CN104459781B - The random noise attenuation method of three-dimensional earthquake data before superposition - Google Patents
The random noise attenuation method of three-dimensional earthquake data before superposition Download PDFInfo
- Publication number
- CN104459781B CN104459781B CN201410749090.4A CN201410749090A CN104459781B CN 104459781 B CN104459781 B CN 104459781B CN 201410749090 A CN201410749090 A CN 201410749090A CN 104459781 B CN104459781 B CN 104459781B
- Authority
- CN
- China
- Prior art keywords
- node
- space
- output
- time small
- small data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
The invention provides a kind of random noise attenuation method of three-dimensional earthquake data before superposition, comprise the following steps:Each nodal parallel receives and is distributed to the space-time small data block of itself in multiple nodes, and the space-time small data block is to split into three-dimensional X, Y, Z earthquake data before superposition according to three directions of geophone offset or Taoist monastic name in the line in interconnection direction, CMP CMP, CMP trace gather in main profile direction;Each node the space-time small data block for being distributed to itself is carried out parallel prestack random noise attenuation RNA calculate, overlap processing and output in node, and pass through to communicate between adjacent node and parallel carry out overlap processing and output between adjacent node.The present invention improves the efficiency of three-dimensional earthquake data before superposition RNA processing, shortens the time needed for three-dimensional earthquake data before superposition RNA processing.
Description
Technical field
The present invention relates to Seismic Exploration Data Processing technical field, more particularly, to a kind of three-dimensional earthquake data before superposition with
Machine noise reduction method.
Background technology
In the gatherer process of geological data, due to the complexity of work area surface conditions, and residing for focus and geophone station
Often contain many random noises, such as microseism, ambient interferences in the influence of external environment condition, seismic data.These are in earthquake record
On to show as the distribution of rambling random noise very wide, had a strong impact on the signal to noise ratio of data, the processing solution to seismic data
Release work and bring great difficulty.Therefore, people are seeking the effective ways for removing random noise always.Based on prediction theory
Random noise attenuation (Random noise attenuation, RNA) be one of relatively effective method, it earliest by
Canales proposed the filtering of F-X domains in 1984, and with the conspicuousness in its theoretic tightness and actual effect quickly on ground
It is used widely in shake data digital processing.By long-term constantly development, mesh is developed into from the two-dimentional F-X domains of early stage
Preceding four-dimensional F-XYZ domains prediction noise-removed technology.
When the occurrence of underground medium is not level, the lineups of reflection seismic can shift, it is impossible to reflecting at this
The true occurrence of lower medium;At this moment require to carry out migration in seismic data process.Skew has two ways:Before superposition
Offset after skew and superposition, the former is just prestack, and the latter is poststack.Equally, three-dimensional earthquake data before superposition is also required to carry out
RNA。
In the prior art, all it is to use single node serial process during three-dimensional earthquake data before superposition carries out RNA processing.
However, with high covering, the popularization and application of high-density acquisition technique, three-dimensional earthquake data before superposition generally needed during RNA processing
Logarithm is wanted to be calculated with TB grades of mass data, this can spend longer operation time, so as to have a strong impact on it in production
Application.
The content of the invention
It is an object of the invention to provide a kind of random noise attenuation method of three-dimensional earthquake data before superposition, to improve prestack
The efficiency of geological data RNA processing.
To reach above-mentioned purpose, the invention provides a kind of random noise attenuation method of three-dimensional earthquake data before superposition, bag
Include following steps:
Each nodal parallel receives and is distributed to the space-time small data block of itself in multiple nodes, and the space-time is small
Data block is according to the line in interconnection direction, the CMP (Common in main profile direction by three-dimensional X, Y, Z earthquake data before superposition
Middle Point, CMP), three directions of the geophone offset in CMP trace gathers or Taoist monastic name split into;
Each node carries out random noise attenuation RNA meters parallel to the space-time small data block for being distributed to itself
Overlap processing and output in calculation, node, and pass through overlap processing and output between the parallel progress adjacent node of communication between adjacent node.
A control is enabled in the random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, each node
Thread, for task computation, the I/O resources controls of this node, the control of overlap processing state and output processing state control.
