diff options
| author | Richard Henderson <rth@redhat.com> | 2007-04-24 08:28:21 -0700 |
|---|---|---|
| committer | Richard Henderson <rth@gcc.gnu.org> | 2007-04-24 08:28:21 -0700 |
| commit | 5fb54b91980565f03b96cadfc55dff8ddf5236ca (patch) | |
| tree | 02de3c14605545c31467f9847eec2443b8c6117d /gcc/libgcc2.h | |
| parent | 688518053c61ccb6d52aa647d6ba6b04fdc1b04c (diff) | |
| download | gcc-5fb54b91980565f03b96cadfc55dff8ddf5236ca.zip gcc-5fb54b91980565f03b96cadfc55dff8ddf5236ca.tar.gz gcc-5fb54b91980565f03b96cadfc55dff8ddf5236ca.tar.bz2 | |
libgcc2.h (AVOID_FP_TYPE_CONVERSION): Rename from IS_IBM_EXTENDED.
* libgcc2.h (AVOID_FP_TYPE_CONVERSION): Rename from
IS_IBM_EXTENDED. Also define in terms of WIDEST_HARDWARE_FP_SIZE.
* libgcc2.c (__floatdisf): Avoid double-word arithmetic when
looking for non-zero bits shifted out. Avoid a recursive call
when constructing the scalar.
(__floatundisf): Likewise.
From-SVN: r124106
Diffstat (limited to 'gcc/libgcc2.h')
| -rw-r--r-- | gcc/libgcc2.h | 10 |
1 files changed, 8 insertions, 2 deletions
diff --git a/gcc/libgcc2.h b/gcc/libgcc2.h index b1c749f..c6084dc 100644 --- a/gcc/libgcc2.h +++ b/gcc/libgcc2.h @@ -115,10 +115,16 @@ extern void __eprintf (const char *, const char *, unsigned int, const char *) /* FIXME: This #ifdef probably should be removed, ie. enable the test for mips too. */ +/* Don't use IBM Extended Double TFmode for TI->SF calculations. + The conversion from long double to float suffers from double + rounding, because we convert via double. In other cases, going + through the software fp routines is much slower than the fallback. */ #ifdef __powerpc__ -#define IS_IBM_EXTENDED(SIZE) (SIZE == 106) +#define AVOID_FP_TYPE_CONVERSION(SIZE) (SIZE == 106) +#elif defined(WIDEST_HARDWARE_FP_SIZE) +#define AVOID_FP_TYPE_CONVERSION(SIZE) (SIZE > WIDEST_HARDWARE_FP_SIZE) #else -#define IS_IBM_EXTENDED(SIZE) 0 +#define AVOID_FP_TYPE_CONVERSION(SIZE) 0 #endif /* In the first part of this file, we are interfacing to calls generated |