[PATCH 0/3] fenv support round 2

[PATCH 0/3] fenv support round 2

[joel]

From: Joel Sherrill <joel@rtems.org>

There are three patches in this set and hopefully two will be uncontroversial.

(1) The common include/fenv.h was copied from riscv, so delete the 
copy in the libc/machine/riscv directory.

(2) Add glibc extension FE_NOMASK_ENV to template sys/fenv.h.
Code was copied from Cygwin fenv.h.

(3) Add x86_64 and i386 fenv support by copying from Cygwin. No
functional changes were made. This needs some help.

   (a) x86_64 isn't picking up fenv.c. Not sure what isn't bootstrapping right.
       Help needed to get it integrated into libm build.
   (b) I added a copyright to fenv.c since it just said "part of Cygwin".
       I hope it is right.
   (c) Any other comments.

After (3) is merged, a Cygwin maintainer will need to clean up winsup
and make sure all is good.

Thanks.

--joel

Joel Sherrill (3):
  riscv/include/fenv.h: Use shared fenv.h.
  libc/include/sys/fenv.h: Add glibc extension of FE_NOMASK_ENV.
  Add i386 and x86_64 fenv support from Cygwin.

 newlib/libc/include/sys/fenv.h           |   9 +
 newlib/libc/machine/i386/sys/fenv.h      |   1 +
 newlib/libc/machine/riscv/include/fenv.h |  42 ---
 newlib/libc/machine/x86_64/sys/fenv.h    | 150 ++++++++++
 newlib/libm/machine/i386/Makefile.am     |   2 +-
 newlib/libm/machine/i386/fenv.c          |   1 +
 newlib/libm/machine/x86_64/Makefile.am   |  18 ++
 newlib/libm/machine/x86_64/fenv.c        | 485 +++++++++++++++++++++++++++++++
 8 files changed, 665 insertions(+), 43 deletions(-)
 create mode 120000 newlib/libc/machine/i386/sys/fenv.h
 delete mode 100644 newlib/libc/machine/riscv/include/fenv.h
 create mode 100644 newlib/libc/machine/x86_64/sys/fenv.h
 create mode 120000 newlib/libm/machine/i386/fenv.c
 create mode 100644 newlib/libm/machine/x86_64/Makefile.am
 create mode 100644 newlib/libm/machine/x86_64/fenv.c

-- 
1.8.3.1

[PATCH 3/3] Add i386 and x86_64 fenv support from Cygwin.

[joel]

From: Joel Sherrill <joel@rtems.org>

---
 newlib/libc/machine/i386/sys/fenv.h    |   1 +
 newlib/libc/machine/x86_64/sys/fenv.h  | 150 ++++++++++
 newlib/libm/machine/i386/Makefile.am   |   2 +-
 newlib/libm/machine/i386/fenv.c        |   1 +
 newlib/libm/machine/x86_64/Makefile.am |  18 ++
 newlib/libm/machine/x86_64/fenv.c      | 485 +++++++++++++++++++++++++++++++++
 6 files changed, 656 insertions(+), 1 deletion(-)
 create mode 120000 newlib/libc/machine/i386/sys/fenv.h
 create mode 100644 newlib/libc/machine/x86_64/sys/fenv.h
 create mode 120000 newlib/libm/machine/i386/fenv.c
 create mode 100644 newlib/libm/machine/x86_64/Makefile.am
 create mode 100644 newlib/libm/machine/x86_64/fenv.c

diff --git a/newlib/libc/machine/i386/sys/fenv.h b/newlib/libc/machine/i386/sys/fenv.h
new file mode 120000
index 0000000..2180578
--- /dev/null
+++ b/newlib/libc/machine/i386/sys/fenv.h
@@ -0,0 +1 @@
+../../x86_64/sys/fenv.h
\ No newline at end of file
diff --git a/newlib/libc/machine/x86_64/sys/fenv.h b/newlib/libc/machine/x86_64/sys/fenv.h
new file mode 100644
index 0000000..69f7bef
--- /dev/null
+++ b/newlib/libc/machine/x86_64/sys/fenv.h
@@ -0,0 +1,150 @@
+/* fenv.h
+
+This file is part of Cygwin.
+
+This software is a copyrighted work licensed under the terms of the
+Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
+details. */
+
+#ifndef _SYS_FENV_H
+#define _SYS_FENV_H 1
+
+#include <sys/cdefs.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/* Primary sources:
+
+     The Open Group Base Specifications Issue 6:
+   http://www.opengroup.org/onlinepubs/000095399/basedefs/fenv.h.html
+
+     C99 Language spec (draft n1256):
+   <url unknown>
+
+     Intel(R) 64 and IA-32 Architectures Software Developer's Manuals:
+   http://www.intel.com/products/processor/manuals/
+
+     GNU C library manual pages:
+   http://www.gnu.org/software/libc/manual/html_node/Control-Functions.html
+   http://www.gnu.org/software/libc/manual/html_node/Rounding.html
+   http://www.gnu.org/software/libc/manual/html_node/FP-Exceptions.html
+   http://www.gnu.org/software/libc/manual/html_node/Status-bit-operations.html
+
+     Linux online man page(s):
+   http://linux.die.net/man/3/fegetexcept
+
+    The documentation quotes these sources for reference.  All definitions and
+   code have been developed solely based on the information from these specs.
+
+*/
+
+/*  Represents the entire floating-point environment. The floating-point
+   environment refers collectively to any floating-point status flags and
+   control modes supported by the implementation.
+    In this implementation, the struct contains the state information from
+   the fstenv/fnstenv instructions and a copy of the SSE MXCSR, since GCC
+   uses SSE for a lot of floating-point operations.  (Cygwin assumes i686
+   or above these days, as does the compiler.)  */
+
+typedef struct _fenv_t
+{
+  struct _fpu_env_info {
+    unsigned int _fpu_cw;	/* low 16 bits only. */
+    unsigned int _fpu_sw;	/* low 16 bits only. */
+    unsigned int _fpu_tagw;	/* low 16 bits only. */
+    unsigned int _fpu_ipoff;
+    unsigned int _fpu_ipsel;
+    unsigned int _fpu_opoff;
+    unsigned int _fpu_opsel;	/* low 16 bits only. */
+  } _fpu;
+  unsigned int _sse_mxcsr;
+} fenv_t;
+
+/*  Represents the floating-point status flags collectively, including
+   any status the implementation associates with the flags. A floating-point
+   status flag is a system variable whose value is set (but never cleared)
+   when a floating-point exception is raised, which occurs as a side effect
+   of exceptional floating-point arithmetic to provide auxiliary information.
+    A floating-point control mode is a system variable whose value may be
+   set by the user to affect the subsequent behavior of floating-point
+   arithmetic. */
+
+typedef __uint32_t fexcept_t;
+
+/*  The <fenv.h> header shall define the following constants if and only
+   if the implementation supports the floating-point exception by means
+   of the floating-point functions feclearexcept(), fegetexceptflag(),
+   feraiseexcept(), fesetexceptflag(), and fetestexcept(). Each expands to
+   an integer constant expression with values such that bitwise-inclusive
+   ORs of all combinations of the constants result in distinct values.  */
+
+#define FE_DIVBYZERO	(1 << 2)
+#define FE_INEXACT	(1 << 5)
+#define FE_INVALID	(1 << 0)
+#define FE_OVERFLOW	(1 << 3)
+#define FE_UNDERFLOW	(1 << 4)
+
+/*  The <fenv.h> header shall define the following constant, which is
+   simply the bitwise-inclusive OR of all floating-point exception
+   constants defined above:  */
+
+/* in agreement w/ Linux the subnormal exception will always be masked */
+#define FE_ALL_EXCEPT \
+  (FE_INEXACT | FE_UNDERFLOW | FE_OVERFLOW | FE_DIVBYZERO | FE_INVALID)
+
+/*  The <fenv.h> header shall define the following constants if and only
+   if the implementation supports getting and setting the represented
+   rounding direction by means of the fegetround() and fesetround()
+   functions. Each expands to an integer constant expression whose values
+   are distinct non-negative vales.  */
+
+#define FE_DOWNWARD	(1)
+#define FE_TONEAREST	(0)
+#define FE_TOWARDZERO	(3)
+#define FE_UPWARD	(2)
+
+/* Only Solaris and QNX implement fegetprec/fesetprec.  As Solaris, use the
+   values defined by http://www.open-std.org/jtc1/sc22//WG14/www/docs/n752.htm
+   QNX defines different values. */
+#if __MISC_VISIBLE
+#define FE_FLTPREC	(0)
+#define FE_DBLPREC	(2)
+#define FE_LDBLPREC	(3)
+#endif
+
+/*  The <fenv.h> header shall define the following constant, which
+   represents the default floating-point environment (that is, the one
+   installed at program startup) and has type pointer to const-qualified
+   fenv_t. It can be used as an argument to the functions within the
+   <fenv.h> header that manage the floating-point environment.  */
+
+extern const fenv_t *_fe_dfl_env;
+#define FE_DFL_ENV (_fe_dfl_env)
+
+/*  Additional implementation-defined environments, with macro
+   definitions beginning with FE_ and an uppercase letter,and having
+   type "pointer to const-qualified fenv_t",may also be specified by
+   the implementation.  */
+
+#if __GNU_VISIBLE
+/*  If possible, the GNU C Library defines a macro FE_NOMASK_ENV which
+   represents an environment where every exception raised causes a trap
+   to occur. You can test for this macro using #ifdef. It is only defined
+   if _GNU_SOURCE is defined.  */
+extern const fenv_t *_fe_nomask_env;
+#define FE_NOMASK_ENV (_fe_nomask_env)
+#endif /* __GNU_VISIBLE */
+
+#ifdef __INSIDE_CYGWIN__
+/* This is Cygwin-custom, not from the standard, for use in the Cygwin CRT.  */
+extern void _feinitialise ();
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _FENV_H */
diff --git a/newlib/libm/machine/i386/Makefile.am b/newlib/libm/machine/i386/Makefile.am
index 6fade2d..8880ff1 100644
--- a/newlib/libm/machine/i386/Makefile.am
+++ b/newlib/libm/machine/i386/Makefile.am
@@ -12,7 +12,7 @@ LIB_SOURCES = \
 	f_log.S f_logf.S f_log10.S f_log10f.S \
 	f_ldexp.S f_ldexpf.S f_lrint.c f_lrintf.c f_lrintl.c \
 	f_pow.c f_powf.c f_rint.c f_rintf.c f_rintl.c \
-	f_tan.S f_tanf.S f_math.h \
+	f_tan.S f_tanf.S f_math.h fenv.c \
 	i386mach.h 
 
