A critical path delay optimization method based on logic netlist
By constructing and optimizing the logic vertebra in the microprocessor design, the problem of critical path delay optimization in high-main frequency design is solved, the equivalence and delay optimization of logic functions are achieved, the performance of microprocessors is improved, and the international Performance gap in products of large companies.
Patent Information
- Application Number
- CN202410165405.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-05
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-02-05
AI Technical Summary
In the design of high-main frequency microprocessors, it is difficult to effectively optimize critical path delays based on the semi-customized design process based on the commercial standard unit library, resulting in a large gap between the performance of microprocessors and the fully customized integrated circuit products of international large companies. .
By obtaining the logical netlist and critical path delay information obtained by logical synthesis, we can deeply analyze the characteristics of the critical paths in the microprocessor logical netlist, construct the logical vertebra of the critical path, and perform logical vertebra decomposition and reconstruction, using dynamic planning and graphs The segmentation method optimizes the logic diagram to achieve the reduction of logical series and unit nodes, ensuring the equivalence of logic functions and delay optimization.
Effectively reduce critical path delays, improve the main frequency performance of microprocessors, reduce design costs, and improve the performance of microprocessors close to the level of fully customized integrated circuits of international large companies.
Smart Images

Figure CN118171609B_ABST
Abstract
Description
Technical Field
[0001] The invention mainly relates to the technical field of microprocessor design, and in particular to a critical path delay optimization method based on a logic netlist. Background Art
[0002] Since the performance of a microprocessor depends largely on its main frequency, based on a certain semiconductor process and microprocessor architecture, in order to develop a higher performance microprocessor, it is crucial to further increase the main frequency of the microprocessor.
[0003] The main frequency is the clock frequency of the processor, which determines the number of clock cycles the processor can run per second; the higher the main frequency, the higher the performance of the microprocessor. Optimizing the critical path delay in the microprocessor circuit design process can directly increase the main frequency of the microprocessor and improve the performance of the microprocessor.
[0004] At present, most high-performance microprocessor designs adopt a semi-custom design method based on a commercial standard cell library. According to the instruction-level architecture of the microprocessor, the register transfer level (RTL) detailed design is completed using a hardware description language. Then, the RTL design that has passed functional verification is used to generate a logic netlist based on a commercial standard cell library using a logic synthesis tool. However, under the design constraints of high main frequency, there are usually a small number of logic circuit paths in the resulting logic netlist. Even with the optimal solution of the logic synthesis tool, the timing design constraints of the high main frequency cannot be met. These logic circuit paths are called "critical paths."
[0005] Currently, there are limited ways to solve these critical paths based on certain semiconductor processes. Some large international companies, such as Intel, AMD, and Apple, use full-custom circuit technology, which has high labor costs on the one hand, and poor reusability and long design cycles on the other. Limited by the capital and time costs of full-custom design, current IC design companies generally use semi-custom design processes based on commercial standard cell libraries, and the main frequency of high-performance microprocessors they develop has a large performance gap with the full-custom integrated circuit (IC) products of large international companies.
[0006] In the semi-custom design process based on commercial standard cell libraries, microprocessor designers are usually only able to perform timing optimization in the architecture and RTL design stages due to limitations of commercial standard cell libraries and logic synthesis tools. However, it is difficult to perform further timing optimization for the critical paths in the logic synthesis netlist that cannot meet the timing requirements. Summary of the invention
[0007] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a critical path delay optimization method based on a logic netlist which has a simple principle, can reduce delay, and increase the main frequency of the processor.
[0008] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0009] A critical path delay optimization method based on a logic netlist, comprising:
[0010] Step S1: Obtaining a logic netlist and a critical path delay information file obtained by logic synthesis;
[0011] Step S2: deeply analyzing the characteristics of the critical path in the microprocessor logic netlist, constructing the logic cone of the critical path, and extracting the corresponding logic netlist;
[0012] Step S3: Decomposing the logical cone using the constructed logical cone;
[0013] Step S4: logically reconstruct the directed acyclic subgraph obtained by logic cone decomposition, and repeatedly iterate the Boolean logic graph, while extracting the optimized logic netlist;
[0014] Step S5: verifying the functional equivalence of the optimized logic;
[0015] Step S6: Use the delay evaluation model to evaluate the delay optimization effect.
