Subnormal: Difference between revisions
Jump to navigation
Jump to search
imported>Niek Sanders (Initial content.) |
imported>Joe Quick m (subpages) |
||
| Line 1: | Line 1: | ||
{{subpages}} | |||
'''Subnormal''' [[floating point]] values (also known as '''denormal''') are too small to be normalized yet can still be represented [http://www.intel.com/software/products/compilers/docs/fmac/doc_files/source/extfile/fpops_for/common/fpops_denormal_num.htm]. Depending on the specific computation, they may provide additional, useful precision. However, managing these values typically causes major slowdowns in a [[Central processing unit|CPU]]'s [[floating point unit]] [http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/pentium4_hh/advice4_hh/found_x87_denormal_numbers_.htm]. | '''Subnormal''' [[floating point]] values (also known as '''denormal''') are too small to be normalized yet can still be represented [http://www.intel.com/software/products/compilers/docs/fmac/doc_files/source/extfile/fpops_for/common/fpops_denormal_num.htm]. Depending on the specific computation, they may provide additional, useful precision. However, managing these values typically causes major slowdowns in a [[Central processing unit|CPU]]'s [[floating point unit]] [http://www.intel.com/software/products/documentation/vlin/mergedprojects/analyzer_ec/mergedprojects/reference_olh/pentium4_hh/advice4_hh/found_x87_denormal_numbers_.htm]. | ||
| Line 4: | Line 5: | ||
For some CPU architectures, it is possible to control the treatment of subnormals. For instance, these values can automatically be rounded to zero. | For some CPU architectures, it is possible to control the treatment of subnormals. For instance, these values can automatically be rounded to zero. | ||
Latest revision as of 20:04, 20 December 2007
Subnormal floating point values (also known as denormal) are too small to be normalized yet can still be represented [1]. Depending on the specific computation, they may provide additional, useful precision. However, managing these values typically causes major slowdowns in a CPU's floating point unit [2].
For the Intel x86 architecture, floats between 1.17549429E-38 and 1.40129846E-45 are denormalized while for doubles the range is 2.2250738585072008D-308 and 4.94065645841246544D-324. [3]
For some CPU architectures, it is possible to control the treatment of subnormals. For instance, these values can automatically be rounded to zero.