Method for operating an information technology system, information technology system, and vehicle

EP4457611A1Pending Publication Date: 2024-11-06MERCEDES BENZ GROUP AG
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
EP2023798436
Authority / Receiving Office
EP · EP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2022-11-21
Filing Date
2023-10-26
Publication Date
2024-11-06

AI Technical Summary

Technical Problem

The challenge is to enhance the performance and efficiency of computing components in vehicles while minimizing space and manufacturing costs, as high-performance hardware is required for time-critical tasks and user experience, but there is limited space and a need to optimize existing methods like parallelization and SIMD for better computing speed.

Method used

A method where program code is divided into function blocks, and a compiler determines their execution order to enable parallel processing, assigning similar instructions to the same execution unit for simultaneous processing, even if loops have different iterations, using placeholder instructions to synchronize execution.

Benefits of technology

This approach increases the degree of parallelization, saving cycle time and improving processing speed by allowing simultaneous execution of sub-functions within the same task, thus enhancing computing efficiency and performance in vehicles with limited space and cost constraints.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1.1
    Figure 1.1
Patent Text Reader

Abstract

The invention relates to a method for operating an information technology system. In order to complete at least one task (1.1, 1.2), a program code which can be used by a processor is divided into at least two function blocks (2) by a compiler in a compiling step, and the function blocks are arranged in a data flow graph (3). The compiler analyzes the data flow graph (3) in order to ascertain an execution sequence for each function block (2) by means of the processor. The method according to the invention is characterized in that the compiler ascertains the function blocks (2) which can be executed in parallel by means of the processor when completing the at least one task (1.1, 1.2), for each group (4) of function blocks (2) which can be executed in parallel, the compiler checks whether at least one program code section of at least two function blocks (2) require the execution of the same instruction (5) by means of the processor, and the compiler assigns said program code sections to the same execution unit of the processor for simultaneous completion according to a single instruction / multiple data operation.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Method for operating an information technology system, information technology system and vehicle

[0002] The invention relates to a method for operating an information technology system according to the type defined in more detail in the preamble of claim 1, a corresponding information technology system and a vehicle with such an information technology system.

[0003] Information technology systems such as computers, embedded systems, or similar computing devices have become indispensable in solving a wide variety of everyday problems. One development goal is to increase the performance of hardware components while simultaneously miniaturizing them. This enables the solution of complex tasks, even in the context of mobile applications.

[0004] With increasing digitalization, the importance of computer systems in vehicles is also increasing. Modern vehicles are equipped with a wide variety of control units, for example for powertrain control, navigation route calculation, or the vehicle's communication connection to the internet. Powerful hardware components are essential for calculating time-critical control signals sufficiently quickly, for maintaining a pleasant user experience by processing user inputs sufficiently quickly and outputting corresponding responses sufficiently quickly, and so on. Furthermore, there is only limited space available in vehicles for the integration of computing components. Furthermore, the integration of such computing devices must not cause the manufacturing costs of vehicles to increase disproportionately. This makes the development of computer systems intended for the automotive sector particularly challenging.In summary, there is a need to continually improve the performance and efficiency of computing components. A well-known approach to accelerating computational problem-solving is parallelization. Modern processors can have multiple execution units, such as processing cores, which allows for multi-threading—the simultaneous processing of different tasks by one and the same processor.

[0005] Another way to increase computing efficiency is Single-Instruction Multiple-Data (SIMD). SIMD enables computer systems to perform the same arithmetic operation on multiple data points simultaneously. For example, in a PC, this reduces the required number of read and write accesses to the main memory, as well as the number of arithmetic operations performed by the processor (CPU). Accordingly, the time required by a program to solve a task can be reduced. For example, D. Nuzman et al., "Vapor SIMD: Auto-vectorize once, run everywhere," International Symposium on Code Generation and Optimization (CGO 2011), 2011, pp. 151-160, doi:

[0006] 10.1109 / CGO.2011.5764683 how compilers transform loops to encode multiple iterations of the same loop in parallel in SIMD instructions.

[0007] Furthermore, WO 2007 / 113369 A1 discloses a method for generating a parallel executable program. The method describes compiling a computer program from source code into executable code, wherein a dataflow graph is generated from the corresponding source code and examined to investigate data dependencies. The executable program thus generated is divided into individual packets, which are distributed among the individual execution units of the processor of the computing device used. The execution sequence is determined as late as possible to account for the computing architecture of the currently used computing device. This enables particularly efficient load distribution among the individual execution units of processors, regardless of the variety of computing devices used to process the program.Here, too, the increase in computing speed is based on distributing individual program sections across multiple execution units of one or more processors to parallelize program execution. The present invention is based on the object of providing an improved method for operating an information technology system, with the aid of which the performance of information technology systems can be further increased.

