Enviar pesquisa
Carregar
2010 bristol q1_coverage-closure
•
1 gostou
•
221 visualizações
O
Obsidian Software
Seguir
Denunciar
Compartilhar
Denunciar
Compartilhar
1 de 35
Baixar agora
Baixar para ler offline
Recomendados
Meyers clp
Meyers clp
Ball State University - Elementary Education - Ford
Oeding
Oeding
Ball State University - Elementary Education - Ford
Kitt clp
Kitt clp
Ball State University - Elementary Education - Ford
Latest Project
Latest Project
thegilvinyl
Frasier clp
Frasier clp
Ball State University - Elementary Education - Ford
Cahoe clp
Cahoe clp
Ball State University - Elementary Education - Ford
Dayal rtp q2_07
Dayal rtp q2_07
Obsidian Software
Brian bailey
Brian bailey
Obsidian Software
Recomendados
Meyers clp
Meyers clp
Ball State University - Elementary Education - Ford
Oeding
Oeding
Ball State University - Elementary Education - Ford
Kitt clp
Kitt clp
Ball State University - Elementary Education - Ford
Latest Project
Latest Project
thegilvinyl
Frasier clp
Frasier clp
Ball State University - Elementary Education - Ford
Cahoe clp
Cahoe clp
Ball State University - Elementary Education - Ford
Dayal rtp q2_07
Dayal rtp q2_07
Obsidian Software
Brian bailey
Brian bailey
Obsidian Software
Zhang rtp q307
Zhang rtp q307
Obsidian Software
Zehr dv club_12052006
Zehr dv club_12052006
Obsidian Software
Yang greenstein part_2
Yang greenstein part_2
Obsidian Software
Yang greenstein part_1
Yang greenstein part_1
Obsidian Software
Williamson arm validation metrics
Williamson arm validation metrics
Obsidian Software
Whipp q3 2008_sv
Whipp q3 2008_sv
Obsidian Software
Vishakantaiah validating
Vishakantaiah validating
Obsidian Software
Validation and-design-in-a-small-team-environment
Validation and-design-in-a-small-team-environment
Obsidian Software
Tobin verification isglobal
Tobin verification isglobal
Obsidian Software
Tierney bq207
Tierney bq207
Obsidian Software
The validation attitude
The validation attitude
Obsidian Software
Thaker q3 2008
Thaker q3 2008
Obsidian Software
Thaker q3 2008
Thaker q3 2008
Obsidian Software
Strickland dvclub
Strickland dvclub
Obsidian Software
Stinson post si and verification
Stinson post si and verification
Obsidian Software
Shultz dallas q108
Shultz dallas q108
Obsidian Software
Shreeve dv club_ams
Shreeve dv club_ams
Obsidian Software
Sharam salamian
Sharam salamian
Obsidian Software
Schulz sv q2_2009
Schulz sv q2_2009
Obsidian Software
Schulz dallas q1_2008
Schulz dallas q1_2008
Obsidian Software
Mais conteúdo relacionado
Mais de Obsidian Software
Zhang rtp q307
Zhang rtp q307
Obsidian Software
Zehr dv club_12052006
Zehr dv club_12052006
Obsidian Software
Yang greenstein part_2
Yang greenstein part_2
Obsidian Software
Yang greenstein part_1
Yang greenstein part_1
Obsidian Software
Williamson arm validation metrics
Williamson arm validation metrics
Obsidian Software
Whipp q3 2008_sv
Whipp q3 2008_sv
Obsidian Software
Vishakantaiah validating
Vishakantaiah validating
Obsidian Software
Validation and-design-in-a-small-team-environment
Validation and-design-in-a-small-team-environment
Obsidian Software
Tobin verification isglobal
Tobin verification isglobal
Obsidian Software
Tierney bq207
Tierney bq207
Obsidian Software
The validation attitude
The validation attitude
Obsidian Software
Thaker q3 2008
Thaker q3 2008
Obsidian Software
Thaker q3 2008
Thaker q3 2008
Obsidian Software
Strickland dvclub
Strickland dvclub
Obsidian Software
Stinson post si and verification
Stinson post si and verification
Obsidian Software
Shultz dallas q108
Shultz dallas q108
Obsidian Software
Shreeve dv club_ams
Shreeve dv club_ams
Obsidian Software
Sharam salamian
Sharam salamian
Obsidian Software
Schulz sv q2_2009
Schulz sv q2_2009
Obsidian Software
Schulz dallas q1_2008
Schulz dallas q1_2008
Obsidian Software
Mais de Obsidian Software
(20)
Zhang rtp q307
Zhang rtp q307
Zehr dv club_12052006
Zehr dv club_12052006
Yang greenstein part_2
Yang greenstein part_2
Yang greenstein part_1
Yang greenstein part_1
Williamson arm validation metrics
Williamson arm validation metrics
Whipp q3 2008_sv
Whipp q3 2008_sv
Vishakantaiah validating
Vishakantaiah validating
Validation and-design-in-a-small-team-environment
Validation and-design-in-a-small-team-environment
Tobin verification isglobal
Tobin verification isglobal
Tierney bq207
Tierney bq207
The validation attitude
The validation attitude
Thaker q3 2008
Thaker q3 2008
Thaker q3 2008
Thaker q3 2008
Strickland dvclub
Strickland dvclub
Stinson post si and verification
Stinson post si and verification
Shultz dallas q108
Shultz dallas q108
Shreeve dv club_ams
Shreeve dv club_ams
Sharam salamian
Sharam salamian
Schulz sv q2_2009
Schulz sv q2_2009
Schulz dallas q1_2008
Schulz dallas q1_2008
2010 bristol q1_coverage-closure
1.