 libi386_la_LDFLAGS = -Xcompiler -nostdlib
diff --git a/newlib/libm/machine/i386/fenv.c b/newlib/libm/machine/i386/fenv.c
new file mode 120000
index 0000000..1d7c7a1
--- /dev/null
+++ b/newlib/libm/machine/i386/fenv.c
@@ -0,0 +1 @@
+../x86_64/fenv.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/Makefile.am b/newlib/libm/machine/x86_64/Makefile.am
new file mode 100644
index 0000000..f33100f
--- /dev/null
+++ b/newlib/libm/machine/x86_64/Makefile.am
@@ -0,0 +1,18 @@
+## Process this file with automake to generate Makefile.in
+
+INCLUDES = -I $(newlib_basedir)/../newlib/libm/common $(NEWLIB_CFLAGS) \
+	$(CROSS_CFLAGS) $(TARGET_CFLAGS)
+
+LIB_SOURCES = \
+	fenv.c
+
+noinst_LIBRARIES = lib.a
+lib_a_SOURCES = $(LIB_SOURCES)
+lib_a_CFLAGS = $(AM_CFLAGS)
+lib_a_CCASFLAGS = $(AM_CCASFLAGS)
+noinst_DATA =
+
+include $(srcdir)/../../../Makefile.shared
+
+ACLOCAL_AMFLAGS = -I ../../.. -I ../../../..
+CONFIG_STATUS_DEPENDENCIES = $(newlib_basedir)/configure.host
diff --git a/newlib/libm/machine/x86_64/fenv.c b/newlib/libm/machine/x86_64/fenv.c
new file mode 100644
index 0000000..19716c9
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fenv.c
@@ -0,0 +1,485 @@
+/*
+ * Copyright (c) 2010-2019 Red Hat, Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <fenv.h>
+#include <errno.h>
+#include <string.h>        // for memcpy 
+#include <stdbool.h>
+
+/*  x87 supports subnormal numbers so we need it below. */
+#define __FE_DENORM	(1 << 1)
+/* mask (= 0x3f) to disable all exceptions at initialization */
+#define __FE_ALL_EXCEPT_X86 (FE_ALL_EXCEPT | __FE_DENORM)
+
+/*  Mask and shift amount for rounding bits.  */
+#define FE_CW_ROUND_MASK	(0x0c00)
+#define FE_CW_ROUND_SHIFT	(10)
+/*  Same, for SSE MXCSR.  */
+#define FE_MXCSR_ROUND_MASK	(0x6000)
+#define FE_MXCSR_ROUND_SHIFT	(13)
+
+/*  Mask and shift amount for precision bits.  */
+#define FE_CW_PREC_MASK		(0x0300)
+#define FE_CW_PREC_SHIFT	(8)
+
+/*  In x87, exception status bits and mask bits occupy
+   corresponding bit positions in the status and control
+   registers, respectively.  In SSE, they are both located
+   in the control-and-status register, with the status bits
+   corresponding to the x87 positions, and the mask bits
+   shifted by this amount to the left.  */
+#define FE_SSE_EXCEPT_MASK_SHIFT (7)
+
+/* These are writable so we can initialise them at startup.  */
+static fenv_t fe_nomask_env;
+
+/* These pointers provide the outside world with read-only access to them.  */
+const fenv_t *_fe_nomask_env = &fe_nomask_env;
+
+/*  Although Cygwin assumes i686 or above (hence SSE available) these
+   days, and the compiler feels free to use it (depending on compile-
+   time flags of course), we should avoid needlessly breaking any
+   purely integer mode apps (or apps compiled with -mno-sse), so we
+   only manage SSE state in this fenv module if we detect that SSE
+   instructions are available at runtime.  If we didn't do this, all
+   applications run on older machines would bomb out with an invalid
+   instruction exception right at startup; let's not be *that* WJM!  */
+static bool use_sse = false;
+
+/*  This function enables traps for each of the exceptions as indicated
+   by the parameter except. The individual exceptions are described in
+   [ ... glibc manual xref elided ...]. Only the specified exceptions are
+   enabled, the status of the other exceptions is not changed.
+    The function returns the previous enabled exceptions in case the
+   operation was successful, -1 otherwise.  */
+int
+feenableexcept (int excepts)
+{
+  unsigned short cw, old_cw;
+  unsigned int mxcsr = 0;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return -1;
+
+  /* Get control words.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (old_cw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+
+  /* Enable exceptions by clearing mask bits.  */
+  cw = old_cw & ~excepts;
+  mxcsr &= ~(excepts << FE_SSE_EXCEPT_MASK_SHIFT);
+
+  /* Store updated control words.  */
+  __asm__ volatile ("fldcw %0" :: "m" (cw));
