Heterogeneous processor platform testing and performance grading method

Abstract

The invention discloses a heterogeneous processor platform testing and performance grading method, which relates to the technical field of heterogeneous processor testing, and comprises the following steps: S1, a test environment preparation stage; s2, a test plan stage; s3, a test execution stage; and S4, a comprehensive score and grade evaluation stage. According to the heterogeneous processor platform testing and performance grading method, a testing method system and a performance level grading strategy of a heterogeneous processor platform are given, index values can be calculated through testing results, and finally the performance condition of the heterogeneous processor platform is obtained and is set to be one of the first-level performance level, the second-level performance level and the third-level performance level according to grading requirements. According to the method, a heterogeneous processor platform testing system is constructed by taking testing as a means, and an enterprise is helped to clarify the capability level of the heterogeneous processor through a grading evaluation mode, so that the performance of the existing heterogeneous processor is optimized in a targeted manner, and the application depth and breadth of the heterogeneous processor are accelerated.

Description

Technical Field

[0001] The present invention relates to the technical field of heterogeneous processor testing, and in particular to a heterogeneous processor platform testing and performance grading method. Background Art

[0002] At present, as various processor design concepts and innovative architectures specifically for artificial intelligence applications gradually take shape, more than ten emerging AI accelerators suitable for different application scenarios continue to emerge from Huawei, Cambrian, Suiyuan Technology, etc., and how to efficiently and reliably utilize these diversified processors, heterogeneity has become the core technology of integrated optimization. Heterogeneous processors maximize computing efficiency by coordinating various computing powers, thereby improving processor performance. However, different types of processors have different testing methods, and different testing methods can meet the same testing requirements. Therefore, there are problems in processor testing, such as inconsistent methods, incomplete architecture, and unclear functions. As a new form, heterogeneous processors have higher requirements for specific and clear testing methods, unified and effective testing systems, and clear and understandable capability levels.

[0003] In the prior art, SPEC CPU is a CPU computing performance test benchmark suite launched by the SPEC organization. It focuses on the CPU's integer computing power, floating-point computing power, integer concurrency rate, and floating-point concurrency rate, and can objectively reflect the CPU's own computing performance. However, the SPEC CPU test itself has problems such as insufficient memory access pressure and limited coverage scenarios. When faced with heterogeneous processors with higher memory access requirements and wider application scenarios, this limitation will be further magnified. At the same time, SPECCPU mainly focuses on computing performance, while ignoring the capabilities of memory, system, etc.

[0004] CINEBENCH was developed by Maxon Computer and uses CPU to render high-definition images. It supports single-thread and multi-thread testing and is a CPU benchmark performance testing standard widely used in the industry. However, CINEBENCH only has one test with a relatively single test dimension, mainly focusing on core performance, which is inconsistent with the systematized, multi-scenario, and multi-dimensional testing requirements for heterogeneous processors.

[0005] In summary, existing testing methods such as SPEC CPU and CINEBENCH mainly focus on CPU capabilities, and pay less attention to GPUs and heterogeneous processors. At the same time, the dimensions of existing testing methods are relatively single and limited, unable to meet diversified testing requirements, and have not yet formed a systematic and quantitative testing framework.

[0006] Therefore, there is an urgent need to improve this shortcoming. The present invention studies and improves the existing technology and its shortcomings, and provides a heterogeneous processor platform testing and performance grading method. Summary of the invention

[0007] The purpose of the present invention is to provide a heterogeneous processor platform testing and performance grading method to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solution: a heterogeneous processor platform testing and performance grading method, comprising the following steps:

[0009] S1. Test environment preparation stage:

[0010] S11. Hardware environment:

[0011] In terms of CPU: Intel Xeon E5-2620 server is selected, with a base frequency of 2.0GHz, dual processors with a main frequency of 2.00GHz and 48GB of memory;

[0012] In terms of GPU: deploy two NVIDIA Tesla K20c with 5G video memory;

[0013] S12, Software environment: Linux CentOS7.0;

[0014] S13. Test samples and equipment:

[0015] Test samples: three heterogeneous multi-core processor platforms of the same model;

[0016] Test equipment: Choose a computer module that is compatible with the test sample interface and a carrier board;

[0017] S2, test planning stage:

[0018] The test purpose is to test the performance of the CPU-GPU heterogeneous processor platform in terms of CPU, GPU, memory, system, etc. and determine the heterogeneous capability level, determine the test items accordingly, and select the corresponding test tools for different test items;

[0019] S3, test execution phase:

[0020] S31. If the test sample supports DVFS, this function must be disabled before testing; if it does not support DVFS, this step can be ignored;

[0021] S32, CPU performance test and stress test

[0022] Use nbench to test CPU performance and obtain MEM, INT, and FP. The MEM index is memory performance, INT reflects fixed-point performance, and FP is double-precision floating-point performance. Then run the stress tool on the three samples to test CPU stress and obtain uptime that reflects the average system load.

