CN114116005B - Immediate data storage method based on AIGPU architecture - Google Patents
Immediate data storage method based on AIGPU architecture Download PDFInfo
- Publication number
- CN114116005B CN114116005B CN202111432949.5A CN202111432949A CN114116005B CN 114116005 B CN114116005 B CN 114116005B CN 202111432949 A CN202111432949 A CN 202111432949A CN 114116005 B CN114116005 B CN 114116005B
- Authority
- CN
- China
- Prior art keywords
- instruction
- immediate
- temporary
- helper
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/30181—Instruction operation extension or modification
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
- G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
- G06F7/57—Arithmetic logic units [ALU], i.e. arrangements or devices for performing two or more of the operations covered by groups G06F7/483 – G06F7/556 or for performing logical operations
- G06F7/575—Basic arithmetic logic units, i.e. devices selectable to perform either addition, subtraction or one of several logical operations, using, at least partially, the same circuitry
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
- G06F9/3885—Concurrent instruction execution, e.g. pipeline or look ahead using a plurality of independent parallel functional units
- G06F9/3887—Concurrent instruction execution, e.g. pipeline or look ahead using a plurality of independent parallel functional units controlled by a single instruction for multiple data lanes [SIMD]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/20—Processor architectures; Processor configuration, e.g. pipelining
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- Executing Machine-Instructions (AREA)
Abstract
The invention discloses an immediate data storage method based on an AIGPU architecture.A register file stores the state of the AIGPU architecture, and an instruction reads data from the register file and writes a calculation result into a target register; in the calculation process, the instruction uses the immediate value contained in the instruction to calculate; the method is characterized in that an assistant instruction is added, the assistant instruction and the instruction are combined to obtain a composite instruction, a temporary immediate number is loaded in the composite instruction, and the temporary immediate number is used for replacing the immediate number to perform instruction calculation. The length of the immediate in a normal instruction is short, while the immediate in a helper instruction may be longer. The helper instruction provided by the invention is mainly used for expanding the immediate number in the instruction, so the implementation is simple.
Description
Technical Field
The invention relates to the field of data storage, in particular to an immediate data storage method based on an AIGPU architecture.
Background
SIMT parallel computing
SIMT (single instruction stream multithreading) is a high performance computer architecture particularly suited for data parallel computing. SIMT was first introduced by nVidia corporation and is an extension of the SIMD (single instruction stream multiple data stream) architecture.
The basic execution unit is a thread (thread), and multiple threads can be combined into one block. All threads in a block execute the same instruction sequence.
SIMT treats a sequence of instructions operating on each element of a long vector as a thread, and threads in a block execute the same instructions synchronously, thus achieving data parallel computation.
A data computation channel in SIMT may also execute several threads concurrently, each belonging to a different warp or a different block, each with its own register file.
SIMT computers allow for branch execution of a single thread, with the sequence of instructions executed by each thread differing by the state of the respective thread.
SIMT corresponds to a data parallel programming model, and enjoys a flexible mapping mode from any vector to a fixed structure machine. For performance considerations, it is desirable to reduce thread branching in SIMT computations, thereby maintaining system utilization and reducing power consumption. The SIMT architecture can reduce access latency through warp switching and stream processing.
AIGPU architecture profiling
The top level architecture of the AIGPU chip is shown in FIG. 1. The AIGPU chip is characterized in that,
it employs on-chip virtual storage, a distinctive storage system.
It uses coarse-grained computational partitioning and data management to resolve data dependencies, unlike superscalar hardware machines or codelet software data streams.
The way of dependency management of AIGPU coarse grain allows implementation using commands, providing a means for software architects and programmers to autonomously control and optimize programs.
The chip is controlled by the host (host) and its operation. The host sends the commands to be executed to the AIGPU device, where the SP analyzes the commands and then hands over a series of operations to the PE and DMA for execution.
The AIGPU has the important characteristics that the on-chip virtual storage adopts a segmented structure, tensor storage can be directly supported, storage data is structured, and the storage data has definite tensor semantics. A tensor is typically stored in a segment. A Kernel program using several tensors can have several segments. One implementation is 16 segments per program.
