Nuclear classification method, mapping method, data transmission method and device, and chip

By classifying cores in multi-core chips, determining invalid and effective scheduling, and executing algorithms only on effective scheduling, the chip mapping and maintenance cost problems caused by faulty cores are solved, and the allowable failure in the case of faultless cores is achieved Quantity and efficient operation.

CN113010354BActive Publication Date: 2024-11-01LYNXI TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110303317.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-03-22
Publication Date
2024-11-01
Estimated Expiration
2041-03-22

AI Technical Summary

Technical Problem

In multi-core chips, when a faulty core occurs, the prior art requires remapping or replacing the entire chip, resulting in high cost and low efficiency, and recompiling in the absence of faulty cores to reserve the allowable fault amount. Affects the normal operation of the chip.

Method used

Through a kernel classification method, the kernels arranged along the preset topology direction are determined as invalid rows and effective rows, ensuring that the algorithm is mapped to the cores with effective rows to execute, avoiding the invalid rows to execute the algorithm, and pre-order without faulty cores Leave invalid scheduling as backup to provide the allowable amount of faults.

Benefits of technology

It realizes that when a faulty core occurs in a multi-core chip, it does not need to be recompiled or replaced, ensuring that the chip works properly, saving maintenance costs, and providing the allowable fault amount in the absence of faulty cores to avoid recompilation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113010354B_ABST
    Figure CN113010354B_ABST
Patent Text Reader

Abstract

The present disclosure provides a core classification method for a chip, wherein the chip includes multiple cores, and the addresses of the multiple cores are arranged in a preset arrangement; the method includes: determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein the other rows of cores are all valid rows of cores. The maintenance cost required when a faulty core appears in a multi-core chip can be saved. The present disclosure also provides a mapping method, a data transmission method, a core classification device, a mapping device, a chip, and a computer-readable medium.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence chip technology, and in particular to a nuclear classification method, a mapping method, a data transmission method, a nuclear classification device, a mapping device, a chip, and a computer-readable medium. Background Art

[0002] Chips (such as artificial intelligence chips) can be composed of one or more processors. A processor usually integrates multiple complete computing engines (cores), and the cores within a processor or between multiple processors can work together.

[0003] When a faulty core appears in the chip, although the faulty core cannot execute the algorithm, the probability of routing damage is low. Therefore, although these faulty cores cannot map the algorithm, they still have the routing function. At present, when the above fault occurs, since the faulty core cannot execute the algorithm, it may be necessary to remap the entire chip so that the algorithm is mapped to a core that can execute the algorithm, or directly replace the entire chip with a normal chip, making the entire chip invalid, which is costly. Summary of the invention

[0004] The present disclosure provides a nuclear classification method, a mapping method, a data transmission method, a nuclear classification device, a mapping device, a chip, and a computer-readable medium.

[0005] In a first aspect, the present disclosure provides a core classification method for a chip, wherein the chip includes a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement; the core classification method includes:

[0006] Some rows of cores in each row of cores arranged along a preset topological direction are determined as invalid rows of cores, wherein the other rows of cores are all valid rows of cores.

[0007] In some embodiments, before determining the cores of some rows of cores arranged along the preset topological direction as invalid rows of cores, the method further includes:

[0008] Determine the preset topology direction.

[0009] In some embodiments, determining a preset topological direction includes:

[0010] In the case that there is no faulty core among the multiple cores, a preset topological direction is used as the preset topological direction, and a preset number of invalid rows is used as the invalid amount of the preset topological direction.

[0011] In some embodiments, determining a preset topological direction includes:

[0012] In the case where there is a faulty core among the multiple cores, determining, according to a preset size, the number of allowed faulty rows in each topological direction when each topological direction is a preset topological direction;

[0013] Determine the actual number of faulty rows in each topological direction, where the actual number of faulty rows is the number of rows including faulty cores in each row of cores arranged along the corresponding topological direction;

[0014] The topology direction in which the actual number of fault rows does not exceed the allowed number of fault rows is the preset topology direction.

[0015] In some embodiments, determining a preset topological direction includes:

[0016] In the case where there is a faulty core among the multiple cores, determining an actual number of faulty rows in each topological direction, wherein the actual number of faulty rows is the number of rows including the faulty core in each row of cores arranged along the corresponding topological direction;

[0017] The topology direction with the smallest number of actual fault rows is determined as the preset topology direction.

[0018] In some embodiments, determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores includes:

[0019] When there is no faulty core in any row of cores arranged along a preset topological direction, an invalid number of rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid rows of cores; wherein the invalid number is preset.

[0020] In some embodiments, the invalid number of rows of cores located at the predetermined position are the last invalid number of rows of cores along a topological direction perpendicular to the preset topological direction.

[0021] In some embodiments, determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores includes:

[0022] When there is a faulty core in any row of cores arranged along a preset topological direction, at least the cores of the row including the faulty core among the rows of cores arranged along the preset topological direction are determined as invalid rows of cores.

[0023] In some embodiments, determining at least the cores of the row including the faulty core in each row of cores arranged along the preset topological direction as the cores of the invalid row includes:

[0024] When the number of rows including faulty cores in each row of cores arranged along a preset topological direction is less than the invalid amount in the preset topological direction by N, the cores in the rows including faulty cores in each row of cores arranged along the preset topological direction are determined as invalid rows of cores, and N rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid rows of cores; wherein the invalid amount is set in advance.

[0025] In some embodiments, the N rows of cores located at predetermined positions are the last N rows of cores that do not include a faulty core along a topological direction perpendicular to the preset topological direction.

[0026] In some embodiments, after determining that cores of some rows of cores in each row of cores arranged along a preset topological direction are invalid rows of cores, the method further includes:

[0027] According to the original addresses of the cores in the invalid row, the original addresses of the cores in the valid row are updated to updated addresses, wherein the arrangement of the updated addresses is the same as that of the original addresses, and the cores in the invalid row have no updated addresses.

[0028] In some embodiments, after determining cores of some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, the arrangement of the addresses of the plurality of cores remains unchanged.

[0029] In a second aspect, the present disclosure provides a mapping method for a chip whose cores have been classified according to the aforementioned core classification method, the mapping method comprising:

[0030] According to the classification of the cores in the chip, multiple tasks are mapped to the cores of the effective row.

[0031] In some embodiments, before mapping the plurality of tasks to the cores of the effective row according to the classification of the cores in the chip, the method further includes:

[0032] According to the classification of the cores in the chip, the problems to be processed are compiled to obtain the multiple tasks.

