we should define semantics for languages, not for tm tim harris (msr cambridge)

We should define semantics for languages, not for TM

Tim Harris (MSR Cambridge)

The main argument

Language implementation

ProgramThreads,atomic blocks,retry, OrElse

TM

StartTx,CommitTx,ReadTx, WriteTx

H/W

Read, Write, CAS

We should focus on defining this programmer-visible interface, rather than the internal interface used by the language implementation

An analogy

Language implementation

Program Garbage collected“infinite” memory

GC

Low-level, broad,platform-specific API,no canonical def.

H/W

How: strong semantics for race-free programs

Strong semantics: simple interleaved model of multi-threaded execution

T

1 2 3 4 5

Thread 4 in an atomic block

Data race: concurrent accesses to the same location, at least one a write

Race-free: no data races (under strong semantics)

Write(x)

Write(x)

Implementation 1: “classical” atomic blocks on TM

Language implementation

ProgramThreads,atomic blocks,retry, OrElse

Strong TM

Simple transformation

Lazy update, opacity,ordering guarantees...

H/W

Language implementation

ProgramThreads,atomic blocks,retry, OrElse

StartTx, CommitTx,ValidateTx,ReadTx(addr)->val,WriteTx(addr, val)

Implementation 2: very weak TM

Very weak STM

• No:– Guarantee that reads see a

consistent view of memory– Guarantee that reads/writes don’t

“spill over” to adjacent memory– Detection of tx/non-tx conflicts– Publication/privatization safety

• Yes:– Strict serializability in a tx-only

workload

Sandboxing for

zombies

Isolation of tx via MMU

Program analyses

GC support

H/W

Implementation 3: lock inference

Language implementation

ProgramThreads,atomic blocks,retry, OrElse

LocksLock, unlock

Lock inference analysis

H/W

Define semantics for languages, not for TM

• We should focus on the interface between the language and the programmer– Define it without talking about tx– Permit a range of tx and non-tx

implementations

• In a high-perf implementation, it is hard to cleanly separate the TM from other components– Interface is likely to be broad, platform-specific,

and may appear weaker than the language constructs