Searcharxiv⌕ Search

arXiv subjects

Jingwen Leng

Publications and source records attributed to Jingwen Leng.

At least 19 recordsLinked to original sources

Accelerating 3D Gaussian Splatting Training through Selective Freezing and Computation Caching

3D Gaussian Splatting (3DGS) becomes a key primitive for immersive 3D experiences. However, its applicability is limited by its long training overhead. Although prior work has shown that gradient accumulation is the dominant bottleneck in training, existing works either have limited speedup or degrade reconstruction quality. To address this issue, we present ElsaGS, an algorithm-hardware co-designed framework that accelerates 3DGS training with negligible accuracy degradation. Our key observation is that most Gaussians converge early, while later iterations only refine a small subset of Gaussians. Thus, we propose a technique, progressive Gaussian freezing, which dynamically freezes well-converged Gaussians and caches their cumulative contributions, thereby reducing computations in both forward rasterization and backward gradient accumulation. To make caching efficient, we further propose an architecture, ElsaGS core, which integrates a pixel-parallel rasterization array with a set-associative frozen Gaussian cache, FGcache. Across three 3DGS datasets, ElsaGS freezes up to 90% of Gaussians with negligible quality loss and achieves up to 20.6x end-to-end speedup and 99.4% energy reduction over GPUs.

cs.AR↗

CueSplat: Learning to Schedule Sparse Deformation for Dynamic Gaussian Splatting on Edge Devices

Deformation-based 4D Gaussian splatting (4DGS) enables dynamic scene rendering. However, evaluating deformation networks for all Gaussians at every frame is compute-intensive on edge devices. We find that Gaussian trajectories are often locally predictable, yet prediction reliability varies across Gaussians and time. Thus, we propose CueSplat, a post-training framework that exploits local temporal regularity while keeping the pretrained backbone frozen and retaining its Gaussian set. Specifically, CueSplat predicts Gaussian attributes from recent history and uses a lightweight learned router to select exact deformation updates within a scene-specific budget calibrated offline. Selection accounts for prediction reliability and rendering contribution. Selected indices are reused across nearby frames to reduce routing overhead, while selected Gaussians still receive exact updates each frame. To control accumulated prediction error, we use drift signals to schedule three consecutive full-deformation frames that rebuild the prediction history. Evaluated across four mainstream 4DGS algorithms, CueSplat achieves up to $6.74\times$ and $8.15\times$ speedups on Jetson AGX Orin and Orin NX, respectively, with a negligible quality drop (at most $0.16$ dB in dataset-average PSNR).

cs.CV↗

Efficient VLA Inference via Retrieval-Based Speculation and Hardware Acceleration

Vision-language-action (VLA) models enable robots to understand and execute complex instructions, but their autoregressive (AR) action generation introduces significant latency on resource-constrained robotic platforms. Speculative decoding can accelerate AR-based inference by parallelizing action generation and verification; however, existing approaches rely on costly draft models, which are inefficient on edge platforms. In this paper, we propose Zora, a hardware-software co-design framework for accelerating AR-based VLA inference. Specifically, Zora introduces a lightweight retrieval-augmented generation (RAG)-based draft model that exploits robotic dynamics without additional neural draft models. We further reduce batch verification overhead through a kinematics-guided action-acceptance strategy and a KV-cache-sharing attention mechanism. To address the retrieval overhead introduced by RAG-based drafting, we co-design dedicated hardware support, Zora core, to address the memory irregularity in RAG-based drafting. We evaluate Zora on four VLA models with the LIBERO benchmark. Zora retains comparable accuracy while achieving up to 3.4$\times$ speedup with software-only optimization and up to 7.6$\times$ speedup with hardware support over GPUs.

cs.AR↗

InfiniLoRA: Disaggregated Multi-LoRA Serving for Large Language Models

LoRA enables efficient customization of LLMs and is widely used in multi-tenant and multi-task serving. However, emerging model architectures such as MoE significantly increase LoRA memory cost, making existing coupled LoRA serving designs poorly scalable and prone to tail-latency inflation. We present InfiniLoRA, a disaggregated LoRA serving system that decouples LoRA execution from base-model inference. InfiniLoRA introduces a shared LoRA Server with parallelism-aware execution, SLO-driven provisioning, and critical-path optimizations, including GPU-initiated communication and hardware-specialized LoRA kernels. Experiments show that InfiniLoRA can achieve an average $3.05\times$ increase in serviceable request rate under strict latency SLOs, and improve the percentage of LoRA adapters satisfying the SLO requirement by 54.0\%.