[0016] As a further improvement of the method of the present invention: when constructing the critical path logic cone, the module hierarchy and the connections within the module are not retained, and only the standard unit modules and ports, as well as the logical connections between the basic units are retained.
[0017] As a further improvement of the method of the present invention: constructing the logic cone of the critical path in the opposite direction to the direction of logic evaluation, gradually backtracking from the end point of the critical path to the output port of the sequential unit, obtaining all the minimum logic units between the end point of the critical path and the sequential unit and the connection relationship between the minimum logic units, and taking each minimum logic unit as a node; after completing the construction of the logic cone of the critical path, extracting the logic netlist corresponding to the logic cone of the critical path.
[0018] As a further improvement of the method of the present invention: when constructing the critical path logic cone, define the relevant data structure, obtain the critical path end point and the starting point from the critical path file reported by the logic synthesis, and search for the corresponding critical path node and other nodes associated with the critical path from the end point of a critical path; after constructing the critical path logic cone, extract the corresponding critical path logic netlist;
[0019] The method for finding the corresponding critical path node and other nodes associated with the critical path is: by finding the driver nodes corresponding to all input ports of the node, the found driver nodes must be the basic units of the flattened hierarchy, and a link relationship is constructed for the found nodes, which are stored in the form of a directed graph.
[0020] As a further improvement of the method of the present invention: the termination condition for finding a timing path is to satisfy any one of the following conditions:
[0021] (1) Output port of the sequential unit;
[0022] (2) The input port of the top-level module in the netlist.
[0023] As a further improvement of the method of the present invention: performing logical cone decomposition includes:
[0024] A proper logic cone segmentation involves two directions of the logic graph, and each segmentation requires an evaluation;
[0025] For the critical path direction: the partitioning is a dynamic planning process, and the partitioning is continuously improved through iteration;
[0026] For non-critical path driven nodes with critical path nodes as vertices: partition selection is a graph partitioning process;
[0027] Any suitable partitioning needs to satisfy the requirement that the minimum delay in the critical path direction after the partitioning is less than the delay in the critical path direction before the partitioning.
[0028] Iterate continuously for different partitioning methods until a partitioning method that meets the timing optimization requirements is found.
[0029] As a further improvement of the method of the present invention: the logic reconstruction includes:
[0030] Based on the information of the critical path, the logic of the critical path is optimized by Boolean logic, and the subgraph obtained by decomposition is minimized by the two-level logic method of AND or OR to obtain a multi-level logic structure that realizes logic minimization;
[0031] The optimized two-input AND-not-graph AIG is expanded to a multi-input AND gate structure, and most logic gates are extracted for full-custom design, represented by MIG. After optimization, each node in MIG is optimal.
[0032] For the reconstructed and optimized logic cone, the corresponding logic netlist for implementing logic optimization is extracted.
[0033] As a further improvement of the method of the present invention: the equivalence test includes:
[0034] Use the logic netlist corresponding to the critical path logic cone and the logic netlist after the logic optimization to perform equivalence check, and use the logic netlist corresponding to the constructed critical path logic cone as a reference model to verify the equivalence of the logic before and after reconstruction;
[0035] When the logical functions corresponding to the two netlists are equivalent, it is proved that the optimized logic cone is equivalent to the original logic cone.
[0036] As a further improvement of the method of the present invention: in the step S6, using the delay evaluation model includes: performing delay evaluation on the logically equivalent reconstructed logic to determine whether the optimization goal is met, and if not, jumping to execute step S3, otherwise obtaining the optimized critical path delay.
[0037] Compared with the prior art, the advantages of the present invention are:
[0038] 1. The critical path delay optimization method based on the logic netlist of the present invention has a simple principle, can reduce delay and improve the main frequency of the processor. The present invention proposes a critical path logic reconstruction method and a logic level simplification method, which optimizes the number of logic levels and unit nodes on the critical path while ensuring the equivalence of logical functions, can effectively reduce the critical path delay and improve the main frequency of the microprocessor.
