Program Slicing Notes

src/teaching/cse-662/2019fa/index.md

@@ -1,5 +1,8 @@
 ---
 title: CSE 662 - Languages and Runtimes for Big Data - Fall 2019
+paper_ideas:
+  - name: Adaptive Functional Programming
+    url: https://www.cs.cmu.edu/~guyb/papers/popl02.pdf
 ---
 
 

src/teaching/cse-662/2019fa/slide/2019-09-16-ProgramSlicing.erb

@@ -0,0 +1,79 @@
+---
+title: Program Slicing
+supplements:
+  - name: A Survey of Program Slicing Techniques (Frank Tip)
+    url: "https://wiki.epfl.ch/provenance2011/documents/10.1.1.93.5469.pdf"
+---
+
+- Announcements:
+  - Progress report Thursday / Talk by Learned Index (45 min + 15 + 15)
+  - Sept 25 deadline for Checkpoint 1:
+    - What is the specific challenge that you will solve?
+    - What metrics will you use to evaluate success?
+    - What deliverables will you produce?
+    - What resources have you collected / what have you learned?
+    - How do you plan to approach the problem?
+    
+- Typo in the paper: 
+  - EQN defining B_C^0
+
+- Overview
+  - Delta query: Come up with a query that computes the difference in results given a change to the output.
+  - Related: Provenance query: What inputs affected the output
+  - Program slice: What instructions affected the output.
+- Background
+  - Program
+    - A sequence of instructions w/ flow patterns (not quite a lattice -- but could be multiple tops)
+    - View 1: (As Memory) Instructions listed in order
+    - View 2: (Control Flow Graph) Previous/next instruction (in some possible trace) "Predecessor", "Successor"
+  - Trace
+    - A sequence of instructions in execution order 
+    - e.g., State Trajectory in paper: Instructions + Corresponding Input State
+  - Flow Dependence
+    - i flows to j if
+      (i) There is a variable x assigned to by i, and 
+      (ii) x is read from by j, and
+      (iii) There is *some* path from i to j on which x is not read.
+  - Post-dominance (Inverse dominator):
+    - A node i in the CFG is post-dominatedby a node by j if all paths from i to STOP pass through j.  
+    - A node j is control dependent on a node i if 
+      (i) there exists a path P from i to j such that j post-dominates every node in P, excluding i and j, and 
+      (ii) i is not post-dominated by j.  <- Q: Because of loops
+- Static Slicing:
+  - Input: A pair (n, V)
+    - n: A line of the program
+    - V: a set of variables of interest.
+  - Output: A (not minimal) subset of the program that can still compute everything in V (assuming deterministic evaluation)
+  - Use cases:
+    - Dead Code Elimination
+    - Highlight Dependent Instructions
+    - Delta Programs
+    - Optimization
+      - Cache/JIT compilation via hotspot prediction
+      - Lazy evaluation
+      - Parallel Execution
+  - Weiser: 
+    - Delete instructions from the program that can not influence the output
+    - R^0(n): "relevant" variables at a line.  Start with V being relevant for the line n, and work backwards through the CFG
+      - Remove variables on lines that overwrite them
+      - When a variable is removed, add all variables read by the line.
+    - S^0: Lines that modify a variable in the line's relevant variable set
+    - B^0: Lines on which some line in S^0 has a control-dependency
+    - {R, S, B}^{n>0}: Add the inputs of everything in B^n-1 to the relevant variable set, and then add lines with a dependency.
+  - Information Flow Relations (Bergeretti & Carre) 
+    - Provenance P10 / Fig 7
+  - Intraprocedural slicing:
+    - Figure out which variables the code reads from / writes to.
+    -  Weiser’s  algorithm  does  not  take  into  accountwhichoutput  parameters  aredependent onwhichinput parameters is a source of imprecision
+    - Bergeretti & Carre: 
+- Dynamic Slicing: 
+  - Input: A triple (x, n, V) and a state trajectory/trace
+    - x: program input
+    - n: step in the trajectory/trace
+    - V: subset of variables of interest
+  - For each step, look at the dependencies:
+    - DU: (Statement that used a variable, Last Statement to Modify It)
+    - TC: (Statement that issues a control predicate, Statement Dependent on it)
+    - IR: (Statement, Any other occurrence of the same statement in the trace)
+      - Needed to capture forward dependencies.
+  - Extend the trace with dependencies in the trace, then map to statements.