cs.DC↗

LiquidCP: Decoupling Attention Tasks from Communication and Kernel Boundaries in Context Parallelism

Context parallelism (CP) partitions long sequences across GPUs for LLM training and inference. For variable-length training and full prefill, attention-balanced placement can leave token loads uneven, while ring execution repeats overhead at each computation and communication step. In stage-serialized decode CP (DCP), long-running tasks delay dependent work and leave substantial GPU hardware idle. We propose LiquidCP, which unifies training and inference through a common model of Transfer, Compute, and Reduction tasks with tile-level dependencies, decoupling attention tasks from communication and kernel boundaries. For training and full prefill, MegaRing incorporates BuddyRing placement to balance attention and token loads, and coalesces ready KV regions during forward attention to amortize overhead. For decoding and mixed batches, MegaDCP selects history-KV splits by jointly modeling the completion of attention and its dependent reductions. A shared persistent GPU-resident engine executes both designs. Compared with the fastest baseline per configuration on eight H100 GPUs, MegaRing achieves geometric-mean core-attention speedups of $1.23\times$ forward and $1.18\times$ backward, and accelerates a Transformer layer by $1.08$--$1.16\times$. MegaDCP achieves geometric-mean core-attention speedups of $1.20\times$ for decode-only and $1.35\times$ for mixed batches. In vLLM serving with KV prefix cache, it achieves an average time-to-first-token speedup of $1.22\times$.

cs.DC↗

ByteSplat: Efficient Distributed 3D Gaussian Splatting Training via Intra- and Inter-GPU communication reduction

3D Gaussian Splatting (3DGS) enables photorealistic scene reconstruction, but training large-scale scenes requires substantial memory and computation. Distributing training across multiple GPUs increases available memory capacity, yet its performance is strictly constrained by data movement. We identify two dominant bottlenecks: repeated off-chip memory accesses during forward and backward rasterization, and inter-GPU communication of partial Gaussian gradients. To address these bottlenecks, we present ByteSplat, a distributed 3DGS training framework that jointly reduces intra- and inter-GPU data movement. First, ByteSplat fuses forward rasterization and backward rasterization into a single GPU kernel, retaining intermediate results on-chip to eliminate redundant off-chip transfers. Second, to alleviate the increased on-chip storage by fused rasterization, ByteSplat introduces hardware-aware pruning that considers both rendering quality and per-tile shared-memory constraints, increasing the overall performance of fused execution. Lastly, ByteSplat exploits gradient sparsity to eliminate zero partial gradients from inter-GPU communication. GPU-efficient encoder and decoder kernels compact the remaining records and directly aggregate the received gradients on their owner GPUs, reducing both communication volume and local aggregation overhead. We evaluate ByteSplat across six datasets. Compared with the baseline, ByteSplat reduces intra-GPU off-chip traffic and backward inter-GPU communication volume by 63.4% and 65.8%, respectively. On eight GPUs, ByteSplat achieves up to 6.1$\times$ training speedup while preserving reconstruction quality.

cs.DC↗

Shared Phase Arithmetic for Parallel Quantum Rotations

A layer of diagonal quantum gates can be specified by an integer-valued function on computational basis states. This description suggests implementing the layer by evaluating that function reversibly and translating its value into a phase using a shared quantum register. We develop this viewpoint for parallel phase kickback (PPK): the weighted sum of several rotation parameters is computed coherently, added once to a phase-gradient state, and then uncomputed. The construction separates the cost of representing a phase function from the cost of applying it. We prove correctness, give explicit error bounds, and analyze logical gate counts for general weights, pairs of rotations, and weights with disjoint binary support. With Clifford-only encoding, a batch uses at most $4(b-1)$ T gates at $b$-bit phase resolution, excluding initialization. The first phase reference is prepared with independent rotations synthesized by Gridsynth; additional phase-gradient states then follow from the disjoint-support construction with $O(b)$ T gates each. The resulting bounds identify when shared phase arithmetic reduces amortized rotation cost and when encoding overhead limits that advantage.

quant-ph↗

Weave: Fine-Grained Dynamic SM Scheduling in an MoE Megakernel for Compute-Communication Overlap

