|
From: Florian K. <br...@ac...> - 2013-06-03 18:41:32
|
Hi, we currently have two rounding modes in VEX: IRRoundingMode to specify the semantics of conversions between - binary floating point values and integer values and - binary floating point values of different precision and IRRoundingModeDFP to specify the semantics of conversions between - decimal floating point values and integer values and - decimal floating point values of different precision To complete the support for DFP on s390 we need to add a few IROps to convert between binary floating point and decimal floating point values. None of the existing rounding modes can be used for those and adding a 3rd rounding mode feels wrong. I'm proposing that we eliminate IRRoundingModeDFP on grounds that the rounding mode is an abstraction and is independent of a particular format. The attached patch shows the changes to libvex_ir.h. Basically, it merges the additional rounding modes from IRRoundingModeDFP into IRRoundongMode. Note, that the encoding of the rounding mode values is preserved so the change is completely transparent. Only the s390 port needs to be adapted. I regtested the patch on PPC-64 with no new regressions. Any objections? Florian |