstruction is a method of particular interest. Solutions leads to determinate sound/vibration power and surface velocity distribution from sparse acoustics samples. It has been shown that the quality of the results depends on Tikhonov regularization parameter. This inverse method is based on the radiation resistance matrix and we show that a single instruction multiple threads computing approach for graphics processing unit device exhibit better speed performance than common approaches to achieve the solution. We compare four regularization and three estimating methods for regularization parameters. We use a similarity measure to the simulated cases in three frequencies. Tikhonov regularization exhibits best reconstruction results. However, truncated singular vector decomposition also shows good performance with the advantage of not using a regularization parameter. Graphics processing unit implementation reduce reconstruction's computing time at least in a half.

Mesoscopic Modeling and Rapid Simulation of Incremental Changes in Epidemic Scenarios on GPUs: Fast What-If Analyses of Localized and Dynamic Effects.

In simulation-based studies and analyses of epidemics, a major challenge lies in resolving the conflict between fidelity of models and the speed of their simulation. Another related challenge arises in dealing with the large number of what–if scenarios that need to be explored. Here, we describe new computational methods that together provide an approach to dealing with both challenges. A mesoscopic modeling approach is described that strikes a middle ground between macroscopic models based on coupled differential equations and microscopic models built on fine-grained behaviors at the individual entity level. The mesoscopic approach offers the ability to incorporate complex compositions of multiple layers of dynamics even while retaining the potential for aggregate behaviors at varying levels. It also is an excellent match to the accelerator-based architectures of modern computing platforms in which graphical processing units (GPUs) can be exploited for fast simulation via the parallel execution mode of single instruction multiple thread (SIMT). The challenge of simulating a large number of scenarios is addressed via a method of sharing model state and computation across a tree of what–if scenarios that are localized, incremental changes to a large base simulation. A combination of the mesoscopic modeling approach and the incremental what–if scenario tree evaluation has been implemented in the software on modern GPUs. Synthetic simulation scenarios are presented to demonstrate the computational characteristics of our approach. Results from the experiments with large population data, including USA, UK, and India, illustrate the modeling methodology and computational performance on thousands of synthetically generated what–if scenarios. Execution of our implementation scaled to 8192 GPUs of supercomputing platforms demonstrates the ability to rapidly evaluate what–if scenarios several orders of magnitude faster than the conventional methods.

A simple method for rejection sampling efficiency improvement on SIMT architectures

We derive a probability distribution for the possible number of iterations required for a SIMT (single instruction multiple thread) program using rejection sampling to finish creating a sample across all threads. This distribution is found to match a recently proposed distribution from Chakraborty and Gupta (in: Communications in statistics: theory and methods, 2015) that was shown as a good approximation of certain datasets. This work demonstrates an exact application of this distribution. The distribution can be used to evaluate the relative merit of some sampling methods on the GPU without resort to numerical tests. The distribution reduces to the expected geometric distribution in the single thread per warp limit. A simplified formula to approximate the expected number of iterations required to obtain rejection iteration samples is provided. With this new result, a simple, efficient layout for assigning sampling tasks to threads on a GPU is found as a function of the rejection probability without recourse to more complicated rejection sampling methods.

Adaptive Heterogeneous Transient Analysis of Wind Farm Integrated Comprehensive AC/DC Grids

The increasingly complex AC/DC network as a result of the massive integration of wind farms manifests the significance of a comprehensive transient study. In this work, the wind turbine (WT) and the DC grid are modeled in detail for the electromagnetic transient (EMT) simulation to maximize its fidelity, whilst the AC grid transient stability is analyzed by dynamic simulation (DS). An interactive EMT-DS interface is thus introduced to enable their concurrency and subsequently form a co-simulation. The CPU which is dominant in system study faces a tremendous challenge in handling a great number of components albeit they exhibit homogeneity. The many-core graphics processing unit (GPU) featuring massive parallelism is therefore exploited and following the definition of an adaptive computing boundary, a flexible heterogeneous sequential-parallel processing architecture is proposed for efficient analysis of the wind-farm-integrated AC/DC grid. Topological reconfiguration of WTs is specifically carried out to reduce the numerical order whilst enhancing system homogeneity that enables the GPU to thoroughly utilize its peculiar property of single-instruction multiple-thread (SIMT) compute paradigm. Consequently, significant speedups can be attained by the proposed computing framework over pure CPU computation, while its accuracy is validated by the commercial EMT and dynamic security analysis tools PSCAD/EMTDC and DSATools, respectively.

