AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-11-24 14:00:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

Simplify use of AVX instructions in internal math macros

Browse Source
This commit is contained in:
Ulrich Drepper 2012-01-28 11:19:06 -05:00
parent d73f93a461
commit 73139a7628
2 changed files with 24 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
ChangeLog
View File

@ -1,3 +1,7 @@
2012-01-28 Ulrich Drepper <drepper@gmail.com>
* sysdeps/x86_64/fpu/math_private.h: Simplify use of AVX instructions.
2012-01-19 Adhemerval Zanella <azanella@linux.vnet.ibm.com> 2012-01-19 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
* sysdeps/powerpc/powerpc32/a2/memcpy.S: Fix for when cache line * sysdeps/powerpc/powerpc32/a2/memcpy.S: Fix for when cache line

79
sysdeps/x86_64/fpu/math_private.h
View File

@ -21,8 +21,12 @@
#ifdef __AVX__ #ifdef __AVX__
# define MOVD "vmovd" # define MOVD "vmovd"
# define STMXCSR "vstmxcsr"
# define LDMXCSR "vldmxcsr"
#else #else
# define MOVD "movd" # define MOVD "movd"
# define STMXCSR "stmxcsr"
# define LDMXCSR "ldmxcsr"
#endif #endif
/* Direct movement of float into integer register. */ /* Direct movement of float into integer register. */
@ -173,7 +177,7 @@
#define libc_fegetround() \ #define libc_fegetround() \
({ \ ({ \
unsigned int mxcsr; \ unsigned int mxcsr; \
asm volatile ("stmxcsr %0" : "=m" (*&mxcsr)); \ asm volatile (STMXCSR " %0" : "=m" (*&mxcsr)); \
(mxcsr & 0x6000) >> 3; \ (mxcsr & 0x6000) >> 3; \
}) })
#undef libc_fegetroundf #undef libc_fegetroundf
@ -184,106 +188,63 @@
#define libc_fesetround(r) \ #define libc_fesetround(r) \
do { \ do { \
unsigned int mxcsr; \ unsigned int mxcsr; \
asm ("stmxcsr %0" : "=m" (*&mxcsr)); \ asm (STMXCSR " %0" : "=m" (*&mxcsr)); \
mxcsr = (mxcsr & ~0x6000) | ((r) << 3); \ mxcsr = (mxcsr & ~0x6000) | ((r) << 3); \
asm volatile ("ldmxcsr %0" : : "m" (*&mxcsr)); \ asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr)); \
} while (0) } while (0)
#undef libc_fesetroundf #undef libc_fesetroundf
#define libc_fesetroundf(r) libc_fesetround (r) #define libc_fesetroundf(r) libc_fesetround (r)
// #define libc_fesetroundl(r) (void) fesetround (r) // #define libc_fesetroundl(r) (void) fesetround (r)
#undef libc_feholdexcept #undef libc_feholdexcept
#ifdef __AVX__ #define libc_feholdexcept(e) \
# define libc_feholdexcept(e) \
do { \ do { \
unsigned int mxcsr; \ unsigned int mxcsr; \
asm ("vstmxcsr %0" : "=m" (*&mxcsr)); \ asm (STMXCSR " %0" : "=m" (*&mxcsr)); \
(e)->__mxcsr = mxcsr; \ (e)->__mxcsr = mxcsr; \
mxcsr = (mxcsr | 0x1f80) & ~0x3f; \ mxcsr = (mxcsr | 0x1f80) & ~0x3f; \
asm volatile ("vldmxcsr %0" : : "m" (*&mxcsr)); \ asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr)); \
} while (0) } while (0)
#else
# define libc_feholdexcept(e) \
do { \
unsigned int mxcsr; \
asm ("stmxcsr %0" : "=m" (*&mxcsr)); \
(e)->__mxcsr = mxcsr; \
mxcsr = (mxcsr | 0x1f80) & ~0x3f; \
asm volatile ("ldmxcsr %0" : : "m" (*&mxcsr)); \
} while (0)
#endif
#undef libc_feholdexceptf #undef libc_feholdexceptf
#define libc_feholdexceptf(e) libc_feholdexcept (e) #define libc_feholdexceptf(e) libc_feholdexcept (e)
// #define libc_feholdexceptl(e) (void) feholdexcept (e) // #define libc_feholdexceptl(e) (void) feholdexcept (e)
#undef libc_feholdexcept_setround #undef libc_feholdexcept_setround
#ifdef __AVX__ #define libc_feholdexcept_setround(e, r) \
# define libc_feholdexcept_setround(e, r) \
do { \ do { \
unsigned int mxcsr; \ unsigned int mxcsr; \
asm ("vstmxcsr %0" : "=m" (*&mxcsr)); \ asm (STMXCSR " %0" : "=m" (*&mxcsr)); \
(e)->__mxcsr = mxcsr; \ (e)->__mxcsr = mxcsr; \
mxcsr = ((mxcsr | 0x1f80) & ~0x603f) | ((r) << 3); \ mxcsr = ((mxcsr | 0x1f80) & ~0x603f) | ((r) << 3); \
asm volatile ("vldmxcsr %0" : : "m" (*&mxcsr)); \ asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr)); \
} while (0) } while (0)
#else
# define libc_feholdexcept_setround(e, r) \
do { \
unsigned int mxcsr; \
asm ("stmxcsr %0" : "=m" (*&mxcsr)); \
(e)->__mxcsr = mxcsr; \
mxcsr = ((mxcsr | 0x1f80) & ~0x603f) | ((r) << 3); \
asm volatile ("ldmxcsr %0" : : "m" (*&mxcsr)); \
} while (0)
#endif
#undef libc_feholdexcept_setroundf #undef libc_feholdexcept_setroundf
#define libc_feholdexcept_setroundf(e, r) libc_feholdexcept_setround (e, r) #define libc_feholdexcept_setroundf(e, r) libc_feholdexcept_setround (e, r)
// #define libc_feholdexcept_setroundl(e, r) ... // #define libc_feholdexcept_setroundl(e, r) ...
#undef libc_fetestexcept #undef libc_fetestexcept
#ifdef __AVX__ #define libc_fetestexcept(e) \
# define libc_fetestexcept(e) \ ({ unsigned int mxcsr; asm volatile (STMXCSR " %0" : "=m" (*&mxcsr)); \
({ unsigned int mxcsr; asm volatile ("vstmxcsr %0" : "=m" (*&mxcsr)); \
mxcsr & (e) & FE_ALL_EXCEPT; }) mxcsr & (e) & FE_ALL_EXCEPT; })
#else
# define libc_fetestexcept(e) \
({ unsigned int mxcsr; asm volatile ("stmxcsr %0" : "=m" (*&mxcsr)); \
mxcsr & (e) & FE_ALL_EXCEPT; })
#endif
#undef libc_fetestexceptf #undef libc_fetestexceptf
#define libc_fetestexceptf(e) libc_fetestexcept (e) #define libc_fetestexceptf(e) libc_fetestexcept (e)
// #define libc_fetestexceptl(e) fetestexcept (e) // #define libc_fetestexceptl(e) fetestexcept (e)
#undef libc_fesetenv #undef libc_fesetenv
#ifdef __AVX__ #define libc_fesetenv(e) \
# define libc_fesetenv(e) \ asm volatile (LDMXCSR " %0" : : "m" ((e)->__mxcsr))
asm volatile ("vldmxcsr %0" : : "m" ((e)->__mxcsr))
#else
# define libc_fesetenv(e) \
asm volatile ("ldmxcsr %0" : : "m" ((e)->__mxcsr))
#endif
#undef libc_fesetenvf #undef libc_fesetenvf
#define libc_fesetenvf(e) libc_fesetenv (e) #define libc_fesetenvf(e) libc_fesetenv (e)
// #define libc_fesetenvl(e) (void) fesetenv (e) // #define libc_fesetenvl(e) (void) fesetenv (e)
#undef libc_feupdateenv #undef libc_feupdateenv
#ifdef __AVX__ #define libc_feupdateenv(e) \
# define libc_feupdateenv(e) \
do { \ do { \
unsigned int mxcsr; \ unsigned int mxcsr; \
asm volatile ("vstmxcsr %0" : "=m" (*&mxcsr)); \ asm volatile (STMXCSR " %0" : "=m" (*&mxcsr)); \
asm volatile ("vldmxcsr %0" : : "m" ((e)->__mxcsr)); \ asm volatile (LDMXCSR " %0" : : "m" ((e)->__mxcsr)); \
__feraiseexcept (mxcsr & FE_ALL_EXCEPT); \ __feraiseexcept (mxcsr & FE_ALL_EXCEPT); \
} while (0) } while (0)
#else
# define libc_feupdateenv(e) \
do { \
unsigned int mxcsr; \
asm volatile ("stmxcsr %0" : "=m" (*&mxcsr)); \
asm volatile ("ldmxcsr %0" : : "m" ((e)->__mxcsr)); \
__feraiseexcept (mxcsr & FE_ALL_EXCEPT); \
} while (0)
#endif
#undef libc_feupdateenvf #undef libc_feupdateenvf
#define libc_feupdateenvf(e) libc_feupdateenv (e) #define libc_feupdateenvf(e) libc_feupdateenv (e)
// #define libc_feupdateenvl(e) (void) feupdateenv (e) // #define libc_feupdateenvl(e) (void) feupdateenv (e)