glibc/sysdeps/i386/fpu/fenv_private.h
Joseph Myers c86ed71d63 Add float128 support for x86_64, x86.
This patch enables float128 support for x86_64 and x86.  All GCC
versions that can build glibc provide the required support, but since
GCC 6 and before don't provide __builtin_nanq / __builtin_nansq, sNaN
tests and some tests of NaN payloads need to be disabled with such
compilers (this does not affect the generated glibc binaries at all,
just the tests).  bits/floatn.h declares float128 support to be
available for GCC versions that provide the required libgcc support
(4.3 for x86_64, 4.4 for i386 GNU/Linux, 4.5 for i386 GNU/Hurd);
compilation-only support was present some time before then, but not
really useful without the libgcc functions.

fenv_private.h needed updating to avoid trying to put _Float128 values
in registers.  I make no assertion of optimality of the
math_opt_barrier / math_force_eval definitions for this case; they are
simply intended to be sufficient to work correctly.

Tested for x86_64 and x86, with GCC 7 and GCC 6.  (Testing for x32 was
compilation tests only with build-many-glibcs.py to verify the ABI
baseline updates.  I have not done any testing for Hurd, although the
float128 support is enabled there as for GNU/Linux.)

	* sysdeps/i386/Implies: Add ieee754/float128.
	* sysdeps/x86_64/Implies: Likewise.
	* sysdeps/x86/bits/floatn.h: New file.
	* sysdeps/x86/float128-abi.h: Likewise.
	* manual/math.texi (Mathematics): Document support for _Float128
	on x86_64 and x86.
	* sysdeps/i386/fpu/fenv_private.h: Include <bits/floatn.h>.
	(math_opt_barrier): Do not put _Float128 values in floating-point
	registers.
	(math_force_eval): Likewise.
	[__x86_64__] (SET_RESTORE_ROUNDF128): New macro.
	* sysdeps/x86/fpu/Makefile [$(subdir) = math] (CPPFLAGS): Append
	to Makefile variable.
	* sysdeps/x86/fpu/e_sqrtf128.c: New file.
	* sysdeps/x86/fpu/sfp-machine.h: Likewise.  Based on libgcc.
	* sysdeps/x86/math-tests.h: New file.
	* math/libm-test-support.h (XFAIL_FLOAT128_PAYLOAD): New macro.
	* math/libm-test-getpayload.inc (getpayload_test_data): Use
	XFAIL_FLOAT128_PAYLOAD.
	* math/libm-test-setpayload.inc (setpayload_test_data): Likewise.
	* math/libm-test-totalorder.inc (totalorder_test_data): Likewise.
	* math/libm-test-totalordermag.inc (totalordermag_test_data):
	Likewise.
	* sysdeps/unix/sysv/linux/i386/libc.abilist: Update.
	* sysdeps/unix/sysv/linux/i386/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/64/libc.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/64/libm.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/x32/libc.abilist: Likewise.
	* sysdeps/unix/sysv/linux/x86_64/x32/libm.abilist: Likewise.
	* sysdeps/i386/fpu/libm-test-ulps: Likewise.
	* sysdeps/i386/i686/fpu/multiarch/libm-test-ulps: Likewise.
	* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.
2017-06-26 22:02:24 +00:00

519 lines
15 KiB
C

#ifndef FENV_PRIVATE_H
#define FENV_PRIVATE_H 1
#include <bits/floatn.h>
#include <fenv.h>
#include <fpu_control.h>
#ifdef __SSE2_MATH__
# define math_opt_barrier(x) \
({ __typeof(x) __x; \
if (sizeof (x) <= sizeof (double) \
|| __builtin_types_compatible_p (__typeof (x), _Float128)) \
__asm ("" : "=x" (__x) : "0" (x)); \
else \
__asm ("" : "=t" (__x) : "0" (x)); \
__x; })
# define math_force_eval(x) \
do { \
if (sizeof (x) <= sizeof (double) \
|| __builtin_types_compatible_p (__typeof (x), _Float128)) \
__asm __volatile ("" : : "x" (x)); \
else \
__asm __volatile ("" : : "f" (x)); \
} while (0)
#else
# define math_opt_barrier(x) \
({ __typeof (x) __x; \
if (__builtin_types_compatible_p (__typeof (x), _Float128)) \
{ \
__x = (x); \
__asm ("" : "+m" (__x)); \
} \
else \
__asm ("" : "=t" (__x) : "0" (x)); \
__x; })
# define math_force_eval(x) \
do { \
__typeof (x) __x = (x); \
if (sizeof (x) <= sizeof (double) \
|| __builtin_types_compatible_p (__typeof (x), _Float128)) \
__asm __volatile ("" : : "m" (__x)); \
else \
__asm __volatile ("" : : "f" (__x)); \
} while (0)
#endif
/* This file is used by both the 32- and 64-bit ports. The 64-bit port
has a field in the fenv_t for the mxcsr; the 32-bit port does not.