The random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, the I/O resources controls tool of described node
Body includes:
The space-time small data block of this node will be distributed to according to tri- directions of X, Y, Z successively label, in each node
Inside, the calculating of the space-time small data block is carried out successively according to the label.
The random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, the overlap processing status controlling packet is included:
Each node records the overlap processing state of all space-time small data blocks in this node using a token;
The overlap processing state of each space-time small data block has three kinds:0 represents not do overlap processing, and 1 represents to do overlap processing, 2
Expression is stayed to be processed later;The overlap processing order of each space-time small data block is carried out according to label order, for
The overlap processing of state 0 is wait;Overlap processing for state 1 is to do overlap processing, and the overlap processing for state 2 is jump
Cross.
The random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, the output processing status controlling packet is included:
Each node records the output state of all space-time small data blocks in this node using a token, each
The output state of space-time small data block has two kinds:0 represents not exportable, and 1 represents output;Each space-time small data block
Output order according to the label order carry out, for state 0 output processing state control for wait, for state 1
The control of processing state is exported directly to export.
Multiple calculating are enabled in the random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, each node
Thread, the prestack RNA that each computational threads are used to complete a space-time small data block is calculated.
The random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, enable in each node one it is overlapping
Thread, for the overlap processing between adjacent space-time small data block in this node.
An output is enabled in the random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, each node
Thread, for data output in this node.
Enabling one in the random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, each node is used for
The process of the communication of overlapped data and overlapping output between adjacent node, the process, which includes one, is used for overlapped data between adjacent node
Communication thread, and one be used for adjacent node between overlapped data output processing overlapping output thread.
The random noise attenuation method of the three-dimensional earthquake data before superposition of the present invention, the multiple node is Linux clusters.
The embodiment of the present invention is by three-dimensional X, Y, Z earthquake data before superposition according to the line in interconnection direction, main profile direction
Three directions of geophone offset or Taoist monastic name in CMP, CMP trace gather split some space-time small data blocks and are distributed on multiple nodes,
Then calculated using multi-core parallel concurrent in multi -CPU between node and node and realize acceleration RNA processing, so as to improve three-dimensional pre-stack seismic
The efficiency of data RNA processing, shortens the time needed for three-dimensional earthquake data before superposition RNA processing.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, constitutes the part of the application, not
Constitute limitation of the invention.In the accompanying drawings:
Fig. 1 is the random noise attenuation method flow diagram of the three-dimensional earthquake data before superposition of the embodiment of the present invention;
Fig. 2 is data distribution schematic diagram in the embodiment of the present invention;
Fig. 3 is the Parallel Design schematic diagram of individual node in the embodiment of the present invention;
Fig. 4 a are prior art serial computing result schematic diagram;
Fig. 4 b are parallel computation result schematic diagram of the embodiment of the present invention.
Embodiment
For the object, technical solutions and advantages of the present invention are more clearly understood, with reference to embodiment and accompanying drawing, to this
Invention is described in further details.Here, the schematic description and description of the present invention is used to explain the present invention, but do not make
For limitation of the invention.
Below in conjunction with the accompanying drawings, the embodiment to the present invention is described in further detail.
With reference to shown in Fig. 1, the random noise attenuation method of the three-dimensional earthquake data before superposition of the embodiment of the present invention is including following
Step:
Each nodal parallel receives and is distributed to the space-time small data block of itself in step S101, multiple nodes.Wherein,
Space-time small data block be by three-dimensional X, Y, Z earthquake data before superposition according to the line in interconnection direction, the CMP in main profile direction,
What three directions of geophone offset or Taoist monastic name in CMP trace gathers were split into;As shown in Fig. 2 system first by data traversal once, it is traversed
A task parallelism of each node reads space-time small data block in journey, if then will according to the task splitting scheme of setting
A dry space-time small data block is distributed to each node, wherein, it is as average as possible for the suitable node of data-handling capacity
Space-time small data block is averagely distributed to each node.Than if any M Y-direction space-time small data block, N number of section
Point, then the number of tasks of each node be:M/N, which is rounded, Jia one.
