On 15 May 2011 15:13, Aurelien Jarno <aurel...@aurel32.net> wrote: > Now that target-i386 uses softfloat, floatx80 is always available and > there is no need anymore to have code handling both float64 and floax80. > > Signed-off-by: Aurelien Jarno <aurel...@aurel32.net>
This patch is OK in terms of how it leaves the code, but I think some parts of it are out of sequence with the rest of the patchset. For instance: > -#ifdef FLOATX80 > -#define USE_X86LDOUBLE > -#endif We've already removed the FLOATX80 define in a previous patch, so if we don't delete the x86 use of it until this patch then the behaviour will briefly flip-flop as you go through the patch stack, which could be bad for bisection. > -#if defined(CONFIG_SOFTFLOAT) > -# define floatx_lg2 make_floatx80( 0x3ffd, 0x9a209a84fbcff799LL ) > -# define floatx_l2e make_floatx80( 0x3fff, 0xb8aa3b295c17f0bcLL ) > -# define floatx_l2t make_floatx80( 0x4000, 0xd49a784bcd1b8afeLL ) > -#else > -# define floatx_lg2 (0.30102999566398119523L) > -# define floatx_l2e (1.44269504088896340739L) > -# define floatx_l2t (3.32192809488736234781L) > -#endif Similarly, this #ifdeffery should have gone away when we took out CONFIG_SOFTFLOAT, not later. (Also the patch was a bit of a pig to review because it combines several distinct mostly-mechanical transformations.) -- PMM
