International Business Machines Corporation

Globalfoundries Inc.

Xiaotong Zhuang

Kai-Ting Amy Wang

Alexandre E Eichenberger

Kathryn Mary O'Brien

Thomas M Gooding

John Kevin O'Brien

Raul Esteban Silvera

Mauricio J Serrano

Peng Wu

John K P O'Brien

Alan G Gara

Harold Wade Cain, Iii

Martin Ohmacht

Mark Edwin Giampapa

Jong-Deok Choi

Mark Yamashita

Dong Chen

Burkhard Steinmacher-Burow

Matthias Augustin Blumrich

Lakshminarayanan Renganarayana

Christoph Von Praun

Luis H Ceze

Yangchun Luo

Phlip Heidelberger

In a multiprocessor system, a conflict checking mechanism is implemented in the L2 cache memory. Different versions of speculative writes are maintained in different ways of the cache. A record of speculative writes is maintained in the cache directory. Conflict checking occurs as part of directory lookup. Speculative versions that do not conflict are aggregated into an aggregated version in a different way of the cache. Speculative memory access requests do not go to main memory.

Cache as point of coherence in multiprocessor system

In an embodiment, if a self thread has more than one conflict, a transaction of the self thread is aborted and restarted. If the self thread has only one conflict and an enemy thread of the self thread has more than one conflict, the transaction of the self thread is committed. If the self thread only conflicts with the enemy thread and the enemy thread only conflicts with the self thread and the self thread has a key that has a higher priority than a key of the enemy thread, the transaction of the self thread is committed. If the self thread only conflicts with the enemy thread, the enemy thread only conflicts with the self thread, and the self thread has a key that has a lower priority than the key of the enemy thread, the transaction of the self thread is aborted.

Speculative thread execution with hardware transactional memory

A mechanism is provided for path-sensitive analysis for reducing rollback overheads. The mechanism receives, in a compiler, program code to be compiled to form compiled code. The mechanism divides the code into basic blocks. The mechanism then determines a restore register set for each of the one or more basic blocks to form one or more restore register sets. The mechanism then stores the one or more register sets such that responsive to a rollback during execution of the compiled code. A rollback routine identifies a restore register set from the one or more restore register sets and restores registers identified in the identified restore register set.

Path-sensitive analysis for reducing rollback overheads

In an embodiment, asynchronous conflict events are received during a previous rollback period. Each of the asynchronous conflict events represent conflicts encountered by speculative execution of a first plurality of work units and may be received out-of-order. During a current rollback period, a first work unit is determined whose speculative execution raised one of the asynchronous conflict events, and the first work unit is older than all other of the first plurality of work units. A second plurality of work units are determined, whose ages are equal to or older than the first work unit, wherein each of the second plurality of work units are assigned to respective executing threads. Rollbacks of the second plurality of work units are performed. After the rollbacks of the second plurality of work units are performed, speculative executions of the second plurality of work units are initiated in age order, from oldest to youngest.

Speculative thread execution and asynchronous conflict events

Mechanisms for building approximate data dependences using a moving look-back window are provided. The mechanisms track dependence information for memory accesses over iterations of execution of a portion of code. The mechanisms receive a memory access of an iteration of the portion of code, the memory access having an address for access the memory and an access type indicating at least one of a read or a write access type. An entry in a moving look-back window data structure is generated corresponding to a memory location accessed by the memory access. The entry comprises at least an identification of the address, the access type, and an iteration number corresponding to the iteration of the memory access. The moving look-back window data structure is utilized to determine dependence information for memory accesses over a plurality of iterations of the portion of code.