Step S102, each node carry out random noise attenuation parallel to the space-time small data block for being distributed to itself
RNA is calculated, overlap processing and output in node, and by communicated between adjacent node it is parallel carry out between adjacent node overlap processing and
Output.The parallel processing mechanism of each node is illustrated with reference to Fig. 3:
In the embodiment of the present invention, a control thread is enabled in each node, the I/O for task computation, this node is provided
Source control, the control of overlap processing state and output processing state control.
The I/O resources controls of this node are specifically included:The space-time small data block of this node will be distributed to according to X, Y, Z
Three directions label successively, in each intra-node, the calculating of the space-time small data block is entered successively according to the label
OK.
Overlap processing status controlling packet is included:Each node records all space-time small datas in this node using a token
The overlap processing state of block;The overlap processing state of each space-time small data block has three kinds:0 represents not do overlap processing, 1
Overlap processing is done in expression, and 2 represent to stay to be processed later;The overlap processing order of each space-time small data block is according to the mark
Number order carry out, for state 0 overlap processing for wait;Overlap processing for state 1 is to do overlap processing, for state
2 overlap processing is to skip.
Output processing status controlling packet is included:Each node records all space-time small datas in this node using a token
The output state of block, the output state of each space-time small data block has two kinds:0 represents not exportable, and 1 represents output;Often
The output order of individual space-time small data block is carried out according to label order, for the output processing state control of state 0
To wait, the output processing state control for state 1 is directly output.
In the embodiment of the present invention, multiple computational threads are enabled in each node, each computational threads are used to complete one
The prestack RNA of individual space-time small data block is calculated, specifically:
(a) FFT, is carried out parallel to multiple space-time small data blocks using multiple threads, so that time domain is four-dimensional
Space small data block is transformed into frequency domain space-time small data block, now, the different seismic channel of each thread process;
(b) predictive filtering operator, is asked for using multiple thread parallels, now, the different frequency of each thread process, for
Each frequency asks for its predictive filtering operator by solving the equation group of a complex coefficient matrix;
(c) convolution calculating, is carried out to multiple predictive filtering operators using multiple thread parallels, by multiple predictive operators pair
On the corresponding frequency domain space-time small data block of each frequency of application answered;
(d) it is, similar with the FFT of step (a), the output result of step (c) is subjected to FFT inverse transformations parallel, so that
Filtered frequency domain space-time small data block is converted back into time domain.
In the embodiment of the present invention, an overlapping thread is enabled in each node, it is small for adjacent space-time in this node
Overlap processing between data block.The space-time small data block overlap processing of such as Z-direction can be carried out all in this node;Y
On direction the space-time small data block at two ends it is overlapping we remain last processing, remaining is directly carried out in this node;X side
Upwards, a part of space-time small data block needs adjacent other node datas, and this part space-time small data block is left for
Overlapping and output process is handled between adjacent node, and remaining space-time small data block is carried out in this intra-node.
In the embodiment of the present invention, an output thread is enabled in each node, for data output in this node.
In the embodiment of the present invention, one is enabled in each node is used between adjacent node the communication of overlapped data and overlapping defeated
The process gone out, the process, which includes one, is used for the communication thread of overlapped data between adjacent node, and one is used for adjacent node
Between overlapped data output processing overlapping output thread.
Below, the effect of the embodiment of the present invention is illustrated by taking the Linux clusters that 6 nodes are constituted as an example, specifically:
(a) test environment and test data
Test cluster environment includes hardware environment, software environment, runs software, and wherein hardware environment is 6 nodes
Linux clusters, each node includes between 24 core CPU, node connecting (two netting twines are bound) by gigabit networking;Operation is soft
Part includes serial RNA programs and parallel RNA programs;Test data uses certain three-dimensional work area earthquake data before superposition, specific every ginseng
Shown in number table 1.
The test environment of table 1
Node hardware environment | CPU:8X Intel(R)Xeon(R)CPU [email protected] |
Internal memory:The internal memories of 6 × 4GB DDR3ECC REG 1333 | |
Network:2 gigabit Ethernets of each node | |
Software environment | OS:Red Hat Enterprise Linux server release 5.5 |
Compiler:Intel compilers V11.0 | |
Runs software | Serial RNA modules |
Parallel RNA modules |
(b) the performance test results
In order to ensure the stability and confidence level of test performance result, we are advised using different work parameter, different pieces of information
Mould, different nodes are tested, and concrete outcome is as shown in table 2:
The test result of table 2
(c) the performance test results
Understand that parallel RNA result and serial RNA are completely the same, still, the embodiment of the present invention with reference to Fig. 4 a and Fig. 4 b
Parallel RNA efficiency is then much higher.
Three-dimensional X, Y, Z earthquake data before superposition is total to by the embodiment of the present invention according to the line in interconnection direction, main profile direction
Three directions of geophone offset or Taoist monastic name in central point CMP, CMP trace gather split into some space-time small data blocks be distributed to it is multiple
On node, then calculated using multi-core parallel concurrent in multi -CPU between node and node and realize acceleration RNA processing, so as to improve three-dimensional
The efficiency of earthquake data before superposition RNA processing, shortens the time needed for three-dimensional earthquake data before superposition RNA processing.
Those skilled in the art will also be appreciated that various illustrative components, blocks, unit and the step that the embodiment of the present invention is listed
Suddenly it can be realized by the combination of hardware, software or both.Realized to hardware or software is then passed through depending on specific
Application and whole system design requirement.Those skilled in the art be able to can use various for every kind of specific application
Method realizes described function, but this realization is understood not to the scope beyond protection of the embodiment of the present invention.
Various illustrative logical blocks described in the embodiment of the present invention, or unit can by general processor,
Digital signal processor, application specific integrated circuit (ASIC), field programmable gate array or other programmable logic devices, discrete gate
Or the design of transistor logic, discrete hardware components, or any of the above described combination is come the function described by realizing or operate.General place
It can be microprocessor to manage device, and alternatively, the general processor can also be any traditional processor, controller, microcontroller
Device or state machine.Processor can also be realized by the combination of computing device, such as digital signal processor and microprocessor,
Multi-microprocessor, one or more microprocessors combine a Digital Signal Processor Core, or any other like configuration
To realize.
The step of method described in the embodiment of the present invention or algorithm can be directly embedded into hardware, computing device it is soft
Part module or the combination of both.Software module can be stored in RAM memory, flash memory, ROM memory, EPROM storages
Other any form of storage media in device, eeprom memory, register, hard disk, moveable magnetic disc, CD-ROM or this area
In.Exemplarily, storage medium can be connected with processor, to allow processor to read information from storage medium, and
Write information can be deposited to storage medium.Alternatively, storage medium can also be integrated into processor.Processor and storage medium can
To be arranged in ASIC, ASIC can be arranged in user terminal.Alternatively, processor and storage medium can also be arranged at use
In different parts in the terminal of family.
In one or more exemplary designs, above-mentioned functions described by the embodiment of the present invention can be in hardware, soft
Part, firmware or any combination of this three are realized.If realized in software, these functions can be stored and computer-readable
On medium, or with it is one or more instruction or code form be transmitted on the medium of computer-readable.Computer readable medium includes electricity
Brain stores medium and is easy to so that allowing computer program to be transferred to other local telecommunication medias from a place.Storing medium can be with
It is that any general or special computer can be with the useable medium of access.For example, such computer readable media can include but
It is not limited to RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or other
What can be used for carrying or store with instruct or data structure and it is other can be by general or special computer or general or specially treated
Device reads the medium of the program code of form.In addition, any connection can be properly termed computer readable medium, example
Such as, if software is to pass through a coaxial cable, fiber optic cables, double from web-site, server or other remote resources
Twisted wire, Digital Subscriber Line (DSL) or with defined in being also contained in of the wireless way for transmitting such as infrared, wireless and microwave
In computer readable medium.Described disk (disk) and disk (disc) include Zip disk, radium-shine disk, CD, DVD, floppy disk
And Blu-ray Disc, disk is generally with magnetic duplication data, and disk generally carries out optical reproduction data with laser.Combinations of the above
It can also be included in computer readable medium.