Easy and Hard
Ways to Reach Coverage Closure Avi Ziv Simulation-based Verification Technologies IBM Haifa research Lab IBM Labs in Haifa © 2010 IBM Corporation
2.
IBM Labs in
Haifa Scope of This Talk Verification Design Checking, Plan Under Assertions Verification Biased-Random Test Pass Directives Stimuli Simulator Fail Generator Test Coverage Information Coverage Reports Coverage Analysis Tool 2 © 2010 IBM Corporation
3.
IBM Labs in
Haifa The Truth About Coverage Analysis The main goals of the coverage process are: Monitor the quality of the verification process Identify unverified and lightly verified areas Help understanding of the verification process This leads to conflicting goals Want to collect as much data as possible So we do not miss important events User needs concise and informative reports So we do not drown in too much detail Coverage analysis help to close the loop from coverage measurement to the verification plan and verification environment adaptation 3 © 2010 IBM Corporation
4.
IBM Labs in
Haifa The Truth About Coverage Analysis Normal projects contains thousands or even millions coverage events Even with high coverage this means many hundreds (or thousands or millions) uncovered events We never have enough resources to deal with all of them Need to extract important information out of them Ints Result RM Rnd Count First Last Fadd +0 0 Y 2 3/4/07 6/6/07 fadd Fadd Fsub +0 +Norm -∞ 0 +∞ +∞ Y Y N 2 21 3 3/4/07 1/6/07 14/2/07 6/6/07 9/6/07 24/4/07 fsqrt qNaN Near Y 0 - - fadd +Norm + ∞ Y 21 1/6/07 9/6/07 fmul +∞ -∞ N 9 31/1/07 20/6/07 fdivs -Norm Near Y 1 22/2/07 22/2/07 Fadd +0 0 Y 2 3/4/07 6/6/07 Fadd +Norm +∞ Y 21 1/6/07 9/6/07 Fsub -∞ +∞ N 3 14/2/07 24/4/07 fsub fsqrt fmul -∞ qNaN +∞ +∞ Near -∞ N Y N 3 0 9 14/2/07 - 31/1/07 24/4/07 - 20/6/07 fdivs -Norm Near Y 1 22/2/07 22/2/07 fnabs -0 Near N 11 3/4/07 11/5/07 fsqrt Fadd Fadd qNaN +0 +Norm Near 0 +∞ Y Y Y 0 2 21 - 3/4/07 1/6/07 - 6/6/07 9/6/07 Fsub -∞ +∞ N 3 14/2/07 24/4/07 fmul N 9 31/1/07 20/6/07 fsqrt qNaN Near Y 0 - - fmul fdivs +∞ +∞ -Norm -∞ -∞ Near N Y 9 1 31/1/07 22/2/07 20/6/07 22/2/07 fnabs -0 Near N 11 3/4/07 11/5/07 Fadd +0 0 Y 2 3/4/07 6/6/07 fdivs Fadd Fsub -Norm +Norm -∞ Near +∞ +∞ Y Y N 1 21 3 22/2/07 1/6/07 14/2/07 22/2/07 9/6/07 24/4/07 fsqrt qNaN Near Y 0 - - fnabs fmul fdivs -0 +∞ -Norm Near -∞ Near N N Y 11 9 1 3/4/07 31/1/07 22/2/07 11/5/07 20/6/07 22/2/07 fnabs -0 Near N 11 3/4/07 11/5/07 Fadd +0 0 Y 2 3/4/07 6/6/07 Fadd +Norm +∞ Y 21 1/6/07 9/6/07 Fsub -∞ +∞ N 3 14/2/07 24/4/07 fsqrt qNaN Near Y 0 - - fmul +∞ -∞ N 9 31/1/07 20/6/07 4 fdivs -Norm Near Y 1 22/2/07 22/2/07 © 2010 IBM Corporation Fadd +0 0 Y 2 3/4/07 6/6/07 Fadd +Norm +∞ Y 21 1/6/07 9/6/07 Fsub -∞ +∞ N 3 14/2/07 24/4/07 fsqrt qNaN Near Y 0 - - fmul +∞ -∞ N 9 31/1/07 20/6/07 fdivs -Norm Near Y 1 22/2/07 22/2/07
5.
IBM Labs in
Haifa (Too) Detailed Status Reports Detailed status reports can provide too much details even for a moderate coverage models Hard to focus on the areas in the coverage model we are currently interested in Hard to understand the meaning of the coverage information Solution – advanced coverage analysis techniques Allow the user to focus on the current area of interest and look at the coverage data with the appropriate level of detail Two basic operations Select important events Group events together Three analysis techniques Manual analysis – coverage views and navigation Automatic analysis – hole analysis and quasi-holes Semi-automatic analysis – hole queries 5 © 2010 IBM Corporation
6.
IBM Labs in
Haifa Manual Analysis Goals Allow the user to focus on the current area of interest and look at the coverage data with the appropriate level of detail Provide means for navigating between coverage reports to extract the useful information Solution – coverage views Dynamically define the events to look at and granularity of the report 6 © 2010 IBM Corporation
7.
IBM Labs in
Haifa Projection Project the n dimension coverage space onto an m (< n) subspace Allow users to concentrate on a specific set of attributes Help in understanding some of things leading up to the big picture Instruction Count Density fadd 12321 127/136 fsub 10923 122/136 fmul 4232 94/136 fsqrt 13288 40/56 fabs 9835 38/40 7 © 2010 IBM Corporation
8.
IBM Labs in
Haifa Automatic Coverage Analysis Detailed status reports do no always reveal interesting information hidden in the coverage data You need to know where to look at You need to know which questions to ask the coverage tool Specifically, it is hard to find large areas of uncovered events in the coverage model 8 © 2010 IBM Corporation
9.
IBM Labs in
Haifa Large Holes Example All combinations of two attributes, X and Y Possible values 0 – 9 for both (100 coverage events) After a period of testing, 70% coverage is achieved Uncovered events 2D Visualization Y X Y X Y X Y 0 2 4 4 7 6 9 0 3 5 2 7 7 8 1 2 5 8 7 8 7 1 4 6 2 8 2 6 2 1 6 6 8 6 5 2 2 6 7 8 7 4 2 6 6 8 8 8 3 3 2 7 2 8 9 2 3 7 7 3 9 2 4 2 7 4 9 9 1 0 X 0 1 2 3 4 5 6 7 8 9 9 © 2010 IBM Corporation
10.
IBM Labs in
Haifa Finding Large Holes 2D visualization can be useful, but only in a limited number of cases Handling spaces with higher dimensionality is difficult Handling attributes with large number of values is difficult Handling unordered values is difficult Finding non-trivial patterns is difficult Automatic techniques can overcome these 9 problems 8 7 Try to find large areas in the coverage space 6 that are not covered 5 Use basic techniques to combine sets of 4 uncovered events into large meaningful holes 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 © 2010 IBM Corporation
11.
IBM Labs in
Haifa Notes on Hole Analysis Algorithms Handling irrelevant (not interesting or not legal) events makes the algorithm conceptually more complex What If The Hole Is Not Pure? Hole analysis produces large set of small holes There is a large area that is lightly covered This area can be more significant than the small holes Covered Uncovered Irrelevant 11 © 2010 IBM Corporation
12.
IBM Labs in
Haifa Automatic Analysis and Adaptation Analysis of coverage data and adapting the verification plan and process according to them is one of the main bottlenecks of the verification process Need to handle huge amount of data Process is tedious and time consuming Requires expertise to: Identify important pieces of information Understand the root causes for them Help fix these root causes Motivation Coverage analysis tools can assess the quality of the verification process, but not recommend how to improve it Objectives Introduce an automatic mechanism to tune stimulus generation Stimulate hard-to-reach coverage points Improve rate of coverage Control coverage space distribution 12 © 2010 IBM Corporation
13.
IBM Labs in
Haifa Closing The Loop from Coverage to Stimuli The problem: Given a coverage event that we want to hit, how to create a stimuli that reaches the requested event In general, this is a very hard problem to solve because of the possible big distance between stimuli and coverage In terms of abstraction In terms of languages In terms of time 13 © 2010 IBM Corporation
14.
IBM Labs in
Haifa How to Hit Uncovered Coverage Events The manual approach Based on understanding of the DUV and its environment Understand the target event and how to reach it Design the stimuli that reaches the event Break the problem into smaller problems Solve each problem separately Combine solutions Measure quality of solutions Iterate and improve until target reached Automatic solution schemes basically follow the same approaches 14 © 2010 IBM Corporation
15.
IBM Labs in
Haifa Model-based Coverage Directed Generation The basic idea Create a model of the DUV and query it on how to reach the target event Model requirements Simple Accurate Supporting queries An important and often difficult part of the solution is translation of the abstract test provided by the model into a concrete one 15 © 2010 IBM Corporation
16.
IBM Labs in
Haifa (Conceptual) Example Build an abstract state machine that Emulate the operation of the DUV Has the target coverage event as a state or transition Use model checker to find a path from the initial state to the target state By challenging it to proof that the state is unreachable Convert the path into a concrete test process 1 6 request 0 2 5 request 4 3 16 © 2010 IBM Corporation
17.
IBM Labs in
Haifa Examples Actually, I am not going to give specific examples There are many papers published with the same basic idea. They present innovation in How to build the model How to traverse the model How to convert the abstract traversal to a concrete test Most of this work is coming from academia and is working on small examples This approach is not adapted in industry 17 © 2010 IBM Corporation
18.
IBM Labs in
Haifa The Model Is Everything If the model is accurate this approach works very well It is (almost) guaranteed to generate tests that reach the target events But small inaccuracies can lead to big degradation in performance Building and maintaining an accurate model can be a big problem Endless number of end cases to take care of Constant changes to the DUV 18 © 2010 IBM Corporation
19.
IBM Labs in
Haifa Exceptions Automatic construction of the model from the implementation For example, by ignoring some of the state variables Here, the big problem is converting the abstract test into a concrete one Similar to abstraction-refinement in formal verification Another possible exception are tools such as Trek by Breker and inFact by Mentor Source: Breker 19 © 2010 IBM Corporation
20.
IBM Labs in
Haifa Data-Driven Coverage Directed Generation Biased-Random Test DUV Pass Directives Stimuli Fail Test Simulator Generator Coverage Information CDG Coverage Engine Analysis Tool Replace the power of the accurate model of model-driven CDG with the ability to learn/adapt based on observed data of stimuli (or directives) and the resulting coverage 20 © 2010 IBM Corporation
21.
IBM Labs in
Haifa How Data-Driven CDG Works The CDG engine is fed pairs of inputs (directives) and outputs (coverage data) These pairs are often called training data The CDG engine “understands” the relations between inputs and outputs and can answer queries about the relations What directive can lead to a requested coverage event? Two levels of understanding Memorizing Generalization In CDG we are usually interested in pairs not seen in the training data Specifically, how to reach uncovered events Generalization is the key to success 21 © 2010 IBM Corporation
22.
IBM Labs in
Haifa How to Generalize Need to know the relations between items in the output space And similarly in the input space Example – ordering rules (<, >, =) Example – similarity Usually means breaking the item into sub-items 22 © 2010 IBM Corporation
23.
IBM Labs in
Haifa Cross-Product Coverage and Generalization Cross-product coverage is a natural form for generalization in the coverage space Break up the output space along the attribute’s axis Understand the input-output relations for each attribute Generalize by combining the understandings But life is not that simple Attributes are related Conflicting understanding ? Randomness … 23 © 2010 IBM Corporation
24.
IBM Labs in
Haifa CDG Using Bayesian Networks Model the CDG process rather than the design under test Cast CDG as a statistical inference framework Use Bayesian networks to represent relations among the CDG ingredients A natural and compact representation of the distribution space Enables encoding of essential domain knowledge cp_cmd_enable[][] = { // mode weight { 0x8, 30-35,}, { 0xE, 1-10,} }; cp_core_enable[][] = CP Core Pipe Cmd { // mode weight 0 0 0 0E { 0x2, 10-100,}, { 0x3, 10-100,} * 1 0 18 }; 2 0 * 1D 24 © 2010 IBM Corporation
25.
