Archive for March, 2016

Habits are the peripheral vision of the mind

March 24th, 2016 3 comments

Achieving a basic proficiency in a new skill requires an investment of conscious cognitive effort, i.e., thinking a lot. Students are constantly in the process of achieving basic proficiency in new skills and conclude that thinking is required for all intellectual activities (an incorrect assumption also held by many teachers).

To get past the conscious thinking stage lots of time has to be spent performing the skill. Repetition provides the opportunity for performance via conscious thought to migrate to subconscious performance (driving being a common example).

Real-time performance requires fluency, that is, being able to handle technical details without having to think about them. Thinking (i.e., conscious thought) is slow and requires lots of effort. It is best held in reserve for the important stuff.

To paraphrase Alfred Whitehead: “Software development advances by extending the number of important operations which we can perform without thinking about them.”

Somebody who has spent 100 hours or so (an hour or two a week for a year) learning to code has the same level of fluency as I have in communicating in a foreign language using a phrase book, or Google translate.

After a 1,000 hours of programming a person should be a very fluent coder.

It is said that becoming an expert requires 10,000 hours of practice. The kind of practice involved is deliberate practice, not unconscious use of what is already known. Becoming an expert requires learning lots of new things, not constantly applying what is already known. Old habits have to be broken and new ones acquired.

Programming is not Zen, although it contains elements that are. Why would a developer want to create a program without conscious thought (that is what scripts are for)?

I used to run ‘advanced’ programming courses for professional developers with 2+ years in industry. In many ways the material was a rerun of what they had learned at the start of their programming career. The difference was that this time around they could ignore the mechanics of writing code, now an ingrained habit, and concentrate on the higher level stuff. The course had to have advanced in its title because experienced developers would never sign up for an introductory course. Most of my one-on-one tutoring effort went on talking people out of bad habits they had picked up over time.

Perhaps live coding can be done with a Zen mind, probably why I don’t regard it as real programming (which I think requires some conscious thought).

Talking about details and high level material in the same breath is what beginners do because they have not yet learned to tell the two apart and be able to ignore one of them.

Like life, programs are mostly built from sequences of commonly occurring patterns. Our minds have evolved to subconsciously detect and take advantage of patterns. Programmers don’t know what the common source code patterns are any more than a native speaker can specify the syntax rules of the language they speak.

cpu+FPGA: applications can soon have bespoke instructions

March 21st, 2016 2 comments

Compiler writers are always frustrated that the cpu they are currently targeting does not contain the one instruction that would enable them to generate really efficient code. If only it were possible to add new instructions to the cpu. Well, it looks like this will soon be possible; Intel have added an on chip FPGA to their Broadwell processor (available circa 2017).

Having custom instructions on a FPGA (they would be loaded at program startup) is not the same as having the instructions on the cpu itself, there will be communication overhead when the data operated on by the custom instruction get transferred back and forth between cpu/FPGA (being on-chip means this will be low). To make the exercise worthwhile the custom instruction has to do something that takes very many cycles on the cpu and either speeds it up or reduces the power consumed (the Catapult project at Microsoft has a rack of FPGA enhanced machines speeding up/reducing the power of matching search engine queries to documents).

A CPU+FPGA is like CPU+GPU, except that FPGAs are programmed at a much lower level, i.e., there is little in the way of abstraction between what the hardware does and what the coder sees.

Does the world need a FPGA attached to their cpu? Most don’t but there are probably a few customers who do, e.g., data centers with systems performing dedicated tasks and anybody into serious bit twiddling. Other considerations include Intel needing to add new bells and whistles to its product so that customers who have been trained over the years to buy the very latest product (which has the largest margins) stay on the buying treadmill. The FPGA is also a differentiator, not that Intel would ever think of AMD as a serious competitor.

Initially the obvious use case is libraries performing commonly occurring functionality. No, not matrix multiple and inverse, FPGA are predominantly integer operation units (there are approaches using non-standard floating-point formats that can be used if your FPGA unit does not have floating-point support).

From the compiler perspective the use case is spotting cpu intensive loops, where all the data can be held on the FPGA until processing is complete. Will there be enough of these loops to make it a worthwhile implementation target? I suspect not. But then I can see many PhDs being written on this topic and one of them could produce a viable implementation that bootstraps itself into one of the popular open source compilers.

Interpreters have to do a lot of housekeeping work. Perhaps programs written in Java or R could be executed on the FPGA that uses the cpu as a slave processor. It is claimed that most R programs spend their time in library functions that have been implemented in C and Fortran, but I’m seeing more and more code that appears to be all R. For some programs an R-machine implemented in hardware could produce orders of magnitude speed improvements.

The next generation of cryptocurrency proof-of-work algorithms are being designed to be memory intensive, so they cannot be efficiently implemented using ASIC-proof (this prevents mining being concentrated in a few groups who have built bespoke mining operations). The analysis I have seen is based on ‘conventional’ cpu and ASIC designs. A cpu+FPGA is a very different kind of beast and one that might require another round of cryptocurrency design.

These cpu+FPGA processors have the potential to dramatically upend existing approaches to structuring programs. Very interesting times ahead!

Tags: , , ,

A survey of opinions on the behavior of various C constructs

March 10th, 2016 No comments

The Cerberus project, researching C semantics, has written up the results of their survey of ‘expert’ C users (short version and long detailed version). I took part in the original survey and at times found myself having to second guess what the questioner was asking; the people involved were/are still learning how C works. Anyway, many of the replies provide interesting insights into current developer interpretation of the behavior of various C constructs (while many of the respondents were compiler writers, it looks like some of them were not C compiler writers).

Some of those working on the Cerberus project are proposing changes to the C standard based on issues they encountered while writing a formal specification for parts of C and are bolstering their argument, in part, using the results of their survey. In many ways the content of the C Standard was derived from a survey of those attending WG14 meetings (or rather x3j11 meetings back in the day).

I think there is zero probability that any of these proposed changes will make it into a revised C standard; none of the reasons are technical and include:

  • If it isn’t broken, don’t fix it. Lots of people have successfully implemented compilers based on the text of the standard, which is the purpose of the document. Where is the cost/benefit of changing the wording to enable a formal specification using one particular mathematical notation?
  • WG14 receives lots of requests for changes to the C Standard and has an implicit filtering process. If the person making the request thinks the change is important, they will:
    • put the effort into wording the proposal in the stylized form used for language change proposals (i.e., not intersperse changes in a long document discussing another matter),
    • be regular attendees of WG14 meetings, working with committee members on committee business and helping to navigate their proposals through the process (turning up to part of a meeting will see your proposal disappear as soon as you leave the building; the next WG14 meeting is in London during April).

It could be argued that having to attend many meetings around the world favors those working for large companies. In practice only a few large companies see any benefit in sending an employee to a standard’s meeting for a week to work on something that may be of long term benefit them (sometimes a hardware company who wants to make sure that C can be compiled efficiently to their processors).

The standard’s creation process is about stability (don’t break existing code; many years ago a company voted against a revision to the Cobol standard because they had lost the source code to one of their products and could not check whether the proposed updates would break this code) and broad appeal (not narrow interests).

Update: Herb Sutter’s C++ trip report gives an interesting overview of the process adopted by WG21.

Tags: , ,