by Pavel Panchekha with Ian Briggs and Yash Lad on Jun 3, 2024 | Tags: accuracy, approximation, awards, correctness, optimization, type systems
Numerical code needs to carefully balance accuracy and performance. A new DSL, MegaLibm, makes this easier by checking for numerical correctness and offering flexible compilation to efficient code. By reconceptualizing classic numerical techniques like polynomial approximation and range reduction as type-theoric operations like casts and binding operators, MegaLibm makes numerical code composable, and its ability to plug in mathematical an numerical tools (like Sollya) makes it easier and safer to write such code.
Read more...
by Santosh Nagarakatte and Jay Lim on Apr 28, 2022 | Tags: awards, correct math libraries, optimization, runtimes
To create a single polynomial approximation that produces correct results for multiple representations and rounding modes, we propose to generate a polynomial that generates the correctly rounded result of f(x) using the non-standard round-to-odd rounding mode with 2 additional precision bits compared to the largest floating point representation that we wish to support. We provide a proof that this method produces correctly rounded results for multiple representations and for all the standard rounding modes. More detailed explanation of our approach can be found in our POPL 2022 distinguished paper.
Read more...
by Jay Lim and Santosh Nagarakatte on Aug 26, 2021 | Tags: awards, correct math libraries, optimization, program synthesis, runtimes
Everyone uses math libraries. Surprisingly, mainstream math libraries do not produce correct results for several thousands of inputs. Developers are seldom aware of them, which affects reproducibility and portability. We describe our new method for synthesizing elementary functions that produces correct results for all inputs but is also significantly faster than the state of the art libraries, which have been optimized for decades.
Read more...
by Alejandro Aguirre on Jul 15, 2021 | Tags: awards, formal verification, probabilistic programming
Sensitivity properties study how the output of a program changes when we make small changes to its input. How can we reason about sensitivity of programs that have probabilistic behavior?
Read more...
by Craig Disselkoen and Marco Vassena on Apr 21, 2021 | Tags: awards, computer security, program analysis, Spectre
Defending cryptographic code from Spectre attacks is difficult. Blade is a fully automatic approach to eliminate speculative leaks provably and efficiently.
Read more...
