jzmaddock

>Does the Halley iterate succeed? In general I find the Halley iterate has a more sane basin of convergence and moreover it can degrade gracefully to a newton iterate. It...

I've pulled some low-hanging fruit (files that shouldn't have been in there in the first place as they're too large) and chopped it down to 104Mb. Better but still not...

@NAThompson by default the Jamfile that builds the Math TR1 library uses PCH's. It can be disabled by building with `b2 pch=off`. On a practical note, I'm not sure that...

Not so fast guys - the comparison to a literal zero is a red-herring here, and genuinely what we want to do anyway as: `x >= 0` is orders of...

zero is often handled as a special case yes (though not as thoroughly as I'd hoped, needs investigating more). However, for types like mpfr_float, all comparisons to int go through...

Changing the code to measure > 2 time I see the same results: ``` Benchmark Time CPU Iterations ----------------------------------------------------------------------------------------------- BM_equal_int 120 ns 119 ns 7466667 BM_equal_int 117 ns 114 ns...

We need to look at this - when that function was written (and the library started), there was negligible performance difference between double and the old x87 long double, so...

>If your thought is to use std::lgamma No there's too much variation between platforms IMO.

A related issue for the whittaker shannon test: ``` Error at libs/math/test/whittaker_shannon_test.cpp:test_bump:117: Mollified relative error in __float128 precision is 2.05e-29, which exceeds 1.93e-33, error/tol = 1.07e+04. Expected: +0.360803088275448657029385009164666528 = +0x1.71765d61c162afbf43d7d05ea90ap-2...

I don't think so: `__float128` *is a* `long double` so it should be supported and well defined. What well may happen though, is for any given precision, we get slightly...