arXiv subjects

Julien Langou

Publications and source records attributed to Julien Langou.

At least 19 recordsLinked to original sources

arXiv2025

2024 NSF CSSI-Cybertraining-SCIPE PI Meeting August 12 to 13, 2024, Charlotte, NC

The second annual NSF, OAC CSSI, CyberTraining and related programs PI meeting was held August 12 to 13 in Charlotte, NC, with participation from PIs or representatives of all major awards. Keynotes, panels, breakouts, and poster sessions allowed PIs to engage with each other, NSF staff, and invited experts. The 286 attendees represented 292 awards across CSSI, CyberTraining, OAC Core, CIP, SCIPE CDSE, and related programs, and presented over 250 posters. This report documents the meetings structure, findings, and recommendations, offering a snapshot of current community perspectives on cyberinfrastructure. A key takeaway is a vibrant, engaged community advancing science through CI. AI-driven research modalities complement established HPC and data centric tools. Workforce development efforts align well with the CSSI community.

cs.ET↗

arXiv2024

Communication efficient application of sequences of planar rotations to a matrix

We present an efficient algorithm for the application of sequences of planar rotations to a matrix. Applying such sequences efficiently is important in many numerical linear algebra algorithms for eigenvalues. Our algorithm is novel in three main ways. First, we introduce a new kernel that is optimized for register reuse in a novel way. Second, we introduce a blocking and packing scheme that improves the cache efficiency of the algorithm. Finally, we thoroughly analyze the memory operations of the algorithm which leads to important theoretical insights and makes it easier to select good parameters. Numerical experiments show that our algorithm outperforms the state-of-the-art and achieves a flop rate close to the theoretical peak on modern hardware.

cs.PF↗

arXiv2024

Probabilistic Analysis of Least Squares, Orthogonal Projection, and QR Factorization Algorithms Subject to Gaussian Noise

We consider the effect of Gaussian perturbations on least-squares residuals, orthogonal projections, and QR-type algorithms. The problem that motivated our investigations is as follows: suppose that a full column-rank matrix $B\in\mathbb{R}^{m\times n}$ has already been computed, and suppose that a new normalized column $q=(x+y)/\|x+y\|_2$ is to be appended to $B$, where $x\perp\operatorname{span}(B)$ is the ideal orthogonal component and $y$ represents the orthogonalization error. How large can the condition number $\kappa([B,q])$ of the resulting matrix $[B,q]$ become? While we provide a Weyl-type bound on the singular values of $[B,q]$, in terms of the extremal singular values of $B$ and the quantity $\|B^T y\|_2/\|x+y\|_2$, we also derive exact probability laws for norms and projection residuals under Gaussian perturbations. Finally, we use these probability laws to derive probabilistic condition-number bounds for QR-type processes with imperfect orthogonalization and exact normalization.

math.NA↗

arXiv2024

Tightening I/O Lower Bounds through the Hourglass Dependency Pattern

When designing an algorithm, one cares about arithmetic/computational complexity, but data movement (I/O) complexity plays an increasingly important role that highly impacts performance and energy consumption. For a given algorithm and a given I/O model, scheduling strategies such as loop tiling can reduce the required I/O down to a limit, called the I/O complexity, inherent to the algorithm itself. The objective of I/O complexity analysis is to compute, for a given program, its minimal I/O requirement among all valid schedules. We consider a sequential execution model with two memories, an infinite one, and a small one of size S on which the computations retrieve and produce data. The I/O is the number of reads and writes between the two memories. We identify a common "hourglass pattern" in the dependency graphs of several common linear algebra kernels. Using the properties of this pattern, we mathematically prove tighter lower bounds on their I/O complexity, which improves the previous state-of-the-art bound by a parametric ratio. This proof was integrated inside the IOLB automatic lower bound derivation tool.

cs.CC↗

arXiv2023

A new deflation criterion for the QZ algorithm

The QZ algorithm computes the Schur form of a matrix pencil. It is an iterative algorithm and at some point, it must decide that an eigenvalue has converged and move on with another one. Choosing a criterion that makes this decision is nontrivial. If it is too strict, the algorithm might waste iterations on already converged eigenvalues. If it is not strict enough, the computed eigenvalues might be inaccurate. Additionally, the criterion should not be computationally expensive to evaluate. This paper introduces a new criterion based on the size of and the gap between the eigenvalues. This is similar to the work of Ahues and Tissuer for the QR algorithm. Theoretical arguments and numerical experiments suggest that it outperforms the most popular criteria in terms of accuracy. Additionally, this paper evaluates some commonly used criteria for infinite eigenvalues.