Exploring Applications of STT-RAM in GPU Architectures

Use of modern GPUs has been extended from traditional 3D graphic processing to computing acceleration of many scientific, engineering, and enterprise applications. In modern GPUs, on-chip memory capacity keeps increasing to support thousands of chip-resident threads. For example, a large register file is needed in order to efficiently process highly-parallel threads in single instruction multiple thread (SIMT) fashion, and a large shared memory is often implemented to allow data sharing among the threads on the chip. On-chip memory capacity of GPUs, however, is highly constrained by large memory cell area and high static power consumption of conventional SRAM implementation. In this work, we propose to utilize the emerging multi-level cell (MLC) spin-transfer torque RAM (STT-RAM) technology to implement register file and shared memory in GPUs. Compared to SRAM, MLC STT-RAM (or MLC-STT) has a much smaller cell area as well as ultra-low standby power, thanks to the non-volatility of MLC-STT technology. Hence, the footprint and leakage power of the implemented memory components are substantially reduced. Moreover, in light of asymmetric performance of soft and hard bits of a MLC-STT cell, we propose a dynamic data remapping strategy in register file and shared memory implementations that allows a flexible tradeoff between the memory access time and the available capacity: frequently-accessed data is always mapped to the fast rows built with the soft bits of the MLC-STT cells while the slow rows composed of the hard bits are used only when a larger capacity is critically needed. We also develop a novel rescheduling scheme to minimize the waiting time of the issued warps to access register banks in the register file, which is induced by the long writeback operations through the reordering of the issued warps. Finally, an early termination technology is also applied to save the write energy of the shared memory if the bits of the memory do not flip. Experimental results on benchmarks of ISPASS2009, Rodinia, Parboil, and CUDA show that on average, MLC-STT register file can achieve 3.28% system performance improvement, 9.48% energy reduction, and 38.9% energy efficiency improvement compared to conventional SRAM-based design. Meanwhile, MLC-STT shared memory leads to 3.45% system performance improvement, 49.3% energy reduction, and 116% energy efficiency improvement.

Boosting Parallel Influence-Maximization Kernels for Undirected Networks With Fusing and Vectorization

Influence maximization (IM) is the problem of finding a seed vertex set which is expected to incur the maximum influence spread on a graph. It has various applications in practice such as devising an effective and efficient approach to disseminate information, news or ad within a social network. The problem is shown to be NP-hard and approximation algorithms with provable quality guarantees exist in the literature. However, these algorithms are computationally expensive even for medium-scaled graphs. Furthermore, graph algorithms usually suffer from spatial and temporal irregularities during memory accesses, and this adds an extra cost on top of the already expensive IM kernels. In this article we leverage fused sampling, memoization, and vectorization to restructure, parallelize and boost their performance on undirected networks. The proposed approach employs a pseudo-random function and performs multiple Monte-Carlo simulations in parallel to exploit the SIMD lanes effectively and efficiently. In addition, it significantly reduces the number of edge traversals, hence the amount of data brought from the memory, which is critical for almost all memory-bound graph kernels. We apply the proposed approach to the traditional MIXGREEDY algorithm and propose INFUSER-MG which is more than 3000χ fasterthan the greedy approaches and can run on large graphs that have been considered as too large in the literature. For instance, the new algorithm runs in 2.09, 0.08, 0.36 seconds on graphs Amazon, NetHEP, NetPhy with 16 threads where the sequential baseline takes 141.3, 259.1 and 1725.2 seconds, respectively. To compare INFUSER-MG with the state-of-the-art approximation algorithms, we conduct a thorough experimental analysis with various influence settings. The results on real-life, undirected networks show that on 16 threads, INFUSER-MG is 2:3χ-173:8χ faster than state-of-the-art while being superior in terms of influence scores, and using a comparable amount of memory.