A kernel may be distributed to several PEs for execution, each PE performing a partial computation and using a portion of the segment.
Each segment has its own independent address space.
Each segment may define a 1D,2D,3D or 4D tensor storage structure.
The storage address is formed by segment _ id RFoffset immediatefoffset one segment containing several pages.
Each page is multidimensional, 1D,2D,3D, or 4D; is a tensor that is the same dimension as the segment.
Each page has an entry in the page table that can be used for address translation and to check if the page is ready; after a PE has used the page, the value of v-counter is decremented.
Segment store Structure-the store segment structure is used by Kernel programs.
1. The virtual storage is organized in segments (segments), and each Kernel may use several segments.
2. Each segment may contain a number of pages (pages), the number of pages within a segment being variable, as determined by the application.
3. To take full advantage of on-chip storage, the size of the page is variable, determined by the application.
The chip contains page tables to manage on-chip virtual storage. A typical program execution includes the following steps,
1. transporting data on off-chip storage to on-chip storage
2. Building page table contents and initial state
3. Starting and executing a program
4. Writing result data to external storage when program execution is completed
The execution steps of the program are similar to the program execution of a classic GPU, but the exchange of data is between the external storage and the on-chip storage of the device. Of course, the resulting data is sometimes written to the host side storage. The advantages of on-chip virtual storage include,
1. the access of data can be optimized by controlling the data transfer between on-chip storage and off-chip storage by a program.
2. Hardware manages the use of page tables.
3. Dynamic overlap of data store allocation with program execution (streaming) may improve performance.
The management of page tables and data usage status are updated by the SP and PE according to program execution. The address within each segment is computed independently of the linear address, with the starting linear address defined by an anchor (ankor) and the address within the segment defined relative to the anchor, and each PE may have its own relative anchor.
Immediate load and helper instruction Immediate temp and helper instruction
The register file stores the architectural state of the processor. The instruction reads data from the register file and writes the result of the computation into the target register. Instructions may also be calculated using immediate values contained in the instructions, but the size range of the immediate is limited by the length of the instruction.
Disclosure of Invention
The present invention addresses the problem of the background art in which the size range of immediate values is limited by the length of instructions, and proposes that in the AIGPU, an immediate value be loaded using an instruction that has a short timeout state and is called a "tentative immediate temp," and a "helper" instruction.
The technical scheme is as follows:
a immediate data storage method based on AIGPU architecture, the register file stores the state of AIGPU architecture, the instruction reads the data from the register file, and write the calculation result into the target register; in the calculation process, the instruction uses the immediate value contained in the instruction to calculate;
and adding a helper instruction, combining the helper instruction and the instruction to obtain a composite instruction, loading a temporary immediate number in the composite instruction, and performing instruction calculation by using the temporary immediate number instead of the immediate number.
The instruction and helper instruction are determined by a particular flag bit: an instruction is when a particular bit is 0 and a helper instruction is when it is 1.
As a first instruction/helper instruction structure, the instruction includes an operand B for specifying a helper instruction; the helper instruction comprises a temporary immediate of long data, and the length of the temporary immediate meets the range required by program data calculation;
combining the instruction with the helper instruction based on operand B, and performing instruction calculation by using a temporary immediate in the helper instruction instead of the immediate.
As a second instruction/helper instruction structure, the instruction further includes: result register R D Temporary immediate data in helper instruction is fed into result register R D The instruction calculation is performed instead of the immediate.
The first implementation method for loading the immediate, wherein the instruction calculation using the temporary immediate instead of the immediate comprises the following steps:
SA1, writing variable data into a temporary immediate data by an assistant instruction, and setting the temporary immediate data to be effective;
SA2, the next instruction replaces the immediate contained in the instruction with the temporary immediate and consumes the temporary immediate, which is set to invalid.
In another case: SA2, the temporary immediate is not used in the next instruction, and the temporary immediate is set to be invalid.
As a third instruction/helper instruction structure, the instructions may also includeComprises the following steps: an opcode ALU, an operand A representing an immediate, an operand B specifying a helper instruction, a result register R D The temporary immediate is obtained after the immediate in the instruction is compounded with the immediate in the helper instruction, and the temporary immediate is sent to the result register R D 。
The third method for implementing loading immediate, said using temporary immediate to replace immediate to calculate instruction includes the following steps:
SB1, an assistant instruction reads the immediate number;
SB2, next instruction reads immediate value of complete helper instruction, simultaneously reads immediate value of instruction operand A, uses the two data to calculate instruction to obtain temporary immediate value, and sends it into result register R D 。
As a fourth instruction/helper instruction structure, the helper instruction further includes: an operation code ALU, an operand A, an operand B and a result register; and the plurality of assistant instructions perform composite operation of corresponding operands according to the operation codes ALU, finally obtain immediate data and send the immediate data into a result register.
The helper instruction comprises a temporary immediate number of long data, the temporary immediate number is written into a temporary immediate number register, and the temporary immediate number replaces the immediate number to carry out instruction calculation.
The invention has the advantages of
The length of the immediate in a normal instruction is short, while the immediate in a helper instruction may be longer. The helper instruction provided by the invention is mainly used for expanding the immediate number in the instruction, and the method is simple to implement, improves the storage efficiency and can expand the calculation range and precision.
This approach can also be used to expand other instructions to form more complex compound instructions.
Drawings
FIG. 1 is a diagram of the top level architecture of the AIGPU in the background art
Detailed Description
The invention is further illustrated by the following examples, without limiting the scope of the invention:
the present invention provides a new immediate data loading mechanism and a related (assistant) instruction, and the mechanism can be used for expanding other instructions to have more formats, so that the flexibility of the instruction is greatly increased, and only small hardware changes are needed.
The method of the present invention using the close coupling of helper instruction and other instruction can provide longer immediate, and can also be used to form more complex compound instruction, which is a method of extended instruction in a broad sense.
There are several ways to load the immediate, and specific two implementation examples are given below.
In the first example of implementation, the helper instruction writes the variable data into a temporary immediate (the data contained in the helper instruction is long enough); the next instruction uses this temporary immediate to replace the immediate contained in the instruction and consumes the temporary immediate. The temporary immediate becomes invalid after it is used. The temporary immediate is also set to valid when written by the helper instruction and the next instruction is considered to always use this temporary value regardless of whether the immediate is used in the next instruction. The temporary immediate is always used and invalidated by the next instruction.
An immediate grow complex instruction. The original instruction set has a limited number of immediate bits and cannot reach the required immediate length. For example, the original instruction length is 32 bits, and the immediate portion is only 8 bits, which cannot satisfy the calculation requirement. At this point, a "helper" instruction may be used, where the immediate portion is 24 bits, which may satisfy the computational requirements. The "helper" instruction and the "helped" instruction may be in the following format,
in the above table, H is the flag of the helper instruction, and the type is usually 1-3 bits; n is the number of assisted commands that are to be commanded, may be most arithmetic logic instructions (+, -, ×, ÷ div,>、＝＝、<、<<、>>Etc., the operation code is expressed by ALU), R A Is one operand of the ALU instruction, the other operand beingFlag is marked to specify that it is a 24-bit immediate in the helper instruction, the result being fed into register R D In (1). Our composition is broad, not just some specific instructions.
In a second implementation example, the helper instruction reads the source data from the register RF and writes it to a temporary immediate. The next instruction may read the full immediate value instead of the partial immediate value contained in the instruction, and simultaneously read the operands in the RF to perform the operation using the two data.
And compounding the comparison instruction. Many programs will combine the results of multiple comparisons using logical operations as a decision condition for conditional statements, such as the following conditional statements,
if(A>B&&C<＝D&&E＝＝F)
x＝y+1；
else
x＝y；
the above conditions contain three comparisons, the results of which are "and" together.
| Instructions | Types of | Operation code | Operand A | Operand B | Result register |
| Compound instruction | 2 | > | R A | R B | Temp&& |
| Compound instruction | 2 | <＝ | R C | R D | Temp&& |
| End instruction | 0 | ＝＝ | R E | R F | R G |
The first two compare instructions are of type 2, indicating that their results are AND 'ed' with the results of one instruction.
Helper instructions may be marked with a special opcode or generic instructions and helper instructions may be marked with a specific bit. With the former method, only one immediate is needed, and neither the result register is needed, since the helper immediate is placed in a specific temporary register. When the second method is used, the specific bit is 0, which is a normal instruction, and 1, which is a helper instruction.
The temporary immediate is a method for expanding other instructions, and the helper instruction of the temporary immediate can be combined with other various instructions to compound more instruction formats, so that the flexibility of the instructions is greatly increased. The assistant instruction method is different from the multi-length instruction set of ARM and RISC-V and the mode of CISC (Complex instruction), the application range is wider, and the hardware realization is very simple.
The temporary immediate helper instruction method couples a simple helper instruction next-hop instruction tightly to form a composite instruction. The helper instruction in the invention is mainly used for expanding the immediate number in the instruction, so the implementation is simple. This method of issuing may also be used to expand other instructions to form more complex compound instructions.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments, or alternatives may be employed, by those skilled in the art, without departing from the spirit or ambit of the invention as defined in the appended claims.
Claims (5)
1. A immediate data storage method based on AIGPU architecture, the register file stores the state of AIGPU architecture, the instruction reads the data from the register file, and write the calculation result into the target register; in the calculation process, the instruction uses the immediate value contained in the instruction to calculate;
adding an assistant instruction, combining the assistant instruction with the instruction to obtain a composite instruction, loading a temporary immediate number in the composite instruction, and using the temporary immediate number to replace the immediate number for instruction calculation; the instructions and helper instructions are determined by a particular flag bit: when the specific bit is 0, the instruction is a helper instruction, and when the specific bit is 1;
the instruction comprises an operand B used for specifying a helper instruction; the helper instruction comprises a temporary immediate of long data;
combining the instruction with the helper instruction based on the operand B, and performing instruction calculation by using a temporary immediate number in the helper instruction instead of the immediate number;
the instructions also include: opcode ALU, operand A representing immediate, result register R D The temporary immediate is obtained after the immediate in the instruction is compounded with the immediate in the helper instruction, and the temporary immediate is sent to the result register R D ；
The instruction calculation by using the temporary immediate in the helper instruction to replace the immediate comprises the following steps:
SB1, assistant command reads immediate;
SB2, the next instruction reads the immediate value of the complete helper instruction, and simultaneously reads the immediate value of the instruction operand A, and the temporary immediate value is obtained by using the two data to calculate the instruction and is sent to the result register R D 。
2. The method according to claim 1 wherein the instruction calculation using the temporary immediate in the helper instruction instead of the immediate comprises the steps of:
SA1, writing variable data into a temporary immediate data by an assistant instruction, and setting the temporary immediate data to be effective;
SA2, the next instruction replaces the immediate contained in the instruction with the temporary immediate and consumes the temporary immediate, which is set to invalid.
3. The method of claim 2, wherein the temporary immediate is not used in the next instruction in SA2, and wherein the temporary immediate is set to invalid.
4. The method according to claim 1, wherein the helper instructions further comprise: an operation code ALU, an operand A, an operand B and a result register; and the plurality of assistant instructions perform composite operation of corresponding operands according to the operation codes ALU, finally obtain immediate data and send the immediate data into a result register.
5. The method of claim 1 wherein the helper instruction includes a temporary immediate of the long data, the temporary immediate written to a temporary immediate register, the temporary immediate replacing the immediate for instruction computation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111432949.5A CN114116005B (en) | 2021-11-29 | 2021-11-29 | Immediate data storage method based on AIGPU architecture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111432949.5A CN114116005B (en) | 2021-11-29 | 2021-11-29 | Immediate data storage method based on AIGPU architecture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114116005A CN114116005A (en) | 2022-03-01 |
| CN114116005B true CN114116005B (en) | 2022-12-23 |
Family
ID=80371331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111432949.5A Active CN114116005B (en) | 2021-11-29 | 2021-11-29 | Immediate data storage method based on AIGPU architecture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114116005B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5201056A (en) * | 1990-05-02 | 1993-04-06 | Motorola, Inc. | RISC microprocessor architecture with multi-bit tag extended instructions for selectively attaching tag from either instruction or input data to arithmetic operation output |
| US5925122A (en) * | 1996-08-30 | 1999-07-20 | Nec Corporation | Data processing unit which pre-fetches instructions of different lengths to conduct processing |
| US6539470B1 (en) * | 1999-11-16 | 2003-03-25 | Advanced Micro Devices, Inc. | Instruction decode unit producing instruction operand information in the order in which the operands are identified, and systems including same |
| CN1477520A (en) * | 2002-08-21 | 2004-02-25 | 先进数字芯片株式会社 | Central processor with extended instruction |
| CN1625731A (en) * | 2002-01-31 | 2005-06-08 | Arc国际公司 | Configurable data processor with multiple length instruction set architectures |
| US7047396B1 (en) * | 2000-06-22 | 2006-05-16 | Ubicom, Inc. | Fixed length memory to memory arithmetic and architecture for a communications embedded processor system |
| US7415599B1 (en) * | 2005-11-01 | 2008-08-19 | Zilog, Inc. | Instruction operation and operand memory location determined based on preceding instruction operation and operand memory location |
| JP2014160393A (en) * | 2013-02-20 | 2014-09-04 | Casio Comput Co Ltd | Microprocessor and arithmetic processing method |
| CN105677298A (en) * | 2015-12-30 | 2016-06-15 | 李朝波 | Method and device for extending immediate operand in computer instruction |
| CN108304217A (en) * | 2018-03-09 | 2018-07-20 | 中国科学院计算技术研究所 | The method that the instruction of long bit wide operands is converted into short bit wide operands instruction |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040193844A1 (en) * | 2003-03-24 | 2004-09-30 | Sun Microsystems, Inc. | Load and/or store queue emptying technique to facilitate atomicity in processor execution of helper set |
| US20110314263A1 (en) * | 2010-06-22 | 2011-12-22 | International Business Machines Corporation | Instructions for performing an operation on two operands and subsequently storing an original value of operand |
| US9110802B2 (en) * | 2010-11-05 | 2015-08-18 | Advanced Micro Devices, Inc. | Processor and method implemented by a processor to implement mask load and store instructions |
| CN102221987B (en) * | 2011-05-11 | 2014-10-01 | 西安电子科技大学 | Instruction set encoding method based on embedded special instruction set processor |
| US10061580B2 (en) * | 2016-02-25 | 2018-08-28 | International Business Machines Corporation | Implementing a received add program counter immediate shift (ADDPCIS) instruction using a micro-coded or cracked sequence |
| CN113656071B (en) * | 2021-10-18 | 2022-02-08 | 深圳市智想科技有限公司 | RISC architecture based CPU instruction set system and CPU system |
-
2021
- 2021-11-29 CN CN202111432949.5A patent/CN114116005B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5201056A (en) * | 1990-05-02 | 1993-04-06 | Motorola, Inc. | RISC microprocessor architecture with multi-bit tag extended instructions for selectively attaching tag from either instruction or input data to arithmetic operation output |
| US5925122A (en) * | 1996-08-30 | 1999-07-20 | Nec Corporation | Data processing unit which pre-fetches instructions of different lengths to conduct processing |
| US6539470B1 (en) * | 1999-11-16 | 2003-03-25 | Advanced Micro Devices, Inc. | Instruction decode unit producing instruction operand information in the order in which the operands are identified, and systems including same |
| US7047396B1 (en) * | 2000-06-22 | 2006-05-16 | Ubicom, Inc. | Fixed length memory to memory arithmetic and architecture for a communications embedded processor system |
| CN1625731A (en) * | 2002-01-31 | 2005-06-08 | Arc国际公司 | Configurable data processor with multiple length instruction set architectures |
| CN1477520A (en) * | 2002-08-21 | 2004-02-25 | 先进数字芯片株式会社 | Central processor with extended instruction |
| US7415599B1 (en) * | 2005-11-01 | 2008-08-19 | Zilog, Inc. | Instruction operation and operand memory location determined based on preceding instruction operation and operand memory location |
| JP2014160393A (en) * | 2013-02-20 | 2014-09-04 | Casio Comput Co Ltd | Microprocessor and arithmetic processing method |
| CN105677298A (en) * | 2015-12-30 | 2016-06-15 | 李朝波 | Method and device for extending immediate operand in computer instruction |
| CN108304217A (en) * | 2018-03-09 | 2018-07-20 | 中国科学院计算技术研究所 | The method that the instruction of long bit wide operands is converted into short bit wide operands instruction |
Non-Patent Citations (3)
| Title |
|---|
| ARM指令中合法立即数的判断方法;杜俊;《甘肃科技》;20141030(第20期);全文 * |
| GPU的发展历程、未来趋势及研制实践;熊庭刚;《微电子与智能制造》;20200630;第2卷(第2期);第36-40页 * |
| High-performance extendable instruction set computing;Heui Lee 等;《Proceedings 6th Australasian Computer Systems Architecture Conference. ACSAC 2001》;20020807;第89-94页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114116005A (en) | 2022-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9830156B2 (en) | Temporal SIMT execution optimization through elimination of redundant operations | |
| EP2483787B1 (en) | Efficient predicated execution for parallel processors | |
| CN109213522B (en) | Remote atomic operation instructions | |
| US20220004391A1 (en) | Processor having multiple cores, shared core extension logic, and shared core extension utilization instructions | |
| US8069340B2 (en) | Microprocessor with microarchitecture for efficiently executing read/modify/write memory operand instructions | |
| US8458677B2 (en) | Generating code adapted for interlinking legacy scalar code and extended vector code | |
| CN104205042A (en) | Context switching mechanism for processing cores with general purpose CPU cores and tightly coupled accelerators | |
| KR101772299B1 (en) | Instruction to reduce elements in a vector register with strided access pattern | |
| TWI659357B (en) | Manage instruction order in the processor pipeline | |
| Lazo et al. | Adaptable register file organization for vector processors | |
| CN115910207A (en) | Implementation of dedicated instructions for accelerating Smith-Waterman sequence alignment | |
| Wang et al. | Architecture and compiler support for gpus using energy-efficient affine register files | |
| JP5193624B2 (en) | Data processor | |
| CN114116005B (en) | Immediate data storage method based on AIGPU architecture | |
| HK40069733B (en) | Immediate data storage method based on aigpu architecture | |
| US12572357B2 (en) | Vector mask buffers in a vector instruction execution pipeline | |
| HK40069733A (en) | Immediate data storage method based on aigpu architecture | |
| JP4444305B2 (en) | Semiconductor device | |
| CN112130970A (en) | Hardware support for dual memory atomic operations | |
| CN115905786A (en) | Techniques for Accelerated Smith-Wortman Sequence Alignment | |
| CN117136357A (en) | Inter-cluster shared data management in sub-NUMA clusters | |
| CN114281414B (en) | Data writing method of URF register in AIGPU architecture | |
| JP5610551B2 (en) | Data processor | |
| US20260086718A1 (en) | Systems and methods for providing extended memory access in a parallel processor | |
| HK40070897A (en) | Data writing method of urf register in aigpu architecture |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40069733 Country of ref document: HK |
|
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Hou Hongchao Inventor after: Ge Jianming Inventor after: Sang Yongqi Inventor after: Zhao Peng Inventor after: Wang Donghui Inventor after: Yao Fei Inventor after: Qiu Xiaogang Inventor before: Hou Hongchao Inventor before: Ge Jianming Inventor before: Sang Yongqi Inventor before: Zhao Peng Inventor before: Wang Donghui Inventor before: Yao Fei Inventor before: Qiu Xiaogang |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |