Kahan's graph theory compiler
Today (Feb 26, 2020) at Bebop, I presented a very rough draft of a talk that I planned to give in various places in the Spring and Summer of 2020. Velvel Kahal was also there as usual. My title was "Sparse Matrices Beyond Solvers: Graphs, Machine Learning, and Biology" but the bulk of the talk was about GraphBLAS and how it does effectively anything you need with graphs, including direction optimization (as exploited by GraphBLAST).
Velvel asked a lot of questions as usual. Uncharacteristically of him, he looked satisfied by my answers today. At the end of my talk, he actually expressed this verbally by saying "I didn't expect to be entertained". I suppose this is how he shows that he approves something.
Then he said "I envy you" and told us story about his time when he was a young Assistant Professor. The story was about him hacking his IBM compiler back in 1960s, to be able to compile certain graph theory codes using Fortran. This worked once but then IBM upgraded its compiler in anticipation of its next processor architecture. In the process, they made it impossible to modify the compiler. When Kahan complained, IBM told him that the compiler was proprietary and he shouldn't ever expect to be able to modify it. Kahan's complaint went all to the Chancellor. I don't remember the exact wording Kahan used but the Chancellor said something along the lines of "graph theory is a useless obsession of Hungarian mathematicians" and did not support Kahan in his argument with IBM.
Then I asked him whether that was the reason why he became a numerical analyst. He beautifully smiled and said "partly". I must have known Kahan for a decade by then (I started attending Bebop meetings around 2010), and this must be the first time I saw him coming close to a full blown laugh. We obviously know that his answer is at best partly true because he was already doing groundbreaking work in numerical analysis by then.
Velvel asked a lot of questions as usual. Uncharacteristically of him, he looked satisfied by my answers today. At the end of my talk, he actually expressed this verbally by saying "I didn't expect to be entertained". I suppose this is how he shows that he approves something.
Then he said "I envy you" and told us story about his time when he was a young Assistant Professor. The story was about him hacking his IBM compiler back in 1960s, to be able to compile certain graph theory codes using Fortran. This worked once but then IBM upgraded its compiler in anticipation of its next processor architecture. In the process, they made it impossible to modify the compiler. When Kahan complained, IBM told him that the compiler was proprietary and he shouldn't ever expect to be able to modify it. Kahan's complaint went all to the Chancellor. I don't remember the exact wording Kahan used but the Chancellor said something along the lines of "graph theory is a useless obsession of Hungarian mathematicians" and did not support Kahan in his argument with IBM.
Then I asked him whether that was the reason why he became a numerical analyst. He beautifully smiled and said "partly". I must have known Kahan for a decade by then (I started attending Bebop meetings around 2010), and this must be the first time I saw him coming close to a full blown laugh. We obviously know that his answer is at best partly true because he was already doing groundbreaking work in numerical analysis by then.
Comments
Post a Comment