A memory bandwidth optimization circuit for a video encoder

By coordinating the scheduling controller and the dynamic partitioning cache circuit, real-time dynamic resource adaptation for video encoder memory access is achieved, solving the problem of insufficient memory bandwidth optimization in existing technologies and improving the utilization efficiency of storage resources.

CN224959681UActive Publication Date: 2026-11-10SHAANXI LONGZHOU TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202522634500.7
Authority / Receiving Office
CN · China
Patent Type
Utility models(China)
Current Assignee / Owner
Filing Date
2025-12-12
Publication Date
2026-11-10
Estimated Expiration
2035-12-12

AI Technical Summary

Technical Problem

Existing memory bandwidth optimization solutions for video encoding chips cannot adjust storage resource allocation in real time based on encoding data, and lack unified hardware scheduling coordination, resulting in limited caching efficiency and difficulty in optimizing memory bandwidth requirements.

Method used

By employing a collaborative scheduling controller, a dynamic partitioned caching unit, an adaptive data compression unit, and a data prefetching unit, and through the collaboration of hardware collaborative scheduling and dynamic partitioned caching circuitry, real-time analysis and dynamic resource adaptation of video encoding memory access are achieved.

Benefits of technology

It achieves real-time, three-dimensional optimization of memory bandwidth, improves the utilization efficiency of storage resources, and is suitable for video encoding chips with high real-time requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN224959681U_ABST
    Figure CN224959681U_ABST
Patent Text Reader

Abstract

This utility model relates to the field of circuit design, and more particularly to a memory bandwidth optimization circuit for a video encoder, including a collaborative scheduling controller, a dynamic partitioned buffer unit, an adaptive data compression unit, and a data prefetch unit. The collaborative scheduling controller includes a hardware access counter array, an address range matching circuit, and a threshold determination circuit. The control terminal of the adaptive data compression unit is connected to the compression control terminal of the collaborative scheduling controller, and its data terminal is connected between the dynamic partitioned buffer unit and an external memory controller. The dynamic partitioned buffer unit includes a partition management circuit, a configurable static random access memory array, and a region metadata table. The configuration terminal of the partition management circuit constitutes the configuration terminal of the dynamic partitioned buffer unit, which is connected to the partition configuration terminal of the collaborative scheduling controller. This utility model, through the collaboration of hardware collaborative scheduling and the dynamic partitioned buffer circuit, realizes real-time analysis and dynamic resource adaptation of video encoding memory access, thereby comprehensively reducing bandwidth requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This utility model relates to the field of circuit design, and in particular to a memory bandwidth optimization circuit for a video encoder. Background Technology

[0002] In the circuit design of video encoding chips, reducing external storage bandwidth is crucial.

[0003] Existing hardware solutions mostly employ static optimization, but they have the following drawbacks.

[0004] For example, caching circuits with fixed structures cannot dynamically adjust storage resource allocation based on the real-time access frequency of encoded data, thus limiting caching efficiency. In addition, auxiliary circuits such as compression and prefetching usually work independently, lacking unified hardware scheduling coordination based on real-time access patterns, making it difficult to form a synergy for optimization.

[0005] Therefore, it is necessary to further improve the memory bandwidth optimization circuit of a video encoder. Utility Model Content

[0006] The technical problem to be solved by this utility model is to overcome the existing defects and provide a memory bandwidth optimization circuit for a video encoder. Through the coordination of hardware collaborative scheduling and dynamic partitioning caching circuit, it realizes real-time analysis and dynamic resource adaptation of video encoding memory access, thereby reducing bandwidth requirements in a three-dimensional way and effectively solving the problems in the background technology.

[0007] To achieve the above objectives, this utility model provides the following technical solution: a memory bandwidth optimization circuit for a video encoder, comprising a collaborative scheduling controller, a dynamic partitioning buffer unit, an adaptive data compression unit, and a data prefetching unit;

[0008] The collaborative scheduling controller includes a hardware access counter array, an address range matching circuit, and a threshold determination circuit.

