Computer Architecture Seminar
Download as PPTX, PDF0 likes466 views
The document discusses eliminating redundant computation through data-triggered threads (DTT). DTT proposes spawning a separate thread to handle redundant loads caused by silent stores, which are stores that do not change memory contents. This avoids recomputing values for redundant loads. The programming model places redundant code in a separate thread triggered by a store. The architecture adds hardware tables to manage thread status and queues. The ISA is modified with new instructions like tstore and tspawn to generate and spawn threads.
![Eliminating Redundant ComputationSilent Store:Amemory store operation that does not change the contents at that location20-68% of all stores are silent [Lepakand Lipasti]How about eliminating the entire stream of computation surrounding a silent store!](https://image.slidesharecdn.com/namancarg-111018104901-phpapp01/85/Computer-Architecture-Seminar-2-320.jpg)
Computer Architecture Seminar
- 1. Data-Triggered ThreadsEliminating Redundant Computation(HPCA 2011)Hung-Wei Tseng and Dean M. TullsenDepartment of Computer Science and Engineering University of California, San DiegoSeminar by: Naman Kumar for http://carg.uwaterloo.ca
- 2. Eliminating Redundant ComputationSilent Store:Amemory store operation that does not change the contents at that location20-68% of all stores are silent [Lepakand Lipasti]How about eliminating the entire stream of computation surrounding a silent store!
- 3. Eliminating Redundant ComputationRedundant loads: silent stores result in redundant loads (last time this load loaded this address, it fetched the same value) SPEC2000 C:78% of all loads are redundant50% of all instructions depend on redundant loads
- 5. DTT: ImplementationThe Programming ModelPlace redundant computation in a separate thread:Thread is restartableThread may be aborted/restarted multiple timesThread management is through architectural changes.Easy to verify data races as thread life is between time between triggering store and main thread join point.
- 6. DTT: ImplementationThe Programming ModelTrigger is placed in data section, not code section
- 7. DTT: ImplementationThe Programming ModelMain Thread
- 8. DTT: ImplementationThe Programming ModelSupport thread
- 9. DTT: ImplementationArchitectural SupportFollowing tables are all implemented in hardwareThread registry (table) Thread Queue (table)Thread Status Table (table) PC
- 10. DTT: ImplementationArchitectural SupportISA modificationststore – generate thread when mem modified is not silenttspawn – spawn the thread using thread registrytreturn– finish execution of the current threadtcancel – terminate a running thread
Editor's Notes
- #2: Memoization and other techniques save on memory access. This technique proposes a solution to save on accesses and the computation involving the data from these access.
- #4: eg: sum of all nodes in a 100 node linked-list. Each node has to be accessed when say, only 2 have changed. That’s 98 redundant loads.
- #5: If value of SP calculated is diff from what is in memory, then a support thread (S) will be spawned to calculate and B. Main thread will skip code section B since data has already been calculated. Instructions for B will be left as is because support thread may have failed to spawn; skipping the thread, code will be executed by the main thread.
- #6: Programmers implement this with C pragma constructs
- #7: Every time the variable is WRITTEN to, the associated DTThread is executed
- #8: If programmer has a reason to suspect that the thread may crash/be aborted, he can place the #cancel pragma. This will ensure that only the main thread executes this block. Support thread will not be registered.
- #9: This function is triggered in a new thread (support thread) when control reaches “#block xxx”
- #10: Start PC: PC of the skippable code in the main thread.Destination PC denotes the end of the skippable region.Post skip PC is address after the region is skipped.
- #13: CMP only