Data processor core, data processor, data processor device, data processing method, electronic device, computer program and storage medium
The data processor core stabilizes power and voltage fluctuations in AI chips by using preheating commands to gradually increase power consumption, enhancing computational accuracy and chip longevity.
Patent Information
- Application Number
- JP2024077371
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-22
- Filing Date
- 2024-05-10
- Publication Date
- 2024-12-04
- Estimated Expiration
- 2044-05-10
AI Technical Summary
Artificial intelligence chips face issues with sudden changes in power consumption and voltage fluctuations during computational tasks, leading to potential circuit damage and inaccurate calculations due to rapid transitions between idle and operational states.
Implementing a data processor core with a control mechanism that generates preheating commands to gradually increase power consumption, allowing computing means subarrays to transition from an idle state to a preheating state before processing computational commands, thereby stabilizing voltage and reducing sudden power changes.
This approach stabilizes voltage and power consumption, improving computational accuracy and chip longevity by avoiding sudden voltage fluctuations and maintaining high computational power levels.
Smart Images

Figure 2024167889000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to the field of artificial intelligence, in particular to the field of integrated circuit technology and chip technology. More specifically, the present disclosure provides a data processor core, a data processor, a data processor device, a data processing method, an electronic device, a computer program, and a storage medium. [Background technology]
[0002] With the development of artificial intelligence technology, the application scenarios of artificial intelligence chips are increasing. The artificial intelligence chip has many multiply accumulate (MAC) arrays and has strong matrix calculation capabilities. Summary of the Invention
[0003] The present disclosure provides a data processor core, a data processor, a data processor device, a data processing method, an electronic device, and a storage medium.
[0004] According to one aspect of the present disclosure, a data processor core is provided, the data processor core including: control means arranged to generate a pre-heat command in response to detecting a computation command to be processed; and a computation means array including at least one computation means sub-array, the computation means sub-array arranged to receive a pre-heat command instructing the computation means sub-array to perform a read operation and a computation operation, and based on the pre-heat command, enter a pre-heat state, and in response to receiving a target command corresponding to the computation command to be processed, switch from the pre-heat state to a computation state.
[0005] According to another aspect of the present disclosure, there is provided a data processor including at least one data processor core as provided herein.
[0006] According to another aspect of the present disclosure, there is provided a data processing apparatus including a data processor as provided herein.
[0007] According to another aspect of the present disclosure, there is provided an electronic device including a data processing device as provided herein.
[0008] According to one aspect of the present disclosure, there is provided a data processing method, the method including receiving a preheat command, generated upon detecting a computation command to be processed, instructing a computation means sub-array to perform a read operation and a computation operation, entering a preheat state based on the preheat command, and switching from the preheat state to a computation state in response to receiving a target command corresponding to the computation command to be processed.
[0009] According to another aspect of the present disclosure, there is provided an electronic device including at least one processor and a memory communicatively coupled to the at least one processor, the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor such that the at least one processor can perform a method provided in the present disclosure.
[0010] According to another aspect of the present disclosure, a non-transitory computer readable storage medium having computer instructions stored thereon is provided, the computer instructions causing a computer to perform a method provided in the present disclosure.
[0011] According to another aspect of the present disclosure, there is provided a computer program which, when executed by a processor, implements the methods provided in the present disclosure.
[0012] It should be understood that the contents described in this section are not intended to identify key features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will be readily understood from the following specification. [Brief description of the drawings]
[0013] The drawings are provided for a better understanding of the invention and are not intended to limit the disclosure.
[0014] [Figure 1]FIG. 1 is a schematic diagram of the current and voltage of an artificial intelligence chip according to one embodiment of the present disclosure. [Diagram 2] FIG. 2 is a schematic block diagram of a data processor core according to one embodiment of the present disclosure. [Figure 3A] FIG. 3A is a schematic diagram of a computing means sub-array in a preheated state according to one embodiment of the present disclosure. [Figure 3B] FIG. 3B is a schematic diagram of a computing means sub-array in a computing state according to one embodiment of the present disclosure. [Figure 4A] FIG. 4A is a schematic diagram of a computational means array according to one embodiment of the present disclosure. [Figure 4B] FIG. 4B is a schematic diagram of a set of computing means in a preheat state according to one embodiment of the present disclosure. [Figure 4C] FIG. 4C is a schematic diagram of a group of computing means in a preheat state according to one embodiment of the present disclosure. [Figure 4D] FIG. 4D is a schematic diagram of a set of computing means in a preheat state according to one embodiment of the present disclosure. [Figure 4E] FIG. 4E is a schematic diagram of a set of computing means in a preheat state according to one embodiment of the present disclosure. [Figure 4F] FIG. 4F is a schematic diagram of a computing means array in a computing state according to one embodiment of the present disclosure. [Diagram 5] FIG. 5 is a schematic diagram of voltages and currents for a chip including a data processor core according to one embodiment of the present disclosure. [Figure 6] FIG. 6 is a schematic diagram of a data processor according to an embodiment of the present disclosure. [Figure 7] FIG. 7 is a schematic diagram of a data processing device according to an embodiment of the present disclosure. [Figure 8] FIG. 8 is a schematic diagram of an electronic device according to an embodiment of the present disclosure. [Figure 9] FIG. 9 is a schematic flowchart of a data processing method according to an embodiment of the present disclosure. [Figure 10] FIG. 10 is a block diagram of an electronic device to which a data processing method according to an embodiment of the present disclosure can be applied. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0015] Hereinafter, exemplary embodiments of the present disclosure will be described with reference to the drawings, and various details of the embodiments of the present disclosure will be included for ease of understanding, which are merely illustrative. Therefore, as can be understood by those skilled in the art, various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, the following description will omit descriptions of known functions and structures.