[0009] The hardware access counter array consists of multiple counters configured in parallel; the address input terminal of the address range matching circuit is used to receive memory access addresses from the video encoding core, and it has multiple matching output terminals, each of which is connected to the clock enable terminal of a corresponding counter in the hardware access counter array; the threshold determination circuit has multiple count value input terminals and multiple determination result output terminals; the multiple count value input terminals are respectively connected to the count output terminals of the corresponding counters in the hardware access counter array; the multiple determination result output terminals of the threshold determination circuit constitute the signal source for the partition configuration terminal, compression control terminal, and prefetch control terminal of the collaborative scheduling controller;

[0010] The control terminal of the adaptive data compression unit is connected to the compression control terminal of the cooperative scheduling controller, and its data terminal is connected between the dynamic partition cache unit and the external memory controller.

[0011] The dynamic partition cache unit includes a partition management circuit, a configurable static random access memory array, and a region metadata table; the update end of the region metadata is connected to the partition configuration end of the cooperative scheduling controller, and its output end is connected to the partition management circuit; the configuration end of the partition management circuit constitutes the configuration end of the dynamic partition cache unit, which is connected to the partition configuration end of the cooperative scheduling controller and is used to receive partition configuration signals; the address mapping output end and data end of the partition management circuit are respectively connected to the address decoding end and data end of the configurable static random access memory array.

[0012] The control terminal of the data prefetching unit is connected to the prefetching control terminal of the cooperative scheduling controller, and its prefetching data output terminal is connected to the data terminal of the configurable static random access memory array for writing prefetched data into the cache.

[0013] Preferably, the partition management circuit includes a reconfiguration arbitration circuit, a programmable partition configuration register set, and an address remapping circuit;

[0014] The first input of the reconfiguration arbitration circuit is connected to the output of the regional metadata table, the second input is connected to the partition configuration terminal of the cooperative scheduling controller, and the output of the reconfiguration arbitration circuit is connected to the write terminal of the programmable partition configuration register group for outputting partition configuration signals.

[0015] The write end of the programmable partition configuration register group constitutes the configuration end of the partition management circuit; the first input end of the address remapping circuit is connected to the output end of the programmable partition configuration register group and is used to receive partition configuration information; the second input end is used to receive the memory access address from the video encoding core.

[0016] Preferably, the configurable static random access memory array includes at least two independent memory banks; the partition management circuit further includes a memory bank selection decoding circuit; the output terminal of the address remapping circuit is connected to the input terminal of the memory bank selection decoding circuit; the memory bank selection decoding circuit has multiple enable signal output terminals, which are respectively connected to the enable signal terminals of the corresponding memory banks.

[0017] Preferably, the adaptive data compression unit includes a first compression engine circuit, a second compression engine circuit, and a data selector;

[0018] The input terminals of the first compression engine circuit and the second compression engine circuit are connected in parallel and are jointly connected to the data input terminal of the adaptive data compression unit; the output terminal of the first compression engine circuit is connected to the first data input terminal of the data selector, and the output terminal of the second compression engine circuit is connected to the second data input terminal of the data selector; the control terminal of the data selector constitutes the control terminal of the adaptive data compression unit, and its output terminal constitutes the data output terminal of the adaptive data compression unit, which is connected to an external memory controller to complete the writing of the compressed data.

[0019] Preferably, the first compression engine circuit is a lossless compression circuit based on run-length encoding; the second compression engine circuit is a lossy compression circuit based on differential pulse code modulation or dictionary encoding.

[0020] Preferably, the data prefetching unit includes an address sequence buffer and an address increment calculation circuit; the input of the address sequence buffer is used to receive memory access addresses, and its output is connected to the input of the address increment calculation circuit; the output of the address increment calculation circuit is connected to the address input of the dynamic partition cache unit and is used to output the prefetch address.

[0021] Compared with the prior art, the beneficial effects of this utility model are:

[0022] The hardware-implemented collaborative scheduling controller can generate partition configuration, compression control, and prefetch control signals in parallel based on the comparison between real-time access counts and preset thresholds. This synchronously drives the hardware circuits of modules such as dynamic partition caching, adaptive compression, and data prefetching to work together, achieving real-time and three-dimensional optimization of memory bandwidth.

[0023] By incorporating a partition management circuit that includes a programmable partition configuration register group and an address remapping circuit, the configurable static random access memory array can be physically and dynamically divided into different regions such as cache and note-taking memory according to real-time control signals. This enables hardware-level dynamic reconfiguration and efficient reuse of on-chip storage resources, making it suitable for video encoding chips with high real-time requirements. Attached Figure Description

[0024] Figure 1 This is a general structural block diagram of the present invention;

[0025] Figure 2 This is a schematic diagram of the internal structure and signal flow of the dynamic partition cache module of this utility model. Detailed Implementation

[0026] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0027] This utility model circuit is integrated into the video encoder as an independent hardware acceleration module; this utility model circuit is connected to the video encoding core and external memory controller through a standard bus interface.

[0028] Please see Figure 1-2 This utility model provides a technical solution: a memory bandwidth optimization circuit for a video encoder, including a collaborative scheduling controller, a dynamic partitioning cache unit, an adaptive data compression unit, and a data prefetching unit;

[0029] The collaborative scheduling controller includes a hardware access counter array, an address range matching circuit, and a threshold determination circuit.

[0030] The hardware access counter array consists of multiple parallel binary counters, each responsible for counting the number of memory access requests occurring within a specific address range.

[0031] Each counter includes automatic reset logic or saturation hold logic. For example, the counter can stop counting and hold after reaching the maximum count value, or the coordinated scheduling controller can send a global reset signal to clear all counters to zero according to a preset period to prevent statistical function failure due to long-term accumulation.

[0032] The address input terminal of the address range matching circuit is used to receive the memory access address from the video encoding core. It has multiple matching output terminals, each of which is connected to the clock enable terminal of a corresponding counter in the hardware access counter array.

[0033] Specifically, the address range matching circuit can be composed of multiple parallel address comparators. Each comparator has two registers, used to store the upper and lower limits of an address range, respectively. The address inputs of all comparators are connected in parallel, collectively receiving the memory access address from the video encoding core. Each comparator compares the input address with its internally stored address range. When the input address value is greater than or equal to the lower limit and less than or equal to the upper limit, the matching output of that comparator outputs a high level. The matching output of each comparator is directly connected to the clock enable terminal of a corresponding counter in the hardware access counter array. When the matching output is high, the enable signal of the corresponding counter is valid, and counting and accumulation begin on the rising edge of the system clock.

[0034] The threshold determination circuit has multiple count value input terminals and multiple determination result output terminals; the multiple count value input terminals are respectively connected to the count output terminals of the corresponding counters in the hardware access counter array; the multiple determination result output terminals of the threshold determination circuit constitute the signal source of the partition configuration terminal, compression control terminal and prefetch control terminal of the collaborative scheduling controller.

[0035] Specifically, the threshold determination circuit includes a threshold register group, three parallel comparator submodules, an output latch and synchronization module, and an associated address FIFO.

[0036] The threshold register group consists of multiple programmable registers used for hardware storage of preset judgment benchmark values, including: a first threshold corresponding to high-heat data, a second threshold corresponding to medium-heat data, and a step size stability threshold corresponding to access mode stability.

[0037] The input terminals of the three comparator submodules are all connected to the count output terminals of the hardware access counter array.

[0038] The first comparator module contains a set of comparators whose inputs are connected to a counter used for statistical high-temperature data access, and compares the count value with a first threshold in the threshold register group in parallel; when the count value exceeds the first threshold, the module directly outputs the corresponding physical level signal, and this set of signals directly constitutes the control signal line of the partition configuration terminal.

[0039] The second comparator module contains two sets of comparators, whose inputs are connected to a counter used to count accesses to the data to be compressed. This module compares the count value with the first threshold and the second threshold in the threshold register group in parallel. Its output is connected to the output encoding module, which generates a preliminary multi-bit compression type code based on the combined state of the two sets of comparison results (such as "above the first threshold", "between the first and second thresholds", "below the second threshold").

[0040] The third comparator module contains at least one comparator whose input is connected to a counter used for statistical access step-size stabilization period; the module compares the count value with the step-size stabilization threshold in the threshold register group, and generates an initial prefetch enable pulse when the count value reaches the threshold.

[0041] The write end of the associated address FIFO is synchronously connected to the address input end of the address range matching circuit. Whenever a memory access address is processed by the matching circuit and drives the corresponding counter, a copy of the address and its corresponding data type identifier (which can be generated by the matching circuit at the same time) are synchronously pushed into the associated address FIFO. The depth of the associated address FIFO is matched with the typical number of delay cycles required for the hardware access counter to reach a stable determination.

