Enhancing weak consistency

• US 9,594,667 B2
• Filed: 12/11/2014
• Issued: 03/14/2017
• Est. Priority Date: 10/09/2014
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Bates, Cary L., Helgeson, Lee, King, Justin K., Schlicht, Michelle A.
• Primary Examiner(s): ST LEGER, GEOFFREY R
• Application Number:
  US14/567,608
• Publication Number:
  US 20160103752A1
• Time in Patent Office: 824 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 11/362 Software debugging
• CPC Class Codes: G06F 11/3632 of specific synchronisation aspects

Abstract

Methods to discover weak consistency errors in an application, by executing, by a debugger, a first thread of the application, by, determining that a first instruction in the first thread specifies to store a first value at a first memory address, setting a current value stored in the first memory address as an old value for the first memory address in a container for the first thread, executing the first instruction to store the first value at the first memory address, and setting the first value as a new value for the first memory address in the container for the first thread. The debugger then executes a second thread of the application, by restoring old values in the containers for all other threads and restoring a new value for each memory address specified in a container for the second thread to its respective memory address.

Claims

1. A method to discover errors in a parallel application comprising a plurality of threads executed on a computing system, comprising:

scanning, by a debugger, a program code of the parallel application;

setting a respective breakpoint in the program code of the parallel application at each;

(i) sync instruction in the program code, and (ii) instruction determined to stall execution of the program code, wherein the debugger is configured to switch execution from a first thread of the plurality of threads of the parallel application to a second thread of the plurality of threads of the parallel application upon encountering one of the breakpoints in the program code;

executing, by the debugger, the first thread comprising;

upon determining that a first instruction in the first thread specifies to store a first value at a first memory address;

setting a current value stored in the first memory address as an old value for the first memory address in a container for the first thread;

executing the first instruction to store the first value at the first memory address; and

setting the first value as a new value for the first memory address in the container for the first thread, wherein the container for the first thread specifies a new value and an old value for each of a plurality of memory addresses including the first memory address; and

executing, by the debugger, a second thread, comprising;

restoring each old value for the plurality of memory addresses in the container for the first thread to the respective memory address specified in the container for the first thread; and

restoring a new value for each of a plurality of memory addresses specified in a container for the second thread to the respective memory address specified in the container for the second thread.

2. The method of claim 1, wherein the old value for the first memory address is of a plurality of old values for the first memory address stored in the container for the first thread, the method further comprising:

prior to setting the current value stored in the first memory address as the old value for the memory address in the container for the first thread;

upon determining that the container for the first thread does not include a record for the first memory address, creating a record for the first memory address in the container for the first thread,

wherein the record for the first memory address specifies each of the plurality of old values for the first memory address and the new value for the first memory address subsequent to setting the first value as the new value for the first memory address in the container for the first thread and prior to executing the second thread.

3. The method of claim 1, further comprising:

upon determining that the first instruction is a sync instruction, clearing the values in the container for the first thread.

4. The method of claim 1, wherein the second thread is executed by the debugger upon further determining at least one of:

(i) that the debugger has executed a predefined number of instructions of the first thread, and (ii) that an amount of time the debugger has executed the first thread exceeds a predefined time threshold.

5. The method of claim 1, wherein executing the second Thread further comprises:

restoring, for each thread of the parallel application, each old value in a respective container of each thread to a respective memory address associated with each old value prior to restoring the new value for each memory address in the container for the second thread.

6. The method of claim 1, wherein executing the second thread further comprises:

upon determining that a second instruction in the second thread specifies to store a second value at a second memory address;

setting a current value stored in the second memory address as an old value for the second memory address in the container for the second thread;

executing the second instruction to store the second value at the second memory address; and

setting the second value as a new value for the second memory address in the container for the second thread.

7. The method of claim 1, further comprising:

upon determining that executing a second instruction in the second thread causes a weak consistency error, returning an indication of the weak consistency error.