Particular embodiments described above, has been carried out further in detail to the purpose of the present invention, technical scheme and beneficial effect
Describe in detail it is bright, should be understood that the foregoing is only the present invention specific embodiment, the guarantor being not intended to limit the present invention
Scope is protected, within the spirit and principles of the invention, any modification, equivalent substitution and improvements done etc. should be included in this
Within the protection domain of invention.
Claims (7)
1. a kind of random noise attenuation method of three-dimensional earthquake data before superposition, it is characterised in that comprise the following steps:
Each nodal parallel receives and is distributed to the space-time small data block of itself, the space-time small data in multiple nodes
Block is according to the line in interconnection direction, CMP CMP, CMP trace gather in main profile direction by three-dimensional X, Y, Z earthquake data before superposition
What interior three directions of geophone offset or Taoist monastic name were split into;
Each node carries out random noise attenuation RNA calculating, section to the space-time small data block for being distributed to itself parallel
Overlap processing and output in point, and pass through overlap processing and output between the parallel progress adjacent node of communication between adjacent node;Wherein:
A control thread is enabled in each node, for task computation, the I/O resources controls of this node, overlap processing
State is controlled and output processing state control;
The I/O resources controls of described node are specifically included:
The space-time small data block of this node will be distributed to according to tri- directions of X, Y, Z successively label, in each intra-node,
The calculating of the space-time small data block is carried out successively according to the label;
Each node records the overlap processing state of all space-time small data blocks in this node using a token;Each
The overlap processing state of space-time small data block has three kinds:0 represents not do overlap processing, and 1 represents to do overlap processing, and 2 represent
Stay to be processed later;The overlap processing order of each space-time small data block is carried out according to label order, for state 0
Overlap processing for wait;Overlap processing for state 1 is to do overlap processing, and the overlap processing for state 2 is to skip.
2. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described defeated
Go out to handle status controlling packet and include:
Each node records the output state of all space-time small data blocks in this node using a token, each four-dimensional
The output state of space small data block has two kinds:0 represents not exportable, and 1 represents output;Each space-time small data block is defeated
Go out order to carry out according to label order, the output processing state for state 0 is controlled to wait, for the output of state 1
The control of processing state is directly output.
3. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described every
Multiple computational threads are enabled in individual node, each computational threads are used for the prestack for completing a space-time small data block
RNA is calculated.
4. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described every
An overlapping thread is enabled in individual node, for the overlap processing between adjacent space-time small data block in this node.
5. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described every
An output thread is enabled in individual node, for data output in this node.
6. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described every
A process for communication and the overlapping output of overlapped data between adjacent node is enabled in individual node, the process includes a use
The communication thread of overlapped data between adjacent node, and one be used for adjacent node between overlapped data output processing it is overlapping defeated
Go out thread.
7. the random noise attenuation method of three-dimensional earthquake data before superposition according to claim 1, it is characterised in that described many
Individual node is Linux clusters.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410749090.4A CN104459781B (en) | 2014-12-09 | 2014-12-09 | The random noise attenuation method of three-dimensional earthquake data before superposition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410749090.4A CN104459781B (en) | 2014-12-09 | 2014-12-09 | The random noise attenuation method of three-dimensional earthquake data before superposition |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104459781A CN104459781A (en) | 2015-03-25 |