[0033] In a third aspect, the present disclosure provides a data transmission method for a chip whose cores have been classified according to the aforementioned core classification method, the method comprising:

[0034] Acquiring data, wherein the data includes address information of a destination core;

[0035] The data is transmitted to a destination core at least according to the address information.

[0036] In some embodiments, the data transmission method is also used for a chip that has updated the address of a core in a valid row according to the aforementioned core classification method, the address information is displacement information, and the displacement information includes the displacement amount along each topological direction; the data is transmitted to the destination core at least according to the address information, including:

[0037] In an array composed of core addresses, the data is transmitted with corresponding displacements in each topological direction according to the displacement information, wherein when the data passes through an invalid row of cores along a topological direction perpendicular to a preset topological direction, the displacement in the topological direction perpendicular to the preset topological direction is ignored.

[0038] In some embodiments, the data transmission method is also used for a chip that has not updated the address of the core of the valid row, the address information is an update address, and the data is transmitted to the destination core at least according to the address information, including:

[0039] The data is transmitted to the destination core according to the update address of each core.

[0040] In a fourth aspect, the present disclosure provides a core classification device for classifying cores of a chip, wherein the chip includes a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement; the device includes:

[0041] The processing module is used to determine cores of some rows of cores arranged along a preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores.

[0042] In a fifth aspect, the present disclosure provides a mapping device for a chip that has been classified according to the aforementioned core classification method, comprising:

[0043] A mapping module is used to map multiple tasks to cores in a valid row according to the classification of the cores in the chip.

[0044] In a sixth aspect, the present disclosure provides a chip, the chip comprising a plurality of cores, the chip being configured to implement at least one of the following methods:

[0045] Nuclear classification methods as described previously;

[0046] As the mapping method described above;

[0047] Data transmission method as described above.

[0048] In a seventh aspect, the present disclosure provides a computer-readable medium having a computer program stored thereon, wherein the computer program implements at least one of the following methods when executed by a processing core:

[0049] Nuclear classification methods as described previously;

[0050] As the mapping method described above;

[0051] Data transmission method as described above.

[0052] The core classification method, mapping method, data transmission method, core classification device, mapping device, chip, and computer-readable medium provided by the present disclosure determine that some rows of cores in each row of cores arranged along a preset topological direction are invalid rows of cores, and in the case of a faulty core, the faulty core must be an invalid row of cores, while the cores of other rows (excluding the faulty core) are all valid rows of cores. Based on the classification of the cores in the address array, the algorithm can be mapped only to the cores of the valid row (all the cores of the valid row can execute the algorithm), ensuring that the cores of the invalid row (the invalid row of cores may include the faulty core) are not required to execute the algorithm, so that the cores classified as valid rows of cores can still execute the algorithm normally, and both the valid and invalid rows of cores can be routed normally. There is no need to recompile according to the specific fault situation, and there is no need to directly use a normal chip to replace the entire chip and then scrap the entire chip, which can ensure that the chip can still work normally, saving the maintenance cost required when a faulty core appears in a multi-core chip.

[0053] Furthermore, in the absence of a faulty core, some rows of cores in each row of cores arranged along a preset topological direction are determined to be invalid rows of cores, while the cores in other rows are all valid rows of cores. The invalid rows of cores can be reserved as backups, which is equivalent to providing a certain amount of tolerable faults to avoid the need for recompilation when a faulty core occurs.

[0054] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the present disclosure and do not constitute a limitation of the present disclosure. The above and other features and advantages will become more apparent to those skilled in the art by describing detailed example embodiments with reference to the accompanying drawings, in which:

[0056] Figure 1 A flowchart of a nuclear classification method provided in an embodiment of the present disclosure;

[0057] Figure 2 A flowchart of another nuclear classification method provided by an embodiment of the present disclosure;

[0058] Figure 3 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure;

[0059] Figure 4 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure;

[0060] Figure 5 A flowchart of another nuclear classification method provided by an embodiment of the present disclosure;

[0061] Figure 6 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure;

[0062] Figure 7 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure;

[0063] Figure 8 A flowchart of another nuclear classification method provided by an embodiment of the present disclosure;

[0064] Fig. 9 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure;

[0065] Fig.10 A flowchart of a mapping method provided in an embodiment of the present disclosure;

[0066] Fig.11 A flowchart of a data transmission method provided by an embodiment of the present disclosure;

[0067] Fig.12 A flowchart of another data transmission method provided by an embodiment of the present disclosure;

[0068] Fig.13 A flowchart of another data transmission method provided by an embodiment of the present disclosure;

[0069] Fig.14 A schematic diagram of an address array provided in an embodiment of the present disclosure;

[0070] Fig.15 A schematic diagram of a classified address array provided in an embodiment of the present disclosure;

[0071] Fig.16 A schematic diagram of a fault condition address array provided by an embodiment of the present disclosure;

[0072] Fig.17 A schematic diagram of another address array for a fault condition provided by an embodiment of the present disclosure;

[0073] Fig.18 Another schematic diagram of a classified address array provided in an embodiment of the present disclosure;

[0074] Fig.19 A schematic diagram of an update address array provided in an embodiment of the present disclosure;

[0075] Fig. 20 A block diagram of a nuclear classification device provided by an embodiment of the present disclosure;

[0076] Fig.21 A block diagram of a mapping device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0077] In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the exemplary embodiments of the present disclosure are described below in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0078] In the absence of conflict, the various embodiments of the present disclosure and the various features therein may be combined with each other.

[0079] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0080] The terms used herein are only used to describe specific embodiments and are not intended to limit the present disclosure. As used herein, the singular forms "a" and "the" are also intended to include plural forms, unless the context clearly indicates otherwise. It will also be understood that when the terms "including" and / or "made of" are used in this specification, the presence of the features, wholes, steps, operations, elements and / or components is specified, but the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof is not excluded. "Connected" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0081] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless explicitly defined as such herein.

[0082] Figure 1 A flowchart of a nuclear classification method provided in an embodiment of the present disclosure.

[0083] Reference Figure 1 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0084] Each core in the chip has an address (or ID (Identity document)), which can be abstracted as coordinates. The addresses of multiple cores in the chip are arranged in a preset arrangement. For example, the addresses are arranged in a matrix to form a rectangular array composed of rows and columns. Each address is an element in the matrix, but the specific content of the matrix can be different, such as the size of the matrix (the number of rows and columns of the matrix, that is, the number of addresses in the row direction and the column direction). Data can be transmitted between two cores whose addresses are adjacent in the row direction of the matrix or in the column direction of the matrix.

[0085] It should be noted that the addresses can also be arranged in other arrangements such as a three-dimensional body. In this case, each original address is a coordinate in the three-dimensional body, but the size of the three-dimensional body (the length, width, and height of the three-dimensional body, that is, the number of original addresses in the three directions of X-axis, Y-axis, and Z-axis) can be different.

[0086] Reference Fig.14 ,like Fig.14 The figure shows an array of core addresses, with each circle representing a core. The addresses of the 16 cores are arranged in a matrix form. The addresses of the 16 cores are: (1, 1), (2, 1), (3, 1) ... (4, 4). Data can be transmitted between the core with address (1, 1) and the core with address (2, 1), and data can also be transmitted between the core with address (1, 1) and the core with address (1, 2).

[0087] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0088] S101 . Determine cores of some rows of cores arranged along a preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores.

[0089] In the core array (cores and their addresses correspond one to one, currently, the core array is the same as the array composed of the core addresses, so it is hereinafter referred to as the address array), some rows of cores in each row of cores arranged along a preset topological direction are determined as invalid rows of cores, and other rows of cores are determined as valid rows of cores, then all the cores in the address array are divided into two categories, one is invalid rows of cores, and the other is valid rows of cores.

[0090] Reference Fig.14 and Fig.15 ,like Fig.14 The address array before classification is shown. Fig.15 The following table shows the sorted address array. Fig.14 In , each circle represents a core (which may or may not include a faulty core). Fig.15In , each solid circle represents a core of a valid arrangement, and each hollow square represents a core of an invalid arrangement. It can be seen that Fig.15 Among the rows of cores arranged along the row direction shown, the cores in the third row in the column direction (i.e., the four cores with addresses (1, 3), (2, 3), (3, 3), and (4, 3) respectively) are all invalid row cores, and the cores in other rows in the column direction (i.e., the first, second, and fourth rows) are all valid row cores.

[0091] It should be noted that Fig.15 It is only used to indicate the classification of the cores in the address array after classification. The solid circle and the hollow circle only indicate whether the corresponding core is an invalid row core or a valid row core, and does not mean that the function of the core must actually have changed between invalid and valid. Moreover, after the cores in the address array are classified, the physical structure of all the cores remains unchanged.

[0092] It should be noted that when the addresses of the cores are arranged in a three-dimensional arrangement, the cores arranged along the preset topological direction should be multiple planes perpendicular to the preset topological direction, that is, multi-layer cores, and the above rows of cores should also include each layer of cores. At this time, S101 should also include: determining the cores of some layers in each layer of cores arranged along the preset topological direction as invalid layers, where the cores of other layers are all valid rows. In addition, when the addresses of the cores are arranged in other arrangements, the specific meaning of "each row of cores" should also be determined according to the specific arrangement.

[0093] After classifying the cores in the address array, some cores are determined to be cores of the invalid row, and other cores are determined to be cores of the valid row, wherein the cores of the invalid row can be either faulty cores or non-faulty cores, and the cores of the valid row must be non-faulty cores. Non-faulty cores have both algorithm functions and routing functions, while faulty cores do not have algorithm functions but have routing functions. Therefore, the cores of the valid row can both execute algorithms and perform routing, and the cores of the invalid row can perform routing regardless of whether they are faulty cores or non-faulty cores.

[0094] Currently, in the event of a faulty core, the algorithm may be mapped to the faulty core, but in fact the faulty core is no longer able to execute the algorithm. A commonly used solution is to recompile according to the specific fault situation so that the tasks obtained by compiling various pending problems (or algorithms that can solve the pending problems) will only be executed by non-faulty cores.

[0095] Using the core classification method of the embodiment of the present disclosure, some rows of cores in each row of cores arranged along a preset topological direction are determined as invalid rows of cores, and in the case of a faulty core, the faulty core must be an invalid row of cores, while the cores of other rows (excluding the faulty core) are all valid rows of cores. Based on the classification of the cores in the address array, the algorithm can be mapped only to the cores of the valid row (all the cores of the valid row can execute the algorithm), ensuring that the cores of the invalid row (the invalid row of cores may include the faulty core) are not required to execute the algorithm, so that the cores classified as valid rows of cores can still execute the algorithm normally, and both the valid and invalid rows of cores can be routed normally. There is no need to recompile according to the specific fault situation, and there is no need to directly replace the entire chip with a normal chip and then scrap the entire chip, which can ensure that the chip can still work normally, saving the maintenance cost required when a faulty core appears in a multi-core chip.

[0096] Furthermore, in the absence of a faulty core, some rows of cores in each row of cores arranged along a preset topological direction are determined to be invalid rows of cores, while the cores in other rows are all valid rows of cores. The invalid rows of cores can be reserved as backups, which is equivalent to providing a certain amount of tolerable faults to avoid the need for recompilation when a faulty core occurs.

[0097] It should be noted that since only the cores in the valid row are required to execute the algorithm and the cores in the invalid row do not execute the algorithm, the routing topology of the cores in the chip may change during the data transmission process.

[0098] Figure 2 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0099] Reference Figure 2 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0100] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0101] S201: Determine a preset topology direction.

[0102] When determining that some rows of cores in each row of cores arranged along a certain (or certain) topological direction are invalid rows of cores, certain preset conditions need to be met. For example, the preset conditions may be that the invalid rows of cores must include all faulty cores, the number of invalid rows of cores cannot be too large, the size of the array formed by the cores that are not determined as invalid rows of cores (i.e., valid rows of cores) when arranged in a preset arrangement must not be less than a preset size, etc.

[0103] However, if some rows of cores in the rows of cores arranged along a certain (or some) topological directions are determined as invalid rows of cores, the preset conditions may not be met. Therefore, a preset topological direction needs to be determined from the topological directions.

[0104] For example, when the addresses are arranged in a matrix form, the topological direction of the matrix includes a row topological direction and a column topological direction, and the preset topological direction can be determined from the row topological direction and the column topological direction; when the addresses are arranged in a three-dimensional form, the topological direction of the three-dimensional body includes an X-axis direction, a Y-axis direction, and a Z-axis direction, and the preset topological direction can be determined from the X-axis direction, the Y-axis direction, and the Z-axis direction.

[0105] S202: Determine some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein the other rows of cores are all valid rows of cores.

[0106] Figure 3 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0107] Reference Figure 3 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0108] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0109] S301 . In the case that there is no faulty core among the multiple cores, a preset topology direction is used as a preset topology direction, and a preset number of invalid rows is used as an invalid amount of the preset topology direction.

