On Wed, 28 Feb 2024 06:08:18 GMT, Joe Darcy <da...@openjdk.org> wrote:
>> A new paper >> >> "Accuracy of Mathematical Functions in Single, Double, Double Extended, and >> Quadruple Precision" >> by Brian Gladman, Vincenzo Innocente and Paul Zimmermann >> https://members.loria.fr/PZimmermann/papers/accuracy.pdf >> >> details the inputs with generate the worst-case observed errors in different >> math library implementations. The FDLIBM-related worst cases should be added >> to the test suite. > > Joe Darcy has updated the pull request incrementally with one additional > commit since the last revision: > > Respond to review feedback, adjust worst-case bounds. test/jdk/java/lang/Math/WorstCaseTests.java line 35: > 33: > 34: /** > 35: * This test containst two distinct kinds of worst-case inputs: Suggestion: * This test contains two distinct kinds of worst-case inputs: test/jdk/java/lang/Math/WorstCaseTests.java line 39: > 37: * 1) Exact numerical results that are nearly half-way between > 38: * representable numbers or very close to a representable > 39: * number. (Half-way caess are hardest for round to nearest even; Suggestion: * number. (Half-way cases are hardest for round to nearest even; test/jdk/java/lang/Math/WorstCaseTests.java line 43: > 41: * roundings.) > 42: * > 43: * 2) Worst-case errors as observed emprically across different Suggestion: * 2) Worst-case errors as observed empirically across different test/jdk/java/lang/Math/WorstCaseTests.java line 46: > 44: * implementations that are not correctly rounded. > 45: * > 46: * For the first categpory, the "Table Maker's Dilemma" results from Suggestion: * For the first category, the "Table Maker's Dilemma" results from test/jdk/java/lang/Math/WorstCaseTests.java line 48: > 46: * For the first categpory, the "Table Maker's Dilemma" results from > 47: * Jean-Michel Muller and Vincent Lefèvre, are used. > 48: * See http://perso.ens-lyon.fr/jean-michel.muller/TMD.html for original Suggestion: * See https://perso.ens-lyon.fr/jean-michel.muller/TMD.html for original test/jdk/java/lang/Math/WorstCaseTests.java line 50: > 48: * See http://perso.ens-lyon.fr/jean-michel.muller/TMD.html for original > 49: * test vectors from 2000 and see > 50: * http://perso.ens-lyon.fr/jean-michel.muller/TMDworstcases.pdf with Suggestion: * https://perso.ens-lyon.fr/jean-michel.muller/TMDworstcases.pdf with test/jdk/java/lang/Math/WorstCaseTests.java line 245: > 243: {-0x1.842d8ec8f752fp+21, -0x1.6ce864edeaffdp-1}, > 244: {-0x1.07e4c92b5349dp+4, +0x1.6a096375ffb23p-1}, > 245: // {-0x1.13a5ccd87c9bbp+1008, -0x1.27b3964185d8dp-1}, // > check -- need +/- High-precision computations confirm `-0x1.27b3964185d8dp-1` ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239002 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239088 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239189 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239253 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239477 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239558 PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1509239711