[0016] The computational power of the artificial intelligence chip may be proportional to the number of resources of the computing means array. The higher the computational power of the chip, the higher the computational resources required. As a result, the driving current and power consumption of the chip with high computational power are both large. In addition, the operating frequency of the chip is fast, and may be greater than 1 gigahertz (GHz). The chip can reach peak power consumption or peak current in a short time. However, the driving current response speed of the chip's power supply is relatively slow, which will be further described below with reference to FIG. 1.
[0017] FIG. 1 is a schematic diagram of the current and voltage of an artificial intelligence chip according to one embodiment of the present disclosure.
[0018] As shown in FIG. 1, when the peak power consumption of a chip is determined, when switching from an idle current to an operating current quickly, the chip's computing power approaches or reaches the peak computing power, but the supply voltage of the chip power supply decreases abruptly (e.g., decreases to voltage VL11). When switching from an operating current to an idle current quickly, the chip's computing power becomes almost zero, but the supply voltage of the chip power supply increases abruptly (e.g., increases to voltage VH11). Both abrupt increases and decreases in the supply voltage can have adverse effects on the chip's circuitry. For example, abrupt increases in the supply voltage can destroy the circuitry. Abrupt decreases in the supply voltage can prevent the circuitry from obtaining sufficient voltage, preventing the circuitry from operating normally, and potentially resulting in erroneous calculation results.
[0019] In some embodiments, in order to avoid the voltage from increasing or decreasing suddenly, all the computing means of the chip can be divided into blocks, and a plurality of computing means arrays can be obtained. The computing means can be a multiply-accumulate computing means. The computing means array can be arranged to perform a calculation in response to receiving a pulse signal. When the pulse signal is not received, the computing means array can be in an idle state. In the idle state, the computing means array does not perform a read operation, a calculation operation or a write operation. The power consumption of the computing means array can be a standby power consumption. The standby power consumption is a small value, which is smaller than the peak power consumption. This can realize the startup in blocks, and prevent the computing power of the chip from quickly reaching the peak computing power. However, the startup in blocks will affect the computing power of the chip. This is because the multiple computing means arrays of the chip are started in blocks, and in the startup process, the multiple computing means arrays are started sequentially, which limits the computing power, lengthens the computing time of the entire chip, and reduces the performance.
[0020] Based on this, in order to provide chip longevity and keep the chip's computing power high, the present disclosure provides a data processor core, which is further described below.
[0021] FIG. 2 is a schematic block diagram of a data processor core according to one embodiment of the present disclosure.
[0022] As shown in FIG. 2, the data processor core 20 may include a control means 201 and a computing means array 202 .
[0023] The control means 201 may be configured to generate a pre-heat command in response to detecting a computation command to be processed. In an embodiment of the present disclosure, the computation command to be processed may be a command related to a deep learning model and is related to a matrix calculation. For example, the control means generates a pre-heat command after receiving the computation command to be processed.
[0024] The computing means array 202 may include at least one computing means subarray. The computing means subarray may be arranged to receive a preheat command, enter a preheat state in response to the preheat command, and switch from the preheat state to a computing state in response to receiving a target command corresponding to the computing command to be processed.
[0025] In the embodiments of the present disclosure, the computation means array may be various computation means arrays, such as a multiply-add means array, a multiplication means array, or a convolution means array.
[0026] In an embodiment of the present disclosure, the preheat command can instruct the computing means subarray to perform read operations and computation operations. For example, the preheat command may instruct the computing means subarray to perform read operations and computation operations, and may instruct the computing means subarray not to perform write operations.
[0027] In an embodiment of the present disclosure, a target command can be determined based on a command to be processed. For example, the command to be processed can correspond to a processor core or a computing means array. The target command can correspond to a computing means sub-array. The array corresponding to the command to be processed can be divided to determine a sub-array corresponding to the target command.