[0110] Among them, "invalid amount" means that when determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, it is necessary to determine an invalid number of rows of cores in each row of cores arranged along the preset topological direction as invalid rows of cores. "Invalid amount" can also be understood as the number of invalid rows.

[0111] In the case where there is no faulty core, a certain topology direction can be designated (ie, pre-set) as a preset topology direction, and a certain invalid row number can be designated (ie, pre-set) as an invalid amount of the preset topology direction according to actual needs.

[0112] S302: Determine cores of some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores.

[0113] Figure 4 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0114] Reference Figure 4, an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0115] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0116] S401 . When there is a faulty core among multiple cores, determine the number of allowed faulty rows in each topological direction when each topological direction is a preset topological direction according to a preset size.

[0117] The addresses of multiple cores are arranged in a preset arrangement, and the array formed has a certain size. The preset size can be understood as that after the core classification is completed, when the addresses of the cores in the invalid row among the multiple cores are removed and the addresses of the cores in the valid row among the multiple cores are arranged in the preset arrangement, the size of the array formed should not be less than the preset size.

[0118] Here, "arranged in a preset arrangement manner" means that the addresses of multiple cores and the addresses of the cores in the effective row form an array in the same "manner", such as both forming an array composed of rows and columns; however, the size of the array formed by the addresses of multiple cores is different from the size of the array formed by the addresses of the cores in the effective row.

[0119] Among them, "size" refers to the number of core addresses included in the array in each topological direction. The preset size needs to match the preset core compilation arrangement layout, but there can be multiple compilation arrangement layouts. In the disclosed embodiment, before the chip leaves the factory, various types of pending problems have been compiled and a predetermined core compilation arrangement layout has been obtained, so that the tasks obtained by compiling the pending problems can be mapped to each core according to the core compilation arrangement layout. Therefore, the chip has a preset core compilation arrangement layout. There can be multiple core compilation arrangement layouts obtained by compiling a certain type of pending problem, and there can also be multiple core compilation arrangement layouts obtained by compiling multiple types of pending problems, but the core compilation arrangement layout is often smaller than the actual arrangement layout of all core addresses. The purpose is to avoid the need to recompile the pending problems when a faulty core appears later as much as possible.

[0120] In addition, the preset size has a certain relationship with the allowable failure rate of the cores in the chip. The cores that actually execute the algorithm are all validly arranged cores. The size of the array composed of the addresses of the validly arranged cores should not be less than the preset size. When it is less than the preset size, the number of cores that can execute the algorithm is too small, which will affect the overall computing power of the chip. For example, if the total number of cores is 100 and the allowable failure rate is 20%, when the addresses are arranged in a matrix form, the preset size can be 8X10 (indicating 8 rows and 10 columns, or 8 columns and 10 rows), and the same is true for the compilation arrangement layout of the cores.

[0121] The array formed by the addresses of the cores only has relative sizes in each topological direction within the array, but no absolute size. For example, when the addresses are arranged in a matrix form, the size in the row direction is only relative to the size in the column direction. The size of 3 rows and 4 columns is essentially equivalent to the size of 4 rows and 3 columns. Therefore, the preset size is not an absolute size. After the cores are classified, it is only necessary to make the size of the array formed by the addresses of the cores in the valid row not less than the preset size in a certain direction. Therefore, no matter from which topological direction the cores of some rows of the cores arranged along the topological direction are determined as invalid rows of cores, as long as the size of the array formed by the addresses of the cores in the valid row is not less than the preset size.

[0122] The number of allowed fault rows in each topological direction when each topological direction is the preset topological direction can be determined respectively according to the preset size.

[0123] The “allowable number of faulty rows” refers to the maximum number of rows including faulty cores that can be allowed in each row of cores arranged along the corresponding topological direction.

[0124] Assuming that the addresses of multiple cores form a matrix array of 4 rows and 4 columns and the preset size is 3X4, as mentioned above, in terms of the address array, the size of 3 rows and 4 columns is substantially equivalent to the size of 4 rows and 3 columns, then it is feasible to use the row direction as the preset topological direction or the column direction as the preset topological direction. When the row direction is the preset topological direction, the allowed number of faulty rows in the row direction is 1, and the allowed deletion amount in the column direction is 0. When the column direction is the preset topological direction, the allowed number of faulty rows in the row direction is 0, and the allowed number of faulty rows in the column direction is 1. Regardless of whether one row of cores including faulty cores is allowed in each row of cores arranged along the row direction, or one row of cores including faulty cores is allowed in each row of cores arranged along the row direction, the size of the array formed by the addresses of the cores in the other rows can be made not to exceed the preset size of 3X4.

[0125] It should be noted that, when there is no faulty core among the multiple cores, the preset topological direction and the number of invalid rows may also be determined according to the preset size.

[0126] S402: Determine the actual number of faulty rows in each topological direction, where the actual number of faulty rows is the number of rows including faulty cores in each row of cores arranged along the corresponding topological direction.

[0127] Reference Fig.16 ,like Fig.16The figure shows an address array, in which each solid circle represents a non-faulty core, and each hollow circle represents a faulty core. It can be seen that the four cores with addresses (3, 1), (2, 2), (4, 3), and (3, 4) are faulty cores. Among the rows of cores arranged along the row topology direction, the rows including the faulty cores are the first, second, third, and fourth rows in the column direction, that is, the actual number of faulty rows in the row topology direction is 4; among the rows of cores arranged along the column topology direction, the rows including the faulty cores are the second, third, and fourth rows in the row direction, that is, the actual number of faulty rows in the column direction is 3.

[0128] S403: The topology direction in which the actual number of fault rows does not exceed the allowed number of fault rows is used as the preset topology direction.

[0129] Once the actual number of faulty rows in a certain topological direction exceeds the allowed number of faulty rows in that topological direction, all rows of cores including the faulty cores in the rows of cores arranged along the topological direction need to be determined as invalid rows of cores, which will cause the size of the array composed of the cores in the valid rows to be smaller than the preset size, resulting in too few cores that can execute the algorithm, affecting the overall computing power of the chip. Therefore, it is necessary to select the topological direction whose actual number of faulty rows does not exceed the allowed number of faulty rows as the preset topological direction.

[0130] In the embodiments of the present disclosure, the meaning of "allowable number of fault rows" is that when the actual number of fault rows in a certain topological direction exceeds the allowable number of fault rows in the topological direction, the number of cores in the fault row is too large and the number of cores that can execute the algorithm is too small, and the chip may need to be scrapped. Therefore, after determining the preset topological direction, the allowable number of fault rows in the preset topological direction can be directly determined as the invalid amount of the preset topological direction. Of course, the actual number of fault rows in the preset topological direction can also be determined as the invalid amount of the preset topological direction, and the embodiments of the present disclosure do not specifically limit this.