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (mxcsr));
+
+  /* Return old value.  We assume SSE and x87 stay in sync.  Note that
+     we are returning a mask of enabled exceptions, which is the opposite
+     of the flags in the register, which are set to disable (mask) their
+     related exceptions.  */
+  return (~old_cw) & FE_ALL_EXCEPT;
+}
+
+/*  This function disables traps for each of the exceptions as indicated
+   by the parameter except. The individual exceptions are described in
+   [ ... glibc manual xref elided ...]. Only the specified exceptions are
+   disabled, the status of the other exceptions is not changed.
+    The function returns the previous enabled exceptions in case the
+   operation was successful, -1 otherwise.  */
+int
+fedisableexcept (int excepts)
+{
+  unsigned short cw, old_cw;
+  unsigned int mxcsr = 0;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return -1;
+
+  /* Get control words.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (old_cw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+
+  /* Disable exceptions by setting mask bits.  */
+  cw = old_cw | excepts;
+  mxcsr |= (excepts << FE_SSE_EXCEPT_MASK_SHIFT);
+
+  /* Store updated control words.  */
+  __asm__ volatile ("fldcw %0" :: "m" (cw));
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (mxcsr));
+
+  /* Return old value.  We assume SSE and x87 stay in sync.  Note that
+     we are returning a mask of enabled exceptions, which is the opposite
+     of the flags in the register, which are set to disable (mask) their
+     related exceptions.  */
+  return (~old_cw) & FE_ALL_EXCEPT;
+}
+
+/*  This function returns a bitmask of all currently enabled exceptions. It
+   returns -1 in case of failure.  */
+int
+fegetexcept (void)
+{
+  unsigned short cw;
+
+  /* Get control word.  We assume SSE and x87 stay in sync.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (cw) : );
+
+  /* Exception is *dis*abled when mask bit is set.  */
+  return (~cw) & FE_ALL_EXCEPT;
+}
+
+/*  Store the floating-point environment in the variable pointed to by envp.
+   The function returns zero in case the operation was successful, a non-zero
+   value otherwise.  */
+int
+fegetenv (fenv_t *envp)
+{
+  /* fnstenv disables all exceptions in the x87 FPU; as this is not what is
+     desired here, reload the cfg saved from the x87 FPU, back to the FPU */
+  __asm__ volatile ("fnstenv %0\n\
+                     fldenv %0"
+		    : "=m" (envp->_fpu) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (envp->_sse_mxcsr) : );
+  return 0;
+}
+
+/*  Store the current floating-point environment in the object pointed to
+   by envp. Then clear all exception flags, and set the FPU to trap no
+   exceptions.  Not all FPUs support trapping no exceptions; if feholdexcept
+   cannot set this mode, it returns nonzero value.  If it succeeds, it
+   returns zero.  */
+int
+feholdexcept (fenv_t *envp)
+{
+  unsigned int mxcsr;
+  fegetenv (envp);
+  mxcsr = envp->_sse_mxcsr & ~FE_ALL_EXCEPT;
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (mxcsr));
+  __asm__ volatile ("fnclex");
+  fedisableexcept (FE_ALL_EXCEPT);
+  return 0;
+}
+
+/*  Set the floating-point environment to that described by envp.  The
+   function returns zero in case the operation was successful, a non-zero
+   value otherwise.  */
+int
+fesetenv (const fenv_t *envp)
+{
+  __asm__ volatile ("fldenv %0" :: "m" (envp->_fpu) );
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (envp->_sse_mxcsr));
+  return 0;
+}
+
+/*  Like fesetenv, this function sets the floating-point environment to
+   that described by envp. However, if any exceptions were flagged in the
+   status word before feupdateenv was called, they remain flagged after
+   the call.  In other words, after feupdateenv is called, the status
+   word is the bitwise OR of the previous status word and the one saved
+   in envp.  The function returns zero in case the operation was successful,
+   a non-zero value otherwise.  */