[0028] According to the embodiment of the present disclosure, the subarray of the computing means enters the preheating state based on the preheating command, so that the computing cores of the entire chip gradually reach or approach the peak power consumption. This can reduce the sudden change in the power consumption of the chip when processing the computing command, and can avoid a sudden increase in the power consumption and a sudden drop in the voltage, which is helpful to improve the performance and life of the chip. In addition, before receiving the target command, the voltage of the chip is relatively stable, which can improve the accuracy of the calculation, which can maintain the computing power of the chip at a high level and help improve the accuracy of the calculation result.
[0029] As can be appreciated, the above describes the processor core of the present disclosure, and the following further describes the preheat and computation states of the present disclosure.
[0030] In an embodiment of the present disclosure, when the control means does not detect a computation command to be processed, the computation means sub-array is in an idle state, in which the computation means sub-array does not perform any read or computation operations.
[0031] In an embodiment of the present disclosure, when the control means does not detect a calculation command to be processed, the calculation means can receive a pre-heat command generated by the control means. The pre-heat command may correspond to the first data. This will be further described below in conjunction with FIG. 3A.
[0032] FIG. 3A is a schematic diagram of a computing means sub-array in a preheated state according to one embodiment of the present disclosure.
[0033] In the embodiment of the present disclosure, the computing means subarray may be configured to read the first data and enter a preheat state. Execute a calculation operation based on the first data. As shown in FIG. 3A, the computing means subarray 3021 may read the first data. Based on the first data D31, the computing means subarray 3021 may execute a calculation operation to obtain a first calculation result corresponding to the preheat command. The first calculation result may not be written to a register or address space. In the process of executing the read operation and the calculation operation, the computing means subarray 3021 may be in a preheat state. As can be understood, the state of the computing means subarray may include an idle state and an operating state. In the embodiment of the present disclosure, the operating state may include a preheat state. The preheat state may be an operating state that executes a read operation and a calculation operation based on the first data. According to the embodiment of the present disclosure, the read operation and the calculation operation are executed based on the first data, which can gradually increase the power consumption of the chip, and is also convenient for realizing the preheat state.
[0034] As can be appreciated, above the preheat state of the present disclosure has been described, and below the calculation state of the present disclosure will be further described.
[0035] FIG. 3B is a schematic diagram of a computing means sub-array in a computing state according to one embodiment of the present disclosure.
[0036] In an embodiment of the present disclosure, the target command corresponds to the second data, and the target command instructs the computing means sub-array to perform a read operation, a computation operation, and a write operation. For example, the second data may be a sub-array associated with the computation command to be processed.
[0037] In an embodiment of the present disclosure, the computing means sub-array may be arranged to read the second data. As shown in Fig. 3B, the computing means sub-array 3021 in the pre-heating state can read the second data D32.
[0038] In an embodiment of the present disclosure, the computing means sub-array may be further configured to overwrite a result obtained by performing a computing operation based on the first data with second data corresponding to the target command. For example, the second data D32 may overwrite the first computation result mentioned above. According to an embodiment of the present disclosure, the second data overwrites the first computation result corresponding to the pre-heat command, so as to efficiently process the target command and save the computation resources of the processor core.
[0039] In the embodiment of the present disclosure, the computing means sub-array may be further configured to perform a computing operation according to the second data to obtain a target computing result. For example, according to the second data D32, the computing means sub-array 3021 can perform a computing operation to obtain a second computing result corresponding to the target command as a target computing result.
[0040] In the embodiment of the present disclosure, the computing means sub-array may be further configured to write the target computation result to an address space corresponding to the target command, and switch from the preheat state to a computation state. For example, the target command may correspond to the address space Addr31. The target computation result may be written to the address space Addr31. In the process of performing the read operation, the computation operation, and the write operation, the computing means sub-array 3021 may be in a computation state. As can be understood, the state of the computing means sub-array may include an idle state and an operating state. In the embodiment of the present disclosure, the operating state may further include a computation state. The computation state may perform the operating states of the read operation, the computation operation, and the write operation based on the second data.
[0041] As can be understood, the pre-heating state and operating state calculation means sub-arrays of the present disclosure have been described above, and the first data and second data of the present disclosure will be further described below.
[0042] In the embodiment of the present disclosure, the data type of the first data and the data type of the second data may be the same. For example, if the second data is floating-point data, the first data may be floating-point data. For example, if the second data is fixed-point data, the first data may be fixed-point data.
[0043] In the embodiment of the present disclosure, the scale difference between the first data and the second data is less than or equal to a predetermined scale difference threshold. For example, if the scale of the second data is [n, c, h, w], the scale of the first data may be [n, c, h, w], where n, c, h and w may be positive integers. According to the embodiment of the present disclosure, the type and scale of the first data and the second data are the same or similar, so that the computing means sub-array can quickly adapt to the data type of the actual computing command, which can further improve the computing power of the chip and improve the computing efficiency.
[0044] As can be appreciated, above the first data and second data of the present disclosure are described, below the control means and computing means sub-arrays of the present disclosure are further described.
[0045] In the embodiment of the present disclosure, the computing means array may include M computing means groups. The computing means groups may include at least one computing means sub-array. M may be an integer equal to or greater than 1. The following description will be given with reference to FIG. 4A.
[0046] FIG. 4A is a schematic diagram of a computational means array according to one embodiment of the present disclosure.
[0047] 4A, the computing means array 402 may include a computing means subarray 40201, a computing means subarray 40202, a computing means subarray 40203, a computing means subarray 40204, a computing means subarray 40205, a computing means subarray 40206, a computing means subarray 40207, a computing means subarray 40208, a computing means subarray 40209, a computing means subarray 40210, a computing means subarray 40211, a computing means subarray 40212, a computing means subarray 40213, a computing means subarray 40214, a computing means subarray 40215, and a computing means subarray 40216. As can be understood, the detailed description of the computing means subarray 40201 to the computing means subarray 40216 can be referred to the above-mentioned computing means subarray 3021, and the detailed description is omitted in this disclosure.
[0048] As shown in FIG. 4A, the computing means array may include four computing means groups. The first computing means group may include computing means subarray 40201 to computing means subarray 40204. The second computing means group may include computing means subarray 40205 to computing means subarray 40208. The third computing means group may include computing means subarray 40209 to computing means subarray 40212. The fourth computing means group may include computing means subarray 40213 to computing means subarray 40216. As can be seen, in this embodiment, M=4.
[0049] As shown in FIG. 4A, the computing means subarrays 40201 to 40216 may be in an idle state.
[0050] In an embodiment of the present disclosure, the control means may be arranged to generate a pre-heat command in response to detecting a computation command to be processed. The control means may then be arranged to output the pre-heat command to at least one computation means sub-array, as further described below with reference to Figures 4B to 4E.
[0051] 4B to 4E are schematic diagrams of a group of computing means in a preheating state according to an embodiment of the present disclosure.
[0052] In an embodiment of the present disclosure, the control means may be configured to output a pre-heat command to the m-th computing means group of the M computing means groups. For example, four pre-heat commands may be output to four computing means sub-arrays of the first computing means group, respectively. The computing means sub-array 40201 to the computing means sub-array 40204 may each receive a pre-heat command and enter a pre-heat state based on the pre-heat command, respectively. The computing means sub-array 40201 to the computing means sub-array 40204 in the pre-heat state are shown in FIG. 4B.
[0053] In the preheat state, the sum of the first power consumptions for the computing means subarrays 40201 to 40204 may be, for example, 20% of the peak power consumption of the computing means array 402.
[0054] In an embodiment of the present disclosure, the control means may be further configured to output a preheat command to the m+1-th computing means group of the M computing means groups in response to determining that the N computing means subarrays in the m-th computing means group enter a preheat state. m may be an integer greater than or equal to 1 and less than M. For example, after determining that the four computing means subarrays of the first computing means group enter a preheat state, the control means may output a preheat command to the four computing means subarrays of the second computing means group. The computing means subarrays 40205 to 40208 each receive a preheat command and each enter a preheat state based on the preheat command. The computing means subarrays 40205 to 40208 in the preheat state are shown in FIG. 4C. According to an embodiment of the present disclosure, the preheat command can be sequentially provided to different computing means subarrays, thereby realizing an increase in power consumption, avoiding abrupt changes in power consumption, and avoiding a voltage that is too low.
[0055] In the preheat state, the sum of the first power consumptions for the computing means subarrays 40201 to 40208 may be, for example, 40% of the peak power consumption of the computing means array 402.
[0056] Next, the control means may output a pre-heat command to the four calculation means sub-arrays of the third calculation means group after determining that the four calculation means sub-arrays of the second calculation means group have entered the pre-heat state. The calculation means sub-arrays 40209 to 40212 each receive the pre-heat command and each enter the pre-heat state based on the pre-heat command. The calculation means sub-arrays 40209 to 40212 in the pre-heat state are shown in FIG. 4D.
[0057] In the preheat state, the sum of the first power consumptions for the computing means subarrays 40201 to 40212 may be, for example, 60% of the peak power consumption of the computing means array 402.
[0058] Next, the control means may output a pre-heat command to the four calculation means sub-arrays of the fourth calculation means group after determining that the four calculation means sub-arrays of the third calculation means group have entered the pre-heat state. The calculation means sub-arrays 40213 to 40216 each receive the pre-heat command and each enter the pre-heat state based on the pre-heat command. The calculation means sub-arrays 40213 to 40216 in the pre-heat state are shown in FIG. 4E.
[0059] In the pre-heat state, for the computing means sub-arrays 40201 to 40216, the sum of the first power consumption may be, for example, 80% of the peak power consumption of the computing means array 402. As can be understood, the sum of the first power consumption is exemplary, and in some other embodiments, in the pre-heat state, for the computing means sub-arrays 40201 to 40216, the sum of the first power consumption may be, for example, equal to the peak power consumption of the computing means array 402.
[0060] As can be seen, above we describe several computing means sub-arrays entering a preheat state, and below we describe several forms of switching to a computing state.
[0061] FIG. 4F is a schematic diagram of a computing means array in a computing state according to one embodiment of the present disclosure.
[0062] In an embodiment of the present disclosure, the control means may be further configured to output a target command to at least one target computing means subarray in the computing means subarray in the preheating state in response to determining that the number of computing means subarrays in the preheating state is equal to or greater than the predetermined number threshold. For example, the predetermined number threshold may be related to the number of computing means subarrays in the computing means array. As shown in FIG. 4E, the computing means array 402 includes 16 computing means subarrays. The predetermined number threshold may be, for example, 12 or 16. Taking the predetermined number threshold as an example that the computing means subarrays 40201 to 40216 all enter the preheating state, the control means outputs a target command to the computing means subarrays 40201 to 40216 respectively, and the computing means subarrays 40201 to 40216 respectively read the second data based on the respective target commands, and further perform a calculation operation and a write operation respectively, and enter a calculation state. The computing means sub-arrays 40201 to 40216 in the computing state are shown in FIG. 4F.
[0063] In an embodiment of the present disclosure, the power consumption difference between the first power consumption corresponding to the preheat state and the second power consumption corresponding to the calculation state is less than or equal to a predetermined power consumption difference threshold. For example, in the calculation state, for the calculation means subarray 40201 to the calculation means subarray 40216, the sum of the second power consumption may be, for example, 100% of the peak power consumption of the calculation means array 402. The predetermined power consumption difference threshold may be, for example, 25%, and the power consumption difference (20%) between the sum of the second power consumption and the sum of the first power consumption is less than the predetermined power consumption difference threshold (25%).
[0064] For understanding, the above describes the present disclosure by taking the control means 201 and the computing means array 202 as examples, but the present disclosure is not limited thereto, and the control means may be at least one, and the computing means array may be at least one, as will be further described below.
[0065] In the embodiment of the present disclosure, the control means may be multiple, the computing means array may be multiple, and the control means may correspond to one computing means array. For example, the multiple control means may include a first control means and a second control means. The multiple computing means array may include a first computing means array and a second computing means array. The first control means may correspond to the first computing means array and generate a pre-heat command for the first computing means array. The second control means may correspond to the second computing means array and generate a pre-heat command for the second computing means array.
[0066] In some other embodiments of the present disclosure, the control means may correspond to multiple computing means arrays, and multiple control means may correspond to one computing means array, and the present disclosure is not limited thereto.
[0067] As can be appreciated, the control means and computing means array of the present disclosure have been described above, and the control means of the present disclosure will be further described below.
[0068] In an embodiment of the present disclosure, the control means may be further configured to, in response to determining that the target command has been executed, output a pre-heat command to a target computing means subarray in at least one computing state of the computing means subarrays. For example, the control means may output a pre-heat command to the computing means subarrays 40201 to 40216 so that the computing means subarrays 40201 to 40216 are switched from a computing state to a pre-heat state, respectively.
[0069] In the embodiment of the present disclosure, the control means may be further arranged to output a predetermined command to the m-th computing means group of the M computing means groups. For example, four predetermined commands may be output to four computing means subarrays of the first computing means group, respectively. The computing means subarrays 40201 to 40204 each receive a predetermined command and can switch from the preheat state to the idle state based on the predetermined command.
[0070] In an embodiment of the present disclosure, the control means may be further arranged to output a predetermined command to the m+1-th computing means group of the M computing means groups in response to determining that the N computing means subarrays in the m-th computing means group are switched to the idle state. For example, after it is determined that the four computing means subarrays of the first computing means group are switched to the idle state, the control means may output a predetermined command to the four computing means subarrays of the second computing means group. The computing means subarrays 40205 to 40208 may each receive a predetermined command and each switch to the idle state based on the predetermined command.
[0071] Next, the control means may output a predetermined command to the four computing means subarrays of the third computing means group after determining that the four computing means subarrays of the second computing means group have been switched to the idle state. The computing means subarrays 40209 to 40212 each receive the predetermined command and switch to the idle state based on the predetermined command.
[0072] Next, the control means may output a predetermined command to the four computing means subarrays of the fourth computing means group after determining that the four computing means subarrays of the third computing means group are switched to the idle state. The computing means subarrays 40213 to 40216 each receive a predetermined command and switch to the idle state based on the preheat command. According to the embodiment of the present disclosure, the computing means arrays can be controlled to gradually reduce the power consumption from a predetermined power consumption value to standby power consumption. This can further reduce the sudden change in the power consumption of the chip after completing the calculation, without a large sudden drop in power consumption and without a sudden increase in voltage, which is helpful to further improve the performance and life of the chip.
[0073] As can be appreciated, above is a further explanation of the preheat command of the present disclosure, below is a discussion of voltages and currents of a chip including a data processor core of the present disclosure.
[0074] FIG. 5 is a schematic diagram of voltages and currents for a chip including a data processor core according to one embodiment of the present disclosure.
[0075] As described above, when the peak power consumption of the chip is determined, when the idle current is quickly switched to the operating current, the computing power of the chip approaches or reaches the peak computing power, but the power supply voltage of the chip power supply drops suddenly to voltage VL51.
[0076] 5, when the peak power consumption of the chip including, for example, processor core 20 is determined, the chip is gradually switched from an idle current to an operating current based on a preheat command. When the computing power of the chip approaches or reaches the peak computing power, the supply voltage of the chip power supply is abruptly reduced to a voltage VL52.
[0077] As shown in Figure 5, the voltage VL52 is greater than the voltage VL51, so that when the chip gradually switches from the idle current to the working current based on the preheat command, the voltage fluctuation range of the chip is greatly reduced, which is helpful in improving the performance and lifespan of the chip.
[0078] Also, as described above, when switching quickly from the operating current to the idle current, the computing power of the chip becomes almost zero, but the power supply voltage of the chip power supply increases rapidly to VH51.
[0079] 5, when the peak power consumption of the chip including the processor core 20 is determined, the chip gradually switches from the operating current to the idle current based on the preheat command. When the computing power of the chip is almost zero, the supply voltage of the chip power supply also increases rapidly to the voltage VH52.
[0080] As shown in Figure 5, the voltage VH52 is smaller than the voltage VH51, so that when the chip gradually switches from the working current to the idle current based on the preheat command and the above-mentioned predetermined command, the voltage fluctuation range of the chip is greatly reduced, which is helpful in improving the performance and life of the chip.
[0081] As can be appreciated, above, a data processor core of the present disclosure is described, and below, a data processor including the data processor core is described.
[0082] FIG. 6 is a schematic diagram of a data processor according to an embodiment of the present disclosure.
[0083] 6, a data processor P600 may include at least one data processor core 60. The data processor core 60 may be, for example, the data processor core 20 described above.
[0084] As can be appreciated, the disclosure has been described above as a data processor, and hereinafter as a data processing device including a data processor.
[0085] FIG. 7 is a schematic diagram of a data processing device according to an embodiment of the present disclosure.
[0086] 7, the data processing device 700 may include a data processor P700. The data processor P700 may be, for example, the data processor P600 described above.
[0087] As can be appreciated, the above describes a data processing device of the present disclosure, and the following describes an electronic device that includes the data processing device.
[0088] FIG. 8 is a schematic diagram of an electronic device according to an embodiment of the present disclosure.
[0089] 8, an electronic device 8000 may include a data processing device 800. The data processing device 800 may be, for example, the data processing device 700 described above.
[0090] For the sake of understanding, the electronic device of the present disclosure has been described above, and the data processing method of the present disclosure will now be described.
[0091] FIG. 9 is a schematic flowchart of a data processing method according to an embodiment of the present disclosure.
[0092] As shown in FIG. 9, the method 900 may include operations S910 to S930.
[0093] In operation S910, a pre-heat command is received. In an embodiment of the present disclosure, the pre-heat command is generated upon detecting a computation command to be processed.
[0094] In operation S920, a pre-heat state is entered based on a pre-heat command.
[0095] In operation S930, in response to receiving a target command corresponding to a computation command to be processed, the preheat state is switched to a computation state.
[0096] In an embodiment of the present disclosure, the method 900 can be implemented using the above-mentioned computing means sub-array.
[0097] In an embodiment of the present disclosure, the pre-heat command instructs the computing means sub-array to perform read and computing operations. For example, the pre-heat command may be generated by the control means described above.
[0098] In some embodiments, the pre-heat command corresponds to the first data.
[0099] In some embodiments, entering the pre-heat state based on the pre-heat command includes reading first data and entering the pre-heat state. Performing a computation operation based on the first data.
[0100] In some embodiments, the target command corresponds to the second data, and the target command instructs the computing means sub-array to perform a read operation, a computation operation, and a write operation.
[0101] In some embodiments, switching from the preheat state to the computation state includes reading second data and switching from the preheat state to the computation state. Performing a computation operation based on the second data to obtain a target computation result. The target computation result is written to an address space corresponding to the target command.
[0102] In some embodiments, the method 900 further includes overwriting a result obtained by performing a computational operation based on the first data with second data corresponding to the target command.
[0103] In some embodiments, a target command is output to a target computing means subarray in at least one computing means subarray in the preheat state when it is determined that the number of computing means subarrays in the preheat state is equal to or greater than a predetermined number threshold.
[0104] In some embodiments, the computing means array includes M computing means groups, each of which includes at least one computing means sub-array, where M is an integer greater than or equal to one.
[0105] In some embodiments, the method 900 further includes outputting a pre-heat command to an m-th group of computing means of the M group of computing means, and outputting a pre-heat command to an m+1-th group of computing means of the M group of computing means in response to determining that N computing means sub-arrays in the m-th group of computing means enter the pre-heat state, where m is an integer greater than or equal to 1 and less than M.
[0106] In some embodiments, a power consumption difference between the first power consumption corresponding to the preheat state and the second power consumption corresponding to the calculation state is less than or equal to a predetermined power consumption difference threshold.
[0107] In some embodiments, the preheat command corresponds to the first data, the target command corresponds to the second data, a data type of the first data and a data type of the second data match, and a magnitude difference between the first data and the second data is less than or equal to a predetermined magnitude difference threshold.
[0108] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure and application of such user personal information shall all comply with the provisions of relevant laws and regulations, adopt necessary confidentiality measures, and not violate public order and morals.
[0109] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program product.
[0110] 10 shows an exemplary block diagram for implementing an exemplary electronic device 1000 according to an embodiment of the present disclosure. The electronic device is intended to represent various types of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various types of mobile devices, such as personal digital assistants, mobile phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure as described and / or claimed herein.
[0111] As shown in Fig. 10, the electronic device 1000 includes a computing means 1001, which can execute various appropriate operations and processes based on a computer program stored in a read-only memory (ROM) 1002 or loaded from a storage means 1008 into a random access memory (RAM) 1003. The RAM 1003 can further store various programs and data necessary for the operation of the electronic device 1000. The computing means 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.
[0112] A number of components in the electronic device 1000 are connected to an I / O interface 1005, and include input means 1006, such as a keyboard, a mouse, etc., output means 1007, such as various types of displays, speakers, etc., storage means 1008, such as a magnetic disk, an optical disk, etc., and communication means 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication means 1009 enables the electronic device 1000 to exchange information / data with other devices via a computer network, such as the Internet, and / or various types of electric communication networks.
[0113] The computing means 1001 may be various general-purpose and / or dedicated processing modules having processing and computing capabilities. Some examples of the computing means 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various machine learning model algorithm computing means, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing means 1001 executes each of the methods and processes described above, such as the data processing methods. For example, in some embodiments, the data processing methods may be realized as a computer software program tangibly embodied in a machine-readable medium, such as the storage means 1008. In some embodiments, some or all of the computer programs may be loaded and / or installed in the electronic device 1000 via the ROM 1002 and / or the communication means 1009. When the computer programs are loaded into the RAM 1003 and executed by the computing means 1001, one or more steps of the data processing methods described above may be executed. Alternatively, in another embodiment, the computing means 1001 may be configured to perform the data processing method in any other suitable manner (eg via firmware).
[0114] Various embodiments of the systems and techniques described herein may be realized in digital electronic circuitry systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may be embodied in one or more computer programs that may be executed and / or interpreted by a programmable system that includes at least one programmable processor, which may be a special purpose or general purpose programmable processor, and may include a processor that is capable of receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.
[0115] The program codes for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer or other programmable data processing apparatus, so that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are performed. The program codes may be executed entirely on the device, partially on the device, partially on the device as a separate software package and partially on a remote device or entirely on a remote device or server.
[0116] In the context of this disclosure, a machine-readable medium may be a tangible medium, and may include or store a program for use in or in combination with an instruction execution system, apparatus, or electronic device. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or electronic device, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections through one or more wires, portable computer disks, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), optical fibers, compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.
[0117] A computer may implement the systems and techniques described herein to provide interaction with a user, and may include a display device (e.g., a CRT (cathode ray tube) display or LCD (liquid crystal display)) for displaying information to a user, and a keyboard and pointing device (e.g., a mouse or trackball) through which a user may provide input to the computer. Other types of devices may also provide interaction with a user, for example, the feedback provided to the user may be any form of sensing feedback (e.g., visual feedback, auditory feedback, or tactile feedback) and may receive input from the user in any form (including voice input, speech input, or tactile input).
[0118] It should be understood that various types of flows shown above may be used, and steps may be rearranged, added, or deleted. For example, each step described in the present invention may be performed in parallel, sequentially, or in a different order, and the present specification is not limited thereto as long as the desired results of the technical solution of the present disclosure can be achieved.
[0119] The specific embodiments described above do not limit the scope of protection of the present disclosure. Those skilled in the art should understand that various modifications, combinations, subcombinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principle of the present disclosure should be included within the scope of protection of the present disclosure.
Claims
1. 1. A data processor core, comprising: control means arranged to generate a preheat command in response to detecting a computation command to be processed; a computing means array including at least one computing means subarray; The computing means subarray comprises: receiving the preheat command instructing the computing means sub-array to perform a read operation and a computing operation; Entering a preheat state based on the preheat command; and configured to switch from the preheat state to a calculation state in response to receiving a target command corresponding to the calculation command to be processed. Processor core.