CN104459781B true CN104459781B (en) | 2017-09-01 |
Family
ID=52906126
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410749090.4A Active CN104459781B (en) | 2014-12-09 | 2014-12-09 | The random noise attenuation method of three-dimensional earthquake data before superposition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104459781B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106338769A (en) * | 2015-07-07 | 2017-01-18 | 中国石油化工股份有限公司 | Seismic data denoising method and system |
CN106547025A (en) * | 2015-09-22 | 2017-03-29 | 中国石油化工股份有限公司 | A kind of mean value weighting denoising method |
CN106526667B (en) * | 2016-10-26 | 2018-10-16 | 中国石油天然气集团公司 | Prestack time migration method and device |
CN109100804B (en) * | 2018-06-11 | 2020-10-09 | 中国石油天然气集团有限公司 | Data reconstruction method, device and system for improving seismic data space sampling attribute |
CN112230284A (en) * | 2019-07-15 | 2021-01-15 | 中国石油天然气集团有限公司 | Parallel random noise attenuation method, monitoring method and node for three-dimensional pre-stack data |
CN113126162B (en) * | 2019-12-30 | 2024-05-28 | 中国石油天然气集团有限公司 | Random noise attenuation calculation method and device |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101930079A (en) * | 2009-06-26 | 2010-12-29 | 西安英诺瓦物探装备有限公司 | Method for processing relevant/stack data in seismic prospecting |
CN102243321A (en) * | 2011-03-15 | 2011-11-16 | 浪潮(北京)电子信息产业有限公司 | Method and system for processing seismic pre-stack time migration |
CN102323619A (en) * | 2011-05-25 | 2012-01-18 | 中国石油集团川庆钻探工程有限公司 | Linear denoising method based on multi-core processor |
-
2014
- 2014-12-09 CN CN201410749090.4A patent/CN104459781B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101930079A (en) * | 2009-06-26 | 2010-12-29 | 西安英诺瓦物探装备有限公司 | Method for processing relevant/stack data in seismic prospecting |
CN102243321A (en) * | 2011-03-15 | 2011-11-16 | 浪潮(北京)电子信息产业有限公司 | Method and system for processing seismic pre-stack time migration |
CN102323619A (en) * | 2011-05-25 | 2012-01-18 | 中国石油集团川庆钻探工程有限公司 | Linear denoising method based on multi-core processor |
Non-Patent Citations (2)
Title |
---|
三维叠前深度偏移及其并行实现;王有新等;《石油地球物理勘探》;19970630;第32卷(第3期);第365-369页 * |
叠前随机噪声衰减及其应用技巧;苏贵仕等;《石油地球物理勘探》;20140630;第49卷;第87-92页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104459781A (en) | 2015-03-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104459781B (en) | The random noise attenuation method of three-dimensional earthquake data before superposition | |
CN111242321B (en) | Data processing method and related product | |
JP7485740B2 (en) | Neural Network Instruction Set Architecture | |
Corrigan et al. | Running unstructured grid‐based CFD solvers on modern graphics hardware | |
JP2020517005A (en) | Neural network processor incorporating multi-layer hierarchically aggregated computational and memory elements | |
JP7162074B2 (en) | Method and system for performing machine learning | |
KR102572757B1 (en) | Modifying machine learning models to improve locality | |
US20150347475A1 (en) | Processor and data gathering method | |
WO2020083050A1 (en) | Data stream processing method and related device | |
EP3872638A1 (en) | Operation method of an accelerator and system including the same | |
US20160335322A1 (en) | Automatic generation of multi-source breadth-first search from high-level graph language | |
Yan et al. | Is apache spark scalable to seismic data analytics and computations? | |
CN113760531A (en) | Scheduler, method of operating scheduler, and accelerator apparatus including scheduler | |
US11842220B2 (en) | Parallelization method and apparatus with processing of neural network model for manycore system | |
US8959498B2 (en) | Parallelization method, system and program | |
CN110515917A (en) | A kind of method, apparatus and medium of control reconfiguration speed | |
CN106133709A (en) | For the method and apparatus preventing the bank conflict in memorizer | |
EP3857384B1 (en) | Processing sequential inputs using neural network accelerators | |
Ltaief et al. | Scaling the “memory wall” for multi-dimensional seismic processing with algebraic compression on cerebras cs-2 systems | |
US10223260B2 (en) | Compiler-generated memory mapping hints | |
Assis et al. | Distributed-memory load balancing with cyclic token-based work-stealing applied to reverse time migration | |
US20170091285A1 (en) | Smart tuple stream alteration | |
Weisz et al. | Graphgen for coram: Graph computation on FPGAs | |
CN106933882A (en) | A kind of big data incremental calculation method and device | |
US20210263870A1 (en) | Accelerator, operation method of the accelerator, and an apparatus including the accelerator |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |