Vector processor, vector instruction processing method, device and equipment
By using vector registers based on 2P SRAM in vector processors, the problem of large area and power consumption of vector processors in the prior art is solved, and more efficient resource utilization and lower energy consumption are achieved.
Patent Information
- Application Number
- CN202411108654.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2024-07-25
- Filing Date
- 2024-08-13
- Publication Date
- 2024-11-26
Smart Images

Figure CN119025167A_ABST
Abstract
Description
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS
[0002] This application claims the priority of Chinese patent application 202411009276.6, entitled “Vector processor, method, device, and apparatus for vector instruction processing,” filed on July 25, 2024, the entire contents of which are incorporated herein by reference. Technical Field
[0003] The present invention relates to the field of computers, and in particular to a vector processor, a method, a device, a equipment and a computer program product for processing vector instructions. Background Art
[0004] Currently, vector processor architectures generally use flip-flop-based vector registers, which generally use 24 transistors to store one bit.
[0005] However, in the related art, a vector processor usually requires 32 1024-bit vector registers, so the vector registers require a large number of transistors, resulting in a large area and power consumption of the vector processor. Summary of the invention
[0006] Embodiments of the present disclosure provide a vector processor, a method, an apparatus, a device and a computer program product for processing vector instructions, which can reduce the area and power consumption of vector registers.
[0007] In a first aspect, an embodiment of the present disclosure provides a vector processor, including:
[0008] A scalar processing unit, used for parsing and distributing vector instructions, the scalar processing unit includes an instruction cache module, a data cache module, an instruction fetch module and a decoding module;
[0009] A vector processing unit connected to the scalar processing unit, the vector processing unit comprising a vector instruction emission module, a target processing unit module, and a plurality of vector registers based on 2P SRAM, and configured to execute the vector instruction emitted by the vector instruction emission module through the target processing unit module, and perform a read operation or a write operation on the plurality of vector registers based on 2P SRAM based on the vector instruction;
[0010] The vector memory is used to store data required to be saved when the target processing unit module performs a read operation or a write operation.
[0011] In an achievable implementation, the target processing unit module includes a vector loading module, a vector saving module and a vector arithmetic logic unit module, wherein the vector loading module includes a write port and the vector saving module includes a read port.
[0012] In one implementation, the vector arithmetic logic unit module includes a plurality of vector arithmetic logic unit modules, and each of the plurality of vector arithmetic logic unit modules includes a read port and a write port.
[0013] In one feasible implementation, the vector arithmetic logic unit module includes a first vector arithmetic logic unit module and a second vector arithmetic logic unit module, the first vector arithmetic logic unit module includes two read ports and one write port; the second vector arithmetic logic unit module includes one read port and one write port.
[0014] In one possible implementation, the 2P SRAM-based vector register includes a read port and a write port.
[0015] In a second aspect, an embodiment of the present disclosure provides a method for processing a vector instruction, the method comprising:
[0016] Obtaining, by a vector processing unit in a vector processor, a vector instruction sent by a scalar processing unit;
[0017] The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting sequence;
[0018] A read operation or a write operation is performed on a vector register of the 2P SRAM based on a vector instruction through a target processing unit module.
[0019] In one feasible implementation, when the target processing unit module is a vector arithmetic logic unit module, the pipeline cycle of the vector instruction corresponding to the vector arithmetic logic unit module includes an instruction fetch cycle, a decoding cycle, an execution cycle, an emission cycle, a source operand read cycle, a target processing operation cycle, a delay clock cycle and a write-back cycle.
[0020] In an achievable implementation, the vector instruction includes a vector length aggregation level parameter, and a target processing unit module performs a read operation or a write operation on a vector register of the 2P SRAM based on the vector instruction, including:
[0021] The target processing unit module performs a read operation or a write operation on the vector registers of the 2P SRAM whose number is indicated by the vector length aggregation level parameter.
[0022] In an achievable implementation, performing a read operation on a vector register of an indicated number of 2P SRAMs through a target processing unit module includes:
[0023] The source operands of the vector registers of the 2P SRAM of the indicated number are read respectively by the target processing unit module;
[0024] According to the source operands, target processing operations are performed on all data of the indicated number of 2P SRAM source vector registers to obtain multiple processing results, wherein the cycles of the target processing unit modules performing target processing operations on all data of the indicated number of 2P SRAM vector registers do not overlap.
[0025] In an achievable implementation, the target processing unit module includes a target vector arithmetic logic unit module, the target vector arithmetic logic unit module includes two read ports and one write port, and the vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmission order, including:
[0026] The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting sequence;
[0027] When the target processing unit module corresponding to the first vector instruction in the vector instructions is a target vector arithmetic logic unit module, the second vector instruction is issued with delay, and the issuance order of the second vector instruction is after the first vector instruction.
[0028] In an achievable implementation, the target processing unit module includes a target queue, the target queue is used to store the vector instruction emitted by the vector instruction emission module, and the target processing unit module performs a read operation or a write operation on the vector register of the 2P SRAM based on the vector instruction, including:
[0029] When the vector instruction in the target queue reaches the cycle phase corresponding to the target processing unit module, a read operation or a write operation is performed on the vector register of the 2P SRAM based on the vector instruction in the target queue.
[0030] In a third aspect, an embodiment of the present disclosure provides a device, comprising any equipment of the first aspect.
[0031] In a fourth aspect, an embodiment of the present disclosure provides a device, comprising a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement a method for vector instruction processing as described in any one of the second aspects.
[0032] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which, when executed by a processor, implements a method for vector instruction processing as described in any one of the second aspects.
[0033] The disclosed embodiments provide a vector processor, a method, an apparatus, a device and a computer program product for processing vector instructions, wherein the vector processor includes a scalar processing unit, a vector processing unit and a vector memory. The scalar processing unit is used to parse and distribute vector instructions; the vector processing unit is connected to the scalar processing unit, and is used to execute vector instructions emitted by a vector instruction emission module through a target processing unit module, and perform read operations or write operations on multiple 2P SRAM-based vector registers based on the vector instructions; the vector memory is used to store the data required to be saved when the target processing unit module performs read operations or write operations. The vector processor of the present application includes a vector register of 2PSRAM, which reduces the number of transistors used, and stores the same number of bits through a smaller number of transistors, which can reduce the area and power consumption of the vector processor. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following is a brief introduction to the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0035] Figure 1 is a schematic diagram of a vector processor provided by an embodiment of the present disclosure;
[0036] Figure 2 is a schematic diagram of a target processing unit module and a vector register connected according to an embodiment of the present disclosure;
[0037] Figure 3 is a flowchart of a method for processing vector instructions provided by an embodiment of the present disclosure;
[0038] Figure 4 is a schematic diagram of a pipeline cycle of a vector instruction of a second vector arithmetic logic unit module provided by an embodiment of the present disclosure;
[0039] Figure 5 is a schematic diagram of a vector register width provided by an embodiment of the present disclosure;
[0040] Figure 6 is a schematic diagram of another vector instruction pipeline cycle corresponding to a second vector arithmetic logic unit module provided by an embodiment of the present disclosure;
[0041] Figure 7 It is a schematic diagram of a vector instruction emission module delaying emission of a vector instruction provided by an embodiment of the present disclosure;
[0042] Figure 8 is a schematic diagram of a pipeline cycle of a vector instruction provided by an embodiment of the present disclosure;
[0043] Fig. 9 It is a schematic diagram of a vector instruction emission module that emits a vector instruction without delay provided by an embodiment of the present disclosure;
[0044] Fig.10 is a schematic diagram of a pipeline cycle of another vector instruction provided by an embodiment of the present disclosure;
[0045] Fig.11 is a schematic diagram of another vector instruction emission module delaying emission of vector instructions provided by an embodiment of the present disclosure;
[0046] Fig.12 is a schematic diagram of a pipeline cycle of another vector instruction provided by an embodiment of the present disclosure;
[0047] Fig.13 is a schematic diagram of a time-sharing access to a vector register of a 2P SRAM provided by an embodiment of the present disclosure;
[0048] Fig.14 is a schematic diagram of another time-sharing access to a vector register of a 2P SRAM provided by an embodiment of the present disclosure;
[0049] Fig.15 It is a schematic diagram of a pipeline cycle of a vector instruction for accessing a vector register of a 2P SRAM provided by an embodiment of the present disclosure;
[0050] Fig.16 It is a schematic diagram of another block-based vector instruction pipeline cycle for accessing a vector register of a 2P SRAM provided by an embodiment of the present disclosure;
[0051] Fig.17 It is a structural schematic diagram of a vector instruction processing device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0052] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present invention, rather than to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.
[0053] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
[0054] Before describing the technical solutions provided by the embodiments of the present disclosure, in order to facilitate understanding of the embodiments of the present disclosure, the present disclosure specifically describes the problems existing in the related technologies:
[0055] Currently, vector processor architectures generally use flip-flop-based vector registers. Typically, a flip-flop uses 24 transistors to store one bit. Since vector processors typically require 32 1024-bit vector registers, vector registers require a large number of transistors, resulting in a large area and power consumption for vector processors.
[0056] The embodiments of the present disclosure provide a method, apparatus, device and computer program product for processing vector instructions, which can solve the above-mentioned technical problems existing in the related technologies.
[0057] The disclosed embodiment obtains the vector instruction sent by the scalar processing unit, and then uses the vector instruction emission module to emit the vector instruction to the target processing unit module. Based on the vector instruction, the target processing unit module is used to perform a read operation or a write operation on a vector register based on a two-port random static memory (2P SRAM), and the vector register based on the 2P SRAM uses 4 transistors to store 1 bit. By using a vector register based on the 2P SRAM, the area and power consumption of the vector processor can be reduced.
[0058] The following describes the solutions of the embodiments of the present disclosure in conjunction with the accompanying drawings.
[0059] Figure 1 Schematic diagram of a vector processor provided by an embodiment of the present disclosure. Figure 1 As shown, the vector processor 101 includes a scalar processing unit 102, a vector processing unit 103 and a vector memory 104. The scalar processing unit 102 is used to parse and distribute vector instructions, and includes an instruction cache module 105, a data cache module 106, an instruction fetch module 107 and a decoding module 108.
[0060] The vector processing unit 103 is connected to the scalar processing unit 102, and includes a vector instruction issuing module 109, a target processing unit module 110, and a plurality of 2P SRAM-based vector registers 111. The vector processing unit 103 is used to execute the vector instruction issued by the vector instruction issuing module through the target processing unit module, and perform a read operation or a write operation on the plurality of 2P SRAM-based vector registers based on the vector instruction.
[0061] The vector memory 104 is used to store data required to be saved when the target processing unit module performs a read operation or a write operation.
[0062] The vector processor of the present application includes a vector register of 2P SRAM, which reduces the number of transistors used and stores the same number of bits through a smaller number of transistors, thereby reducing the area and power consumption of the vector processor.
[0063] In one example, the target processing unit module 110 includes a vector arithmetic logic unit module 1101 , a vector loading module 1102 , and a vector saving module 1103 , wherein the vector loading module 1102 includes a write port, and the vector saving module 1103 includes a read port.
[0064] In one example, the vector ALU module 1101 includes a plurality of vector ALU modules, and each of the plurality of vector ALU modules includes a read port and a write port.
[0065] In one example, the vector arithmetic logic unit module includes a first vector arithmetic logic unit module 11011 and a second vector arithmetic logic unit module 11012, the first vector arithmetic logic unit module 11011 includes two read ports and one write port; the second vector arithmetic logic unit module 11012 includes one read port and one write port.
[0066] In an example, the vector register 111 based on the 2P SRAM includes a read port and a write port. The width of the vector register 111 based on the 2P SRAM is not limited and can be 256 bits, 1024 bits, etc.
[0067] In one example, a schematic diagram of connecting a target processing unit module with a 2P SRAM-based vector register is shown in FIG. Figure 2As shown in the figure, the 32 vector registers VR0-VR31 based on 2P SRAM include a read port and a write port, supporting parallel reading and writing. The vector registers VR0-VR31 are respectively connected to the vector storage module (VST), the first vector arithmetic logic unit module (VALU1) and the second vector arithmetic logic unit module (VALU2) through the read data bus (RD BUS). The vector registers VR0-VR31 are respectively connected to the vector loading module (VLD), the first vector arithmetic logic unit module (VALU1) and the second vector arithmetic logic unit module (VALU2) through the write data bus (WT BUS). According to the read and write timing of the four pre-arranged target processing unit modules, in each clock cycle, only the data of a certain target processing unit module will be read out or the processed data will be written back to the vector register based on 2PSRAM based on the address and BUS of the vector register based on 2P SRAM. Since the storage addresses of different target processing unit modules in the vector register based on 2P SRAM are different, it supports simultaneous reading and writing of different addresses of the vector register based on 2P SRAM. This ensures that the vector register based on 2P SRAM is always in a continuous working state within one processing cycle, which can fully utilize the vector register resources based on 2P SRAM and reduce the processing time each time. All target processing unit modules complete the processing and support turning off the clock of the vector register based on 2PSRAM, waiting for the next instruction.
[0068] Figure 3 An embodiment of the present disclosure provides an application Figure 1 FIG. 1 is a flow chart showing a method for processing vector instructions of a vector processor. Figure 3 , the method may include steps S210-S230.
[0069] S210: Acquire a vector instruction sent by a scalar processing unit through a vector processing unit in a vector processor.
[0070] After the scalar processing unit in the vector processor parses and obtains the vector instruction, the vector processing unit in the vector processor obtains the vector instruction sent by the scalar processing unit through the vector instruction queue in the vector instruction issuing module.
[0071] The vector instruction is used to control the target processing unit module corresponding to the vector instruction to perform read and write operations with the vector register based on 2PSRAM.
[0072] In one example, a scalar processing unit in a vector processor selects an instruction from an instruction cache module through an instruction fetch module, and decodes the instruction through a decoding module to obtain a vector instruction. After obtaining the vector instruction, the scalar processing unit sends the vector instruction to the vector processing unit. The vector processing unit obtains the vector instruction sent by the vector instruction queue in the scalar processing unit through the vector instruction emission module.
[0073] S220: The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting sequence.
[0074] After the vector instruction transmitting module receives the vector instruction sent by the scalar processing unit, the vector instruction in the vector instruction queue in the vector instruction transmitting module is transmitted to the target processing unit module according to the preset transmission sequence. The preset transmission sequence is controlled by the scoreboard in the vector instruction transmitting module.
[0075] In one example, the issuance order of the vector instructions may be sequential issuance or out-of-order issuance. The parallelism of the vector issuance may be single issuance, dual issuance or superscalar multiple issuance.
[0076] S230: Performing a read operation or a write operation on a vector register of the 2P SRAM based on the vector instruction through the target processing unit module.
[0077] The target processing unit module performs a read operation or a write operation on the vector register of the 2PSRAM according to the vector instruction issued by the vector instruction issuing module.
[0078] In one example, the read / write port of the target processing unit module is directly connected to the read / write port of the vector register of the 2PSRAM through a BUS.
[0079] The disclosed embodiment obtains the vector instruction sent by the scalar processing unit, and then uses the vector instruction emission module to emit the vector instruction to the target processing unit module. Based on the vector instruction, the target processing unit module is used to perform a read operation or a write operation on a vector register based on 2P SRAM. By using a vector register based on 2P SRAM, the area and power consumption of the vector register can be reduced.
[0080] In one embodiment, Figure 4 As shown, the pipeline cycle of the vector instruction corresponding to the second vector arithmetic logic unit module VALU2 includes IF, ID, EX, VIS, VS1 RD, VS2 RD, VEX1, VEX2, VEX3, VEX4, LEG and WB.
[0081] Among them, IF is the instruction fetch cycle, ID is the decode cycle, and EX is the execution cycle. IF, ID, and EX cycles are pipeline cycles shared with scalar instructions. VIS is the issue cycle, and the issue vector instruction operation is executed in the VIS cycle. VS1 RD and VS2 RD are the source operand read cycles, and the source operands of the VS1 source vector register and the VS2 source vector register are read in the VS1 RD and VS2 RD cycles. VEX1, VEX2, VEX3, and VEX4 are target processing operation cycles, and all data in the source vector register are processed in the VEX1, VEX2, VEX3, and VEX4 cycles to obtain the processing results. LEG is the delayed clock cycle after VEX4 is executed. WB is the write-back cycle, and the obtained processing results are written back to the target register in the WB cycle.
[0082] In one example, a target processing operation cycle of the vector instruction is greater than or equal to a source operand reading cycle of the vector instruction.
[0083] In one example, if Figure 5 As shown, each of the 32 vector registers based on 2P SRAM, VR0 to VR31, includes 8 elements with a width of SEW, and SEW is 32 bits. The total width of each vector register is VLEN, and VLEN is equal to the width of 8 SEWs, which is equal to 256 bits. Among them, the width of every two SEWs in the vector register corresponds to one DLANE, indicating that the target processing unit module can process 2 elements per VEX cycle when processing the data of the vector register. Therefore, it takes 4 VEX cycles for the target processing unit module to process all the data of the vector register.
[0084] In another example, VEX1, VEX2, VEX3, and VEX4 may be a target processing unit module with a width of DLANE to complete a cycle of processing all data in a vector register with a width of VLEN, wherein the width of VLEN is greater than the width of DLANE, and the width of VLEN may be 4 times the width of DLANE.
[0085] The disclosed embodiment sets the pipeline cycle of vector instructions so that target processing unit modules corresponding to different vector instructions can simultaneously read and write access to vector registers based on 2P SRAM, thereby improving the read and write efficiency of vector registers based on 2P SRAM.
[0086] In one embodiment, the vector instruction includes a vector length aggregation level parameter, and performing a read operation or a write operation on a vector register of the 2P SRAM based on the vector instruction by the target processing unit module may include step S231.
[0087] S231: Perform a read operation or a write operation on the vector registers of the 2PSRAM whose number is indicated by the vector length aggregation level parameter through the target processing unit module.
[0088] After acquiring the vector instruction, the target processing unit module determines the indicated number of 2PSRAM vector registers according to the vector length aggregation level parameter in the vector instruction, and performs a read operation or a write operation on the indicated number of 2PSRAM vector registers.
[0089] In one example, after the vector loading module obtains the vector instruction, it is determined that the vector length aggregation level parameter in the vector instruction is 2, and then a write operation is performed on two vector registers.
[0090] The disclosed embodiment can obtain the vector length aggregation level parameter in the vector instruction, thereby performing corresponding operations on vector registers whose number is equal to the vector length aggregation level parameter, thereby improving operation efficiency.
[0091] In one embodiment, a read operation is performed on the vector registers of the indicated number of 2P SRAMs by the target processing unit module, including step S2311 and step S2312.
[0092] S2311: Read the source operands of the vector registers of the 2P SRAM of the indicated number respectively through the target processing unit module.
[0093] The target processing unit module reads source operands of the vector registers of the 2P SRAM in an indicated quantity according to the vector instruction issued by the vector instruction issuing module.
[0094] The source operand may be address data.
[0095] S2312: Perform target processing operations on all data of the indicated number of source vector registers of the 2P SRAM according to the source operands to obtain multiple processing results.
[0096] Based on the pipeline cycle of the vector instruction, target processing operations are performed on all data of the indicated number of 2P SRAM source vector registers according to the read source operands to obtain multiple processing results. The cycles of the target processing unit modules performing target processing operations on all data of the indicated number of 2P SRAM vector registers do not overlap.
[0097] In one example, the vector instruction pipeline cycle corresponding to the second vector arithmetic logic unit module (VALU2) is as follows: Figure 6As shown, the vector length aggregation level parameter (LMUL) of the vector instruction is 4, which indicates that operations are performed on four vector registers, namely VR1, VR2, VR3 and VR4. After the vector instruction is obtained after the three cycles of IF, ID and EX. The vector instruction is emitted to VALU2 in the VIS cycle. VALU2 reads the source operands of the VR1, VR2, VR3 and VR4 vector registers in the VS1 RD and VS2 RD cycles according to the vector instruction. In the VEX1, VEX2, VEX3 and VEX4 cycles, the target processing operations are performed on the VR1, VR2, VR3 and VR4 vector registers respectively to obtain multiple processing results. The delay clock after the VEX4 cycle is executed is set in the LEG cycle. Multiple processing results are written back to the target vector register in the WB cycle. Among them, the VEX1, VEX2, VEX3 and VEX4 cycles in which VALU2 performs target processing operations on the VR1, VR2, VR3 and VR4 vector registers respectively do not overlap.
[0098] In the disclosed embodiment, the cycles of the target processing unit module performing target processing operations on the vector registers of the indicated number of 2P SRAMs do not overlap, thereby ensuring that the target processing unit module operates at full load.
[0099] In one embodiment, the target processing unit module includes a target vector arithmetic logic unit module, the target vector arithmetic logic unit module includes two read ports and one write port, and the vector instruction transmission module transmits the vector instruction to the target processing unit module according to a preset transmission order, which may include steps S221 and S222.
[0100] S221: The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting order.
[0101] The vector instruction transmitting module transmits the vector instructions in the vector instruction queue to the target processing unit module according to the preset transmitting sequence included in the scoreboard.
[0102] S222: When the target processing unit module corresponding to the first vector instruction in the vector instructions is a target vector arithmetic logic unit module, delay issuing a second vector instruction, and the issuing order of the second vector instruction is after the first vector instruction.
[0103] In the process of issuing a vector instruction, if the issued vector instruction is a first vector instruction, that is, the first vector instruction is used to control a target vector arithmetic logic unit module including two read ports and one write port, then after issuing the first vector instruction, a second vector instruction whose issuing order is after the first vector instruction is issued with a delay.
[0104] In one example, a schematic diagram of a vector instruction issuing module delaying issuing a vector instruction is shown in FIG. Figure 7 As shown: the vector instruction emission module emits 4 vector instructions. The first vector instruction controls the vector loading module VLD, the second vector instruction controls the second vector arithmetic logic unit module VALU2, the third vector instruction controls the first vector arithmetic logic unit module VALU1, and the fourth vector instruction controls the vector storage module VST. Since VALU2 includes two read ports, there are two read requirements. After the vector instruction controlling VALU2 is emitted, by setting an emission delay of one clock cycle to delay the emission of subsequent vector instructions, it can ensure that different target processing unit modules access the vector registers of 2PSRAM in a time-sharing manner.
[0105] and Figure 7 The schematic diagram of the pipeline cycle of the corresponding vector instruction is as follows Figure 8 As shown: The vector instruction pipeline cycle corresponding to VLD first passes through the address generation (AG) cycle, and then passes through the WB cycle. The vector instruction pipeline cycle corresponding to VALU2 first passes through the VS1 RD and VS2 RD cycles, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and WB cycle. The vector instruction pipeline cycle corresponding to VALU1 first passes through the VS1 RD cycle, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and WB cycle. The vector instruction pipeline cycle corresponding to VST first passes through the VS3 RD cycle, then passes through the AG cycle, and finally passes through the end (DN) cycle.
[0106] In another example, a schematic diagram of a vector instruction issuing module issuing a vector instruction without delay is shown in FIG. Fig. 9 As shown: the vector instruction emission module emits 4 vector instructions. The first vector instruction controls the second vector arithmetic logic unit module VALU2, the second vector instruction controls the vector loading module VLD, the third vector instruction controls the first vector arithmetic logic unit module VALU1, and the fourth vector instruction controls the vector storage module VST. Since VLD only has a write port and does not require a read of the vector register of 2PSRAM, after emitting the vector instruction controlling VALU2, the vector instruction emission module does not need to delay the emission of subsequent vector instructions and continues to emit the vector instruction controlling VLD.
[0107] and Fig. 9 The schematic diagram of the pipeline cycle of the corresponding vector instruction is as follows Fig.10As shown: The vector instruction pipeline cycle corresponding to VALU2 first passes through the VS1 RD and VS2 RD cycles, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and the WB cycle. The vector instruction pipeline cycle corresponding to VLD first passes through the AG cycle, and then passes through the WB cycle. The vector instruction pipeline cycle corresponding to VALU1 first passes through the VS1 RD cycle, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and the WB cycle. The vector instruction pipeline cycle corresponding to VST first passes through the VS3 RD cycle, then passes through the AG cycle, and finally passes through the DN cycle.
[0108] In another example, a schematic diagram of a vector instruction issuing module delaying issuing a vector instruction is shown in FIG. Fig.11 As shown: the vector instruction emission module emits 3 vector instructions. The first vector instruction controls the second vector arithmetic logic unit module VALU2, the second vector instruction controls the first vector arithmetic logic unit module VALU1, and the third vector instruction controls the vector storage module VST. And the vector length aggregation level parameter of the first vector instruction, the second vector instruction and the third vector instruction is equal to 2. After the vector instruction controlling VALU2 is emitted, the emission delay of one clock cycle is set to delay the emission of subsequent vector instructions, so that different target processing unit modules can access the vector register of the 2P SRAM in a time-sharing manner.
[0109] and Fig.11 The schematic diagram of the pipeline cycle of the corresponding vector instruction is as follows Fig.12 As shown in the figure: According to the vector length aggregation level parameter 2 in the vector instruction, VALU2, VALU1 and VST perform corresponding operations on the two vector registers VR1 and VR2. Among them, the vector instruction pipeline cycle corresponding to VALU2 first passes through the VS1 RD and VS2 RD cycles, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and WB cycle. The vector instruction pipeline cycle corresponding to VALU1 first passes through the VS1 RD cycle, then passes through the VEX1, VEX2, VEX3 and VEX4 cycles, and finally passes through the LEG cycle and WB cycle. The vector instruction pipeline cycle corresponding to VST first passes through the VS3 RD cycle, then passes through the AG cycle, and finally passes through the DN cycle.
[0110] The disclosed embodiment delays the issuance of a subsequent second vector instruction after issuing a first vector instruction for controlling a target vector arithmetic logic unit module, thereby ensuring that different target processing unit modules access the vector registers of the 2P SRAM in a time-sharing manner.
[0111] In one embodiment, the target processing unit module includes a target queue, and the target queue is used to store the vector instruction emitted by the vector instruction emission module. The target processing unit module performs a read operation or a write operation on the vector register of the 2P SRAM based on the vector instruction, which may include step S223:
[0112] S223: When the vector instruction in the target queue reaches the cycle phase corresponding to the target processing unit module, a read operation or a write operation is performed on the vector register of the 2P SRAM based on the vector instruction in the target queue.
[0113] The current cycle phase is detected, and when it is detected that the vector instruction in the target queue reaches the cycle phase corresponding to the target processing unit module, a read operation or a write operation is performed on the vector register of the 2PSRAM according to the vector instruction in the target queue.
[0114] In one example, if Fig.13 As shown, the cycle of accessing the vector register of the 2P SRAM can be divided into four time slots. In the first time slot, the second vector arithmetic logic unit module VALU2 accesses the vector register of the 2P SRAM. In the second time slot, the vector load unit module VLD accesses the vector register of the 2P SRAM. In the third time slot, the first vector arithmetic logic unit module VALU3 accesses the vector register of the 2P SRAM. In the fourth time slot, the vector save unit module VST accesses the vector register of the 2P SRAM.
[0115] Among them, VALU1 stores the corresponding vector instruction to the target queue in the first time slot of the cycle of the first access to the vector register of the 2P SRAM, and executes the vector instruction in the target queue in the third time slot of the cycle according to the cycle phase of the instruction. VALU2 stores the corresponding vector instruction to the target queue in the second time slot of the cycle of the first access to the vector register of the 2PSRAM, and executes the vector instruction in the target queue in the first time slot of the cycle of the second access to the vector register of the 2P SRAM according to the cycle phase of the instruction. VST stores the corresponding vector instruction to the target queue in the third time slot of the cycle of the first access to the vector register of the 2P SRAM, and executes the vector instruction in the target queue in the fourth time slot of the cycle of the second access to the vector register of the 2P SRAM according to the cycle phase of the instruction. VLD stores the corresponding vector instruction to the target queue in the fourth time slot of the cycle of the first access to the vector register of the 2P SRAM, and executes the vector instruction in the target queue in the second time slot of the cycle of the second access to the vector register of the 2P SRAM according to the cycle phase of the instruction. VALU1 stores the corresponding vector instruction into the target queue in the first time slot of the second cycle accessing the vector register of the 2P SRAM, and executes the vector instruction in the target queue in the third time slot of the cycle according to the cycle phase of the instruction.
[0116] In another example, Fig.14 As shown, the cycle of accessing the vector register of the 2P SRAM can be divided into four time slots. In the first time slot, the second vector arithmetic logic unit module VALU2 accesses the vector register of the 2P SRAM. In the second time slot, the vector load unit module VLD accesses the vector register of the 2P SRAM. In the third time slot, the first vector arithmetic logic unit module VALU3 accesses the vector register of the 2P SRAM. In the fourth time slot, the vector save unit module VST accesses the vector register of the 2P SRAM.
[0117] Among them, the vector instruction emission module emits 4 instructions in a parallel emission mode, and these 4 instructions are respectively used to control the second vector arithmetic logic unit module VALU2, the first vector arithmetic logic unit module VALU1, the vector storage module VST and the vector loading module VLD, and these 4 instructions are stored in the target queue of the corresponding target processing unit module in the first time slot of the cycle of the first access to the vector register of the 2P SRAM. According to the cycle phase of the vector instruction in the target queue of VALU1, VALU1 executes the vector instruction in the target queue of VALU1 in the third time slot of the cycle of the first access to the vector register of the 2P SRAM to access the vector register of the 2P SRAM. According to the cycle phase of the vector instruction in the target queue of VALU2, VALU2 executes the vector instruction in the target queue of VALU2 in the first time slot of the cycle of the second access to the vector register of the 2P SRAM to access the vector register of the 2P SRAM. According to the cycle phase of the vector instruction in the target queue of VST, VST executes the vector instruction in the target queue of VST in the fourth time slot of the cycle of the first access to the vector register of the 2PSRAM to access the vector register of the 2P SRAM. According to the cycle phase of the vector instruction in the target queue of the VLD, the VLD executes the vector instruction in the target queue of the VLD in the second time slot of the first cycle of accessing the vector register of the 2P SRAM, thereby accessing the vector register of the 2P SRAM.
[0118] The disclosed embodiment can ensure that different target processing unit modules access the vector registers of the 2PSRAM in a time-sharing manner by allocating a cycle for accessing the vector registers of the 2PSRAM to each target processing unit module.
[0119] In one embodiment, the vector register includes N banks, and performing a read operation or a write operation on the vector register of the 2P SRAM based on a vector instruction by the target processing unit module may include step S232.
[0120] S232: Perform a read operation or a write operation on M target blocks among the N blocks in the vector register based on the vector instruction through the target processing unit module.
[0121] The target processing unit module determines to interact with M target blocks among the N blocks in the vector register according to the vector instruction sent by the vector instruction transmitting module, and performs a read operation or a write operation corresponding to the vector instruction with each of the M target blocks.
[0122] Among them, N≧2, N>M≧1.
[0123] In an example, a vector register based on a 2P SRAM can be physically divided into multiple blocks by adding a Bank interface, and each block stores a portion of the data stored in the physically divided vector register.
[0124] In one example, the vector instruction issuing module issues a vector instruction to VALU2, and the vector instruction is used to control VALU2 to perform a read operation on the blocks in the vector register. The pipeline cycle of the vector instruction is as follows: Fig.15 As shown, the vector register VR1 is divided into 4 blocks, namely VR1_B1, VR1_B2, VR1_B3 and VR1_B4. After the vector instruction is obtained after the three cycles of IF, ID and EX. The vector instruction is sent to VALU2 in the VIS cycle. VALU2 reads the source operand of VR1_B1 according to the vector instruction, and passes through the VR1 and VR2 cycles. Then, the target processing operation is performed on VR1_B1 in the VX1 and VX2 cycles. While the target processing operation is performed on VR1_B1 in the VX1 and VX2 cycles, VALU2 reads the source operand of VR1_B2 according to the vector instruction, and passes through the VR1 and VR2 cycles. Then, the target processing operation is performed on VR1_B2 in the VX3 and VX4 cycles. While the target processing operation is performed on VR1_B2 in the VX3 and VX4 cycles, VALU2 reads the source operand of VR1_B3 according to the vector instruction, and passes through the VR1 and VR2 cycles. Then, the target processing operation is performed on VR1_B3 in VX5 and VX6 cycles. While the target processing operation is performed on VR1_B3 in VX5 and VX6 cycles, VALU2 reads the source operand from VR1_B4 according to the vector instruction, and passes through VR1 and VR2 cycles. Then, the target processing operation is performed on VR1_B4 in VX7 and VX8 cycles. The delay clock after the VX8 cycle is set in the LEG cycle after the VX8 cycle. The processing result is written back to the target vector register in the WB cycle.
[0125] In another example, the vector instruction issuing module issues four vector instructions in sequence. The first vector instruction controls the second vector arithmetic logic unit module VALU2, the second vector instruction controls the vector loading module VLD, the third vector instruction controls the first vector arithmetic logic unit module VALU1, and the fourth vector instruction controls the vector saving module VST. The pipeline cycles of these four vector instructions are as follows: Fig.16As shown, the vector register VR1 is divided into 4 blocks, namely VR1B1, VR1B2, VR1B3 and VR1B4. The vector register VR2 is divided into 4 blocks, namely VR2B1, VR2B2, VR2B3 and VR2B4. VALU2 reads the source operands of VR1B1 and VR2B1 according to the vector instruction, and passes through VR1_B1 and VR2_B1 cycles. Then, target processing operations are performed on VR1B1 and VR2B1 in VX1 and VX2 cycles. While the target processing operations are performed on VR1B1 and VR2B1 in VX1 and VX2 cycles, VALU2 reads the source operands of VR1B2 and VR2B2 according to the vector instruction, and passes through VR1_B2 and VR2_B2 cycles. Then, target processing operations are performed on VR1B2 and VR2B2 in VX3 and VX4 cycles. At the same time as VR1B3 and VR2B3 are being processed by the target in VX3 and VX4 cycles, VALU2 reads the source operands from VR1B3 and VR2B3 according to the vector instruction, and passes through VR1_B3 and VR2_B3 cycles. Then, at the VX5 and VX6 cycles, VR1B3 and VR2B3 are being processed by the target. At the same time as VR1B3 and VR2B3 are being processed by the target in VX5 and VX6 cycles, VALU2 reads the source operands from VR1B4 and VR2B4 according to the vector instruction, and passes through VR1_B4 and VR2_B4 cycles. Then, at the VX7 and VX8 cycles, VR1B4 and VR2B4 are being processed by the target. The LEG cycle after the VX8 cycle sets the delay clock after the VX8 cycle is executed. The processing result is written back to the target vector register at the WB cycle.
[0126] The vector instruction pipeline cycle corresponding to VLD is first through the AG cycle and then through the WB cycle. VALU1 reads the source operand VR2B1 according to the vector instruction and passes through the VR2_B1 cycle. Then, the target processing operation is performed on VR2B1 in the VX1 and VX2 cycles. While the target processing operation is performed on VR2B1 in the VX1 and VX2 cycles, VALU1 reads the source operand VR2B2 according to the vector instruction and passes through the VR2_B2 cycle. Then, the target processing operation is performed on VR2B2 in the VX3 and VX4 cycles. While the target processing operation is performed on VR2B2 in the VX3 and VX4 cycles, VALU1 reads the source operand VR2B3 according to the vector instruction and passes through the VR2_B3 cycle. Then, the target processing operation is performed on VR2B3 in the VX5 and VX6 cycles. While the target processing operation is performed on VR2B3 in the VX5 and VX6 cycles, VALU1 reads the source operand VR2B4 according to the vector instruction and passes through the VR2_B4 cycle. Then, the target processing operation is performed on VR2B3 in the VX7 and VX8 cycles. The delay clock after the VX8 cycle is set in the LEG cycle after the VX8 cycle. The processing result is written back to the target vector register in the WB cycle.
[0127] The pipeline cycle corresponding to VST is first VR2_B1 cycle, then AG cycle, VR2_B2 cycle, VR2_B3 cycle, VR2_B4 cycle and Mem cycle. Among them, VST reads the source operand of VR2B1 in VR2_B1 cycle; VST reads the source operand of VR2B2 in VR2_B2 cycle; VST reads the source operand of VR2B3 in VR2_B3 cycle; VST reads the source operand of VR2B4 in VR2_B4 cycle.
[0128] In summary, the disclosed embodiment can perform a read operation or a write operation on a part of the blocks in the vector register based on a vector instruction through the target processing unit module, so the amount of data read or written becomes less, thereby reducing the size of the temporary register inside the target processing unit module and saving physical resources.
[0129] Fig.17 A schematic diagram of the hardware structure of vector instruction processing provided by an embodiment of the present disclosure is shown.
[0130] The apparatus for vector instruction processing may include a processor 1701 and a memory 1702 storing computer program instructions.
[0131] Specifically, the processor 1701 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.
[0132] Memory 1702 may include a large capacity memory for data or instructions. By way of example and not limitation, memory 1702 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. In appropriate cases, memory 1702 may include removable or non-removable (or fixed) media. In appropriate cases, memory 1702 may be inside or outside the integrated gateway disaster recovery device. In a specific embodiment, memory 1702 is a non-volatile solid-state memory.
[0133] The memory 1702 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Thus, typically, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., a memory device) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to one aspect of the present disclosure.
[0134] The processor 1701 implements any one of the vector instruction processing methods in the above embodiments by reading and executing computer program instructions stored in the memory 1702 .
[0135] In one example, the vector instruction processing device may further include a communication interface 1703 and a bus 1704. Fig.17 As shown, the processor 1701, the memory 1702, and the communication interface 1703 are connected via a bus 1704 and communicate with each other.
[0136] The communication interface 1703 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.
[0137] Bus 1704 includes hardware, software or both, and couples the components of the online data traffic billing device to each other. For example, but not limitation, the bus may include an accelerated graphics port (Accelerated Graphics Port, AGP) or other graphics bus, an enhanced industry standard architecture (Extended Industry Standard Architecture, EISA) bus, a front side bus (Front Side Bus, FSB), a hypertransport (Hyper Transport, HT) interconnection, an industry standard architecture (Industry Standard Architecture, ISA) bus, an infinite bandwidth interconnection, a low pin count (LPC) bus, a memory bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable buses or a combination of two or more of these. In appropriate cases, bus 1704 may include one or more buses. Although the embodiment of the present invention describes and illustrates a specific bus, the present invention considers any suitable bus or interconnection. In addition, in conjunction with the method for vector instruction processing in the above-mentioned embodiment, the embodiment of the present application also provides a computer storage medium to realize. The computer storage medium stores computer program instructions; when the computer program instructions are executed by the processor, any one of the vector instruction processing methods in the above embodiments is implemented.
[0138] An embodiment of the present application also provides a computer program product, including a computer program, which implements any one of the vector instruction processing methods in the above embodiments when the computer program is executed by a processor.
[0139] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.
[0140] The functional blocks shown in the above structural block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), appropriate firmware, plug-in, function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or communication link by a data signal carried in a carrier. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, read-only memory (Read-Only Memory, ROM), flash memory, erasable read-only memory (Erasable ReadOnly Memory, EROM), floppy disks, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), optical discs, hard disks, optical fiber media, radio frequency (Radio Frequency, RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.
[0141] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.
[0142] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.
[0143] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system, module and unit described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be covered within the protection scope of the present invention.
Claims
1. A vector processor, characterized in that: include: A scalar processing unit, used for parsing and distributing vector instructions, the scalar processing unit comprising an instruction cache module, a data cache module, an instruction fetch module and a decoding module; a vector processing unit connected to the scalar processing unit, the vector processing unit comprising a vector instruction emission module, a target processing unit module, and a plurality of vector registers based on 2P SRAM, configured to execute the vector instruction emitted by the vector instruction emission module through the target processing unit module, and perform a read operation or a write operation on the plurality of vector registers based on 2P SRAM based on the vector instruction; The vector memory is used to store data required to be saved when the target processing unit module performs a read operation or a write operation.
2. The vector processor according to claim 1, characterized in that The target processing unit module includes a vector loading module, a vector saving module and a vector arithmetic logic unit module, wherein the vector loading module includes a write port, and the vector saving module includes a read port.
3. The vector processor according to claim 2, characterized in that The vector arithmetic logic unit module includes a plurality of vector arithmetic logic unit modules, and each of the plurality of vector arithmetic logic unit modules includes a read port and a write port.
4. The vector processor according to claim 2, characterized in that: The vector arithmetic logic unit module includes a first vector arithmetic logic unit module and a second vector arithmetic logic unit module, wherein the first vector arithmetic logic unit module includes two read ports and one write port; and the second vector arithmetic logic unit module includes one read port and one write port.
5. The vector processor according to claim 1, characterized in that: The 2PSRAM-based vector register includes a read port and a write port.
6. A method for processing vector instructions applied to the vector processor of claim 1, characterized in that: include: Acquire, by a vector processing unit in the vector processor, a vector instruction sent by a scalar processing unit; The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting order; The target processing unit module performs a read operation or a write operation on a vector register of the 2P SRAM based on the vector instruction.
7. The method according to claim 1, characterized in that In the case where the target processing unit module is a vector arithmetic logic unit module, the pipeline cycle of the vector instruction corresponding to the vector arithmetic logic unit module includes an instruction fetch cycle, a decoding cycle, an execution cycle, an emission cycle, a source operand read cycle, a target processing operation cycle, a delay clock cycle and a write-back cycle.
8. The method according to claim 6, characterized in that The vector instruction includes a vector length aggregation level parameter, and performing a read operation or a write operation on a vector register of the 2PSRAM based on the vector instruction by the target processing unit module includes: The target processing unit module performs a read operation or a write operation on the vector registers of the 2P SRAM whose number is indicated by the vector length aggregation level parameter.
9. The method according to claim 8, characterized in that The performing a read operation on the vector registers of the indicated number of 2P SRAMs through the target processing unit module comprises: Read the source operands of the vector registers of the indicated number of 2P SRAMs respectively through the target processing unit module; According to the source operands, target processing operations are performed on all data of the indicated number of source vector registers of the 2PSRAM to obtain multiple processing results, wherein the cycles of the target processing unit modules performing target processing operations on all data of the indicated number of vector registers of the 2PSRAM do not overlap.
10. The method according to any one of claims 6 to 9, characterized in that: The target processing unit module includes a target vector arithmetic logic unit module, the target vector arithmetic logic unit module includes two read ports and one write port, and the vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmission order, including: The vector instruction transmitting module transmits the vector instruction to the target processing unit module according to a preset transmitting order; When the target processing unit module corresponding to the first vector instruction in the vector instructions is the target vector arithmetic logic unit module, the second vector instruction is issued with delay, and the issuance order of the second vector instruction is after the first vector instruction.
11. The method according to any one of claims 6 to 9, characterized in that: The target processing unit module includes a target queue, the target queue is used to store the vector instruction emitted by the vector instruction emission module, and the performing a read operation or a write operation on the vector register of the 2P SRAM based on the vector instruction by the target processing unit module includes: When the vector instruction in the target queue reaches the cycle phase corresponding to the target processing unit module, a read operation or a write operation is performed on the vector register of the 2P SRAM based on the vector instruction in the target queue.
12. The method according to any one of claims 6 to 9, characterized in that: The vector register includes N blocks, and the performing a read operation or a write operation on the vector register of the 2P SRAM based on the vector instruction by the target processing unit module includes: The target processing unit module performs a read operation or a write operation on M target blocks of the N blocks in the vector register based on the vector instruction, wherein N≧2, N>M≧1.
13. A device, characterized in that: The device comprises the vector processor according to any one of claims 1 to 5.
14. A device, characterized in that The device comprises: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the vector instruction processing method according to any one of claims 6 to 12.
15. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for processing vector instructions according to any one of claims 6 to 12 is implemented.
Citation Information
Cited By
Data loading method and device, processor, electronic equipment and storage medium
CN121166573A
Vector memory-based scalar access system and method
CN121478812A