[0042] The output latch and synchronization module contains multiple latches. Its input receives the preliminary judgment results from three comparator submodules, and its trigger is connected to a delayed synchronization clock that is synchronized with the reading rhythm of the address flow through the associated address FIFO.

[0043] The dynamic partition cache unit includes a partition management circuit, a configurable static random access memory array, and a region metadata table; the region metadata table may be composed of a set of status registers.

[0044] The update end of the regional metadata table is connected to the partition configuration end of the collaborative scheduling controller, and its output end is connected to the partition management circuit; the configuration end of the partition management circuit constitutes the configuration end of the dynamic partition cache unit, which is connected to the partition configuration end of the collaborative scheduling controller and is used to receive partition configuration signals; the address mapping output end and data end of the partition management circuit are respectively connected to the address decoding end and data end of the configurable static random access memory array.

[0045] Furthermore, the partition management circuit includes a reconfiguration arbitration circuit, a programmable partition configuration register set, and an address remapping circuit;

[0046] The first input of the reconfiguration arbitration circuit is connected to the output of the regional metadata table to monitor the status and heat of each region; the second input is connected to the partition configuration terminal of the collaborative scheduling controller to receive the high heat judgment signal for newly accessed data; the output of the reconfiguration arbitration circuit is connected to the write terminal of the programmable partition configuration register group to output partition configuration instructions.

[0047] The write end of the programmable partition configuration register group constitutes the configuration end of the partition management circuit, and its write operation is controlled by the reconfiguration arbitration circuit. The information stored in the register defines the partitioning method of the current configurable static random access memory array. For example, memory bank 0 and memory bank 1 are used as hardware-managed caches, memory bank 2 is used as SPM, and a specific mapping base address X is set for the SPM region.

[0048] The region metadata table corresponds to each storage bank or storage region of the configurable static random access memory array. Each entry records at least the current status of the region as "Cache" or "SPM", an internal heat counter, and the primary data identifier currently stored there. The update terminal of the region metadata table is connected to the partition configuration terminal of the cooperative scheduling controller to receive data heat determination results.

[0049] When the arbitration circuit determines that an SPM region needs to be allocated for new high-temperature data but all SPM regions are full, it initiates a reconfiguration process: based on the internal heat counter in the region metadata table, it selects an existing SPM region for downgrading, changes its status to Cache, and triggers a write-back or invalidation operation for the data in that region; at the same time, the arbitration circuit outputs new partition configuration information to the programmable partition configuration register group to complete the dynamic reconfiguration.

[0050] The first input of the address remapping circuit is connected to the output of the programmable partition configuration register group to receive the latest partition configuration information; the second input is used to receive the memory access address from the video encoding core; the address remapping circuit converts the input linear address into the internal physical address of the target memory according to the configuration information.

[0051] Furthermore, the configurable static random access memory array includes at least two independent memory banks; the partition management circuit also includes a memory bank selection decoding circuit; the output terminal of the address remapping circuit (containing target memory bank identification information) is connected to the input terminal of the memory bank selection decoding circuit; the memory bank selection decoding circuit has multiple enable signal output terminals, and the decoding circuit selects one of its multiple enable signal output terminals according to the identifier, sets it to an active level, and connects it to the enable signal terminal of the corresponding memory bank to activate the memory bank for read and write operations.

[0052] The control terminal of the adaptive data compression unit is connected to the compression control terminal of the cooperative scheduling controller to receive compression control signals. Its data terminal is connected between the dynamic partition cache unit and the external memory controller to selectively compress data written to the external memory according to the signal.

[0053] Furthermore, the adaptive data compression unit includes a first compression engine circuit, a second compression engine circuit, and a data selector;

[0054] The input terminals of the first compression engine circuit and the second compression engine circuit are connected in parallel and are jointly connected to the data input terminal of the adaptive data compression unit; the output terminal of the first compression engine circuit is connected to the first data input terminal of the data selector, and the output terminal of the second compression engine circuit is connected to the second data input terminal of the data selector; the control terminal of the data selector constitutes the control terminal of the adaptive data compression unit, and its output terminal constitutes the data output terminal of the adaptive data compression unit, which is connected to an external memory controller to complete the writing of the compressed data.