Building approximate data dependences with a moving window

A process for check pointing in speculative execution frameworks, identifies calls to a set of setjmp/longjmp instructions to form identified calls to setjmp/longjmp, determines a control flow path between a call to a setjmp and a longjmp pair of instructions in the identified calls to setjmp/longjmp and replaces calls to the setjmp/longjmp pair of instructions with calls to an improved_setjmp and improved_longjmp instruction pair. The process creates a context data structure in memory, computes a non-volatile save/restore set and replaces the call to improved_setjmp of the setjmp/longjmp pair of instructions with instructions to save all required non-volatile and special purpose registers and replaces a call to improved_longjmp of the setjmp/longjmp pair of instructions with instructions to restore all required non-volatile and special purpose registers and to branch to an instruction immediately following a block of code containing the call to improved_setjmp.

setjmp/longjmp for speculative execution frameworks

Mechanisms are provided for data placement optimization during runtime of a computer program. The mechanisms detect cache misses in a cache of the data processing system and collect cache miss information for objects of the computer program. Data context information is generated for an object in an object access sequence of the computer program. The data context information identifies one or more additional objects accessed as part of the object access sequence in association with the object. The cache miss information is correlated with the data context information of the object. Data placement optimization is performed on the object, in the object access sequence, with which the cache miss information is associated. The data placement optimization places connected objects in the object access sequence in close proximity to each other in a memory structure of the data processing system.

Data placement optimization using data context collected during garbage collection

An optimizing compiler device, a method, a computer program product which are capable of performing parallelization of irregular reductions. The method for performing parallelization of irregular reductions includes receiving, at a compiler, a program and selecting, at compile time, at least one unit of work (UW) from the program, each UW configured to operate on at least one reduction operation, where at least one reduction operation in the UW operates on a reduction variable whose address is determinable when running the program at a run-time. At run time, for each successive current UW, a list of reduction operations accessed by that unit of work is recorded. Further, it is determined at run time whether reduction operations accessed by a current UW conflict with any reduction operations recorded as having been accessed by prior selected units of work, and assigning the unit of work as a conflict free unit of work (CFUW) when no conflicts are found. Finally, there is scheduled, for parallel run-time operation, at least two or more processing threads to process a respective the at least two or more assigned CFUWs.

Parallelization of irregular reductions via parallel building and exploitation of conflict-free units of work at runtime

Mechanisms are provided for tracking dependencies of threads in a multi-threaded computer program execution. The mechanisms detect a dependency of a first thread's execution on results of a second thread's execution in an execution flow of the multi-threaded computer program. The mechanisms further store, in a hardware thread dependency vector storage associated with the first thread's execution, an identifier of the dependency by setting at least one bit in the hardware thread dependency vector storage corresponding to the second thread. Moreover, the mechanisms schedule tasks performed by the multi-threaded computer program based on the hardware thread dependency vector storage to minimize squashing of threads.

Tracking and detecting thread dependencies using speculative versioning cache

A runtime dependence-aware scheduling of dependent iterations mechanism is provided. Computation is performed for one or more iterations of computer executable code by a main thread. Dependence information is determined for a plurality of memory accesses within the computer executable code using modified executable code using a set of dependence threads. Using the dependence information, a determination is made as to whether a subset of a set of uncompleted iterations in the plurality of iterations is capable of being executed ahead-of-time by the one or more available threads in the data processing system. If the subset of the set of uncompleted iterations in the plurality of iterations is capable of being executed ahead-of-time, the main thread is signaled to skip the subset of the set of uncompleted iterations and the set of assist threads is signaled to execute the subset of the set of uncompleted iterations.

Runtime dependence-aware scheduling using assist thread

Thread speculative execution and asynchronous conflict

Mechanisms are provided for controlling version pressure on a speculative versioning cache. Raw version pressure data is collected based on one or more threads accessing cache lines of the speculative versioning cache. One or more statistical measures of version pressure are generated based on the collected raw version pressure data. A determination is made as to whether one or more modifications to an operation of a data processing system are to be performed based on the one or more statistical measures of version pressure, the one or more modifications affecting version pressure exerted on the speculative versioning cache. An operation of the data processing system is modified based on the one or more determined modifications, in response to a determination that one or more modifications to the operation of the data processing system are to be performed, to affect the version pressure exerted on the speculative versioning cache.

Version pressure feedback mechanisms for speculative versioning caches

Computer implemented method, system and computer usable program code for profiling the execution of an application that is both space- and time-efficient and highly accurate. A computer implemented method for profiling the execution of an application includes sampling execution characteristics of the application at a plurality of sampling points to provide samples, and deriving a calling context of the samples. The application is continuously executed between sampling points while additional profiling data is gathered.

Growing community of inventors

White Plains, NY, United States of America

Xiaotong Zhuang