math.NA↗

arXiv2022

Numerical analysis of Givens rotation

Generating 2-by-2 unitary matrices in floating-precision arithmetic is a delicate task. One way to reduce the accumulation error is to use less floating-point operations to compute each of the entries in the 2-by-2 unitary matrix. This paper shows an algorithm that reduces the number of operations to compute the entries of a Givens rotation. Overall, the new algorithm has more operations in total when compared to algorithms in different releases of LAPACK, but less operations per entry. Numerical tests show that the new algorithm is more accurate on average.

math.NA↗

arXiv2022

Proposed Consistent Exception Handling for the BLAS and LAPACK

Numerical exceptions, which may be caused by overflow, operations like division by 0 or sqrt(-1), or convergence failures, are unavoidable in many cases, in particular when software is used on unforeseen and difficult inputs. As more aspects of society become automated, e.g., self-driving cars, health monitors, and cyber-physical systems more generally, it is becoming increasingly important to design software that is resilient to exceptions, and that responds to them in a consistent way. Consistency is needed to allow users to build higher-level software that is also resilient and consistent (and so on recursively). In this paper we explore the design space of consistent exception handling for the widely used BLAS and LAPACK linear algebra libraries, pointing out a variety of instances of inconsistent exception handling in the current versions, and propose a new design that balances consistency, complexity, ease of use, and performance. Some compromises are needed, because there are preexisting inconsistencies that are outside our control, including in or between existing vendor BLAS implementations, different programming languages, and even compilers for the same programming language. And user requests from our surveys are quite diverse. We also propose our design as a possible model for other numerical software, and welcome comments on our design choices.

cs.MS↗

arXiv2022

I/O-Optimal Algorithms for Symmetric Linear Algebra Kernels

In this paper, we consider two fundamental symmetric kernels in linear algebra: the Cholesky factorization and the symmetric rank-$k$ update (SYRK), with the classical three nested loops algorithms for these kernels. In addition, we consider a machine model with a fast memory of size $S$ and an unbounded slow memory. In this model, all computations must be performed on operands in fast memory, and the goal is to minimize the amount of communication between slow and fast memories. As the set of computations is fixed by the choice of the algorithm, only the ordering of the computations (the schedule) directly influences the volume of communications.We prove lower bounds of $\frac{1}{3\sqrt{2}}\frac{N^3}{\sqrt{S}}$ for the communication volume of the Cholesky factorization of an $N\times N$ symmetric positive definite matrix, and of $\frac{1}{\sqrt{2}}\frac{N^2M}{\sqrt{S}}$ for the SYRK computation of $\mat{A}\cdot\transpose{\mat{A}}$, where $\mathbf{A}$ is an $N\times M$ matrix. Both bounds improve the best known lower bounds from the literature by a factor $\sqrt{2}$.In addition, we present two out-of-core, sequential algorithms with matching communication volume: \TBS for SYRK, with a volume of $\frac{1}{\sqrt{2}}\frac{N^2M}{\sqrt{S}} + \bigo{NM\log N}$, and \LBC for Cholesky, with a volume of $\frac{1}{3\sqrt{2}}\frac{N^3}{\sqrt{S}} + \bigo{N^{5/2}}$. Both algorithms improve over the best known algorithms from the literature by a factor $\sqrt{2}$, and prove that the leading terms in our lower bounds cannot be improved further. This work shows that the operational intensity of symmetric kernels like SYRK or Cholesky is intrinsically higher (by a factor $\sqrt{2}$) than that of corresponding non-symmetric kernels (GEMM and LU factorization).

cs.DC↗

arXiv2021

Low-Synch Gram-Schmidt with Delayed Reorthogonalization for Krylov Solvers

The parallel strong-scaling of Krylov iterative methods is largely determined by the number of global reductions required at each iteration. The GMRES and Krylov-Schur algorithms employ the Arnoldi algorithm for nonsymmetric matrices. The underlying orthogonalization scheme is left-looking and processes one column at a time. Thus, at least one global reduction is required per iteration. The traditional algorithm for generating the orthogonal Krylov basis vectors for the Krylov-Schur algorithm is classical Gram Schmidt applied twice with reorthogonalization (CGS2), requiring three global reductions per step. A new variant of CGS2 that requires only one reduction per iteration is applied to the Arnoldi-QR iteration. Strong-scaling results are presented for finding eigenvalue-pairs of nonsymmetric matrices. A preliminary attempt to derive a similar algorithm (one reduction per Arnoldi iteration with a robust orthogonalization scheme) was presented by Hernandez et al.(2007). Unlike our approach, their method is not forward stable for eigenvalues.

math.NA↗

arXiv2019

Automated Derivation of Parametric Data Movement Lower Bounds for Affine Programs