+int
+feupdateenv (const fenv_t *envp)
+{
+  fenv_t envcopy;
+  unsigned int mxcsr = 0;
+  unsigned short sw;
+
+  /* Don't want to modify *envp, but want to update environment atomically,
+     so take a copy and merge the existing exceptions into it.  */
+  memcpy (&envcopy, envp, sizeof *envp);
+  __asm__ volatile ("fnstsw %0" : "=m" (sw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+  envcopy._fpu._fpu_sw |= (sw & FE_ALL_EXCEPT);
+  envcopy._sse_mxcsr |= (mxcsr & FE_ALL_EXCEPT);
+
+  return fesetenv (&envcopy);
+}
+
+/*  This function clears all of the supported exception flags indicated by
+   excepts.  The function returns zero in case the operation was successful,
+   a non-zero value otherwise.  */
+int
+feclearexcept (int excepts)
+{
+  fenv_t fenv;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return EINVAL;
+
+  /* Need to save/restore whole environment to modify status word.  */
+  fegetenv (&fenv);
+
+  /* Mask undesired bits out.  */
+  fenv._fpu._fpu_sw &= ~excepts;
+  fenv._sse_mxcsr &= ~excepts;
+
+  /* Set back into FPU state.  */
+  return fesetenv (&fenv);
+}
+
+/*  This function raises the supported exceptions indicated by
+   excepts.  If more than one exception bit in excepts is set the order
+   in which the exceptions are raised is undefined except that overflow
+   (FE_OVERFLOW) or underflow (FE_UNDERFLOW) are raised before inexact
+   (FE_INEXACT). Whether for overflow or underflow the inexact exception
+   is also raised is also implementation dependent.  The function returns
+   zero in case the operation was successful, a non-zero value otherwise.  */
+int
+feraiseexcept (int excepts)
+{
+  fenv_t fenv;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return EINVAL;
+
+  /* Need to save/restore whole environment to modify status word.  */
+  __asm__ volatile ("fnstenv %0" : "=m" (fenv) : );
+
+  /* Set desired exception bits.  */
+  fenv._fpu._fpu_sw |= excepts;
+
+  /* Set back into FPU state.  */
+  __asm__ volatile ("fldenv %0" :: "m" (fenv));
+
+  /* And trigger them - whichever are unmasked.  */
+  __asm__ volatile ("fwait");
+
+  return 0;
+}
+
+/*  Test whether the exception flags indicated by the parameter except
+   are currently set. If any of them are, a nonzero value is returned
+   which specifies which exceptions are set. Otherwise the result is zero.  */
+int
+fetestexcept (int excepts)
+{
+  unsigned short sw;
+  unsigned int mxcsr = 0;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return EINVAL;
+
+  /* Get status registers.  */
+  __asm__ volatile ("fnstsw %0" : "=m" (sw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+
+  /* Mask undesired bits out and return result.  */
+  return (sw | mxcsr) & excepts;
+}
+/*  This function stores in the variable pointed to by flagp an
+   implementation-defined value representing the current setting of the
+   exception flags indicated by excepts.  The function returns zero in
+   case the operation was successful, a non-zero value otherwise.  */
+int
+fegetexceptflag (fexcept_t *flagp, int excepts)
+{
+  unsigned short sw;
+  unsigned int mxcsr = 0;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return EINVAL;
+
+  /* Get status registers.  */
+  __asm__ volatile ("fnstsw %0" : "=m" (sw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+
+  /* Mask undesired bits out and set result.  */
+  *flagp = (sw | mxcsr) & excepts;
+
+  return 0;
+}
+
+/*  This function restores the flags for the exceptions indicated by
+   excepts to the values stored in the variable pointed to by flagp.  */
+int
+fesetexceptflag (const fexcept_t *flagp, int excepts)
+{
+  fenv_t fenv;
+
+  if (excepts & ~FE_ALL_EXCEPT)
+    return EINVAL;
+
+  /* Need to save/restore whole environment to modify status word.  */
+  fegetenv (&fenv);
+
+  /* Set/Clear desired exception bits.  */
+  fenv._fpu._fpu_sw &= ~excepts;
+  fenv._fpu._fpu_sw |= excepts & *flagp;
+  fenv._sse_mxcsr &= ~excepts;
+  fenv._sse_mxcsr |= excepts & *flagp;
+
+  /* Set back into FPU state.  */