2. The preheat command corresponds to first data. The processor core of claim 1 .
3. The computing means subarray further comprises: reading the first data and entering the preheat state; configured to perform a computational operation based on the first data. The processor core of claim 2.
4. the target command corresponds to second data; The target commands instruct the computation means sub-arrays to perform read operations, computation operations, and write operations. The processor core of claim 1 .
5. The computing means subarray further comprises: reading the second data and switching from the preheating state to the calculation state; Perform a calculation operation based on the second data to obtain a target calculation result; and arranged to write said target computation result into an address space corresponding to said target command. The processor core of claim 4.
6. The computing means subarray further comprises: and configured to overwrite a result obtained by performing a computational operation based on said first data based on second data corresponding to said target command. The processor core of claim 2.
7. The control means further comprises: and configured to output the target command to a target computing means subarray in at least one of the computing means subarrays in the preheat state in response to determining that the number of the computing means subarrays in the preheat state is equal to or greater than a predetermined number threshold. The processor core of claim 1 .
8. The computing means array includes M computing means groups (M is an integer equal to or greater than 1), and the computing means groups include at least one of the computing means subarrays; The control means further comprises: outputting the preheat command to an m-th (m is an integer equal to or greater than 1 and less than M) calculation means group of the M calculation means; is configured to output the pre-heat command to an m+1-th computing means group of M computing means in response to the N computing means sub-arrays in the m computing means group determining to enter the pre-heat state. The processor core of claim 1 .
9. A power consumption difference between a first power consumption corresponding to the preheat state and a second power consumption corresponding to the calculation state is less than or equal to a predetermined power consumption difference threshold. The processor core of claim 1 .
10. The preheat command corresponds to first data, The target command corresponds to second data; a data type of the first data corresponds to a data type of the second data; A magnitude difference between the first data and the second data is less than or equal to a predetermined magnitude difference threshold. The processor core of claim 1 .
11. The control means is a plurality of means, The computing means array is a plurality of arrays, and the control means corresponds to one of the computing means arrays. The processor core of claim 1 .
12. Comprising at least one data processor core according to any one of claims 1 to 11. Data processor.
13. A data processor according to claim 12. Data processing device.
14. A data processing device comprising: electronic equipment.
15. receiving a preheat command generated upon detecting a computation command to be processed, the preheat command directing the computation means subarray to perform a read operation and a computation operation; entering a preheat state based on the preheat command; and switching from the preheat state to a computation state in response to receiving a target command corresponding to the computation command to be processed. Data processing methods.
16. The preheat command corresponds to first data. The method of claim 15.
17. Entering a pre-heat state based on the pre-heat command includes: reading the first data and entering the preheat state; performing a computation operation based on the first data.
17. The method of claim 16.
18. The target command corresponds to second data, and the target command directs the computing means sub-array to perform a read operation, a computation operation, and a write operation. The method of claim 15.
19. Switching from the preheat state to a calculation state comprises: reading the second data and switching from the preheating state to the calculating state; performing a calculation operation based on the second data to obtain a target calculation result; writing the target computation result into an address space corresponding to the target command.
20. The method of claim 18.
20. and overwriting a result obtained by performing a calculation operation based on the first data based on second data corresponding to the target command.
17. The method of claim 16.
21. The target command is output to a target computing means subarray in at least one of the computing means subarrays in the preheating state when it is determined that the number of the computing means subarrays in the preheating state is equal to or greater than a predetermined number threshold. The method of claim 15.
22. the computing means array includes M computing means groups (M is an integer equal to or greater than 1), the computing means groups including at least one of the computing means subarrays; The method comprises: outputting the preheat command to an m-th (m is an integer equal to or greater than 1 and less than M) calculation means group of the M calculation means; In response to determining that the N computing means subarrays in the m computing means group enter the preheat state, outputting the preheat command to the m+1-th computing means group of the M computing means group. The method of claim 15.
23. A power consumption difference between a first power consumption corresponding to the preheat state and a second power consumption corresponding to the calculation state is less than or equal to a predetermined power consumption difference threshold. The method of claim 15.
24. The preheat command corresponds to first data, The target command corresponds to second data; a data type of the first data and a data type of the second data are the same; A magnitude difference between the first data and the second data is less than or equal to a predetermined magnitude difference threshold. The method of claim 15.
25. At least one processor; a memory in communication with the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor such that the at least one processor can perform the method according to any one of claims 15 to 24. electronic equipment.
26. A non-transitory computer-readable storage medium having computer instructions stored thereon, comprising: The computer instructions cause a computer to carry out a method according to any one of claims 15 to 24. A non-transitory computer-readable storage medium having computer instructions stored thereon.
27. A computer program which, when executed by a processor, implements the method according to any one of claims 15 to 24.
Citation Information
Patent Citations
Processor and method for driving the same
JP2013229019A
Semiconductor integrated circuit and compiler
JP2014059761A
