- Aug 05, 2014
-
-
Raymond Toy authored
-
Raymond Toy authored
Netbsd apparently doesn't like this, so we're moving that option to the linux and darwin config files. Reported by Robert Swindells.
-
- Aug 04, 2014
-
-
Raymond Toy authored
-
Raymond Toy authored
Remove the variables i0 and i1 since they weren't used anywhere anyway. They weren't used in the original code either.
-
Raymond Toy authored
I think the offending code in each case is trying to extract the low word if x. I think it works because on a little-endian machine, *(unsigned*)&one is 0 since the low word is stored first. On a big-endian machine, *(unsigned*)&one is the high word which, when right shifted by 29, gives 1. That is added to the address of x to get the low word of x.
-
Raymond Toy authored
linux that they might be used uninitialized.
-
Raymond Toy authored
4.4.1 on Linux.
-
Raymond Toy authored
-
- Aug 03, 2014
-
-
Raymond Toy authored
processors.
-
Raymond Toy authored
instructions. This is completely untested.
-
Raymond Toy authored
since they're in GNUMakefile now.
-
Raymond Toy authored
* Add the appropriate gcc options to enable sse2 instructions. * Remove the fdlibm routines since they're in GNUMakefile now.
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
GNUMakefile now.
-
Raymond Toy authored
* GNUMakefile: * List all of the fdlibm files here instead of the individual configs because we use fdlibm everywhere now. * Config.x86_common: * Remove e_rem_pio2 and k_rem_pio2 which is in GNUMakefile now. * Config.x86_darwin: * Remove all of the fdlibm files since they're in GNUMakefile now. * Config.x86_linux: * Remove all of the fdlibm files since they're in GNUMakefile now.
-
Raymond Toy authored
update to include the rest of the fdlibm routines. With this change, all of the C code should now use sse2 instructions as needed, which means no more -ffloat-store, and the couple of test cases like tan.exceptions no longer fail because we signal execptions as expected.
-
- Aug 02, 2014
-
-
Raymond Toy authored
.
-
Raymond Toy authored
* fdlibm.h: * Declare fdlibm_atan * e_atan2.c: * Use fdlibm_atan instead of atan.
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
* fdlibm.h: * Declare fdlibm_log1p and __ieee754_log * e_acosh.c: * Use unions to get the parts of a double. * Use fdlibm_log1p instead of log1p. * e_atanh.c: * Use unions to get the parts of a double. * Use fdlibm_log1p instead of log1p. * s_asinh.c: * Rename from asinh to fdlibm_asinh. * Use unions to get the parts of a double. * Use fdlibm_log1p instead of log1p.
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
routine names. * fdlibm.h: * Correct the names for sin, cos, and tan; they have fdlibm_ prefix. * Declare fdlibm_expm1 and __ieee754_exp. * e_cosh.c: * Use unions * Call fdlibm_expm1 instead of expm1. * e_sinh.c: * Use unions * Call fdlibm_expm1 instead of expm1. * s_tanh.c: * Use unions * Call fdlibm_expm1 instead of expm1.
-
Raymond Toy authored
-
Raymond Toy authored
statements.
-
Raymond Toy authored
statements.
-
Raymond Toy authored
-
Raymond Toy authored
The compiler thinks k might not be initialized because the if on line 143 might not be true, leaving t uninitialized. We know that the condition is always true because |x| < -2^-28 in the previous line.
-
Raymond Toy authored
The compiler thinks t might not be initialized because the if on line 90 might not be true, leaving t uninitialized. We know that the condition is always true because |x| < -2^-27 in the previous line.
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-
Raymond Toy authored
-