IEEE 754R minutes from 11 November, 2003

Attendance

Note taker: Jeff Kidder, Intel Corporation

Testing

Have a good start from John Hauser (and some tests) Jeff Kidder offered to help support development of a reference implementation. What language? C++, C99, Java, GCC. The source should live for a long time. Should separate the numeric content from the language issues. Numeric tests should be defined as conditions in a file that gets swizzled into whatever test environment. Mike's tests are this way, as are Coonen.

gcc: compiles in Linux and many environments, extendable
java: compiling against a virtual machine

Should we specify things we cannot test? Issues: How do we test extended types. Would be easier if we know the parameters of the extended type.

Issue: Testing expression evaluation Need to test compile time as well as run time.

We should focus on the changes. A portable edifice is not likely possible even though we are trying to specify ways to direct languages to make these features portable. Look at UCBTest as an example.

Should figure out how to represent the expression evaluation in a portable file format similar to the numeric tests. Something that looks at expressions and churns out code in some language to run and check.

We can't be all things to all people. Our job is to come up with a format for test vectors and an interpreter for some environment. Should the test development happen in the whole group or in a subgroup. Could we have a gcc reference implementation?

  1. "the easy part": numeric tests
  2. "are the features right": what is a test, what is passing
  3. "reference implementation":

The languages use external test vendors and examples the illustrate how things are done. We could use examples in some language that people could use for comparison. Paranoia is another approach -- but harder to extend.

You need to tempt the compiler to make a mistake. "You can put your effort into writing the test or grading the answers." It is very had to get this stuff right. Dan Zuras will coordinate the next step. Please send him your email address if you want to be involved in the testing effort. Jeff Kidder will look at Hauser's C (SoftFloat) reference for 754r.

Concurrency

Ivan Godard's Presentation
Modes:
Speculation: example ((x > 0.0) ? sqrt(x) : 0.0)
Ivan will look at David's expression proposal and respond with how it does or doesn't address his presentation.

Jim proposed that people sign up as sponsors for one the various topics. Form a subcommittee of those interested in the topic. Arrange a time to present to the group. After the discussion their will be a resolution or vote to continue in subcommittee or not.

Dan wants to put together an agenda to meet the goal of getting a vote in 12/04, we need to try to get the standard buttoned up by 9/04. David, Jim, and Dan will attempt to order the issues list in some reasonable order. Jim will propose a process to allow more parallel work.

754 | revision | FAQ | references