+  return fesetenv (&fenv);
+}
+
+/*  Returns the currently selected rounding mode, represented by one of the
+   values of the defined rounding mode macros.  */
+int
+fegetround (void)
+{
+  unsigned short cw;
+
+  /* Get control word.  We assume SSE and x87 stay in sync.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (cw) : );
+
+  return (cw & FE_CW_ROUND_MASK) >> FE_CW_ROUND_SHIFT;
+}
+
+/*  Changes the currently selected rounding mode to round. If round does
+   not correspond to one of the supported rounding modes nothing is changed.
+   fesetround returns zero if it changed the rounding mode, a nonzero value
+   if the mode is not supported.  */
+int
+fesetround (int round)
+{
+  unsigned short cw;
+  unsigned int mxcsr = 0;
+
+  /* Will succeed for any valid value of the input parameter.  */
+  if (round < FE_TONEAREST || round > FE_TOWARDZERO)
+    return EINVAL;
+
+  /* Get control words.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (cw) : );
+  if (use_sse)
+    __asm__ volatile ("stmxcsr %0" : "=m" (mxcsr) : );
+
+  /* Twiddle bits.  */
+  cw &= ~FE_CW_ROUND_MASK;
+  cw |= (round << FE_CW_ROUND_SHIFT);
+  mxcsr &= ~FE_MXCSR_ROUND_MASK;
+  mxcsr |= (round << FE_MXCSR_ROUND_SHIFT);
+
+  /* Set back into FPU state.  */
+  __asm__ volatile ("fldcw %0" :: "m" (cw));
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (mxcsr));
+
+  /* Indicate success.  */
+  return 0;
+}
+
+#if defined(__CYGWIN__)
+/*  Returns the currently selected precision, represented by one of the
+   values of the defined precision macros.  */
+int
+fegetprec (void)
+{
+  unsigned short cw;
+
+  /* Get control word.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (cw) : );
+
+  return (cw & FE_CW_PREC_MASK) >> FE_CW_PREC_SHIFT;
+}
+
+/* http://www.open-std.org/jtc1/sc22//WG14/www/docs/n752.htm:
+
+   The fesetprec function establishes the precision represented by its
+   argument prec.  If the argument does not match a precision macro, the
+   precision is not changed.
+
+   The fesetprec function returns a nonzero value if and only if the
+   argument matches a precision macro (that is, if and only if the requested
+   precision can be established). */
+int
+fesetprec (int prec)
+{
+  unsigned short cw;
+
+  /* Will succeed for any valid value of the input parameter.  */
+  switch (prec)
+    {
+    case FE_FLTPREC:
+    case FE_DBLPREC:
+    case FE_LDBLPREC:
+      break;
+    default:
+      return 0;
+    }
+
+  /* Get control word.  */
+  __asm__ volatile ("fnstcw %0" : "=m" (cw) : );
+
+  /* Twiddle bits.  */
+  cw &= ~FE_CW_PREC_MASK;
+  cw |= (prec << FE_CW_PREC_SHIFT);
+
+  /* Set back into FPU state.  */
+  __asm__ volatile ("fldcw %0" :: "m" (cw));
+
+  /* Indicate success.  */
+  return 1;
+}
+#endif
+
+/*  Set up the FPU and SSE environment at the start of execution.  */
+void
+_feinitialise (void)
+{
+  unsigned int edx, eax;
+  extern fenv_t __fe_dfl_env;
+
+  /* Check for presence of SSE: invoke CPUID #1, check EDX bit 25.  */
+  eax = 1;
+  __asm__ volatile ("cpuid" : "=d" (edx), "+a" (eax) :: "%ecx", "%ebx");
+  /* If this flag isn't set we'll avoid trying to execute any SSE.  */
+  if ((edx & (1 << 25)) != 0)
+    use_sse = true;
+
+  /* Reset FPU: extended prec, all exceptions cleared and masked off.  */
+  __asm__ volatile ("fninit");
+  /* The default cw value, 0x37f, is rounding mode zero.  The MXCSR has
+     no precision control, so the only thing to do is set the exception
+     mask bits.  */
+
+  /* initialize the MXCSR register: mask all exceptions */
+  unsigned int mxcsr = __FE_ALL_EXCEPT_X86 << FE_SSE_EXCEPT_MASK_SHIFT;
+  if (use_sse)
+    __asm__ volatile ("ldmxcsr %0" :: "m" (mxcsr));
+
+  /* Setup unmasked environment, but leave __FE_DENORM masked.  */
+  feenableexcept (FE_ALL_EXCEPT);
+  fegetenv (&fe_nomask_env);
+
+  /* Restore default exception masking (all masked).  */
+  fedisableexcept (FE_ALL_EXCEPT);
+
+  /* Finally cache state as default environment. */
+  fegetenv (&__fe_dfl_env);
+}
+
-- 
1.8.3.1