[0008] According to the invention, this object is achieved by a method for operating an information technology system having the features of claim 1. Advantageous embodiments and further developments as well as a corresponding information technology system suitable for carrying out the method according to the invention and a vehicle with such an information technology system emerge from the dependent claims.

[0009] A generic method for operating an information technology system, wherein program code which can be used by a processor to process at least one task is divided into at least two function blocks by a compiler in a compilation step and these are arranged in a data flow graph and the compiler analyses the data flow graph to determine an execution sequence of the respective function blocks by the processor, is further developed according to the invention in that the compiler determines function blocks which can be executed in parallel by the processor when processing the at least one output, checks for each group of function blocks which can be executed in parallel whether at least program code parts of at least two function blocks require the execution of the same instruction by the processor and assigns these program code parts to the same execution unit of the processor for simultaneous processing in accordance with single-instruction-multiple-data.

[0010] With the help of the method according to the invention, the degree of program parallelization can be increased even further. This not only makes it possible to execute different programs simultaneously with a processor having multiple execution units, but one and the same execution unit of the processor is also enabled to execute different parts of the same program simultaneously in one computing unit. The simultaneous execution of subfunctions of one and the same task allows clock time on the processor to be saved and thus the underlying task to be solved more quickly, i.e. the corresponding program to be executed more quickly. The processor can have multiple execution units, for example processor cores. The processor does not necessarily have to have multiple physical execution units. The processor can also be designed to perform multi-threading.The processor can be a central processing unit (CPU), a graphics processor (GPU), or another dedicated programmable computing unit, for example integrated on a system-on-a-chip (SoC).

[0011] To determine which function blocks can be executed in parallel, the compiler checks the dependencies between the individual function blocks in the data flow graph. A function block typically has an input interface and an output interface. The input interface provides the required data to the function block, and the output interface provides the result generated by the respective function block. If a function block requires a specific result from another function block, or if the function block passes its calculated result to another function block, these function blocks cannot be executed in parallel. However, if two function blocks have no dependencies with regard to their respective input and output data, they can generally be executed in parallel.

[0012] A particularly efficient application of the method according to the invention is possible if, in the case of loops containing function blocks, these function blocks or the respective loops have the same depth and particularly preferably the same upper bound.

[0013] An advantageous development of the method provides that the compiler determines the number of instructions to be processed by the processor for each of the parallel-executable function blocks and, for a respective group of parallel-executable function blocks, incorporates placeholder instructions into the respective function blocks for those function blocks that contain fewer instructions than the function block with the maximum number of instructions in the respective group, so that all function blocks in a respective group contain the same number of instructions. If the function blocks in a group of parallel-executable function blocks contain the same number of instructions to be executed by the processor, a particularly simple and efficient assignment of the program code sections to the same execution unit of the processor is possible according to single-instruction-multiple-data.In this context, instructions refer to the presence of the same instruction in all function blocks of the group of parallel-executable function blocks. For example, this instruction could be the "Add" or "Multiply" instruction. If individual program code sections deviate from this instruction or are missing, for example, if the "Multiply" instruction is used instead of "Add" or a command line is missing entirely, a placeholder instruction must be inserted at the appropriate point in the respective functionality block to enable the assignment of the program code sections according to the single-instruction-multiple-data principle.

[0014] A further advantageous embodiment of the method according to the invention further provides that if the program code parts contained by the function blocks of a group of parallel executable function blocks comprise loops:

[0015] - the compiler determines the number of iterations of the loops;

[0016] - the compiler checks whether the loops of the function blocks containing the program code parts assignable to the same execution unit of the processor according to Single-Instruction-Multiple-Data include a different number of iterations; and if this is the case:

[0017] - the compiler ensures that the respective program code parts of the function blocks corresponding to the lowest number of iterations are processed simultaneously by the same execution unit of the processor and those program code parts of the function blocks with a higher number of iterations are processed sequentially by the processor according to the remaining iterations.

[0018] A particularly simple and efficient implementation of the method according to the invention is possible if the program code sections to be assigned to the same execution unit of the processor according to single-instruction-multiple-data consist of only one instruction of the same type. However, the corresponding program code sections of the function blocks can also contain loops. If several function blocks have a different number of iterations of the corresponding loops, only those program code sections can be parallelized according to single-instruction-multiple-data that are the same instructions in the function blocks of the group of parallel-executable function blocks, corresponding to the number of iterations of the loop with the lowest number of iterations.In other words, for the program code sections of the function block with the higher number of iterations, no corresponding instructions from the remaining function block are available to be simultaneously distributed to the processor's execution unit. The corresponding surplus is then processed sequentially by the processor.

[0019] Preferably, in a multi-threading-capable processor, at least two different tasks are processed in parallel on at least two different execution units of the processor. This allows the degree of parallelization to be increased even further. It is also conceivable that the inventive method for assigning the program code sections according to single-instruction-multiple-data is applied individually for each task on each execution unit of the processor. Thus, multiple tasks can be processed simultaneously on multiple execution units, with parallelization taking place on each individual execution unit according to the inventive method.

[0020] According to the invention, an information technology system is configured to execute a method described above. The information technology system can be a known computing device such as a PC, an embedded system, a SoC, or the like. The method according to the invention can be used particularly profitably in the context of embedded systems, since comparatively weak hardware components are installed here, which makes it necessary to use other measures to increase computing efficiency.

[0021] Such an information technology system is particularly preferably integrated into a vehicle according to the invention. The vehicle can be any vehicle such as a car, truck, van, bus, or the like. The information technology system can form a central on-board computer, a control unit of a vehicle subsystem, or the like. Several such information technology systems can also be integrated into the vehicle. As already mentioned at the beginning, there are limits to the integration of powerful computing components in vehicles. By applying the method according to the invention and integrating such an information technology system according to the invention, the computing efficiency and thus the performance of the installed computing devices in the vehicle can be increased, thereby utilizing the available installation space in a space-saving manner and ensuring the cost-effective production of the vehicle.

[0022] Further advantageous embodiments of the method according to the invention for operating the information technology system also emerge from the exemplary embodiments which are described in more detail below with reference to the figures.

[0023] Showing:

[0024] Fig. 1 shows a representation of a program code divided into individual function blocks as a data flow graph; and

[0025] Fig. 2 Pseudocode of the function blocks C and D shown in Fig. 1 and an assignment of instructions contained in the pseudocode to the same execution unit of a processor according to a method according to the invention.

[0026] Figure 1 shows, in an abstract representation as a data flow graph 3, a program code executed on an information technology system to solve one or more tasks 1.1, 1.2, divided into individual function blocks 2. A function block 2 corresponds to a separately calculable sub-problem of the respective task

[0027] 1.1 , 1.2. The processor of the information technology system can be configured to execute multi-threading, which enables the parallel processing of several tasks 1.1,

[0028] 1.2, i.e., multiple execution threads, are permitted. The execution threads can have the same or a different execution frequency. Each task 1.1, 1.2 is assigned to an individual execution unit of the processor, for example, to separate processing cores. This represents a first possibility for parallelizing programs. To maintain clarity, only a few of the similar reference objects are provided with reference symbols in the figures.

[0029] The individual function blocks 2 have input interfaces 2.E for reading input data and output interfaces 2.A for outputting calculation results. Taking into account corresponding data dependencies, an execution sequence of the function blocks 2 can be derived. In a prior art information technology system, the individual function blocks 2 of the respective tasks 1.1 and 1.2 are each executed sequentially one after the other. However, this wastes computing capacity, since individual function blocks 2 or program code sections contained in the respective function blocks 2 could generally also be executed in parallel.

[0030] To further increase computing efficiency, a method according to the invention is used which allows the instructions 5 contained in the individual function blocks 2, shown in more detail in Figure 2, to be distributed across the processor in such a way that parts of a single task 1.1 can also be parallelized. The method according to the invention provides that similar instructions 5 from parallel-executable function blocks 2 are assigned to the same execution unit of the processor according to Single-Instruction-Multiple-Data (SIMD). This means that the same instruction is applied to the different register areas of the same register in one computing cycle. In this context, "similar" means the same type of instruction, for example that the code lines to be parallelized all contain the "Add" instruction, which is a fundamental prerequisite for SIMD.

[0031] The method according to the invention provides that a corresponding compiler analyzes the data flow graph 3 shown in Figure 1 and determines the said execution sequence of the respective function blocks 2. Function blocks 2 can be executed in parallel if there are no dependencies between the data exchanged via the input interfaces 2.E and output interfaces 2.A. This is the case in Figure 1, for example, for function blocks C and D as well as C and B. Furthermore, function blocks A and H can also be parallelized. Parallelization for function blocks E and F, as well as for task 1.2 for function blocks M, N, O, and Q, is not possible due to the dependencies.

[0032] In the following, it is assumed that function blocks C and D can be executed in parallel, which then forms a group of 4 parallel-executable function blocks 2, as shown in Figure 2. Depending on the complexity of the program code, such a group 4 could also contain three, four, or even more parallel-executable function blocks 2.

[0033] Figure 2 shows, abstracted as pseudocode, the program code sections comprised by function blocks C and D. Depending on the design of the information technology system and the problem to be addressed, any conceivable programming language can generally be used to implement the method according to the invention.

[0034] The "Mov" instruction describes the movement of a data item from one register to another or from memory to a register. Accordingly, the number 42 or 30 is assigned to the register rO.

[0035] “:Loop” describes the beginning of a loop and “Bnz loop” describes the end of a loop, assuming that rO = 0 has been reached.

[0036] The "Mov" and "Add" instructions are instructions to be executed by the processor of the information technology system. Instead of an addition, a multiplication or similar operation could also be requested. For example, the instruction "Add r1, [r2], + rO" describes that the sum of the contents of register r2 plus the value of rO is assigned to register r1. The loop shown in function block C counts backward from 42 to 0. The countdown is implemented by the code line "Add rO, rO, -1."

[0037] The two code lines "Add r1, [r2], + r0" from function block C and "Add r1, [r2], 0" from function block D correspond to the same instruction 5 for the processor and access the same register areas. Thus, they are suitable for applying the inventive assignment of instructions 5 to the same execution unit of the processor of the information technology system according to single-instruction-multiple-data according to the inventive method.

[0038] However, the loops contained in function blocks C and D have a different number of iterations and a different number of instructions 5. To accommodate this, a so-called placeholder instruction 6 is first inserted into function block C. Function blocks C and D then have the same number of instructions 5. For example, the designation "nop" is used as placeholder instruction 6, causing the processor to not perform any arithmetic operations when reading placeholder instruction 6.

[0039] Because the number of iterations of the loops of function blocks C and D differs, the program code portions of function block C are executed sequentially on the processor first, followed by parallel execution of the program code portions of function blocks C and D. The loop of function block C has 42 iterations, and the loop of function block D has 30 iterations. Thus, the loop of function block C is first executed sequentially 12 times, before the program code of function blocks C and D is then executed together 30 times on the same processor execution unit. Depending on the design of the program code and the underlying programming language, parallel execution could also occur first, followed by sequential execution.

[0040] The program code "Add r1a, [r2a], 0 & r1b, [r2b], +r0," marked by a dashed box in Figure 2, describes the part of the parallelized program code that actually leads to the computational efficiency gain. Furthermore, the part "Add r0, r0, -1" also needs to be executed less frequently, namely 12 times in the sequential loop and 30 times in the parallelized loop (for a total of 42 times), instead of 42 times sequentially and 30 times sequentially (for a total of 72 times) according to the state-of-the-art.

[0041] Register r1 and register r2 are each divided into a register area a and a register area b. The program code therefore provides that register areas b of the first and second registers are accessed according to function block C, and register areas a of the first and second registers are accessed according to the program code of function block D.

Claims

Patent claims Method for operating an information technology system, wherein program code which can be used by a processor to process at least one task (1.1, 1.2) is divided into at least two function blocks (2) by a compiler in a compilation step and these are arranged in a data flow graph (3), and the compiler analyses the data flow graph (3) to determine an execution sequence of the respective function blocks (2) by the processor, characterized in that the compiler is divided into at least two function blocks (2) by the processor when processing the at least one task (1.1, 1.2).2) determines parallel executable function blocks (2), checks for each group (4) of parallel executable function blocks (2) whether at least program code parts of at least two function blocks (2) require the execution of the same instruction (5) by the processor and assigns these program code parts according to single-instruction-multiple-data to the same execution unit of the processor for simultaneous processing.Method according to claim 1, characterized in that the compiler determines the number of instructions (5) to be processed by the processor for each of the parallel-executable function blocks (2) and, for a respective group (4) of parallel-executable function blocks (2), incorporates placeholder instructions (6) into the respective function blocks (C) for those function blocks (C) which contain fewer instructions (5) than the function block (D) with the maximum number of instructions in the respective group (4), so that all function blocks (2) in a respective group (4) contain the same number of instructions (5). Method according to claim 1 or 2, characterized in that. if the program code parts contained by the function blocks (2) of a group (4) of parallel executable function blocks (2) include loops: - the compiler determines the number of iterations of the loops; - the compiler checks whether the loops of the function blocks (2) containing the program code parts assignable to the same execution unit of the processor according to Single-Instruction-Multiple-Data comprise a different number of iterations; and if this is the case: - the compiler causes respective program code portions of the function blocks (C, D) corresponding to the lowest number of iterations to be processed simultaneously by the same execution unit of the processor, and those program code portions of the function blocks (C) with a higher number of iterations to be processed sequentially by the processor according to the remaining iterations. Method according to one of claims 1 to 3, characterized in that, with a processor capable of multithreading, at least two different tasks (1.1, 1.2) are processed in parallel on at least two different execution units of the processor. Information technology system, characterized by a device for executing a method according to one of claims 1 to 4. Vehicle, characterized by an information technology system according to claim 5.