[0039] 2. The critical path delay optimization method based on logic netlist of the present invention is based on the high main frequency design requirements of the microprocessor under certain process conditions, deeply analyzes the characteristics of the critical path in the microprocessor logic netlist, and uses precise synthesis and logic reconstruction methods to optimize the number of logic levels and unit nodes on the critical timing path through a small number of fully customized designs of most logic gate units. Under the premise of ensuring the equivalence of logical functions, it effectively reduces the critical path delay in the logic synthesis process and improves the main frequency of the microprocessor.
[0040] 3. The critical path delay optimization method based on the logic netlist of the present invention provides a method for constructing a critical path logic cone to address the problems that the amount of textual information in the logic netlist is large and the connection relationship between modules is not intuitive. The method extracts key information from the netlist and quickly constructs a logic cone that is essentially a weighted directed acyclic graph. The method describes all instance standard units and their connection relationships on the critical path in the netlist, which is convenient for subsequent logic analysis and optimization.
[0041] 4. The critical path delay optimization method based on the logic netlist of the present invention uses dynamic programming and graph partitioning methods to quickly decompose the logic cone, partitions the constructed logic graph into multiple directed acyclic subgraphs, and converts the logic network optimization problem into a subgraph logic minimization problem.
[0042] 5. The critical path delay optimization method based on the logic netlist of the present invention accurately extracts a small amount of common logic for the partitioned subgraphs in the critical path, performs logic optimization on the logic of the critical path, and replaces the unoptimized logic network with an equivalent optimal logic network by performing template pattern matching, thereby reducing the number of nodes and logic depth, thereby effectively optimizing the critical path delay.
[0043] 6. The critical path delay optimization method based on logic netlist of the present invention can be easily embedded into the conventional IC design process, effectively reduce the delay of the critical path in the logic synthesis netlist, and efficiently realize the improvement of the main frequency of the microprocessor based on a specific process. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a schematic flow diagram of the method of the present invention.
[0045] Figure 2 It is a schematic diagram of the structural principle of a critical path logic column constructed based on a logic netlist in a specific application example of the present invention.
[0046] Figure 3 It is a flow chart of a method for constructing a critical path logic cone based on a logic netlist in a specific application example of the present invention.
[0047] Figure 4 It is a schematic diagram of the logical cone decomposition of the present invention in a specific application example.
[0048] Figure 5 It is a flow chart of the logic cone decomposition method in a specific application example of the present invention.
[0049] Figure 6 It is a flow chart of a method for performing logic reconstruction based on a logic subgraph obtained by division in a specific application example of the present invention. DETAILED DESCRIPTION
[0050] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0051] like Figure 1 As shown, the critical path delay optimization method based on the logic netlist of the present invention comprises the following steps:
[0052] Step S1: Obtaining a logic netlist and a critical path delay information file obtained by logic synthesis;
[0053] That is: obtaining the logic netlist and critical path delay information file obtained by top-level logic design synthesis, and deeply analyzing the characteristics of the microprocessor logic netlist file and the critical path delay information file;
[0054] Step S2: deeply analyzing the characteristics of the critical path in the microprocessor logic netlist, constructing the logic cone of the critical path, and extracting the corresponding logic netlist;
[0055] That is, construct the critical path logic cone based on the logic netlist, and extract the corresponding logic netlist at the same time. Each node in the logic cone represents a standard unit instance. The module hierarchy and the internal connections of the module are not retained. Only the standard unit modules and ports, as well as the logical connections between the basic units are retained.
[0056] Step S3: Decomposing the logical cone using the constructed logical cone;
[0057] That is: the constructed logical cone is divided into multiple directed acyclic subgraphs, each of which is balanced in size and has no overlap between the subgraphs;
[0058] Step S4: logically reconstruct the directed acyclic subgraph obtained by logic cone decomposition, and repeatedly iterate the Boolean logic graph, while extracting the optimized logic netlist;
[0059] That is, the sub-graphs obtained by the division are logically reconstructed, Boolean logic optimization is performed based on the information of the key timing paths, a batch of majority logic units are extracted for full-custom design, and finally, based on the optimized AIG, the conversion to MIG is realized;
[0060] Step S5: verifying the functional equivalence of the optimized logic;
[0061] That is: using the logic netlist corresponding to the constructed critical path logic cone as a reference model, verify the equivalence of the logic before and after reconstruction;
[0062] Step S6: Use the delay evaluation model to evaluate the delay optimization effect.