[0131] S404: Determine some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein the other rows of cores are all valid rows of cores.

[0132] Figure 5 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0133] Reference Figure 5 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0134] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0135] S501 . When there is a faulty core among the multiple cores, determine the actual number of faulty rows in each topological direction, where the actual number of faulty rows is the number of rows including the faulty core in each row of cores arranged along the corresponding topological direction.

[0136] In the case of a faulty core, the actual number of faulty rows in each topological direction is determined separately.

[0137] S502: Determine the topology direction with the smallest actual number of fault rows as the preset topology direction.

[0138] The minimum actual fault row number is determined among the actual fault row numbers in each topology direction, and the topology direction corresponding to the minimum actual fault row number is determined as the preset topology direction.

[0139] Reference Fig.17 ,like Fig.17 The figure shows an address array, where each solid circle represents a non-faulty core, and each hollow circle represents a faulty core. It can be seen that the two cores with addresses (3, 1) and (3, 4) are faulty cores. Then, among the rows of cores arranged along the row topology direction, the rows including the faulty cores are the first and fourth rows in the column direction, that is, the actual number of faulty rows in the row topology direction is 2; among the rows of cores arranged along the column topology direction, the row including the faulty cores is only the third row in the row direction, that is, the actual number of faulty rows in the column direction is 1. At this time, the topology direction with the smallest actual number of faulty rows is the column direction, and the column direction can be determined as the preset topology direction.

[0140] It should be understood that if the topological direction with the smallest actual number of faulty rows is determined as the preset topological direction, then the actual number of faulty rows in this topological direction should also be smaller than the number of rows of cores arranged along this topological direction. Otherwise, it is meaningless to use this topological direction as the preset topological direction, and the chip may still need to be scrapped.

[0141] After the preset topology direction is determined, the actual number of faulty rows in the preset topology direction may be directly determined as the invalid amount of the preset topology direction.

[0142] S503: Determine cores of some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores.

[0143] Using the core classification method of the embodiment of the present disclosure, the topological direction with the smallest number of actual faulty rows is determined as the preset topological direction from each topological direction. As long as at least one row of cores in each row of cores arranged along the preset topological direction does not include faulty cores, the chip does not need to be directly scrapped. When the number of rows of cores in each row of cores arranged along the preset topological direction that do not include faulty cores is too small (but greater than 0), it is possible to consider changing the application scenario of the chip and applying the chip to any other scenario that only requires a small number of cores to work together, rather than directly scrapping the entire chip.

[0144] Figure 6 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0145] Reference Figure 6 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0146] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0147] S601. When there is no faulty core in any row of cores arranged along a preset topological direction, an invalid number of rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid row of cores; wherein the invalid number is preset, and the cores in other rows are all valid row of cores.

[0148] When there is no faulty core in the chip, the invalid number of rows of cores can be determined as invalid row of cores among the rows of cores arranged along the preset topological direction according to the preset topological direction and invalid number, but the invalid number of rows of cores need to be located at a predetermined position.

[0149] In some embodiments, the invalid number of rows of cores located at the predetermined position are the last invalid number of rows of cores along a topological direction perpendicular to a preset topological direction.

[0150] When there is no faulty core, the last invalid rows of cores in each row of cores arranged along the preset topological direction may be preferentially determined as invalid rows of cores, where “last” refers to a topological direction perpendicular to the preset topological direction.

[0151] Reference Fig.18 ,like Fig.18 The figure shows the classified address array, where each solid circle represents a core in a valid row, and each hollow square represents a core in an invalid row. Taking the preset topological direction as the column direction and the invalidity amount in the preset topological direction as 1 as an example, it can be seen that in the absence of a faulty core, the core of the last row along the topological direction perpendicular to the column direction (i.e., along the row direction) in each row of cores arranged along the column direction can be determined as the core of an invalid row, and the original addresses of the cores of the invalid row are: (4, 1), (4, 2), (4, 3), (4, 4), respectively.

[0152] It should be noted that the invalid number of rows of cores located at predetermined positions are cores that do not include faulty cores in the starting invalid number of rows along the topological direction perpendicular to the preset topological direction, or cores that do not include faulty cores in the middle invalid number of rows along the topological direction perpendicular to the preset topological direction, both of which are feasible.

[0153] Figure 7 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0154] Reference Figure 7, an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0155] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0156] S701: When there is a faulty core in any row of cores arranged along a preset topological direction, at least the cores in the row including the faulty core among the rows of cores arranged along the preset topological direction are determined as invalid rows of cores, wherein the cores in other rows are all valid rows of cores.

[0157] When there is a faulty core in any row of cores arranged along a preset topological direction, at least all the cores in the row including the faulty core can be determined as invalid row cores among the rows of cores arranged along the preset topological direction according to the preset topological direction, and all other cores that have not been determined as invalid row cores are valid row cores.

[0158] Figure 8 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0159] Reference Figure 8 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0160] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0161] S801. When there is a faulty core in any row of cores arranged along a preset topological direction, when the number of rows including the faulty core in each row of cores arranged along the preset topological direction is less than the invalid amount in the preset topological direction by N, the cores in the row including the faulty core in each row of cores arranged along the preset topological direction are determined as invalid rows of cores, and N rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid rows of cores.

[0162] Among them, the invalid amount is pre-set, and the cores in other rows are all valid rows of cores.

[0163] When there is a faulty core in any row of cores arranged along a preset topological direction and the invalid amount in the preset topological direction is preset, if, in each row of cores arranged along the preset topological direction, the number of rows including the faulty core is less than the invalid amount in the preset topological direction by N, then not only all the cores in the row including the faulty core need to be determined as invalid row cores, but also the cores in N rows located at predetermined positions need to be determined as invalid row cores, and all other cores that have not been determined as invalid row cores are valid row cores.

[0164] In some embodiments, the N rows of cores located at predetermined positions are the last N rows of cores that do not include a faulty core along a topological direction perpendicular to a preset topological direction.

[0165] The last invalid number of rows of cores in each row of cores arranged along the preset topological direction may be determined as the invalid row of cores, where “last” refers to the topological direction perpendicular to the preset topological direction.

[0166] When there is a faulty core in any row of cores arranged along the preset topological direction and the invalid amount of the preset topological direction is preset, if the number of rows including faulty cores in each row of cores arranged along the preset topological direction is less than the invalid amount of the preset topological direction by N, then not only all the cores in the row including the faulty core need to be determined as invalid row cores, but also the last N rows of cores in each row of cores arranged along the preset topological direction need to be determined as invalid row cores, where "last" is relative to the topological direction perpendicular to the preset topological direction. All other cores that are not determined as invalid row cores are valid row cores.