[0055] Furthermore, the first compression engine circuit is a lossless compression circuit based on run-length encoding, which is suitable for compressing structured data such as motion vectors; the second compression engine circuit is a lossy compression circuit based on differential pulse code modulation or dictionary encoding, which is suitable for compressing image residual data at a higher ratio under controllable distortion.

[0056] The control terminal of the data prefetching unit is connected to the prefetching control terminal of the cooperative scheduling controller, and its prefetching data output terminal is connected to the data terminal of the configurable static random access memory array, which is used to actively load the predicted data from the external memory into the dynamic partition cache unit according to the prefetching control.

[0057] Furthermore, the data prefetching unit includes an address sequence buffer and an address increment calculation circuit; the address sequence buffer may be a configurable first-in-first-out memory circuit, whose input is used to receive memory access addresses, and whose output is connected to the input of the address increment calculation circuit; the output of the address increment calculation circuit is connected to the address input of the dynamic partition cache unit and is used to output the prefetch address.

[0058] The data prefetching unit is configured to, when its control terminal is activated, asynchronously initiate a read operation to the external memory controller, independent of the current access request of the video encoding core. The prefetched data is written to the buffer area allocated for prefetched data in the configurable static random access memory array through a dedicated or arbitrated data path.

[0059] When the video encoding core initiates a memory request, the access address first enters this optimized circuit.

[0060] The coordinated scheduling controller captures the address in real time and identifies the data type of the address through its internal address range matching circuit, thereby enabling the corresponding hardware access counter to accumulate. At the same time, a copy of the address is stored in the associated address FIFO within the controller.

[0061] The threshold determination circuit compares the real-time count values ​​of each counter with the preset threshold in parallel, and the determination result is synchronized with the original address stream that triggered the statistical trend.

[0062] When an address (denoted as address A) is read from the associated address FIFO, the output latch and synchronization module in the threshold determination circuit captures the statistical determination results (such as popularity, compression type, prefetch enable) accumulated from address A and its subsequent similar accesses, and latches these determination results with address A. Subsequently, the controller uses address A as an index to issue three precisely associated cooperative control signals:

[0063] (1) Based on the data popularity determination associated with address A, a partition configuration signal is generated and sent to the partition management circuit. This signal not only guides the circuit to route the current data accessed by address A, but also serves as the basis for updating the relevant popularity information in the area metadata table.

[0064] The reconfiguration arbitration circuit within the partition management circuit will make a decision based on this signal and the status of the metadata table: if it is determined that SPM resources need to be allocated for new high-hot data but the resources are full, the dynamic reconfiguration process will be automatically triggered, that is, an existing SPM region will be selected according to the policy to be downgraded to a cache, the partition configuration register group will be updated, and then the data at address A will be routed to the newly allocated SPM region.

[0065] (2) For data that needs to be written back to external memory, based on the data type determination associated with address A, a compression control signal is generated and sent to the adaptive data compression unit, instructing its data selector to select the corresponding compression engine circuit on the write-back path to compress the data of the type to which address A belongs in real time.

[0066] (3) Based on the stability determination of the access mode associated with address A, the prefetch control signal is activated when the sequential access mode is identified. The signal starts the data prefetch unit, calculates the prefetch address based on address A, and loads the subsequent data from the external memory into the dynamic partition cache unit in advance.

[0067] Through the coordinated operation of the aforementioned hardware circuits with timing synchronization mechanisms, accurate statistics of memory access patterns and dynamic adaptation of hardware resource latency are achieved, thereby realizing three-dimensional optimization of memory bandwidth.

[0068] The circuit of this utility model can be a dedicated integrated circuit module with a defined physical form and connection relationship. Its product form can be an independently packaged integrated circuit chip, whose pins include address and data bus interfaces for communication with the video encoding core, interfaces for communication with the off-chip memory controller, power supply and clock pins, etc.