[0063] That is, using the delay evaluation model, the delay evaluation is performed on the logically equivalent reconstructed logic to determine whether the optimization goal is met. If not, the process jumps to step S3, otherwise the optimized critical path delay is obtained.
[0064] As can be seen from the above, in the critical path delay optimization method based on the logic netlist provided by the present invention, the logic cone of the critical path is first constructed based on the logic netlist, without retaining the module hierarchy and the internal connections of the module, and only retaining the standard unit modules and ports contained in the commercial standard unit library, as well as the connection relationship between the standard units; then logical division is performed based on the constructed logic cone; the logic on the critical path is optimized by Boolean logic using the divided subgraph to reduce the number of nodes and the number of logic levels, and the conversion to MIG is achieved on the basis of AIG optimization; the optimization effect is evaluated using the delay optimization model, and any suitable division needs to satisfy that the minimum delay value of the critical path after the division is less than the delay of the critical path before the division; different division methods are continuously iterated until a division method that meets the timing optimization requirements is found and stopped, and finally the optimization of the critical path is achieved.
[0065] Furthermore, in a preferred embodiment, in actual application, the method of the present invention can be used to perform loop optimization on the logic netlist multiple times when necessary to reduce the critical path delay as much as possible.
[0066] Furthermore, based on a given logic netlist, the present invention does not retain the module hierarchy and the connections within the module, but only retains the standard unit modules and ports, as well as the logical connections between the standard units, constructs the logic cone of the critical path, and provides a corresponding logic netlist.
[0067] Furthermore, the process of constructing the logic cone of the critical path of the present invention is exactly opposite to the direction of the logic evaluation of the critical path. When constructing the logic cone, the critical path is gradually traced back to the output port of the sequential unit, and all the minimum logic units (i.e., the standard units in the commercial standard unit library) between the critical path end point and the sequential unit and the connection relationship between the minimum logic units are obtained, and each minimum logic unit is regarded as a node.
[0068] In a specific application example, according to actual needs, the construction of the critical path logic cone specifically includes: not retaining the module hierarchy and the internal connections of the module, only retaining the standard unit module and port, and the logical connections between the basic units. The construction of the critical path logic cone is exactly the opposite of the direction of the logic evaluation, and it gradually traces back from the end point of the critical path to the output port of the sequential unit (here the sequential unit is a trigger, static random access memory (SRAM) or some intellectual property IP hard core in the logic synthesis process library), and obtains all the minimum logic units (i.e., basic units) between the critical path end point and the sequential unit, as well as the connection relationship between the minimum logic units, and each minimum logic unit is regarded as a node. After completing the construction of the critical path logic cone, the logic netlist corresponding to the critical path logic cone is extracted.
[0069] In a specific application example, according to actual needs, the detailed process of logical cone decomposition may include:
[0070] A proper logic cone segmentation involves two directions of the logic graph, and each segmentation requires an evaluation;
[0071] For the critical path direction: the partitioning is a dynamic planning process, and the partitioning is continuously improved through iteration;
[0072] For non-critical path driven nodes with critical path nodes as vertices: Partition selection is a graph partitioning process.
[0073] Any suitable partitioning needs to satisfy the requirement that the minimum delay in the critical path direction after the partitioning is less than the delay in the critical path direction before the partitioning. Iterate continuously for different partitioning methods until a partitioning method that meets the timing optimization requirements is found.
[0074] In a specific application example, according to actual needs, the detailed process of logic reconstruction includes:
[0075] Based on the information of the critical path, the logic of the critical path is optimized by Boolean logic, and the subgraph obtained by decomposition is minimized by the two-level logic method of AND or OR to obtain a multi-level logic structure that realizes logic minimization;
[0076] The optimized two-input AND-INVERTER Graph (AIG) is expanded to a multi-input AND gate structure, and a small number of majority logic gates are extracted for full-custom design, represented by MIG (majority-inverter graph). After optimization, each node in MIG is optimal.