Mixture-of-Experts (MoE) inference under expert parallelism (EP) turns each MoE layer into a distributed computation with costly dispatch and combine communication. State-of-the-art systems reduce this cost through communication-computation overlap, splitting the GPU's SMs for communication and computation respectively. However, this approach still leaves GPU resources wasted along two dimensions. Spatially, the best SM split is determined by each layer's routing result and varies across layers and GPUs, so fixed policies mismatch the workload and waste either NVLink bandwidth or compute throughput. Temporally, complex MoE data dependencies introduce bubbles that leave SMs idle. We present Weave, to our knowledge the first MoE overlap system that performs fine-grained dynamic SM scheduling - deciding per layer and per GPU by routing results at runtime. Once routing completes, each layer's communication and computation volumes become known; Weave exploits this predictability through a lightweight cost model running inside the persistent megakernel: a spatial scheduler partitions SMs into communication workers and computation workers to match the communication/computation throughput ratio, and a temporal scheduler coordinates the two worker groups to minimize SM idleness. On 4x H100 SXM GPUs across six mainstream MoE models, Weave achieves a 2.89x geometric-mean MoE-layer speedup and a 1.33x geometric-mean end-to-end speedup over five state-of-the-art baselines.

cs.DC↗

Atlas: Algorithm-Hardware Co-Design for On-Device City-Scale 3D Gaussian Splatting in VR

3D Gaussian splatting (3DGS) has drawn significant attention in the architectural community recently. However, enabling city scale 3DGS on mobile VR devices remains challenging, as the memory requirement of large scale scenes far exceeds the memory capacity of today's mobile GPUs. This paper presents Atlas, an on device city scale 3DGS rendering framework that enables scalable rendering without runtime Internet access. The key insight is that although the full 3DGS model is massive, each frame only requires a small subset of Gaussians under the current pose and level of detail requirement. Based on this insight, Atlas introduces a hierarchical memory offloading mechanism that dynamically loads only necessary Gaussian data into device memory. To further improve performance, Atlas proposes temporal aware LoD search and stereo rasterization to avoid redundant computation in VR. We further show that our technique can be integrated with existing 3DGS accelerators with negligible hardware overhead. Overall, Atlas achieves 18.5x speedup over the GPU baseline and 3.9x speedup over the state of the art 3DGS accelerators, with 92.4% energy savings.

cs.AR↗

A Thread-Register Decoupled GPU Execution Model for Efficient Tensor Computation

Modern GPUs increasingly integrate Tensor Cores into the execution pipeline. Although aggregate tensor throughput continues to grow, aided by an operand supply that has evolved from register-based in Ampere to redundancy-free, memory-based in Hopper and Blackwell, efficiently orchestrating the complete tensor compute pipeline for the modern AI workloads remains challenging. We identify the fundamental bottlenecks as fixed parallelism and coarse-grained scheduling, both of which are exposed by modern AI workloads that interleave diverse non-GEMM operations with GEMM. To orchestrate tensor computation efficiently, we propose FIBER, a new architecture that extends the GPU SIMT (single instruction, multiple thread) model. Its basic execution instance, the \emph{fiber}, is decoupled from private register ownership, carrying only minimal control state while accessing an SM's registers through a shared view. This enables dynamic parallelism scaling, fine-grained register-level dataflow scheduling, and offers a redundancy-free alternative for matrix operand supply. We extend the ISA, microarchitecture, and compiler to realize shared-register addressing, conflict-free operand delivery, and fiber-based program mapping. Under a typical mixed-precision LLM serving scenario, FIBER achieves a 2.25x end-to-end speedup on Ampere (1.15x for the original FP16 computation), with 1.8x and 2.09x on Hopper and Blackwell respectively, and kernel-level gains up to 2.49x.

cs.AR↗

Deltoris: Enabling Real-time VLA Inference in Embodied AI via Bit-level Sparsity and Speculative Inference

Vision-language-action (VLA) models have emerged as a key component in embodied AI. Among existing approaches, diffusion-based VLA models achieve superior motion quality and generalization. However, diffusion-based VLA models are compute-intensive and must run at high control frequency, e.g., 50-200 Hz. Thus, it imposes strict latency and energy constraints on edge devices. In this work, we present Deltoris, an algorithm-hardware co-design framework for efficient diffusion-based VLA inference. First, we exploit the temporal similarity of consecutive inputs and propose a \textit{temporal-aware bit-sparsity} algorithm that computes only the differences between consecutive inputs, eliminating redundant bit-level operations. To further address the extra off-chip traffic introduced by our algorithm, we propose a \textit{speculative inference} technique, which amortizes data loading across multiple control steps. Lastly, to support these techniques, we co-design a dedicated accelerator with customized 1D systolic bit-serial PE arrays that eliminate PE workload imbalance. Our evaluation shows that Deltoris achieves up to 34.2$\times$ speedup over mobile GPUs and 6.1$\times$ over prior accelerators, while maintaining comparable accuracy.

