diff options
Diffstat (limited to 'contrib/gdtoa/README')
| -rw-r--r-- | contrib/gdtoa/README | 37 |
1 files changed, 25 insertions, 12 deletions
diff --git a/contrib/gdtoa/README b/contrib/gdtoa/README index 6bece75..cf1947a 100644 --- a/contrib/gdtoa/README +++ b/contrib/gdtoa/README @@ -144,18 +144,33 @@ appropriate depends on the compiler; for this to work, it may be necessary to #include "float.h" or another system-dependent header file. -The values returned for NaNs may be signaling NaNs on some systems, -since the rules for distinguishing signaling from quiet NaNs are -system-dependent. You can easily fix this by suitably modifying the -ULto* routines in strtor*.c. +Source file strtodnrp.c gives a strtod that does not require 53-bit +rounding precision on systems (such as Intel IA32 systems) that may +suffer double rounding due to use of extended-precision registers. +For some conversions this variant of strtod is less efficient than the +one in strtod.c when the latter is run with 53-bit rounding precision. + +The values that the strto* routines return for NaNs are determined by +gd_qnan.h, which the makefile generates by running the program whose +source is qnan.c. Note that the rules for distinguishing signaling +from quiet NaNs are system-dependent. For cross-compilation, you need +to determine arith.h and gd_qnan.h suitably, e.g., using the +arithmetic of the target machine. C99's hexadecimal floating-point constants are recognized by the strto* routines (but this feature has not yet been heavily tested). Compiling with NO_HEX_FP #defined disables this feature. -The strto* routines do not (yet) recognize C99's NaN(...) syntax; the -strto* routines simply regard '(' as the first unprocessed input -character. +When compiled with -DINFNAN_CHECK, the strto* routines recognize C99's +NaN and Infinity syntax. Moreover, unless No_Hex_NaN is #defined, the +strto* routines also recognize C99's NaN(...) syntax: they accept +(case insensitively) strings of the form NaN(x), where x is a string +of hexadecimal digits and spaces; if there is only one string of +hexadecimal digits, it is taken for the fraction bits of the resulting +NaN; if there are two or more strings of hexadecimal digits, each +string is assigned to the next available sequence of 32-bit words of +fractions bits (starting with the most significant), right-aligned in +each sequence. For binary -> decimal conversions, I've provided just one family of helper routines: @@ -213,7 +228,7 @@ for intermediate quantities, and MALLOC (see gdtoaimp.h) is called only if the private pool does not suffice. 2000 is large enough that MALLOC is called only under very unusual circumstances (decimal -> binary conversion of very long strings) for conversions to and from double -precision. For systems with preemptivaly scheduled multiple threads +precision. For systems with preemptively scheduled multiple threads or for conversions to extended or quad, it may be appropriate to #define PRIVATE_MEM nnnn, where nnnn is a suitable value > 2000. For extended and quad precisions, -DPRIVATE_MEM=20000 is probably @@ -317,7 +332,5 @@ Compiling g__fmt.c, strtod.c, and strtodg.c with -DUSE_LOCALE causes the decimal-point character to be taken from the current locale; otherwise it is '.'. -Please send comments to - - David M. Gay - dmg@acm.org +Please send comments to David M. Gay (dmg at acm dot org, with " at " +changed at "@" and " dot " changed to "."). |