Re: [PATCH 3/3] Add i386 and x86_64 fenv support from Cygwin.

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 3926 bytes --]

On Aug 28 10:06, joel@rtems.org wrote:
> From: Joel Sherrill <joel@rtems.org>
> 
> ---
>  newlib/libc/machine/i386/sys/fenv.h    |   1 +
>  newlib/libc/machine/x86_64/sys/fenv.h  | 150 ++++++++++
>  newlib/libm/machine/i386/Makefile.am   |   2 +-
>  newlib/libm/machine/i386/fenv.c        |   1 +
>  newlib/libm/machine/x86_64/Makefile.am |  18 ++
>  newlib/libm/machine/x86_64/fenv.c      | 485 +++++++++++++++++++++++++++++++++
>  6 files changed, 656 insertions(+), 1 deletion(-)
>  create mode 120000 newlib/libc/machine/i386/sys/fenv.h
>  create mode 100644 newlib/libc/machine/x86_64/sys/fenv.h
>  create mode 120000 newlib/libm/machine/i386/fenv.c
>  create mode 100644 newlib/libm/machine/x86_64/Makefile.am
>  create mode 100644 newlib/libm/machine/x86_64/fenv.c
> 
> diff --git a/newlib/libc/machine/i386/sys/fenv.h b/newlib/libc/machine/i386/sys/fenv.h
> new file mode 120000
> index 0000000..2180578
> --- /dev/null
> +++ b/newlib/libc/machine/i386/sys/fenv.h
> @@ -0,0 +1 @@
> +../../x86_64/sys/fenv.h
> \ No newline at end of file
> diff --git a/newlib/libc/machine/x86_64/sys/fenv.h b/newlib/libc/machine/x86_64/sys/fenv.h
> new file mode 100644
> index 0000000..69f7bef
> --- /dev/null
> +++ b/newlib/libc/machine/x86_64/sys/fenv.h
> @@ -0,0 +1,150 @@
> +/* fenv.h
> +
> +This file is part of Cygwin.
> +
> +This software is a copyrighted work licensed under the terms of the
> +Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
> +details. */