cs.AR↗

Timeripple: Accelerating vDiTs by Understanding the Spatio-Temporal Correlations in Latent Space

The recent surge in video generation has shown the growing demand for high-quality video synthesis using large vision models. Existing video generation models are predominantly based on the video diffusion transformer (vDiT), however, they suffer from substantial inference delay due to self-attention. While prior studies have focused on reducing redundant computations in self-attention, they often overlook the inherent spatio-temporal correlations in video streams and directly leverage sparsity patterns from large language models to reduce attention computations. In this work, we take a principled approach to accelerate self-attention in vDiTs by leveraging the spatio-temporal correlations in the latent space. We show that the attention patterns within vDiT are primarily due to the dominant spatial and temporal correlations at the token channel level. Based on this insight, we propose a lightweight and adaptive reuse strategy that approximates attention computations by reusing partial attention scores of spatially or temporally correlated tokens along individual channels. We demonstrate that our method achieves significantly higher computational savings (85\%) compared to state-of-the-art techniques over 4 vDiTs, while preserving almost identical video quality ($<$0.06\% loss on VBench).

cs.AR↗

Kaleido: Algorithm-Hardware Co-Design for Video Diffusion Transformers by Exploiting Latent Space Correlations

Video diffusion transformers (vDiTs) generate high quality video but introduce extremely high compute cost due to the long diffusion timesteps and self attention computation. As diffusion timesteps are reduced, the computation cost of self attention becomes the dominant bottleneck. Existing acceleration approaches largely inherit sparse attention techniques from large language models, which fail to consider the unique spatiotemporal correlation of video data. This paper presents Kaleido, an algorithm hardware codesign that accelerates all operations in vDiTs by exploiting channel-wise spatiotemporal correlations in latent space. Based on this insight, we propose a lightweight channelwise reuse algorithm that skips redundant computations by reusing partial results while preserving higher generative quality than prior methods (>17 dB). To efficiently support this algorithm, we design a systolic array like accelerator with reconfigurable processing elements and a lightweight data dispatcher to mitigate irregular sparsity and data access patterns introduced by our reuse algorithm. Evaluations across three mainstream vDiT models show that Kaleido achieves up to 5.9x speedup and 16.0x energy savings over state of the art accelerators.

cs.AR↗

GPU-Tile-Sim: A Tile-Centric GPU Simulation Framework for LLM Hardware-Software Co-Design

Modern LLM (large language model) workloads increasingly rely on optimized GPU kernels through hardware-software co-design. These kernels achieve high-performance through fine-grained dependency scheduling and computation-memory overlap. As such, they incur new challenges on existing GPU performance models. Instruction-driven simulators are costly to adapt to evolving architectures, while analytical models are too coarse to capture kernels' characteristics. We propose GPU-Tile-Sim, a tile-centric GPU simulation framework for LLM hardware-software co-design. The key insight is that modern LLM kernel performance is governed less by individual instruction latency than by the dependency structure that controls execution order and overlap. Accordingly, GTSim represents kernel execution as a warp-level tile graph whose nodes capture tile-level operations and whose edges encode data and ordering constraints. Using this representation, we design an automatic tile-graph frontend and a graph-driven simulation backend. We evaluate GTSim on representative GEMM, attention, and end-to-end LLM inference workloads. On A100 and H100 across both conventional and highly optimized kernels, GTSim achieves high performance-modeling accuracy (MAPE, Mean Absolute Percentage Error, 1.22%--8.71%). We further extend GTSim to Blackwell with preliminary validation, and demonstrate its effectiveness in analyzing software and architectural design choices.

cs.DC↗

GF-DiT: Scheduling Parallelism for Diffusion Transformer Serving

Diffusion Transformers (DiTs) have become the dominant architecture for image and video generation, creating growing demand for efficient DiT serving. Existing systems assign each request a fixed parallel configuration throughout its lifetime. However, DiT workloads exhibit substantial heterogeneity across requests, execution stages, and system conditions, making static parallelism inefficient and often leading to poor GPU utilization and degraded service quality. This paper argues that DiT serving should treat GPU parallelism as a first-class schedulable resource. We present GF-DiT, a policy-programmable runtime for elastic DiT serving that dynamically adapts the parallelism of running requests according to workload demands and service objectives. GF-DiT introduces an asynchronous execution abstraction that decomposes requests into independently schedulable trajectory tasks and enables online GPU reallocation. To make elastic parallelism practical, GF-DiT further proposes group-free collectives, a lightweight communication abstraction that supports low-overhead online formation and reconfiguration of arbitrary execution groups. We implement GF-DiT in vLLM-Omni and evaluate it on representative image and video diffusion workloads. Compared with fixed-pipeline execution with static parallelism, GF-DiT improves throughput by up to 6.01$\times$, reduces mean latency by up to 95%, lowers SLO violation rates by up to 90%, and reduces communication-group setup overhead from 778 ms to approximately 60 $μ$s. Our code is available at https://github.com/SJTU-Liquid/GF-DiT.