IBM Labs in
Haifa Bayesian Networks – Compact Representation of Probability Distributions via Conditional Independence Family of Alarm E B P(A | E,B) Qualitative part: Earthquake Burglary e b .9 .1 Directed acyclic graph (DAG) e b .7 .3 Nodes – random variables Edges – direct influence Radio Alarm e b .8 .2 e b .01 .99 Together: Call Define a unique distribution in a Quantitative part: Set of conditional factored form probability distributions 25 © 2010 IBM Corporation
26.
IBM Labs in
Haifa Employing Bayesian Networks for CDG Test Coverage Directive Simulator Coverage Generator Tool Directive Space Coverage Space D1 C1 D2 C2 Mapping Mapping H1 Dm Cn 26 © 2010 IBM Corporation
27.
IBM Labs in
Haifa Reaching a Specific Coverage Point 1. Map the point to values in the coverage nodes of the Bayesian network C1=c1, C2=c2, …, Cn=cn. 2. Query the network for the most probable explanation, (d1, …, dm) = argmax P(C1=c1,…,Cn=cn | D1, …,Dm) 3. Map the values in the directive nodes to test the directive file Test Directive Specific Coverage File Point Space D1 C1 D2 C2 Mapping Mapping H1 Dm Cn 27 © 2010 IBM Corporation
28.
IBM Labs in
Haifa Advantages and Disadvantages There are other similar approaches For example, the use of Inductive Logic Programming (ILP) to learn the relations between the stimuli and coverage (Hsueh and Eder) All of them are less dependent on the accuracy of the model But they do not guarantee to hit the target event At best they improve the probability of doing so These approaches rely on some structure in the coverage model As is, they cannot work on singular coverage events 28 © 2010 IBM Corporation
29.
IBM Labs in
Haifa Back to The Shower In model-based CDG, the model provides us an accurate solution But we depend on the accuracy of the model In the data-driven approach we showed, we do not need an accurate model But all we get is improved probability of hitting the target Yet another approach is to take existing attempts and iteratively improve them until the target is hit Several such CDG systems exist based on Genetic algorithms Reward functions Path tracing 29 © 2010 IBM Corporation
30.
IBM Labs in
Haifa Genetic Algorithms Algorithmic framework that tries to imitate nature evolution There are several published papers on CDG using GA Mostly for processor verification (stimuli is assembly programs) The basic idea 1. Choose the initial population of individuals 2. Evaluate the fitness of each individual in that population 3. Repeat on this generation until termination: 1. Select the best-fit individuals for reproduction 2. Breed new individuals through crossover and mutation operations to give birth to offspring 3. Evaluate the individual fitness of new individuals 4. Replace least-fit population with new individuals The good and bad about GA is that we do not need to understand why changes improve the next generation, just to know that they do 30 © 2010 IBM Corporation
31.
IBM Labs in
Haifa Source: Nusym Path Tracing Collect information on decisions made during simulation runs In the design and the testbench Identify contributors to the decisions Trace the contributors back to their roots For example, random decisions by the generator Modify the roots to reach desired decisions Properties Not guaranteed to find satisfying path Finds different path Scalable 31 © 2010 IBM Corporation
32.
IBM Labs in
Haifa Source: Nusym Path Tracing Example Random Run Path Tracing Replay a = $random; a = 12; a = 24; a = 24; b = $random; b = 21; b = 21; c = a + b; c = 33; c = 45; c = 45; if (c == 45) => else => then F T d = 1; d = 0; d = 1; d = 0; d = 0; If (d == 0) => then => else => then T $display(“HIT!!”); HIT!! 32 © 2010 IBM Corporation
33.
IBM Labs in
Haifa Source: Nusym Example of CDG Results coverage Conventional Nusym simulation grading 100% IBM CDG 0% Base coverage Lack of coverage because 90% coverage after 55K runs TB over-constrained (40%) CDG results Unreachable code (10%) >95% coverage after 4K runs Dead code (40%) Two large holes identified Tool timeout (10%) 33 © 2010 IBM Corporation
34.
IBM Labs in
Haifa Summary Getting to coverage closure is one of the most difficult and time consuming tasks verification engineers face The task has two important aspects Extracting important information out of the ocean of data Act upon this information to fix issues in activation of the verification environment E.g., hit uncovered events Advanced techniques and automation can help in both aspects We are far away from having an end-to-end working solution But we are making progress 34 © 2010 IBM Corporation
35.
IBM Labs in
Haifa 35 © 2010 IBM Corporation
Baixar agora