[0023] S33, GPU stress test

[0024] Install the Glmark2 tool to stress test the GPU and get the final score of the graphics unit performance such as lighting, buffering, and textures;

[0025] S34, CPU and GPU joint test

[0026] LuxMark 2.0 tool is used to test OpenCL computing performance in CPU+GPU mode;

[0027] S35, memory access latency evaluation and bandwidth test

[0028] The Lmbench tool is used to test the memory access latency and obtain the values ​​of continuous access latency and random access latency. The Stream tool is used to test the memory bandwidth performance and obtain the Copy, Scale, Add, and Triad bandwidth measurement values.

[0029] S36, system comprehensive performance test

[0030] The UnixBench tool was used to test the comprehensive performance of the whole machine equipped with heterogeneous processors. The tests were conducted in single-core and quad-core environments respectively to obtain single-process and multi-process UnixBench values.

[0031] S4. Comprehensive score and grade assessment stage:

[0032] After testing the samples, they are scored according to the score calculation requirements described in the Heterogeneous Processor Platform Score Calculation Requirements Table, and different heterogeneous processor platform levels are divided according to the scores.

[0033] Furthermore, in step S2, the test items specifically include: CPU performance test and stress test, GPU stress test, CPU and GPU joint test, memory access delay evaluation and bandwidth test, and system comprehensive performance test.

[0034] Furthermore, the test tools selected for the test items are as follows:

[0035] 1) CPU performance test and stress test, using Nbench and Stress as test tools;

[0036] 2) GPU stress test, using Glmark2 as the test tool;

[0037] 3) CPU and GPU joint test, using LuxMark 2.0 as the test tool;

[0038] 4) Memory access latency evaluation and bandwidth testing, using Lmbench and Stream as testing tools;

[0039] 5) Comprehensive performance test of the platform, using UnixBench as the system performance testing tool of the platform.

[0040] Furthermore, in step S4, the score calculation requirement table for the heterogeneous processor platform includes five test items: CPU, GPU, CPU+GPU, memory, and system.

[0041] Furthermore, the total score of the CPU test item is 20 points, including four indicators: MEM, INF, FP, and uptime. The sub-items of each indicator are 5 points, and the scoring rules of each indicator are as follows:

[0042] MEM: When MEM ≥ 6.0, the full score is 5 points;

[0043] When MEM is less than 6.0, the score is 5*MEM / 6;

[0044] INF: When INT ≥ 6.5, the full score is 10 points;

[0045] When INT is less than 6.5, the score is 10*INT / 13;

[0046] FP: When FP ≥ 6.5, the full score is 10 points;

[0047] When FP < 6.5, the score is 10*FP / 13;

[0048] Uptime: When uptime>0.7CPU, the score is 0;

[0049] When uptime is between (0.6CPU, 0.7CPU], the score is 1;

[0050] When uptime is between (0.3CPU, 0.6CPU], the score is 3;

[0051] When the uptime is between (0,0.3CPU], the score is 5.

[0052] Furthermore, the total score of the GPU test item is 15 points, specifically including the Glmark2 indicator, and the sub-item of this indicator is 15 points, and the scoring rules are as follows:

[0053] SG When the score is ≥3000, the full score is 15 points;

[0054] S G When <3000, the score is S G / 200;

[0055] Among them, S G This is the Glmark2 test result.

[0056] Furthermore, the total score of the CPU+GPU test item is 15 points, which specifically includes the LuxMark indicator, and the sub-item of this indicator is 15 points. The scoring rules are as follows:

[0057] S L When the score is ≥400, the full score is 15 points;

[0058] S L When <400, the score is 3*S L / 80;

[0059] Among them, S L These are the results of the LuxMark benchmark test.

[0060] Furthermore, the total score of the memory test item is 36 points, which specifically includes six indicators: continuous access delay, random access delay, Copy, Scale, Add, and Triad, and the sub-items and scoring rules of each indicator are as follows:

[0061] Continuous access delay (6 minutes): 0ns≤t cont When ≤10ns, the full score is 6 points;

[0062] t cont When >10ns, the score is 60 / t cont ;

[0063] Among them, t cont Delay duration for consecutive accesses;

[0064] Random access delay (6 minutes): 0ns≤t rand When ≤120ns, the full score is 6 points;

[0065] t rand When >120ns, the score is 720 / t rand ;

[0066] Among them, t rand is the random access delay duration;

[0067] Copy (4 points): When Copy ≥ 10000MB / s, the full score is 4 points;

[0068] When Copy<10000MB / s, the score is Copy / 2500;

[0069] Scale (4 points): When Scale ≥ 8000MB / s, the full score is 4 points;

[0070] When Scale is less than 8000MB / s, the score is Scale / 2000;

[0071] Add (4 points): When Add ≥ 9000MB / s, the full score is 4 points;

[0072] When Add<9000MB / s, the score is Add / 2250;

[0073] Triad (4 points): When Triad ≥ 9000MB / s, the full score is 4 points;

[0074] When Triad<9000MB / s, the score is Triad / 2250.

[0075] Furthermore, the total score of the system test item is 14 points, including two indicators: single process and multi-process. The sub-items of each indicator are 7 points, and the scoring rules of each indicator are as follows:

[0076] Single process: U 单 When ≥1400, the full score is 7 points;

[0077] U 单 When <1400, the score is U 单 / 200;

[0078] Among them, U 单 Represents the single-process UnixBench value;

[0079] Multi-process: U 多 When the speed is ≥3000MB / s, the full score is 7 points;

[0080] U 多 When the speed is less than 3000MB / s, the score is 7*U 多 / 3000;

[0081] Among them, U 多 Represents a multi-process UnixBench value.

[0082] Furthermore, in step S4, there are three levels of heterogeneous processor platforms: from level one to level three, and the level one heterogeneous processor platform has the lowest performance, and the level three heterogeneous processor platform has the highest performance.

[0083] The present invention provides a heterogeneous processor platform testing and performance grading method, which has the following beneficial effects:

[0084] The present invention provides a testing method system and performance level classification strategy for a heterogeneous processor platform. The index value can be calculated through the test results, and the performance of the heterogeneous processor platform can be finally obtained, and it can be classified into a certain performance level among "level one", "level two" and "level three" according to the classification requirements. The present invention uses testing as a means to build a heterogeneous processor platform testing system, and through a hierarchical evaluation method, it helps enterprises to clarify the capability level of heterogeneous processors, thereby optimizing the performance of existing heterogeneous processors in a targeted manner and accelerating the depth and breadth of heterogeneous processor applications.

[0085] The technical solution of the present invention is helpful to clarify the performance level of heterogeneous processors, help builders and users to clarify the performance level of heterogeneous processors, accelerate the improvement of the overall performance of computing power infrastructure, and provide solid equipment guarantee for the promotion of heterogeneous processor applications and the construction of high-performance computing power infrastructure.

[0086] Through the technical solution of the present invention, it is possible to provide ideas for optimizing heterogeneous processors, improve the performance of heterogeneous processors from multiple angles, adopt diversified technical means, focus on CPU, GPU, memory, system and other dimensions, improve processor load capacity and access memory bandwidth, reduce processing latency, and achieve processor performance optimization and efficient processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 A heterogeneous processor platform score calculation requirement table for a heterogeneous processor platform testing and performance grading method of the present invention;

[0088] Figure 2 A heterogeneous processor platform level requirement table for a heterogeneous processor platform testing and performance grading method of the present invention. DETAILED DESCRIPTION

[0089] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0090] like Figure 1-Figure 2 As shown, a heterogeneous processor platform testing and performance grading method includes the following steps:

[0091] S1. Test environment preparation stage:

[0092] S11. Hardware environment:

[0093] In terms of CPU: Intel Xeon E5-2620 server is selected, with a base frequency of 2.0GHz, dual processors with a main frequency of 2.00GHz and 48GB of memory;

[0094] In terms of GPU: deploy two NVIDIA Tesla K20c with 5G video memory;

[0095] S12, Software environment: Linux CentOS7.0;

[0096] S13. Test samples and equipment:

[0097] Test samples: three heterogeneous multi-core processor platforms of the same model;

[0098] Test equipment: Choose a computer module that is compatible with the test sample interface and a carrier board;

[0099] S2, test planning stage:

[0100] The test purpose is to test the performance of the CPU-GPU heterogeneous processor platform in terms of CPU, GPU, memory, system, etc. and determine the heterogeneous capability level;

[0101] Based on this, the test items are determined as follows: CPU performance test and stress test, GPU stress test, CPU and GPU joint test, memory access latency evaluation and bandwidth test, system comprehensive performance test, and corresponding test tools are selected for different test items, as follows:

[0102] 1) CPU performance test and stress test, using Nbench and Stress as test tools;

[0103] 2) GPU stress test, using Glmark2 as the test tool;

[0104] 3) CPU and GPU joint test, using LuxMark 2.0 as the test tool;

[0105] 4) Memory access latency evaluation and bandwidth testing, using Lmbench and Stream as testing tools;

[0106] 5) Comprehensive performance test of the platform, using UnixBench as the system performance test tool of the platform;

[0107] S3, test execution phase:

[0108] S31. If the test sample supports DVFS, this function must be disabled before testing; if it does not support DVFS, this step can be ignored;

[0109] S32, CPU performance test and stress test

[0110] Use nbench to test CPU performance and obtain MEM, INT, and FP. The MEM index is memory performance, INT reflects fixed-point performance, and FP is double-precision floating-point performance. Then run the stress tool on the three samples to test CPU stress and obtain uptime that reflects the average system load.

[0111] S33, GPU stress test

[0112] Install the Glmark2 tool to stress test the GPU and get the final score of the graphics unit performance such as lighting, buffering, and textures;

[0113] S34, CPU and GPU joint test

[0114] LuxMark 2.0 tool is used to test OpenCL computing performance in CPU+GPU mode;

[0115] S35, memory access latency evaluation and bandwidth test

[0116] The Lmbench tool is used to test the memory access latency and obtain the values ​​of continuous access latency and random access latency. The Stream tool is used to test the memory bandwidth performance and obtain the Copy, Scale, Add, and Triad bandwidth measurement values.

[0117] S36, system comprehensive performance test

[0118] The UnixBench tool was used to test the comprehensive performance of the whole machine equipped with heterogeneous processors. The tests were conducted in single-core and quad-core environments respectively to obtain single-process and multi-process UnixBench values.

[0119] S4. Comprehensive score and grade assessment stage:

[0120] After testing the samples, according to the heterogeneous processor platform score calculation requirements table (see Figure 1 ) and divide the heterogeneous processor platform into different levels according to the scores. There are three levels of heterogeneous processor platforms: from level 1 to level 3, and the level 1 heterogeneous processor platform has the lowest performance and the level 3 heterogeneous processor platform has the highest performance (see Figure 2 );

[0121] The score calculation requirement table for heterogeneous processor platforms includes five test items: CPU, GPU, CPU+GPU, memory, and system. The details are as follows:

[0122] 1) The total score of the CPU test item is 20 points, including four indicators: MEM, INF, FP, and uptime. The sub-items of each indicator are 5 points, and the scoring rules for each indicator are as follows:

[0123] MEM: When MEM ≥ 6.0, the full score is 5 points;

[0124] When MEM is less than 6.0, the score is 5*MEM / 6;

[0125] INF: When INT ≥ 6.5, the full score is 10 points;

[0126] When INT is less than 6.5, the score is 10*INT / 13;

[0127] FP: When FP ≥ 6.5, the full score is 10 points;

[0128] When FP < 6.5, the score is 10*FP / 13;

[0129] Uptime: When uptime>0.7CPU, the score is 0;

[0130] When uptime is between (0.6CPU, 0.7CPU], the score is 1;

[0131] When uptime is between (0.3CPU, 0.6CPU], the score is 3;

[0132] When uptime is between (0,0.3CPU], the score is 5;

[0133] 2) The total score of the GPU test item is 15 points, including the Glmark2 indicator, and the sub-item of this indicator is 15 points. The scoring rules are as follows:

[0134] S G When the score is ≥3000, the full score is 15 points;

[0135] S G When <3000, the score is S G / 200;

[0136] Among them, S G This is the Glmark2 test result;

[0137] 3) The total score of the CPU+GPU test item is 15 points, including the LuxMark indicator, and the sub-item of this indicator is 15 points. The scoring rules are as follows:

[0138] S L When the score is ≥400, the full score is 15 points;

[0139] S L When <400, the score is 3*S L / 80;

[0140] Among them, S LThis is the LuxMark benchmark test result;

[0141] 4) The total score of the memory test item is 36 points, including six indicators: continuous access delay, random access delay, Copy, Scale, Add, and Triad. The sub-items and scoring rules of each indicator are as follows:

[0142] Continuous access delay (6 minutes): 0ns≤t cont When ≤10ns, the full score is 6 points;

[0143] t cont When >10ns, the score is 60 / t cont ;

[0144] Among them, t cont Delay duration for consecutive accesses;

[0145] Random access delay (6 minutes): 0ns≤t rand When ≤120ns, the full score is 6 points;

[0146] t rand When >120ns, the score is 720 / t rand ;

[0147] Among them, t rand is the random access delay duration;

[0148] Copy (4 points): When Copy ≥ 10000MB / s, the full score is 4 points;

[0149] When Copy<10000MB / s, the score is Copy / 2500;

[0150] Scale (4 points): When Scale ≥ 8000MB / s, the full score is 4 points;

[0151] When Scale is less than 8000MB / s, the score is Scale / 2000;

[0152] Add (4 points): When Add ≥ 9000MB / s, the full score is 4 points;

[0153] When Add<9000MB / s, the score is Add / 2250;

[0154] Triad (4 points): When Triad ≥ 9000MB / s, the full score is 4 points;

[0155] When Triad is less than 9000MB / s, the score is Triad / 2250;

[0156] 5) The total score of the system test item is 14 points, including two indicators: single process and multi-process. The sub-items of each indicator are 7 points, and the scoring rules of each indicator are as follows:

[0157] Single process: U 单 When ≥1400, the full score is 7 points;

[0158] U 单 When <1400, the score is U 单 / 200;

[0159] Among them, U 单 Represents the single-process UnixBench value;

[0160] Multi-process: U 多 When the speed is ≥3000MB / s, the full score is 7 points;

[0161] U 多 When the speed is less than 3000MB / s, the score is 7*U 多 / 3000;

[0162] Among them, U 多 Represents a multi-process UnixBench value.

[0163] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A method for testing and grading the performance of a heterogeneous processor platform, characterized in that: The following steps are involved: S1. Test environment preparation stage: S11. Hardware environment: In terms of CPU: Intel Xeon E5-2620 server is selected, with a base frequency of 2.0GHz, dual processors with a main frequency of 2.00GHz and 48GB of memory; In terms of GPU: deploy two NVIDIA Tesla K20c with 5G video memory; S12, Software environment: Linux CentOS7.0; S13. Test samples and equipment: Test samples: three heterogeneous multi-core processor platforms of the same model; Test equipment: Choose a computer module that is compatible with the test sample interface and a carrier board; S2, test planning stage: The test purpose is to test the performance of the CPU-GPU heterogeneous processor platform in terms of CPU, GPU, memory, and system, and determine the heterogeneous capability level, determine the test items accordingly, and select the corresponding test tools for different test items; S3, test execution phase: S31. If the test sample supports DVFS, this function must be disabled before testing; if it does not support DVFS, this step can be ignored; S32, CPU performance test and stress test Use nbench to test CPU performance and obtain MEM, INT, and FP. The MEM index is memory performance, INT reflects fixed-point performance, and FP is double-precision floating-point performance. Then run the stress tool on the three samples to test CPU stress and obtain uptime that reflects the average system load. S33, GPU stress test Install the Glmark2 tool to stress test the GPU and get the final score of the graphics unit performance; S34, CPU and GPU joint test LuxMark 2.0 tool is used to test OpenCL computing performance in CPU+GPU mode; S35, memory access latency evaluation and bandwidth test The Lmbench tool is used to test the memory access latency and obtain the values ​​of continuous access latency and random access latency. The Stream tool is used to test the memory bandwidth performance and obtain the Copy, Scale, Add, and Triad bandwidth measurement values. S36, system comprehensive performance test The UnixBench tool was used to test the comprehensive performance of the whole machine equipped with heterogeneous processors. The tests were conducted in single-core and quad-core environments respectively to obtain single-process and multi-process UnixBench values. S4. Comprehensive score and grade assessment stage: After testing the samples, they are scored according to the score calculation requirements described in the Heterogeneous Processor Platform Score Calculation Requirements Table, and different heterogeneous processor platform levels are divided according to the scores.

2. A heterogeneous processor platform testing and performance grading method according to claim 1, characterized in that: In step S2, the test items specifically include: CPU performance test and stress test, GPU stress test, CPU and GPU joint test, memory access latency evaluation and bandwidth test, and system comprehensive performance test.

3. A heterogeneous processor platform testing and performance grading method according to claim 2, characterized in that: The test tools selected for the test items are as follows: 1) CPU performance test and stress test, using Nbench and Stress as test tools; 2) GPU stress test, using Glmark2 as the test tool; 3) CPU and GPU joint test, using LuxMark 2.0 as the test tool; 4) Memory access latency evaluation and bandwidth testing, using Lmbench and Stream as testing tools; 5) Comprehensive performance test of the platform, using UnixBench as the system performance testing tool of the platform.

4. A heterogeneous processor platform testing and performance grading method according to claim 1, characterized in that: In step S4, the score calculation requirement table of the heterogeneous processor platform includes five test items: CPU, GPU, CPU+GPU, memory, and system.

5. A heterogeneous processor platform testing and performance grading method according to claim 4, characterized in that: The total score of the CPU test item is 20 points, including four indicators: MEM, INF, FP, and uptime. The sub-items of each indicator are 5 points, and the scoring rules of each indicator are as follows: MEM: When MEM ≥ 6.0, the full score is 5 points; When MEM is less than 6.0, the score is 5*MEM / 6; INF: When INT ≥ 6.5, the full score is 10 points; When INT is less than 6.5, the score is 10*INT / 13; FP: When FP ≥ 6.5, the full score is 10 points; When FP < 6.5, the score is 10*FP / 13; Uptime: When uptime>0.7CPU, the score is 0; When uptime is between (0.6CPU, 0.7CPU], the score is 1; When uptime is between (0.3CPU, 0.6CPU], the score is 3; When the uptime is between (0,0.3CPU], the score is 5.

6. A heterogeneous processor platform testing and performance grading method according to claim 4, characterized in that: The total score of the GPU test item is 15 points, specifically including the Glmark2 indicator, and the sub-item of this indicator is 15 points. The scoring rules are as follows: S G When the score is ≥3000, the full score is 15 points; S G When <3000, the score is S G / 200; Among them, S G This is the Glmark2 test result.

7. A heterogeneous processor platform testing and performance grading method according to claim 4, characterized in that: The total score of the CPU+GPU test item is 15 points, which specifically includes the LuxMark indicator, and the sub-item of this indicator is 15 points. The scoring rules are as follows: S L When the score is ≥400, the full score is 15 points; S L When <400, the score is 3*S L / 80; Among them, S L These are the results of the LuxMark benchmark test.

8. A heterogeneous processor platform testing and performance grading method according to claim 4, characterized in that: The total score of the memory test item is 36 points, which specifically includes six indicators: continuous access delay, random access delay, Copy, Scale, Add, and Triad. The sub-items and scoring rules of each indicator are as follows: Continuous access delay (6 minutes): 0ns≤t cont When ≤10ns, the full score is 6 points; t cont When >10ns, the score is 60 / t cont ; Among them, t cont Delay duration for consecutive accesses; Random access delay (6 minutes): 0ns≤t rand When ≤120ns, the full score is 6 points; t rand When >120ns, the score is 720 / t rand ; Among them, t rand is the random access delay duration; Copy (4 points): When Copy ≥ 10000MB / s, the full score is 4 points; When Copy<10000MB / s, the score is Copy / 2500; Scale (4 points): When Scale ≥ 8000MB / s, the full score is 4 points; When Scale is less than 8000MB / s, the score is Scale / 2000; Add (4 points): When Add ≥ 9000MB / s, the full score is 4 points; When Add<9000MB / s, the score is Add / 2250; Triad (4 points): When Triad ≥ 9000MB / s, the full score is 4 points; When Triad<9000MB / s, the score is Triad / 2250.

9. A heterogeneous processor platform testing and performance grading method according to claim 4, characterized in that: The total score of the system test item is 14 points, including two indicators: single process and multi-process. The sub-items of each indicator are 7 points, and the scoring rules of each indicator are as follows: Single process: U 单 When ≥1400, the full score is 7 points; U 单 When <1400, the score is U 单 / 200; Among them, U 单 Represents the single-process UnixBench value; Multi-process: U 多 When the speed is ≥3000MB / s, the full score is 7 points; U 多 When the speed is less than 3000MB / s, the score is 7*U 多 / 3000; Among them, U 多 Represents a multi-process UnixBench value.

10. The method for testing and grading the performance of a heterogeneous processor platform according to claim 1, characterized in that: In step S4, there are three levels of heterogeneous processor platforms: from level one to level three, and the level one heterogeneous processor platform has the lowest performance, and the level three heterogeneous processor platform has the highest performance.

Images