cs.DC↗

Accelerating MoE with Dynamic In-Switch Computing on Multi-GPUs

Mixture-of-Experts (MoE) has been adopted by many leading large models to reduce computational requirements. However, frequent inter-GPU communication in MoE expert parallelism (EP) becomes a performance challenge. We observe substantial redundant inter-GPU data transfers in MoE that can be potentially addressed by in-switch computing. Unfortunately, the existing solution, NVLink SHARP (NVLS), can only support static collectives with regular patterns, incapable of dynamic communication with irregular patterns in MoE. To bridge the functionality gap, we propose DySHARP, an integral dynamic in-switch computing solution to accelerate MoE, encompassing both communication primitives and communication-aware scheduling: 1) Dynamic multimem addressing co-designs ISA, architecture, and runtime, as a dynamic extension to NVLS, reducing redundant traffic. However, the resulting traffic reduction is inherently asymmetric between two directions, preventing it from directly translating into speedup. 2) Token-centric kernel fusion deeply fuses the dispatch-computation-combine pipeline, resolving this asymmetry to translate traffic reduction into actual speedup. Compared with the state-of-the-art solution, DySHARP achieves up to 1.79$\times$ speedup.

cs.AR↗

Towards Compute-Aware In-Switch Computing for LLMs Tensor-Parallelism on Multi-GPU Systems

Tensor parallelism (TP) in large-scale LLM inference and training introduces frequent collective operations that dominate inter-GPU communication. While in-switch computing, exemplified by NVLink SHARP (NVLS), accelerates collective operations by reducing redundant data transfer, its communication-centric design philosophy introduces the mismatch between its communication mode and the memory semantic requirement of LLM's computation kernel. Such a mismatch isolates the compute and communication phases, resulting in underutilized resources and limited overlap in multi-GPU systems. To address the limitation, we propose CAIS, the first Compute-Aware In-Switch computing framework that aligns communication modes with computation's memory semantics requirement. CAIS consists of three integral techniques: (1) compute-aware ISA and microarchitecture extension to enable compute-aware in-switch computing. (2) merging-aware TB (Thread Block) coordination to improve the temporal alignment for efficient request merging. (3) graph-level dataflow optimizer to achieve a tight cross-kernel overlap. Evaluations on LLM workloads show that CAIS achieves 1.38$\times$ average end-to-end training speedup over the SOTA NVLS-enabled solution, and 1.61$\times$ over T3, the SOTA compute-communicate overlap solutions but do not leverage NVLS, demonstrating its effectiveness in accelerating TP on multi-GPU systems.

cs.AR↗

MoE-Hub: Taming Software Complexity for Seamless MoE Overlap with Hardware-Accelerated Communication on Multi-GPU Systems

The Mixture-of-Experts (MoE) architecture is crucial for scaling large language models, but its scalability is severely limited by inter-GPU communication bottlenecks in multi-GPU systems. Although overlapping communication with computation is a widely recognized optimization, its effective deployment still remains challenging, both in terms of performance and programmability. In this work, we identify the root cause as a fundamental abstraction mismatch between MoE's dynamic, irregular token-to-expert mapping and the static, address-centric communication model of modern GPUs, which necessitates a complex software mediation phase to resolve addresses before data transfers, limiting performance and software flexibility. To resolve this, we propose MoE-Hub, a hardware-software co-design that introduces a destination-agnostic communication paradigm. MoE-Hub decouples data transmission from address management, allowing producers to send data immediately after routing using only a logical destination, while address allocation and data-flow orchestration are handled transparently by lightweight hardware in the GPU hub. By hardware-accelerating the entire communication control plane, MoE-Hub enables seamless and transparent overlap. Our evaluation shows that MoE-Hub achieves 1.40x-3.08x per-layer and 1.21x-1.98x end-to-end speedup over state-of-the-art systems.

cs.AR↗