A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
Filters
Subword parallelism with MAX-2
1996
IEEE Micro
Preserved Fulltext
Table 5 . 5 Operation parallelism in PA-8000 with MAX-2. ...
MAX speeds this up with parallel shift left and add, and parallel shift right and add. ...
doi:10.1109/40.526925
fatcat:xd575wxq5zh73hz2osunqjo4ta
Multimedia instructions in ia-64
2001
IEEE International Conference on Multimedia and Expo, 2001. ICME 2001.
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
They are both a subset and a superset of the multimedia instructions from the predecessor architectures: MMX, SSE and SSE-2 from the IA-32 architecture, and MAX and MAX-2 from the PA-RISC architecture. ...
These multimedia instructions implement subword parallelism, also called packed parallelism or microSIMD parallelism. ...
These instructions implement the ISA concept of subword parallelism [2, 3] , also called packed parallelism [4] or microSIMD parallelism [5] . ...
doi:10.1109/icme.2001.1237694
dblp:conf/icmcs/LeeFB01
fatcat:hu6cvwfttffxfn7igmzlbi6vji
PLX: An Instruction Set Architecture and Testbed for Multimedia Information Processing
2005
Journal of VLSI Signal Processing Systems for Signal, Image and Video Technology
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
Key characteristics of PLX are a fully subword-parallel architecture with novel features like wordsize scalability from 32-bit to 128-bit words, a new definition of predication, and an innovative set of ...
subword permutation instructions. ...
Operation 1
Instructions
Explanation
Clip all a i at an
arbitrary maximum
value a max , where
0 ≤ a max ≤ 2 15 −1.
padd.s.2 a, a, b Initially all b i contain
2 15 − 1 − a max . ...
doi:10.1007/s11265-005-4940-8
fatcat:vzndq4zbfvdt7cey2yncygiuoi
Subword extensions for video processing on mobile systems
1998
IEEE Concurrency
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
We'll refer to this particular form of micro-SIMD execution as subword execution. ...
execution on data organized in parallel. ...
With both MAX-1 and MAX-2, HP takes a minimalist approach to multimedia extensions. ...
doi:10.1109/4434.708250
fatcat:5uaj7z4luvdl7lffllfpvokgci
Multimedia Instructions in Microprocessors for Native Signal Processing
[chapter]
2001
Signal Processing and Communications
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2010; you can also visit the original URL.
The file type is application/pdf.
Native signal processing is DSP performed in the microprocessor itself, with the addition of general-purpose multimedia instructions. ...
Common usage TM * Through this chapter, the subwords in a register will be indexed from 1 to n, where n will be the number of subwords in that register. ...
MAX-2, although designed simultaneously with MAX-1, was introduced later with the 64-bit PA-RISC 2.0 architecture. ...
doi:10.1201/9780203908068.ch3
fatcat:2jijojacfvaablxttyp3eko5be
Efficiency of microSIMD architectures and index-mapped data for media processors
1998
Media Processors 1999
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2006; you can also visit the original URL.
The file type is application/pdf.
We define alternative mappings of data onto subwords, and show that the index mapping is an ideal mapping for achieving maximal subword parallelism with minimal revamping of the original serial loop code ...
We also show how to convert rapidly between data mappings by using the Mix permutation instructions, first defined in the MAX-2 multimedia extensions for PA-RISC processors. ...
In general, the subword parallelism in figure 3 can be combined with the functional unit parallelism of a uniprocessor in figure 2 , and with the processor parallelism in figure 1 , for even higher ...
doi:10.1117/12.334770
fatcat:xn2hirfmo5ax3h3xgw5an3iktq
External inverse pattern matching
[chapter]
1997
Lecture Notes in Computer Science
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2018; you can also visit the original URL.
The file type is application/pdf.
The entire problem is to nd a patternP MAX 2 m which is not a subword of T and which maximizes the sum of Hamming distances betweenP MAX and all subwords of T of length m. ...
Moreover we discuss a fast parallel implementation of our algorithm on the CREW PRAM model. ...
subwords of P MAX (using LCA queries and table ...
doi:10.1007/3-540-63220-4_53
fatcat:hsoif5hterehdczipwca4io5qq
SoftSIMD - Exploiting Subword Parallelism Using Source Code Transformations
2007
2007 Design, Automation & Test in Europe Conference & Exhibition
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2007; you can also visit the original URL.
The file type is application/pdf.
Vendors often use proprietary platforms which are incompatible with others. Therefore, porting software is a very complex and time consuming task. ...
A, B ∈ {−M max , ..., M max } A bias = A + M max B bias = B + M max C = A bias − B bias = A − B C ∈ {−2 M max , ..., 2 M max }
Figure 2. ...
Thus, with a 32bit multiplication it is only possible to implement a SIMD 2 multiplication for two 8
-bit values in parallel. ...
doi:10.1109/date.2007.364485
fatcat:nsvil2ufczhfvibeu6itpfeb4q
Challenges to combining general-purpose and multimedia processors
1997
Computer
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2004; you can also visit the original URL.
The file type is application/pdf.
In addition to microarchitectural features, the ISA extensions are continuing to evolve (for example, HP has released two generations of extensions, MAX-1 and MAX-2). ...