For most relevant computation, the energy and time needed for data movement dominates that for performing arithmetic operations on all computing systems today. Hence it is of critical importance to understand the minimal total data movement achievable during the execution of an algorithm. The achieved total data movement for different schedules of an algorithm can vary widely depending on how efficiently the cache is used, e.g., untiled versus effectively tiled matrix-matrix multiplication. A significant current challenge is that no existing tool is able to meaningfully quantify the potential reduction to the data movement of a computation that can be achieved by more effective use of the cache through operation rescheduling. Asymptotic parametric expressions of data movement lower bounds have previously been manually derived for a limited number of algorithms, often without scaling constants. In this paper, we present the first compile-time approach for deriving non-asymptotic parametric expressions of data movement lower bounds for arbitrary affine computations. The approach has been implemented in a fully automatic tool (IOLB) that can generate these lower bounds for input affine programs. IOLB's use is demonstrated by exercising it on all the benchmarks of the PolyBench suite. The advantages of IOLB are many: (1) IOLB enables us to derive bounds for few dozens of algorithms for which these lower bounds have never been derived. This reflects an increase of productivity by automation. (2) Anyone is able to obtain these lower bounds through IOLB, no expertise is required. (3) For some of the most well-studied algorithms, the lower bounds obtained by \tool are higher than any previously reported manually derived lower bounds.

cs.CC↗

arXiv2019

A Tight I/O Lower Bound for Matrix Multiplication

A tight lower bound for required I/O when computing an ordinary matrix-matrix multiplication on a processor with two layers of memory is established. Prior work obtained weaker lower bounds by reasoning about the number of segments needed to perform $C:=AB$, for distinct matrices $A$, $B$, and $C$, where each segment is a series of operations involving $M$ reads and writes to and from fast memory, and $M$ is the size of fast memory. A lower bound on the number of segments was then determined by obtaining an upper bound on the number of elementary multiplications performed per segment. This paper follows the same high level approach, but improves the lower bound by (1) transforming algorithms for MMM so that they perform all computation via fused multiply-add instructions (FMAs) and using this to reason about only the cost associated with reading the matrices, and (2) decoupling the per-segment I/O cost from the size of fast memory. For $n \times n$ matrices, the lower bound's leading-order term is $2n^3/\sqrt{M}$. A theoretical algorithm whose leading terms attains this is introduced. To what extent the state-of-the-art Goto's Algorithm attains the lower bound is discussed.

cs.CC↗

arXiv2018

Low synchronization GMRES algorithms

Communication-avoiding and pipelined variants of Krylov solvers are critical for the scalability of linear system solvers on future exascale architectures. We present low synchronization variants of iterated classical (CGS) and modified Gram-Schmidt (MGS) algorithms that require one and two global reduction communication steps. Derivations of low synchronization iterated CGS algorithms are based on previous work by Ruhe. Our main contribution is to introduce a backward normalization lag into the compact $WY$ form of MGS resulting in a ${\cal O}(\eps)\kappa(A)$ stable GMRES algorithm that requires only one global synchronization per iteration. The reduction operations are overlapped with computations and pipelined to optimize performance. Further improvements in performance are achieved by accelerating GMRES BLAS-2 operations on GPUs.

math.NA↗

arXiv2018

Fast Parallel Randomized QR with Column Pivoting Algorithms for Reliable Low-rank Matrix Approximations

Factorizing large matrices by QR with column pivoting (QRCP) is substantially more expensive than QR without pivoting, owing to communication costs required for pivoting decisions. In contrast, randomized QRCP (RQRCP) algorithms have proven themselves empirically to be highly competitive with high-performance implementations of QR in processing time, on uniprocessor and shared memory machines, and as reliable as QRCP in pivot quality. We show that RQRCP algorithms can be as reliable as QRCP with failure probabilities exponentially decaying in oversampling size. We also analyze efficiency differences among different RQRCP algorithms. More importantly, we develop distributed memory implementations of RQRCP that are significantly better than QRCP implementations in ScaLAPACK. As a further development, we introduce the concept of and develop algorithms for computing spectrum-revealing QR factorizations for low-rank matrix approximations, and demonstrate their effectiveness against leading low-rank approximation methods in both theoretical and numerical reliability and efficiency.

math.NA↗

arXiv2016

Bidiagonalization with Parallel Tiled Algorithms

We consider algorithms for going from a "full" matrix to a condensed "band bidiagonal" form using orthogonal transformations. We use the framework of "algorithms by tiles". Within this framework, we study: (i) the tiled bidiagonalization algorithm BiDiag, which is a tiled version of the standard scalar bidiagonalization algorithm; and (ii) the R-bidiagonalization algorithm R-BiDiag, which is a tiled version of the algorithm which consists in first performing the QR factorization of the initial matrix, then performing the band-bidiagonalization of the R-factor. For both bidiagonalization algorithms BiDiag and R-BiDiag, we use four main types of reduction trees, namely FlatTS, FlatTT, Greedy, and a newly introduced auto-adaptive tree, Auto. We provide a study of critical path lengths for these tiled algorithms, which shows that (i) R-BiDiag has a shorter critical path length than BiDiag for tall and skinny matrices, and (ii) Greedy based schemes are much better than earlier proposed variants with unbounded resources. We provide experiments on a single multicore node, and on a few multicore nodes of a parallel distributed shared-memory system, to show the superiority of the new algorithms on a variety of matrix sizes, matrix shapes and core counts.