Side-Channel Attacks With Multi-Thread Mixed Leakage

Side-channel attacks are one of the greatest practical threats to security-related applications, because they are capable of breaking ciphers that are assumed to be mathematically secure. Lots of studies have been devoted to power or electro-magnetic (EM) analysis against desktop CPUs, mobile CPUs (including ARM, MSP, AVR, etc) and FPGAs, but rarely targeted modern GPUs. Modern GPUs feature their special and specific single instruction multiple threads (SIMT) execution fashion, which makes their power/EM leakage more sophisticated in practical scenarios. In this article, we study side-channel attacks with leakage from SIMT systems, and propose leakage models suited to any SIMT systems and specifically to CUDA-enabled GPUs. Afterwards, we instantiate the models with a GPU AES implementation, which is also used for performance evaluations. In addition to the models, we provide optimizations on the attacks that are based on the models. To evaluate the models and optimizations, we run the GPU AES implementation on a CUDA-enabled GPU and, at the same time, collect its EM leakage. The experimental results show that the proposed models are more efficient and the optimizations are effective as well. Our study suggests that GPU-based cryptographic implementations may be much vulnerable to microarchitecture-based side-channel attacks. Therefore, GPU-specific countermeasures should be considered for GPU-based cryptographic implementations in practical applications.

Hot sparing for lifetime-chip-performance and cost improvement in application specific SIMT processors

Redundancy is a well-known technique for replacing components with manufacturing defects, improving yield and reducing cost. Previously, most yield improvement strategies utilized redundant components only when another component had failed (i.e., cold spares). However, utilizing hot spares is becoming popular in commercial products (e.g., the NVIDIA Ti GPU series). Hot spares address manufacturing cost when the components are defective; otherwise, they can be used to improve performance in the field. In this paper, we investigate the effect of hot spares on lifetime-chip-performance (LCP) in multi-core single-instruction, multiple-thread (SIMT) processors. We observe that hot sparing is outstandingly effective for specific types of SIMT processor configurations (small and medium systems) and applications (FFT and FILTER), while improving cost and LCP over other configurations and applications as well. For example, hot-sparing can improve LCP more than 75% compared with conventional methods (i.e., cold sparing), on average, for applications that experience significant performance improvement when adding hot spares (e.g., FFT and FILTER). In particular, microarchitectural hot redundant resources (e.g., hot spare lanes) achieve better LCP improvement than conventional architectural redundancies (e.g., hot spare cores).

GPU-based non-binary LDPC decoder with weighted bit-reliability based algorithm

In this paper, we present a graphics processing unit (GPU)-based implementation of a weighted bit-reliability based (wBRB) decoder for non-binary LDPC (NB-LDPC) codes. To achieve coalesced memory accesses, an efficient data structure for the wBRB algorithm is proposed. Based on the Single-Instruction Multiple-Threads (SIMT) programming model, a novel mapping strategy with high intra-frame parallelism is presented to improve the latency and throughput performance. Moreover, by using Single-Instruction Multiple-Data (SIMD) intrinsics, four 8-bit message elements are packed into a 32-bit unit and simultaneously processed. Experimental results show that the proposed wBRB decoder provides good tradeoff between error performance and throughput for the codes with relatively large column degrees or high rates.

Parallelized Kalman-Filter-Based Reconstruction of Particle Tracks on Many-Core Architectures with the CMS Detector

In the High–Luminosity Large Hadron Collider (HL–LHC), one of the most challenging computational problems is expected to be finding and fitting charged-particle tracks during event reconstruction. The methods currently in use at the LHC are based on the Kalman filter. Such methods have shown to be robust and to provide good physics performance, both in the trigger and offline. In order to improve computational performance, we explored Kalman-filter-based methods for track finding and fitting, adapted for many-core SIMD (single instruction, multiple data) and SIMT (single instruction, multiple thread) architectures. Our adapted Kalman-filter-based software has obtained significant parallel speedups using such processors, e.g., Intel Xeon Phi, Intel Xeon SP (Scalable Processors) and (to a limited degree) NVIDIA GPUs. Recently, an effort has started towards the integration of our software into the CMS software framework, in view of its exploitation for the Run III of the LHC. Prior reports have shown that our software allows in fact for some significant improvements over the existing framework in terms of computational performance with comparable physics performance, even when applied to realistic detector configurations and event complexity. Here, we demonstrate that in such conditions physics performance can be further improved with respect to our prior reports, while retaining the improvements in computational performance, by making use of the knowledge of the detector and its geometry.