These include MAX-2 extensions to Hewlett-Packard PA-RISC, 1 MMX for Intel's x 86, 2,3 UltraSparc's VIS, 4 and MDMX extensions to MIPS V. 5 Processors targeted to embedded multimedia applications-the ...
doi:10.1109/2.642799
fatcat:j2gnaqh77fagbgxqjy5hbolqiq
Character Transformations for Non-Autoregressive GEC Tagging
[article]
2021
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
We propose a character-based nonautoregressive GEC approach, with automatically generated character transformations. ...
correction for languages other than English, with character transformations applied at subwords, inferred automatically from parallel GEC corpus. ...
() = g.strip() 0.5 • LevenshSimilarity(s[i], g) w[i, j] ← max(w[i, j], c+w[i, j +1]) end end end return alignment with weight w[0, 0], as in LCS with an attempt at non-autoregressive grammatical error ...
arXiv:2111.09280v1
fatcat:2zk6abkwufecheqh5b7wat2xai
Cross-language spoken document retrieval using HMM-based retrieval model with multi-scale fusion
2003
ACM Transactions on Asian Language Information Processing
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
In this work the extended HMM-based retrieval model has been applied to an English-Mandarin CL-SDR task, which is to search the Mandarin spoken document collection with English queries at word and subword ...
In addition, this HMM-based CLIR retrieval model is also extended for retrieval at subword scales. ...
D i p(D i |Q) = arg max D i p(Q · D i ) p(Q) = arg max D i p(Q|D i ) p(D i ) p(Q) (2) (2) becomes arg max D i p(D i |Q) = arg max D i p(Q|D i ) p(D i ) (3) If we assume that the a priori probability p( ...
doi:10.1145/964161.964162
fatcat:meyid3zxrjglvpf3zm4nn3u2wu
Overview of research efforts on media ISA extensions and their usage in video coding
2002
IEEE transactions on circuits and systems for video technology (Print)
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2003; you can also visit the original URL.
The file type is application/pdf.
Optimized applications include, in addition to some proprietary methods, all of the major video coding standards such as H.261, H.263, MPEG-4, MPEG-1, and MPEG-2. ...
This paper summarizes the results of over 25 research groups or individual researchers that have presented video coding implementations on general-purpose processors with the new single instruction multiple ...
This subset of subword parallelism is also referred to as packed parallelism. In MMX, MAX, VIS, and MDMX, for example, the subwords are integer subwords, while in SSE, 3DNow! ...
doi:10.1109/tcsvt.2002.800865
fatcat:tf62dj6pozda5e2iqck6iipu2u
Subword Segmentation and a Single Bridge Language Affect Zero-Shot Neural Machine Translation
[article]
2020
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2020; you can also visit the original URL.
The file type is application/pdf.
We show that this bias towards English can be effectively reduced with even a small amount of parallel data in some of the non-English pairs. ...
We find that language-specific subword segmentation results in less subword copying at training time, and leads to better zero-shot performance compared to jointly trained segmentation. ...
To test this hypothesis, we train two models with languagespecific subword segmentation: a) a model with language-specific subword seg-2 See also (Ha et al., 2017; Arivazhagan et al., 2019; Zhang et al ...
arXiv:2011.01703v1
fatcat:qoy243alazfp3nq7z2vtrikcfm
Performance Improvement of Multimedia Kernels by Alleviating Overhead Instructions on SIMD Devices
[chapter]
2009
Lecture Notes in Computer Science
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2017; you can also visit the original URL.
The file type is application/pdf.
The extended subwords, uses four extra bits for every byte in a media register and it provides additional parallelism. ...
SIMD extension is one of the most common and effective technique to exploit data-level parallelism in today's processor designs. ...
GPP with
Multimedia Extension
ISA Name
AltiVec/VMX MAX-1/2 MDMX MMX/
VIS MMX/ SSE
SSE2
SPU ISA
3DNow
SIMD
Company
Motorola/IBM
HP
MIPS
AMD
Sun
Intel
Intel
Intel
IBM/Sony/Toshiba
Instruction ...
doi:10.1007/978-3-642-03644-6_31
fatcat:tvpvpowuzzcqncqefdhl4tkesu
SimAlign: High Quality Word Alignments without Parallel Training Data using Static and Contextualized Embeddings
[article]
2021
arXiv
pre-print
Preserved Fulltext
A copy of this work was available on the public web and has been preserved in the Wayback Machine. The capture dates from 2021; you can also visit the original URL.
The file type is application/pdf.
Other Versions
We find that alignments created from embeddings are superior for four and comparable for two language pairs compared to those produced by traditional statistical aligners, even with abundant parallel data ...
Statistical word aligners perform well, as do methods that extract alignments jointly with translations in NMT. ...
(S, n max , α ∈ [0, 1]) 2: A, M = zeros like(S) 3: for n ∈ [1, . . . , n max ] do 4: ∀i, j : 5: Mij = 1 if max le l=0 A lj , l f l=0 A il = 0 0 if min le l=0 A lj , l f l=0 A il > 0 α otherwise ...
arXiv:2004.08728v4
fatcat:pxp5aq5m2vemtepw4py33yd5ry
« Previous
Showing results 1 — 15 out of 2,932 results