[0077] For the reconstructed and optimized logic cone, the corresponding logic netlist for implementing logic optimization is extracted.
[0078] In specific application examples, according to actual needs, the equivalence test specifically includes:
[0079] Use the logic netlist corresponding to the critical path logic cone and the logic netlist after the logic optimization to perform equivalence check, and use the logic netlist corresponding to the constructed critical path logic cone as a reference model to verify the equivalence of the logic before and after reconstruction;
[0080] When the logical functions corresponding to the two netlists are equivalent, it is proved that the optimized logic cone is equivalent to the original logic cone.
[0081] In specific application examples, according to actual needs, performance evaluation specifically includes:
[0082] Since the delay between basic units is not a fixed value but an interval, the performance evaluation of the present invention utilizes a delay evaluation model to quantitatively analyze the circuit delay before and after optimization.
[0083] like Figure 2 As shown in FIG. 1 , a schematic diagram of the structure of a critical path logic cone instance constructed by the above method of the present invention is shown. The circles in the figure represent the minimum logic unit instance, the lines between the circles represent the connection relationship between the standard unit instances, and the direction of the arrow is the logic evaluation direction. The critical path logic cone includes an end point and multiple starting points (one of which is the starting point of this critical path, represented by a black circle). The black circles in the figure represent the standard units on the critical path, and the path formed by these nodes is the critical path. The remaining white circles represent the standard units associated with the critical path.
[0084] The present invention extracts the corresponding logic netlist based on the constructed critical path logic cone, which is convenient for later verification of logic function equivalence.
[0085] like Figure 3As shown, in a specific example, the detailed process of constructing a critical path logic cone based on a logic netlist includes:
[0086] Define relevant data structures to facilitate subsequent information extraction and storage; then obtain the critical path endpoint and starting point from the critical path file reported by logic synthesis, and start from the end point of a critical path to find the corresponding critical path node and other nodes associated with the critical path. After constructing the critical path logic spine, extract the corresponding critical path logic netlist.
[0087] The method for finding the corresponding critical path node and other nodes associated with the critical path is: by finding the driver nodes corresponding to all input ports of the node, the driver nodes found must be the basic units of the flattened hierarchy, and link relationships are constructed for the found nodes and stored in the form of a directed graph.
[0088] The termination condition for finding a timing path is to satisfy any one of the following conditions:
[0089] (1) Output port of the sequential unit;
[0090] (2) The input port of the top-level module in the netlist.
[0091] In the present invention, under certain constraints, the critical path logic cone composed of basic units is divided into multiple subsets, which can reduce the complexity of the problem and achieve better results, and facilitate the subsequent logical optimization of the directed acyclic subgraph obtained by the division.
[0092] like Figure 4 As shown, in specific applications, there are many ways to segment the critical path, and there are more ways to combine the associated paths. There are many ways to divide a logical cone. The specific ways to divide the logical cone are as follows:
[0093] The logical cone is divided on the basis of satisfying certain constraints and given division targets. The scale of each subgraph obtained after the division is balanced, and there is no overlap between the subgraphs obtained after the division.
[0094] The segmentation in the present invention involves two directions of the logic graph, and each segmentation requires an evaluation:
[0095] (1) For the critical path direction (black node direction): the partitioning is a dynamic planning process, and the partitioning is performed through iterative improvement;
[0096] (2) For non-critical path driven nodes with critical path nodes as vertices (white node direction): partition selection is a graph partitioning process.
[0097] like Figure 5 As shown, in a specific application, the process of dividing the logical cone is as follows:
[0098] The present invention describes the logic cone k partitioning problem (k is an integer greater than 1) as follows: given a critical path logic cone constructed by a logic netlist, a node set is defined , where v i (i is a positive integer, i=1,2,…n) is one of the standard unit nodes, which contains n nodes in total. Here, all modules are divided into k subsets based on the conditions that meet certain constraints and the given partitioning goals, that is, .