Instead, we (ab)use the only 32-bit field extant in the struct. */
#ifndef __x86_64__
# define __mxcsr __eip
#endif
/* All of these functions are private to libm, and are all used in pairs
to save+change the fp state and restore the original state. Thus we
need not care for both the 387 and the sse unit, only the one we're
actually using. */
#if defined __AVX__ || defined SSE2AVX
# define STMXCSR "vstmxcsr"
# define LDMXCSR "vldmxcsr"
#else
# define STMXCSR "stmxcsr"
# define LDMXCSR "ldmxcsr"
#endif
static __always_inline void
libc_feholdexcept_sse (fenv_t *e)
{
unsigned int mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
e->__mxcsr = mxcsr;
mxcsr = (mxcsr | 0x1f80) & ~0x3f;
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
}
static __always_inline void
libc_feholdexcept_387 (fenv_t *e)
{
/* Recall that fnstenv has a side-effect of masking exceptions.
Clobber all of the fp registers so that the TOS field is 0. */
asm volatile ("fnstenv %0; fnclex"
: "=m"(*e)
: : "st", "st(1)", "st(2)", "st(3)",
"st(4)", "st(5)", "st(6)", "st(7)");
}
static __always_inline void
libc_fesetround_sse (int r)
{
unsigned int mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
mxcsr = (mxcsr & ~0x6000) | (r << 3);
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
}
static __always_inline void
libc_fesetround_387 (int r)
{
fpu_control_t cw;
_FPU_GETCW (cw);
cw = (cw & ~0xc00) | r;
_FPU_SETCW (cw);
}
static __always_inline void
libc_feholdexcept_setround_sse (fenv_t *e, int r)
{
unsigned int mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
e->__mxcsr = mxcsr;
mxcsr = ((mxcsr | 0x1f80) & ~0x603f) | (r << 3);
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
}
/* Set both rounding mode and precision. A convenience function for use
by libc_feholdexcept_setround and libc_feholdexcept_setround_53bit. */
static __always_inline void
libc_feholdexcept_setround_387_prec (fenv_t *e, int r)
{
libc_feholdexcept_387 (e);
fpu_control_t cw = e->__control_word;
cw &= ~(_FPU_RC_ZERO | _FPU_EXTENDED);
cw |= r | 0x3f;
_FPU_SETCW (cw);
}
static __always_inline void
libc_feholdexcept_setround_387 (fenv_t *e, int r)
{
libc_feholdexcept_setround_387_prec (e, r | _FPU_EXTENDED);
}
static __always_inline void
libc_feholdexcept_setround_387_53bit (fenv_t *e, int r)
{
libc_feholdexcept_setround_387_prec (e, r | _FPU_DOUBLE);
}
static __always_inline int
libc_fetestexcept_sse (int e)
{
unsigned int mxcsr;
asm volatile (STMXCSR " %0" : "=m" (*&mxcsr));
return mxcsr & e & FE_ALL_EXCEPT;
}
static __always_inline int
libc_fetestexcept_387 (int ex)
{
fexcept_t temp;
asm volatile ("fnstsw %0" : "=a" (temp));
return temp & ex & FE_ALL_EXCEPT;
}
static __always_inline void
libc_fesetenv_sse (fenv_t *e)
{
asm volatile (LDMXCSR " %0" : : "m" (e->__mxcsr));
}
static __always_inline void
libc_fesetenv_387 (fenv_t *e)
{
/* Clobber all fp registers so that the TOS value we saved earlier is
compatible with the current state of the compiler. */
asm volatile ("fldenv %0"
: : "m" (*e)
: "st", "st(1)", "st(2)", "st(3)",
"st(4)", "st(5)", "st(6)", "st(7)");
}
static __always_inline int
libc_feupdateenv_test_sse (fenv_t *e, int ex)
{
unsigned int mxcsr, old_mxcsr, cur_ex;
asm volatile (STMXCSR " %0" : "=m" (*&mxcsr));