[0167] It should be noted that the N rows of cores located at predetermined positions are cores that do not include faulty cores in the starting N rows along the topological direction perpendicular to the preset topological direction, or cores that do not include faulty cores in the middle N rows along the topological direction perpendicular to the preset topological direction, both of which are feasible.

[0168] Fig. 9 A flowchart of another nuclear classification method provided in an embodiment of the present disclosure.

[0169] Reference Fig. 9 , an embodiment of the present disclosure provides a core classification method for a chip, the chip comprising a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement.

[0170] The nuclear classification method of the embodiment of the present disclosure specifically includes:

[0171] S901. Determine cores of some rows of cores arranged along a preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores.

[0172] S902 . Update the original addresses of the cores in the valid row to updated addresses according to the original addresses of the cores in the invalid row, wherein the arrangement of the updated addresses is the same as that of the original addresses, and the cores in the invalid row have no updated addresses.

[0173] The original address is the address that has not been updated. The updated address of the core of the valid row can be the same as its original address or different from its original address.

[0174] Reference Fig.15 and Fig.19 ,like Fig.15 The following is the original address array after classification: Fig.19Shown is an update address array of cores in a valid rank, where each solid circle represents a core in a valid rank, and each hollow square represents a core in an invalid rank. Fig.19 In the example, when there are two brackets above the core of the valid row to represent the address, the upper address is the original address, and the lower address is the updated address (if the original address and the updated address are the same, only one bracket is used to represent it). Fig.15 The original address array shown has an invalid row of cores. Fig.19 The update address array shown does not include cores in an invalid row, indicating that cores in an invalid row do not have update addresses, and original addresses of cores in some valid rows are different from update addresses.

[0175] It should be noted that Fig.19 It only represents the array composed of the updated addresses of the cores of all valid rows after the original addresses of the cores of the valid rows are updated, and does not mean that the actual number of cores is reduced after the original addresses of the cores of the valid rows are updated, that is, the physical structure of all cores remains unchanged after the original addresses of the cores of the valid rows are updated.

[0176] In some embodiments, after determining that cores in some rows of cores arranged along a preset topological direction are invalid rows of cores, the arrangement of the addresses of the plurality of cores remains unchanged.

[0177] After determining that some rows of cores in each row of cores arranged along the preset topological direction are invalid rows of cores, the addresses of the cores in the valid rows may not be updated. Regardless of whether they are valid rows of cores or invalid rows of cores, the addresses of all cores still constitute an address array. After the core classification is completed, the address array does not change.

[0178] Fig.10 A flowchart of a mapping method provided in an embodiment of the present disclosure.

[0179] Reference Fig.10 , an embodiment of the present disclosure provides a mapping method for a chip whose cores have been classified by the aforementioned core classification method.

[0180] The mapping method of the embodiment of the present disclosure specifically includes:

[0181] S1001. Map multiple tasks to cores in a valid row according to the classification of cores in the chip.

[0182] After the chip has classified the cores according to the aforementioned core classification method, multiple tasks can be mapped to the cores of each valid row according to the classification of the cores in the chip. The cores in the chip are classified into cores of valid rows and cores of invalid rows. The cores of valid rows can all execute algorithms. After mapping multiple tasks to the cores of valid rows, each core can process the tasks separately and work together normally.

[0183] In some embodiments, before mapping the multiple tasks to the cores of the effective row according to the classification of the cores in the chip, the problem to be processed may be compiled according to the classification of the cores in the chip to obtain the multiple tasks.

[0184] The cores in the chip are classified into cores in a valid row and cores in an invalid row. The cores in a valid row can all execute algorithms. Therefore, when compiling the problem to be processed (or the algorithm that can solve the problem to be processed), the problem to be processed can be compiled into multiple tasks according to the classification of the cores, so that all tasks can be executed by the cores in a valid row.

[0185] Fig.11 A flowchart of a data transmission method provided in an embodiment of the present disclosure.

[0186] Reference Fig.11 , an embodiment of the present disclosure provides a data transmission method for a chip whose cores have been classified according to the aforementioned core classification method.

[0187] The data transmission method of the embodiment of the present disclosure specifically includes:

[0188] S1101. Acquire data, where the data includes address information of a destination core.

[0189] Regardless of whether there is a faulty core in the chip, the multiple cores in the chip at least have a routing function, so that data can be transmitted normally. In response to obtaining the data, the destination core address information carried in the data can be obtained first.

[0190] S1102. Transmit data to a destination core at least according to the address information.

[0191] After the chip has classified the cores according to the aforementioned core classification method, the cores in the valid row must be able to execute the algorithm. Therefore, the task obtained by compiling the problem to be processed (or the algorithm that can solve the problem to be processed) will be mapped to the cores in the valid row. Therefore, when transmitting data, the destination core is the core in the valid row, and at least based on the destination core address information, the data can be transmitted to the destination core.

[0192] Fig.12 A flowchart of another data transmission method provided by an embodiment of the present disclosure.

[0193] Reference Fig.12 The embodiment of the present disclosure provides a data transmission method for a chip that has classified cores according to the aforementioned core classification method, and after the core classification is completed, the arrangement of the addresses of the multiple cores remains unchanged and the addresses of the cores have not been updated.

[0194] The data transmission method of the embodiment of the present disclosure specifically includes:

[0195] S1201. Acquire data, where the data includes address information of a destination core.

[0196] The address information is displacement information, and the displacement information includes the displacement amount along each topological direction.

[0197] The address information of the destination core can be understood as the displacement that the data needs to pass through in each topological direction in the array composed of the core addresses, starting from the current core, in order to reach the destination core.

[0198] S1202. In an array formed by the addresses of the cores, data is transmitted in each topological direction by a corresponding displacement according to the displacement information, wherein when the data passes through the cores of an invalid row along a topological direction perpendicular to the preset topological direction, the displacement in the topological direction perpendicular to the preset topological direction is ignored.

[0199] The displacement information includes the displacement amount along each topological direction. For example, when the addresses are arranged in a matrix form, the displacement information includes the displacement amount in the row topological direction and the displacement amount in the column topological direction; when the addresses are arranged in a three-dimensional form, the displacement information includes the displacement amount in the X-axis topological direction, the displacement amount in the Y-axis topological direction, and the displacement amount in the Z-axis topological direction.

[0200] In the array composed of the addresses of the cores, starting from the source core, the data needs to be transmitted in each topological direction according to the displacement information. Furthermore, if during the data transmission process, the data needs to pass through the cores of an invalid row along a topological direction perpendicular to the preset topological direction during a certain displacement, the displacement in the topological direction perpendicular to the preset topological direction is not calculated in the displacement.

[0201] Assume that in the process of transmitting data from the source core to the destination core, it needs to pass through core M and core N in turn, and needs to pass through core M first, and then core M transmits the data to core N. Then during the transmission process, when core M determines that the core is going to transmit data to core N, core M can modify the displacement in the corresponding topological direction in the destination core address information, so that the destination core address information obtained by core N is the displacement information of the destination core relative to core N. Of course, if core N is a core in an invalid row, and core M passes through core N in a topological direction perpendicular to the preset topological direction, no modification is required. At this time, if the displacements in the destination core address information obtained by core N are all 0, it means that core N is the destination core and the data transmission is completed.

[0202] Alternatively, core N obtains the data and the destination core address information carried in the data, and determines that the data is transmitted from core M to core N, and core N modifies the displacement in the corresponding topological direction in the destination core address information. The modified destination core address information is the displacement information of the destination core relative to core N. If core N determines that each displacement in the modified destination core address information is 0, it means that core N is the destination core and the data transmission is completed. If core N is not the destination core, core N continues to transmit data.

[0203] Reference Fig.15 ,like Fig.15 The following is the classified address array. Each solid circle represents a core in the valid row, and each hollow square represents a core in the invalid row. Fig.15 Among the rows of cores arranged along the row direction shown, the cores in the third row in the column direction (i.e., the four cores with addresses (1, 3), (2, 3), (3, 3), and (4, 3) respectively) are all invalid row cores, and the cores in other rows in the column direction (i.e., the first, second, and fourth rows) are all valid row cores.

[0204] Assume that the core with address (1, 2) obtains data, and the destination core address information obtained is (dx, dy) = (1, 1). At this time, the destination core address information does not mean that the absolute address of the destination core is (1, 1), but that the displacement information of the destination core relative to the core with address (1, 2) is (1, 1). That is to say, starting from the core with address (1, 2), the data needs to be transmitted by a displacement in both the row topology direction and the column topology direction in order for the data to reach the destination core. The core with address (1, 2) transfers data to the core with address (2, 2), and the displacement information of the destination core relative to the core with address (2, 2) is (0, 1). The core with address (2, 2) transfers data to the core with address (2, 3). Since the data passes through the core with address (2, 3) in the column direction and the core with address (2, 3) is a core of an invalid row, the address information of the destination core is not modified, and the displacement information of the destination core relative to the core with address (2, 3) is still (0, 1). The core with address (2, 3) transfers data to the core with address (2, 4), and the displacement information of the destination core relative to the core with address (2, 4) is (0, 0). The core with address (2, 4) is the destination core, and the data transmission is completed.

[0205] Fig.13 A flowchart of another data transmission method provided by an embodiment of the present disclosure.

[0206] Reference Fig.13The embodiment of the present disclosure provides a data transmission method for a chip that has classified its cores according to the aforementioned core classification method, and after the core classification is completed, only the addresses of the valid cores are updated, and the arrangement of the updated addresses is the same as the arrangement of the original addresses (i.e., the addresses that have not been updated).

[0207] The data transmission method of the embodiment of the present disclosure specifically includes:

[0208] S1301, obtaining data, the data including address information of a destination core, wherein the address information is an update address.

[0209] Only the original addresses of the cores in the valid row in the chip are updated. Only the cores in the valid row have the updated addresses. The cores in the valid row can execute the algorithm, and the tasks will be mapped to the cores in the valid row. Therefore, when transmitting data, the destination core is also the core in the valid row, that is, the core with the updated address. The destination core address carried by the data can be the destination core update address.

[0210] S1302 . According to the update address of each core, transmit the data to the destination core.

[0211] When transmitting data, the destination core is the core of the valid row. The core of the valid row has both the update address and the original address, and the core of the invalid row has the original address. Each core can still be routed according to the original address. The destination core is determined according to the update address, and the data can be transmitted to the destination core.

[0212] Fig. 20 A block diagram of the composition of a nuclear classification device provided in an embodiment of the present disclosure.

[0213] Reference Fig. 20 The embodiment of the present disclosure provides a core classification device 2000 for classifying cores of a chip, wherein the chip includes multiple cores, and the addresses of the multiple cores are arranged in a preset arrangement.

[0214] The nuclear classification device 2000 comprises:

[0215] The processing module 2001 is used to determine cores in some rows of cores arranged along a preset topological direction as invalid rows of cores, wherein the cores in other rows are all valid rows of cores.

[0216] Fig.21 A block diagram of a mapping device provided in an embodiment of the present disclosure.

[0217] Reference Fig.21 , the embodiment of the present disclosure provides a mapping device 3000 for a chip that has been classified according to the aforementioned core classification method.

[0218] The mapping device 3000 comprises:

[0219] The mapping module 3001 is used to map multiple tasks to cores in a valid row according to the classification of the cores in the chip.

[0220] In addition, the present disclosure also provides a chip for implementing at least one of the following methods:

[0221] As mentioned above, the nuclear classification method;

[0222] As the mapping method mentioned above;

[0223] As the data transmission method mentioned above.

[0224] In addition, an embodiment of the present disclosure further provides a computer-readable medium on which a computer program is stored, wherein the computer program implements at least one of the following methods when executed by a processing core:

[0225] As mentioned above, the nuclear classification method;

[0226] As the mapping method mentioned above;

[0227] As the data transmission method mentioned above.

[0228] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0229] Example embodiments have been disclosed herein, and although specific terms are employed, they are used and should be interpreted only in a general illustrative sense and not for limiting purposes. In some instances, it will be apparent to those skilled in the art that, unless otherwise expressly noted, features, characteristics, and / or elements described in conjunction with a particular embodiment may be used alone or in combination with features, characteristics, and / or elements described in conjunction with other embodiments. Therefore, those skilled in the art will appreciate that various changes in form and detail may be made without departing from the scope of the present disclosure as set forth in the appended claims.

Claims

1. A core classification method for a chip, wherein the chip comprises a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement; the method comprises: Determine the preset topology direction; Determine some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores, wherein the other rows of cores are all valid rows of cores; The cores of the valid row include only non-faulty cores; the cores of the invalid row are all faulty cores, or the cores of the invalid row include at least one faulty core and at least one non-faulty core; when the addresses of the cores of the valid row are arranged according to the preset arrangement mode, the size of the array formed by the addresses of the cores of the valid row is greater than or equal to the preset size, and the preset size matches the preset core compilation arrangement layout of the chip; The determining of the preset topology direction includes: In the case where there is a faulty core among the multiple cores, determining, according to a preset size, the number of allowed faulty rows in each topological direction when each topological direction is a preset topological direction; Determine the actual number of faulty rows in each topological direction, where the actual number of faulty rows is the number of rows including faulty cores in each row of cores arranged along the corresponding topological direction; The topology direction in which the actual number of fault rows does not exceed the allowed number of fault rows is the preset topology direction.

2. The nuclear classification method according to claim 1, wherein: The determining of the preset topology direction includes: In the case that there is no faulty core among the multiple cores, a preset topological direction is used as the preset topological direction, and a preset number of invalid rows is used as the invalid amount of the preset topological direction.

3. The nuclear classification method according to claim 1, wherein: The determining of the preset topology direction includes: In the case where there is a faulty core among the multiple cores, determining the actual number of faulty rows in each topological direction; The topology direction with the smallest number of actual fault rows is determined as the preset topology direction.

4. The nuclear classification method according to claim 1, wherein: The step of determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores includes: When there is no faulty core in any row of cores arranged along a preset topological direction, an invalid number of rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid rows of cores; wherein the invalid number is preset.

5. The nuclear classification method according to claim 4, wherein: The invalid number of rows of cores located at the predetermined position are the last invalid number of rows of cores along the topological direction perpendicular to the preset topological direction.

6. The nuclear classification method according to claim 1, wherein: The step of determining some rows of cores in each row of cores arranged along a preset topological direction as invalid rows of cores includes: When there is a faulty core in any row of cores arranged along a preset topological direction, at least the cores of the row including the faulty core among the rows of cores arranged along the preset topological direction are determined as invalid rows of cores.

7. The nuclear classification method according to claim 6, wherein: The step of at least determining the cores of the row including the faulty core in each row of cores arranged along the preset topological direction as the cores of the invalid row includes: When the number of rows including faulty cores in each row of cores arranged along a preset topological direction is less than the invalid amount in the preset topological direction by N, the cores in the rows including faulty cores in each row of cores arranged along the preset topological direction are determined as invalid rows of cores, and N rows of cores located at predetermined positions in each row of cores arranged along the preset topological direction are determined as invalid rows of cores; wherein the invalid amount is set in advance.

8. The nuclear classification method according to claim 7, wherein: The N rows of cores located at predetermined positions are the last N rows of cores along a topological direction perpendicular to the preset topological direction and excluding any faulty cores.

9. The nuclear classification method according to any one of claims 1 to 8, wherein: After determining that cores of some rows of cores in each row of cores arranged along the preset topological direction are invalid rows of cores, the method further includes: According to the original addresses of the cores in the invalid row, the original addresses of the cores in the valid row are updated to updated addresses, wherein the arrangement of the updated addresses is the same as that of the original addresses, and the cores in the invalid row have no updated addresses.

10. The nuclear classification method according to any one of claims 1 to 8, wherein: After determining the cores of some rows of cores in the rows of cores arranged along the preset topological direction as invalid rows of cores, the arrangement manner of the addresses of the plurality of cores remains unchanged.

11. A mapping method for a chip whose cores have been classified according to the core classification method according to any one of claims 1 to 10, the mapping method comprising: According to the classification of the cores in the chip, multiple tasks are mapped to the cores of the effective row.

12. The mapping method according to claim 11, wherein: Before mapping the plurality of tasks to the cores of the effective row according to the classification of the cores in the chip, the method further includes: According to the classification of the cores in the chip, the problem to be processed is compiled to obtain the plurality of tasks.

13. A data transmission method, for use with a chip whose cores have been classified by the core classification method according to any one of claims 1 to 10, the method comprising: Acquiring data, wherein the data includes address information of a destination core; The data is transmitted to a destination core at least based on the address information.

14. The data transmission method according to claim 13, used for a chip whose cores have been classified by the core classification method according to claim 10, wherein the address information is displacement information, and the displacement information includes the displacement amount along each topological direction; the data is transmitted to the destination core at least according to the address information, comprising: In an array composed of core addresses, the data is transmitted with corresponding displacements in each topological direction according to the displacement information, wherein when the data passes through an invalid row of cores along a topological direction perpendicular to a preset topological direction, the displacement in the topological direction perpendicular to the preset topological direction is ignored.

15. The data transmission method according to claim 13, used for a chip whose cores have been classified according to the core classification method according to claim 9, wherein the address information is an update address, and the data is transmitted to a destination core at least according to the address information, comprising: The data is transmitted to the destination core according to the update address of each core.

16. A core classification device for classifying cores of a chip, wherein the chip comprises a plurality of cores, and the addresses of the plurality of cores are arranged in a preset arrangement; the device comprises: A processing module, configured to determine a preset topological direction, and determine cores of some rows of cores arranged along the preset topological direction as invalid rows of cores, wherein cores of other rows are all valid rows of cores; The cores of the valid row include only non-faulty cores; the cores of the invalid row are all faulty cores, or the cores of the invalid row include at least one faulty core and at least one non-faulty core; when the addresses of the cores of the valid row are arranged according to the preset arrangement mode, the size of the array formed by the addresses of the cores of the valid row is greater than or equal to the preset size, and the preset size matches the preset core compilation arrangement layout of the chip; The determining of the preset topology direction includes: In the case where there is a faulty core among the multiple cores, determining, according to a preset size, the number of allowed faulty rows in each topological direction when each topological direction is a preset topological direction; Determine the actual number of faulty rows in each topological direction, where the actual number of faulty rows is the number of rows including faulty cores in each row of cores arranged along the corresponding topological direction; The topology direction in which the actual number of fault rows does not exceed the allowed number of fault rows is the preset topology direction.

17. A mapping device for a chip for classifying nuclei according to the nuclei classification method according to any one of claims 1 to 10, comprising: A mapping module is used to map multiple tasks to cores in a valid row according to the classification of the cores in the chip.

18. A chip comprising a plurality of cores, wherein the chip is configured to implement at least one of the following methods: The nuclear classification method according to any one of claims 1 to 10; The mapping method according to any one of claims 11 to 12; A data transmission method as claimed in any one of claims 13 to 15.

19. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by the processing core, at least one of the following methods is implemented: The nuclear classification method according to any one of claims 1 to 10; The mapping method according to any one of claims 11 to 12; A data transmission method as claimed in any one of claims 13 to 15.