Claims

1. A memory bandwidth optimization circuit for a video encoder, comprising a cooperative scheduling controller, a dynamic partitioning buffer unit, an adaptive data compression unit, and a data prefetching unit, characterized in that: The collaborative scheduling controller includes a hardware access counter array, an address range matching circuit, and a threshold determination circuit. The hardware access counter array consists of multiple counters configured in parallel; the address input terminal of the address range matching circuit is used to receive memory access addresses from the video encoding core, and it has multiple matching output terminals, each of which is connected to the clock enable terminal of a corresponding counter in the hardware access counter array; the threshold determination circuit has multiple count value input terminals and multiple determination result output terminals; the multiple count value input terminals are respectively connected to the count output terminals of the corresponding counters in the hardware access counter array; the multiple determination result output terminals of the threshold determination circuit constitute the signal source for the partition configuration terminal, compression control terminal, and prefetch control terminal of the collaborative scheduling controller; The control terminal of the adaptive data compression unit is connected to the compression control terminal of the cooperative scheduling controller, and its data terminal is connected between the dynamic partition cache unit and the external memory controller. The dynamic partition cache unit includes a partition management circuit, a configurable static random access memory array, and a region metadata table; the update end of the region metadata table is connected to the partition configuration end of the cooperative scheduling controller, and its output end is connected to the partition management circuit; the configuration end of the partition management circuit constitutes the configuration end of the dynamic partition cache unit, which is connected to the partition configuration end of the cooperative scheduling controller, and is used to receive partition configuration signals. The address mapping output terminal and data terminal of the partition management circuit are respectively connected to the address decoding terminal and data terminal of the configurable static random access memory array; The control terminal of the data prefetching unit is connected to the prefetching control terminal of the cooperative scheduling controller, and its prefetching data output terminal is connected to the data terminal of the configurable static random access memory array for writing prefetched data into the cache.

2. The memory bandwidth optimization circuit for a video encoder according to claim 1, characterized in that: The partition management circuit includes a reconfiguration arbitration circuit, a programmable partition configuration register set, and an address remapping circuit. The first input of the reconfiguration arbitration circuit is connected to the output of the regional metadata table, the second input is connected to the partition configuration terminal of the cooperative scheduling controller, and the output of the reconfiguration arbitration circuit is connected to the write terminal of the programmable partition configuration register group for outputting partition configuration signals. The write end of the programmable partition configuration register group constitutes the configuration end of the partition management circuit; the first input end of the address remapping circuit is connected to the output end of the programmable partition configuration register group and is used to receive partition configuration information; the second input end is used to receive the memory access address from the video encoding core.

3. The memory bandwidth optimization circuit for a video encoder according to claim 2, characterized in that: The configurable static random access memory array includes at least two independent memory banks; the partition management circuit further includes a memory bank selection and decoding circuit; the output of the address remapping circuit is connected to the input of the memory bank selection and decoding circuit; the memory bank selection and decoding circuit has multiple enable signal outputs, which are respectively connected to the enable signal terminals of the corresponding memory banks.

4. The memory bandwidth optimization circuit for a video encoder according to claim 1, characterized in that: The adaptive data compression unit includes a first compression engine circuit, a second compression engine circuit, and a data selector; The input terminals of the first compression engine circuit and the second compression engine circuit are connected in parallel and are jointly connected to the data input terminal of the adaptive data compression unit; the output terminal of the first compression engine circuit is connected to the first data input terminal of the data selector, and the output terminal of the second compression engine circuit is connected to the second data input terminal of the data selector; the control terminal of the data selector constitutes the control terminal of the adaptive data compression unit, and its output terminal constitutes the data output terminal of the adaptive data compression unit, which is connected to an external memory controller to complete the writing of the compressed data.

5. The memory bandwidth optimization circuit for a video encoder according to claim 4, characterized in that: The first compression engine circuit is a lossless compression circuit based on run-length encoding; the second compression engine circuit is a lossy compression circuit based on differential pulse code modulation or dictionary encoding.

6. The memory bandwidth optimization circuit for a video encoder according to claim 1, characterized in that: The data prefetching unit includes an address sequence buffer and an address increment calculation circuit; the input of the address sequence buffer is used to receive memory access addresses, and its output is connected to the input of the address increment calculation circuit. The output of the address increment calculation circuit is connected to the address input of the dynamic partition cache unit and is used to output the prefetch address.