]> git.baikalelectronics.ru Git - kernel.git/commit
math-emu: fix floating-point to integer overflow detection
authorJoseph Myers <joseph@codesourcery.com>
Mon, 4 Nov 2013 16:54:15 +0000 (16:54 +0000)
committerScott Wood <scottwood@freescale.com>
Wed, 8 Jan 2014 00:36:24 +0000 (18:36 -0600)
commit1b7d5ab7b0ce200596a1fbd0397119ce40b03211
tree7af31abd819a314106ebc88614a7b0d315b7d87f
parent6d61c4a83b9d8190dc57ef985a39077eb4e31b02
math-emu: fix floating-point to integer overflow detection

On overflow, the math-emu macro _FP_TO_INT_ROUND tries to saturate its
result (subject to the value of rsigned specifying the desired
overflow semantics).  However, if the rounding step has the effect of
increasing the exponent so as to cause overflow (if the rounded result
is 1 larger than the largest positive value with the given number of
bits, allowing for signedness), the overflow does not get detected,
meaning that for unsigned results 0 is produced instead of the maximum
unsigned integer with the give number of bits, without an exception
being raised for overflow, and that for signed results the minimum
(negative) value is produced instead of the maximum (positive) value,
again without an exception.  This patch makes the code check for
rounding increasing the exponent and adjusts the exponent value as
needed for the overflow check.

Signed-off-by: Joseph Myers <joseph@codesourcery.com>
Signed-off-by: Scott Wood <scottwood@freescale.com>
include/math-emu/op-common.h