[0099] The general constraint condition is the area balance constraint, that is, when the area A(v i ), the lower bound of the total area of modules in each subset when the balance factor is β (0<β<1): , upper bound For each subset Area of (j=1, 2, ..., k) Recorded as , requirements to be met .
[0100] The main objective function of circuit partitioning is minimum delay: the delay on the critical path from the input to the output is minimized. A suitable partitioning needs to satisfy the minimum delay in the critical path direction after the partition is less than the delay in the critical path direction before the partition.
[0101] For the critical path logic cone constructed above, define its logical nodes The logic delay is , the total delay of the critical path is: .
[0102] A set of logical subgraphs obtained by logical vertebra k partitioning , where for any m and n ,if , then it satisfies: , that is, there is no intersection between any two sub-graphs obtained by division. After logical reconstruction, The delay in the critical path direction is , the total delay is Any suitable segmentation should satisfy: .
[0103] Iterate continuously for different partitioning methods until a partitioning method that meets the timing optimization requirements is found.
[0104] Furthermore, the directed acyclic subgraph obtained by the logic cone decomposition is logically reconstructed, and the Boolean logic graph is repeatedly iterated, and the corresponding optimized logic netlist is extracted.
[0105] In the present invention, the key to logic reconstruction is to minimize the subgraph logic, obtain a multi-level logic structure that realizes logic level minimization, extract a batch of majority logic function units after realizing logic level minimization, and use the optimized logic structure to replace the original logic structure to realize logic reconstruction. Finally, the optimized two-input AIG is expanded to a multi-input AND gate structure, and a batch of majority logic gates are extracted for full-custom design, which is represented by MIG. After optimization, each node in MIG is optimal.
[0106] like Figure 6 As shown in the figure, in a specific application, the detailed process of logic reconstruction is as follows:
[0107] Subgraph logic minimization includes: one node can be replaced by another node under the premise of functional equivalence, and one node can be removed from the circuit when two nodes are merged; the merging of nodes may produce other redundancies, and it is necessary to continue merging to minimize the circuit. The logical implication merging can be performed using the two-level AND or OR logic method, which mainly includes two steps: finding all prime implications and finding the minimum prime implication set. The rule of logical minimization is: if two original lines related to the result differ only in one condition, then this condition can be considered logically redundant and can be omitted, so that the two lines are merged into a simpler expression.
[0108] Subgraph replacement: Replace the original logical structure with the optimized logical structure.
[0109] Public processing: On all critical paths, nodes that have already appeared in the network are reused for representation, and if the number of nodes or the depth is optimized, they are replaced.
[0110] Convert to MIG: Expand the optimized two-input AIG to a multi-input AND gate structure, extract most logic gates, perform full-custom design, and express it in MIG. After optimization, each node in MIG is optimal.
[0111] Extract the corresponding logic netlist: For the logic cone that has completed the reconstruction optimization, extract the corresponding logic netlist.
[0112] In specific applications, the specific methods of equivalence testing are as follows:
[0113] The equivalence check is performed using the logic netlist corresponding to the critical path logic cone and the logic netlist after logic reconstruction and optimization. The constructed logic netlist corresponding to the critical path logic cone is used as a reference model. On the premise that other irrelevant conditions remain consistent, the equivalence of the logic before and after reconstruction is verified. When the logic functions corresponding to the two netlists are equivalent, it is proved that the optimized logic cone is equivalent to the original logic cone.
[0114] In specific applications, a delay evaluation model is combined to evaluate the delay optimization effect of the critical path. The specific evaluation method is as follows:
[0115] The delay between basic units is not a fixed value, but an interval. The performance evaluation is completed using a delay evaluation model to quantitatively analyze the circuit delay before and after optimization. In the present invention, the critical path delay optimization method based on a logic netlist is provided to meet the requirement that the minimum delay in the critical path direction after optimization is less than the delay in the critical path direction before division.
[0116] Those skilled in the art should understand that the above-mentioned embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the process Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including an instruction device, which implements the functions specified in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.