cur_ex = mxcsr & FE_ALL_EXCEPT;
/* Merge current exceptions with the old environment. */
old_mxcsr = e->__mxcsr;
mxcsr = old_mxcsr | cur_ex;
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
/* Raise SIGFPE for any new exceptions since the hold. Expect that
the normal environment has all exceptions masked. */
if (__glibc_unlikely (~(old_mxcsr >> 7) & cur_ex))
__feraiseexcept (cur_ex);
/* Test for exceptions raised since the hold. */
return cur_ex & ex;
}
static __always_inline int
libc_feupdateenv_test_387 (fenv_t *e, int ex)
{
fexcept_t cur_ex;
/* Save current exceptions. */
asm volatile ("fnstsw %0" : "=a" (cur_ex));
cur_ex &= FE_ALL_EXCEPT;
/* Reload original environment. */
libc_fesetenv_387 (e);
/* Merge current exceptions. */
__feraiseexcept (cur_ex);
/* Test for exceptions raised since the hold. */
return cur_ex & ex;
}
static __always_inline void
libc_feupdateenv_sse (fenv_t *e)
{
libc_feupdateenv_test_sse (e, 0);
}
static __always_inline void
libc_feupdateenv_387 (fenv_t *e)
{
libc_feupdateenv_test_387 (e, 0);
}
static __always_inline void
libc_feholdsetround_sse (fenv_t *e, int r)
{
unsigned int mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
e->__mxcsr = mxcsr;
mxcsr = (mxcsr & ~0x6000) | (r << 3);
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
}
static __always_inline void
libc_feholdsetround_387_prec (fenv_t *e, int r)
{
fpu_control_t cw;
_FPU_GETCW (cw);
e->__control_word = cw;
cw &= ~(_FPU_RC_ZERO | _FPU_EXTENDED);
cw |= r;
_FPU_SETCW (cw);
}
static __always_inline void
libc_feholdsetround_387 (fenv_t *e, int r)
{
libc_feholdsetround_387_prec (e, r | _FPU_EXTENDED);
}
static __always_inline void
libc_feholdsetround_387_53bit (fenv_t *e, int r)
{
libc_feholdsetround_387_prec (e, r | _FPU_DOUBLE);
}
static __always_inline void
libc_feresetround_sse (fenv_t *e)
{
unsigned int mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
mxcsr = (mxcsr & ~0x6000) | (e->__mxcsr & 0x6000);
asm volatile (LDMXCSR " %0" : : "m" (*&mxcsr));
}
static __always_inline void
libc_feresetround_387 (fenv_t *e)
{
_FPU_SETCW (e->__control_word);
}
#ifdef __SSE_MATH__
# define libc_feholdexceptf libc_feholdexcept_sse
# define libc_fesetroundf libc_fesetround_sse
# define libc_feholdexcept_setroundf libc_feholdexcept_setround_sse
# define libc_fetestexceptf libc_fetestexcept_sse
# define libc_fesetenvf libc_fesetenv_sse
# define libc_feupdateenv_testf libc_feupdateenv_test_sse
# define libc_feupdateenvf libc_feupdateenv_sse
# define libc_feholdsetroundf libc_feholdsetround_sse
# define libc_feresetroundf libc_feresetround_sse
#else
# define libc_feholdexceptf libc_feholdexcept_387
# define libc_fesetroundf libc_fesetround_387
# define libc_feholdexcept_setroundf libc_feholdexcept_setround_387
# define libc_fetestexceptf libc_fetestexcept_387
# define libc_fesetenvf libc_fesetenv_387
# define libc_feupdateenv_testf libc_feupdateenv_test_387
# define libc_feupdateenvf libc_feupdateenv_387
# define libc_feholdsetroundf libc_feholdsetround_387
# define libc_feresetroundf libc_feresetround_387
#endif /* __SSE_MATH__ */
#ifdef __SSE2_MATH__
# define libc_feholdexcept libc_feholdexcept_sse
# define libc_fesetround libc_fesetround_sse
# define libc_feholdexcept_setround libc_feholdexcept_setround_sse
# define libc_fetestexcept libc_fetestexcept_sse
# define libc_fesetenv libc_fesetenv_sse
# define libc_feupdateenv_test libc_feupdateenv_test_sse
# define libc_feupdateenv libc_feupdateenv_sse
# define libc_feholdsetround libc_feholdsetround_sse
# define libc_feresetround libc_feresetround_sse
#else
# define libc_feholdexcept libc_feholdexcept_387
# define libc_fesetround libc_fesetround_387
# define libc_feholdexcept_setround libc_feholdexcept_setround_387
# define libc_fetestexcept libc_fetestexcept_387
# define libc_fesetenv libc_fesetenv_387
# define libc_feupdateenv_test libc_feupdateenv_test_387
# define libc_feupdateenv libc_feupdateenv_387
# define libc_feholdsetround libc_feholdsetround_387
# define libc_feresetround libc_feresetround_387
#endif /* __SSE2_MATH__ */
#define libc_feholdexceptl libc_feholdexcept_387
#define libc_fesetroundl libc_fesetround_387
#define libc_feholdexcept_setroundl libc_feholdexcept_setround_387
#define libc_fetestexceptl libc_fetestexcept_387
#define libc_fesetenvl libc_fesetenv_387
#define libc_feupdateenv_testl libc_feupdateenv_test_387
#define libc_feupdateenvl libc_feupdateenv_387
#define libc_feholdsetroundl libc_feholdsetround_387
#define libc_feresetroundl libc_feresetround_387
#ifndef __SSE2_MATH__
# define libc_feholdexcept_setround_53bit libc_feholdexcept_setround_387_53bit
# define libc_feholdsetround_53bit libc_feholdsetround_387_53bit
#endif
#ifdef __x86_64__
/* The SSE rounding mode is used by soft-fp (libgcc and glibc) on
x86_64, so that must be set for float128 computations. */
# define SET_RESTORE_ROUNDF128(RM) \
SET_RESTORE_ROUND_GENERIC (RM, libc_feholdsetround_sse, libc_feresetround_sse)
#endif
/* We have support for rounding mode context. */
#define HAVE_RM_CTX 1
static __always_inline void
libc_feholdexcept_setround_sse_ctx (struct rm_ctx *ctx, int r)
{
unsigned int mxcsr, new_mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
new_mxcsr = ((mxcsr | 0x1f80) & ~0x603f) | (r << 3);
ctx->env.__mxcsr = mxcsr;
if (__glibc_unlikely (mxcsr != new_mxcsr))
{
asm volatile (LDMXCSR " %0" : : "m" (*&new_mxcsr));
ctx->updated_status = true;
}
else
ctx->updated_status = false;
}
/* Unconditional since we want to overwrite any exceptions that occurred in the
context. This is also why all fehold* functions unconditionally write into
ctx->env. */
static __always_inline void
libc_fesetenv_sse_ctx (struct rm_ctx *ctx)
{
libc_fesetenv_sse (&ctx->env);
}
static __always_inline void
libc_feupdateenv_sse_ctx (struct rm_ctx *ctx)
{
if (__glibc_unlikely (ctx->updated_status))
libc_feupdateenv_test_sse (&ctx->env, 0);
}
static __always_inline void
libc_feholdexcept_setround_387_prec_ctx (struct rm_ctx *ctx, int r)
{
libc_feholdexcept_387 (&ctx->env);
fpu_control_t cw = ctx->env.__control_word;
fpu_control_t old_cw = cw;
cw &= ~(_FPU_RC_ZERO | _FPU_EXTENDED);
cw |= r | 0x3f;
if (__glibc_unlikely (old_cw != cw))
{
_FPU_SETCW (cw);
ctx->updated_status = true;
}
else
ctx->updated_status = false;
}
static __always_inline void
libc_feholdexcept_setround_387_ctx (struct rm_ctx *ctx, int r)
{
libc_feholdexcept_setround_387_prec_ctx (ctx, r | _FPU_EXTENDED);
}
static __always_inline void
libc_feholdexcept_setround_387_53bit_ctx (struct rm_ctx *ctx, int r)
{
libc_feholdexcept_setround_387_prec_ctx (ctx, r | _FPU_DOUBLE);
}
static __always_inline void
libc_feholdsetround_387_prec_ctx (struct rm_ctx *ctx, int r)
{
fpu_control_t cw, new_cw;
_FPU_GETCW (cw);
new_cw = cw;
new_cw &= ~(_FPU_RC_ZERO | _FPU_EXTENDED);
new_cw |= r;
ctx->env.__control_word = cw;
if (__glibc_unlikely (new_cw != cw))
{
_FPU_SETCW (new_cw);
ctx->updated_status = true;
}
else
ctx->updated_status = false;
}
static __always_inline void
libc_feholdsetround_387_ctx (struct rm_ctx *ctx, int r)
{
libc_feholdsetround_387_prec_ctx (ctx, r | _FPU_EXTENDED);
}
static __always_inline void
libc_feholdsetround_387_53bit_ctx (struct rm_ctx *ctx, int r)
{
libc_feholdsetround_387_prec_ctx (ctx, r | _FPU_DOUBLE);
}
static __always_inline void
libc_feholdsetround_sse_ctx (struct rm_ctx *ctx, int r)
{
unsigned int mxcsr, new_mxcsr;
asm (STMXCSR " %0" : "=m" (*&mxcsr));
new_mxcsr = (mxcsr & ~0x6000) | (r << 3);
ctx->env.__mxcsr = mxcsr;
if (__glibc_unlikely (new_mxcsr != mxcsr))
{
asm volatile (LDMXCSR " %0" : : "m" (*&new_mxcsr));
ctx->updated_status = true;
}
else
ctx->updated_status = false;
}
static __always_inline void
libc_feresetround_sse_ctx (struct rm_ctx *ctx)
{
if (__glibc_unlikely (ctx->updated_status))
libc_feresetround_sse (&ctx->env);
}
static __always_inline void
libc_feresetround_387_ctx (struct rm_ctx *ctx)
{
if (__glibc_unlikely (ctx->updated_status))
_FPU_SETCW (ctx->env.__control_word);
}
static __always_inline void
libc_feupdateenv_387_ctx (struct rm_ctx *ctx)
{
if (__glibc_unlikely (ctx->updated_status))
libc_feupdateenv_test_387 (&ctx->env, 0);
}
#ifdef __SSE_MATH__
# define libc_feholdexcept_setroundf_ctx libc_feholdexcept_setround_sse_ctx
# define libc_fesetenvf_ctx libc_fesetenv_sse_ctx
# define libc_feupdateenvf_ctx libc_feupdateenv_sse_ctx
# define libc_feholdsetroundf_ctx libc_feholdsetround_sse_ctx
# define libc_feresetroundf_ctx libc_feresetround_sse_ctx
#else
# define libc_feholdexcept_setroundf_ctx libc_feholdexcept_setround_387_ctx
# define libc_feupdateenvf_ctx libc_feupdateenv_387_ctx
# define libc_feholdsetroundf_ctx libc_feholdsetround_387_ctx
# define libc_feresetroundf_ctx libc_feresetround_387_ctx
#endif /* __SSE_MATH__ */
#ifdef __SSE2_MATH__
# define libc_feholdexcept_setround_ctx libc_feholdexcept_setround_sse_ctx
# define libc_fesetenv_ctx libc_fesetenv_sse_ctx
# define libc_feupdateenv_ctx libc_feupdateenv_sse_ctx
# define libc_feholdsetround_ctx libc_feholdsetround_sse_ctx
# define libc_feresetround_ctx libc_feresetround_sse_ctx
#else
# define libc_feholdexcept_setround_ctx libc_feholdexcept_setround_387_ctx
# define libc_feupdateenv_ctx libc_feupdateenv_387_ctx
# define libc_feholdsetround_ctx libc_feholdsetround_387_ctx
# define libc_feresetround_ctx libc_feresetround_387_ctx
#endif /* __SSE2_MATH__ */
#define libc_feholdexcept_setroundl_ctx libc_feholdexcept_setround_387_ctx
#define libc_feupdateenvl_ctx libc_feupdateenv_387_ctx
#define libc_feholdsetroundl_ctx libc_feholdsetround_387_ctx
#define libc_feresetroundl_ctx libc_feresetround_387_ctx
#ifndef __SSE2_MATH__
# define libc_feholdsetround_53bit_ctx libc_feholdsetround_387_53bit_ctx
# define libc_feresetround_53bit_ctx libc_feresetround_387_ctx
#endif
#undef __mxcsr
#endif /* FENV_PRIVATE_H */