cs.MS↗

arXiv2015

A Backward/Forward Recovery Approach for the Preconditioned Conjugate Gradient Method

Several recent papers have introduced a periodic verification mechanism to detect silent errors in iterative solvers. Chen [PPoPP'13, pp. 167--176] has shown how to combine such a verification mechanism (a stability test checking the orthogonality of two vectors and recomputing the residual) with checkpointing: the idea is to verify every $d$ iterations, and to checkpoint every $c \times d$ iterations. When a silent error is detected by the verification mechanism, one can rollback to and re-execute from the last checkpoint. In this paper, we also propose to combine checkpointing and verification, but we use algorithm-based fault tolerance (ABFT) rather than stability tests. ABFT can be used for error detection, but also for error detection and correction, allowing a forward recovery (and no rollback nor re-execution) when a single error is detected. We introduce an abstract performance model to compute the performance of all schemes, and we instantiate it using the preconditioned conjugate gradient algorithm. Finally, we validate our new approach through a set of simulations.

cs.DS↗

arXiv2015

A Makespan Lower Bound for the Scheduling of the Tiled Cholesky Factorization based on ALAP scheduling

Due to the advent of multicore architectures and massive parallelism, the tiled Cholesky factorization algorithm has recently received plenty of attention and is often referenced by practitioners as a case study. It is also implemented in mainstream dense linear algebra libraries. However, we note that theoretical study of the parallelism of this algorithm is currently lacking. In this paper, we present new theoretical results about the tiled Cholesky factorization in the context of a parallel homogeneous model without communication costs. We use standard flop-based weights for the tasks. For a $t$-by-$t$ matrix, we know that the critical path of the tiled Cholesky algorithm is $9t-10$ and that the weight of all tasks is $t^3$. In this context, we prove that no schedule with less than $0.185 t^2$ processing units can finish in a time less than the critical path. In perspective, a naive bound gives $0.11 t^2.$ We then give a schedule which needs less than $0.25 t^2+0.16t+3$ processing units to complete in the time of the critical path. In perspective, a naive schedule gives $0.50 t^2.$ In addition, given a fixed number of processing units, $p$, we give a lower bound on the execution time as follows: $$\max( \frac{t^{3}}{p}, \frac{t^{3}}{p} - 3\frac{t^2}{p} + 6\sqrt{2p} - 7 , 9t-10).$$ The interest of the latter formula lies in the middle term. Our results stem from the observation that the tiled Cholesky factorization is much better behaved when we schedule it with an ALAP (As Late As Possible) heuristic than an ASAP (As Soon As Possible) heuristic. We also provide scheduling heuristics which match closely the lower bound on execution time. We believe that our theoretical results will help practical scheduling studies. Indeed, our results enable to better characterize the quality of a practical schedule with respect to an optimal schedule.

cs.DC↗

arXiv2014

On matrix balancing and eigenvector computation

Balancing a matrix is a preprocessing step while solving the nonsymmetric eigenvalue problem. Balancing a matrix reduces the norm of the matrix and hopefully this will improve the accuracy of the computation. Experiments have shown that balancing can improve the accuracy of the computed eigenval- ues. However, there exists examples where balancing increases the eigenvalue condition number (potential loss in accuracy), deteriorates eigenvector accuracy, and deteriorates the backward error of the eigenvalue decomposition. In this paper we propose a change to the stopping criteria of the LAPACK balancing al- gorithm, GEBAL. The new stopping criteria is better at determining when a matrix is nearly balanced. Our experiments show that the new algorithm is able to maintain good backward error, while improving the eigenvalue accuracy when possible. We present stability analysis, numerical experiments, and a case study to demonstrate the benefit of the new stopping criteria.

math.NA↗

arXiv2014

Stability Analysis of QR factorization in an Oblique Inner Product

In this paper we consider the stability of the QR factorization in an oblique inner product. The oblique inner product is defined by a symmetric positive definite matrix A. We analyze two algorithm that are based a factorization of A and converting the problem to the Euclidean case. The two algorithms we consider use the Cholesky decomposition and the eigenvalue decomposition. We also analyze algorithms that are based on computing the Cholesky factor of the normal equa- tion. We present numerical experiments to show the error bounds are tight. Finally we present performance results for these algorithms as well as Gram-Schmidt methods on parallel architecture. The performance experiments demonstrate the benefit of the communication avoiding algorithms.

math.NA↗