[0117] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A critical path delay optimization method based on a logic netlist, characterized in that: include: Step S1: Obtaining a logic netlist and a critical path delay information file obtained by logic synthesis; Step S2: deeply analyzing the characteristics of the critical path in the microprocessor logic netlist, constructing the logic cone of the critical path, and extracting the corresponding logic netlist; Step S3: Decomposing the logical cone using the constructed logical cone; Step S4: logically reconstruct the directed acyclic subgraph obtained by logic cone decomposition, and repeatedly iterate the Boolean logic graph, while extracting the optimized logic netlist; Step S5: verifying the functional equivalence of the optimized logic; Step S6: using the delay evaluation model to evaluate the delay optimization effect; When constructing the critical path logic cone, the module hierarchy and the internal connections of the modules are not retained, only the standard unit modules and ports, as well as the logical connections between the basic units are retained; the construction of the critical path logic cone is exactly opposite to the direction of the logic evaluation, and it gradually traces back from the end point of the critical path to the output port of the sequential unit, obtains all the minimum logic units between the critical path end point and the sequential unit, and the connection relationship between the minimum logic units, and regards each minimum logic unit as a node; after completing the construction of the critical path logic cone, extract the logic netlist corresponding to the critical path logic cone.
2. The critical path delay optimization method based on logic netlist according to claim 1, characterized in that: When constructing the critical path logic cone, define the relevant data structure, obtain the critical path end point and start point from the critical path file reported by the logic synthesis, and start from the end point of a critical path to find the corresponding critical path node and other nodes associated with the critical path; After constructing the critical path logic cone, extract the corresponding critical path logic netlist; The method for finding the corresponding critical path node and other nodes associated with the critical path is: by finding the driver nodes corresponding to all input ports of the node, the found driver nodes must be the basic units of the flattened hierarchy, and a link relationship is constructed for the found nodes, which are stored in the form of a directed graph.
3. The critical path delay optimization method based on logic netlist according to claim 2 is characterized in that: The termination condition for finding a timing path is to satisfy any one of the following conditions: (1) Output port of the sequential unit; (2) The input port of the top-level module in the netlist.
4. The critical path delay optimization method based on logic netlist according to any one of claims 1 to 3, characterized in that: The logical cone decomposition includes: A proper logic cone segmentation involves two directions of the logic graph, and each segmentation requires an evaluation; For the critical path direction: the partitioning is a dynamic planning process, and the partitioning is continuously improved through iteration; For non-critical path driven nodes with critical path nodes as vertices: partition selection is a graph partitioning process; Any suitable partitioning needs to satisfy the requirement that the minimum delay in the critical path direction after the partitioning is less than the delay in the critical path direction before the partitioning. Iterate continuously for different partitioning methods until a partitioning method that meets the timing optimization requirements is found.
5. The critical path delay optimization method based on logic netlist according to any one of claims 1 to 3, characterized in that: Logical reconstruction includes: Based on the information of the critical path, the logic of the critical path is optimized by Boolean logic, and the subgraph obtained by decomposition is minimized by the two-level logic method of AND or OR to obtain a multi-level logic structure that realizes logic minimization; The optimized two-input AND-not-graph AIG is expanded to a multi-input AND gate structure, and most logic gates are extracted for full-custom design, represented by MIG. After optimization, each node in MIG is optimal. For the reconstructed and optimized logic cone, the corresponding logic netlist for implementing logic optimization is extracted.
6. The critical path delay optimization method based on logic netlist according to any one of claims 1 to 3, characterized in that: Equivalence tests include: Use the logic netlist corresponding to the critical path logic cone and the logic netlist after the logic optimization to perform equivalence check, and use the logic netlist corresponding to the constructed critical path logic cone as a reference model to verify the equivalence of the logic before and after reconstruction; When the logical functions corresponding to the two netlists are equivalent, it is proved that the optimized logic cone is equivalent to the original logic cone.
7. The critical path delay optimization method based on logic netlist according to any one of claims 1 to 3, characterized in that: In the step S6, using the delay evaluation model includes: performing delay evaluation on the logically equivalent reconstructed logic to determine whether the optimization goal is met, and if not, jumping to step S3, otherwise obtaining the optimized critical path delay.
