Patent Application
20170140495
This application claims the benefit of Taiwan application Serial No. 104137577, filed Nov. 13, 2015, the disclosure of which is incorporated by reference herein in its entirety.
The disclosure relates in general to an analyzing method and an analyzing system for graphics process.
To satisfy the demand of meticulousness and gorgeousness, the graphics processing unit is widely used for drawing some 3D scenes. However, the battery capacity in a mobile device is limited. A program developer is needed to know the power consumption of the graphics processing unit when the 3D scenes are drawn. It is required to balance the performance, the representation of the 3D scene and the power consumption. The program developer could try to modify the graphics application program to optimize performance per unit of power consumption.
To estimate the power consumption of the graphics processing unit, the signal information or the states information of the hardware components are detected and used for estimating the power consumption. However, the signal information and the states information is based on the particular hardware architecture, and the supplier of the graphics processing unit may not provide the particular method and equipment for detecting the signal information and the states information. Moreover, the program developer cannot find out the particular block of program codes or the particular 3D scene which causes much power consumption.
Further, the above described problems are also happened on the estimation of the time consumption. The program developer cannot find out the particular block of program codes or the particular 3D scene which causes much time consumption.
The disclosure is directed to an analyzing method and an analyzing system for graphics process.
According to one embodiment, an analyzing method for graphics process is provided. The analyzing method includes the following steps. A plurality of reference graphics application programs and a plurality of reference graphics parameters of each of the reference graphics application programs are provided. The reference graphics application programs are grouped into a plurality of groups. A classifying relationship between the reference application graphics application programs and the groups is trained according to the reference graphics parameters. A relative relationship between a plurality of loadings of a plurality of hardware components of a graphic processing unit and the reference graphics parameters is analyzed.
According to another embodiment, an analyzing method for graphics process is provided. A current graphics application program is provided. A plurality of current graphics parameters of the current graphics application program are obtained. The current graphics application program is classified to be at least one of a plurality of groups according to the current graphics parameters. A plurality of weighting coefficients is obtained. A total loading of a graphics processing unit for performing the current graphics application program is calculated according to the weighting coefficients and the current graphics parameters.
According to another embodiment, an analyzing system for graphics process is provided. The analyzing system includes a parameter capturing unit, a grouping unit, a classifier and a relative information unit. The parameter capturing unit is configured to capture a plurality of reference graphics parameters of each of a plurality of reference graphics application programs. The grouping unit is configured to group the reference graphics application programs into a plurality of groups. The classifier is configured to train a classifying relationship between the reference graphics application programs and the groups according to the reference graphics parameters. The relative information unit is configured to analyze a relative relationship between a plurality of loadings of a plurality of hardware components of a graphic processing unit and the reference graphics parameters.
According to another embodiment, an analyzing system for graphics process is provided. The analyzing system includes a parameter capturing unit, a classifier and a calculating unit. The parameter capturing unit is configured to obtain a plurality of current graphics parameters of a current graphics application program. The classifier is configured to classify the current graphics application program to be at least one of a plurality of groups according to the current graphics parameters and obtain a plurality of weighting coefficients. The calculating unit is configured to calculate a total loading of a graphics processing unit for performing the current graphics application program according to the weighting coefficients and the current graphics parameters.
In the following detailed description, for purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the disclosed embodiments. It will be apparent, however, that one or more embodiments may be practiced without these specific details. In other instances, well-known structures and devices are schematically shown in order to simplify the drawing.
Please refer to
The analyzing system 100 includes a parameter capturing unit 110, a grouping unit 120, a classifier 130, a relative information unit 140 and a calculating unit 150. If the analyzing system 100 is performed at an offline mode, the parameter capturing unit 110, the grouping unit 120, the classifier 130 and the relative information unit 140 create an estimating model according to a plurality of reference graphics application programs. If the analyzing system 100 is performed at an online mode, the parameter capturing unit 110, the classifier 130, the relative information unit 140 and the calculating unit 150 analyze the current graphics application program to obtain the total loading of the graphics processing unit for performing the current graphics application program according to the estimating model.
Please refer to
One or all of the parameter capturing unit 110, the grouping unit 120, the classifier 130, the relative information unit 140 and the calculating unit 150 can be a chip, a circuit, a circuit board, or a storage device storing a plurality of program codes.
Please refer to
According to the following equation (1), the total loading PT is the sum of the product of a plurality of weighting coefficients Wk,j (1≦j≦J) and a plurality of loadings of the hardware components. The subscript “k” indicates the k-th group, the subscript “j” indicates the j-th hardware component. Each group corresponds J weighting coefficients Wk,j (1≦j≦J).
Please refer to table 1. Table 1 shows the relationship between a plurality of reference graphics parameters of each reference graphics application program and the hardware components of the graphics processing unit. The reference graphics parameters can include the general parameter, the geometry parameter, the fragment parameter, and the texture parameter.
The general parameter may include the frame rate. The geometry parameter may include the number of triangles, the number of vertices, the number of batches, the lighting mode, the level-of-detail, the primitive assembly type, the interleaving attributes, the instancing. The fragment parameter may include the resolution, the scissor test/stencil buffer, the alpha test, the anti-aliasing. The texture parameter may include the number of texels, the texture mapping, the texture filtering, the texture compression, and the POT/NPOT.
In table 1, “X” indicates that the reference graphics parameter and the hardware component are highly correlated. As shown in equation (2), there is a relative relationship between each loading Pj and I reference graphics parameters Ri (1≦i≦I). In one example, the relative relationship is illustrated by a linear relationship. The subscript “i” indicates the i-th reference graphics parameter.
That is to say, the equation (1) can be substituted by the equation (3). The equation (3) is the estimating model for the total loading PT.
As shown in
Please refer to
In step S120, the grouping unit 120 groups the reference graphics application programs RPn (1≦n≦N) into a plurality of groups, such as K groups. In this step, the number of the groups can be predetermined. The total loadings PTn (1≦n≦N) of the reference graphics application programs RPn (1≦n≦N) are measured respectively. Then, the weighting coefficients Wn,j (1≦n≦N and 1≦j≦J) of the hardware components are obtained respectively. Afterwards, a grouping algorithm, such as a K-means algorithm, is used to group the reference graphics application programs RPn (1≦n≦N). If some sets of the growth trends of the weighting coefficients Wn,j of the reference graphics application programs RPn (1≦n≦N) are similar, then those reference graphics application programs RPn are grouped into the same group. In each of the groups, the weighting coefficients Wk,j (1≦k≦K and 1≦j≦J) is calculated by linear interpolation. For example, in 30 reference graphics application programs RPn (1≦n≦30), the weighting coefficients Wn,j (1≦n≦10 and 1≦j≦J) of the first to 10-th reference graphics application programs RPn (1≦n≦10) are similar. Therefore, the first to 10-th reference graphics application programs RPn (1≦n≦10) are grouped into a first group, and the weighting coefficients W1,j (1≦j≦J) (not shown) of the first group are calculated by linear interpolation. The weighting coefficients Wn,j (11≦n≦30 and 1≦j≦J) of the 11-th to 30-th reference graphics application programs RPn (11≦n≦30) are similar. Therefore, the 11-th to 30-th reference graphics application programs RPn (11≦n≦30) are grouped into a second group, and the weighting coefficients W2,j (1≦j≦J) (not shown) of the second group are calculated by linear interpolation. In the other embodiments, the grouping algorithm can be an Expectation-maximization algorithm, a Highly Connected Subgraphs clustering algorithm, a Connectivity based clustering algorithm, or a hierarchical clustering algorithm.
After grouping, the grouped reference graphics application programs RPn (1≦n≦N) are used for training. In step S130, the classifier 130 trains a classifying relationship CR between the reference graphics application programs and the groups according to the reference graphics parameters Rn,i (1≦n≦N and 1≦i≦I). In this step, the training procedure is performed according a Machine Learning Algorithm, such as an artificial neural network algorithm (ANN algorithm), or a Support Vector Machine algorithm (SVM algorithm). In this step, the classifying relationship CR is not trained according to the information detected from the hardware components. That is to say, when to determine which group the current graphics application program RP′ (shown in
In step S140, the relative information unit 140 analyzes a relative relationship, such as a linear relationship LR, between the loadings Pn,j (1≦n≦N and 1≦j≦J)) of the reference graphics parameters Rn,i (1≦n≦N and 1≦i≦I) to obtain the slope coefficients αj,i (1≦j≦J and 1≦i≦I) and the shift coefficients βj (1≦j≦J). In other embodiments, the relative relationship can be a Polynomial regression relationship, a Logarithmic regression relationship, or an Exponential regression relationship.
The step S140 can be performed before the steps S110 to S130; or, the step S140 and one of the steps S110 to S130 can be performed at the same time. The step S140 does not interfere with the steps S110 to S130.
Please refer to
In S220, the classifier 130 classifies the current graphics application program to be at least one of the groups, such as the k′-th group, according to the current graphics parameters Ri′ (1≦i≦I), and obtains the weighting coefficients Wk′,j (1≦j≦J) accordingly. In this step, the input of the current graphics parameters Ri′ (1≦i≦I) and the output is at least one group which the current graphics application program RP′ belongs to. If the output is only one group, then the weighting coefficients of this group are obtained. If the output is more than one groups, then, in one embodiment, interpolating the weighting coefficients of those groups.
In step S230, the calculating unit 150 calculates the total loading PT′ of the graphics processing unit for performing the current graphics application program RP′ according to the weighting coefficients Wk′,j (1≦j≦J), the current graphics parameters Ri′ (1≦i≦I), the slope coefficients αj,i (1≦j≦J and 1≦i≦I) the shift coefficients βj (1≦j≦J) via the equation (3).
In one embodiment, the relative relationship can be trained at the offline mode.
In one embodiment, the hardware components of the graphics processing unit performing the current graphics application program are corresponding to the loading individually.
According to the embodiments described above, the graphics parameters are obtained by analyzing the source codes or the API Trace, and the total loading of the graphics processing unit can be calculated accordingly. This method is suitable for the program developer. As such, the program developer can find out the particular block of program codes which causes much power consumption, and improves the program codes accordingly. To obtain graphics parameters, it is no need to detect or simulate the hardware components. And, the obtained graphics parameters are not based on any particular hardware architecture. Therefore, the estimating model can be widely used for any hardware architecture of the graphics processing unit. The analyzing method and the analyzing system have much flexibility and extensibility.
Moreover, in the estimating model, the weighting coefficients can be dynamically selected according to the current graphics application program to adapt to the load imbalance of the hardware components of the graphics processing unit. It will be apparent to those skilled in the art that various modifications and variations can be made to the disclosed embodiments. It is intended that the specification and examples be considered as exemplary only, with a true scope of the disclosure being indicated by the following claims and their equivalents.
| Number | Date | Country | Kind |
|---|---|---|---|
| 104137577 | Nov 2015 | TW | national |