  SPDX-License-Identifier: BSD-2-Clause

not BSD-3-Clause as I wrote in other mail.

> +/*  The <fenv.h> header shall define the following constant, which
> +   represents the default floating-point environment (that is, the one
> +   installed at program startup) and has type pointer to const-qualified
> +   fenv_t. It can be used as an argument to the functions within the
> +   <fenv.h> header that manage the floating-point environment.  */
> +
> +extern const fenv_t *_fe_dfl_env;
> +#define FE_DFL_ENV (_fe_dfl_env)

These can go away, right?  They are already defined in
newlib/libc/include/sys/fenv.h.

> +#if __GNU_VISIBLE
> +/*  If possible, the GNU C Library defines a macro FE_NOMASK_ENV which
> +   represents an environment where every exception raised causes a trap
> +   to occur. You can test for this macro using #ifdef. It is only defined
> +   if _GNU_SOURCE is defined.  */
> +extern const fenv_t *_fe_nomask_env;
> +#define FE_NOMASK_ENV (_fe_nomask_env)
> +#endif /* __GNU_VISIBLE */

And those you just added to newlib/libc/include/sys/fenv.h in patch 2 of
this set.

> +/* These are writable so we can initialise them at startup.  */
> +static fenv_t fe_nomask_env;
> +
> +/* These pointers provide the outside world with read-only access to them.  */
> +const fenv_t *_fe_nomask_env = &fe_nomask_env;

Given these are now declared in a shared header, shouldn't these be
added to their own file, newlib/libm/fenv/fe_nomask_env.c parallel
to newlib/libm/fenv/fe_dfl_env.c?

> +/*  Although Cygwin assumes i686 or above (hence SSE available) these

Please drop Cygwin-specific comments.  They just don't make sense in
common newlib code, except in rare cases to explain a difference to
other targets.

> [...]
> +#if defined(__CYGWIN__)

Great.

> +/*  Returns the currently selected precision, represented by one of the
> +   values of the defined precision macros.  */
> +int
> +fegetprec (void)
> +{
> [...]
> +int
> +fesetprec (int prec)
> +{
> [...]
> +#endif

Uh oh.  What about _feinitialise()?  Cygwin calls this function
right from the initial code, but is it really the right thing
to enforce this for all i386/x86_64 targets?

Any idea how we can generate the default environment on the fly
while maintaining backward compat on Cygwin?


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 3/3] Add i386 and x86_64 fenv support from Cygwin.

[Joel Sherrill]

On Wed, Aug 28, 2019 at 10:41 AM Corinna Vinschen <vinschen@redhat.com> wrote:
>
> On Aug 28 10:06, joel@rtems.org wrote:
> > From: Joel Sherrill <joel@rtems.org>
> >
> > ---
> >  newlib/libc/machine/i386/sys/fenv.h    |   1 +
> >  newlib/libc/machine/x86_64/sys/fenv.h  | 150 ++++++++++
> >  newlib/libm/machine/i386/Makefile.am   |   2 +-
> >  newlib/libm/machine/i386/fenv.c        |   1 +
> >  newlib/libm/machine/x86_64/Makefile.am |  18 ++
> >  newlib/libm/machine/x86_64/fenv.c      | 485 +++++++++++++++++++++++++++++++++
> >  6 files changed, 656 insertions(+), 1 deletion(-)
> >  create mode 120000 newlib/libc/machine/i386/sys/fenv.h
> >  create mode 100644 newlib/libc/machine/x86_64/sys/fenv.h
> >  create mode 120000 newlib/libm/machine/i386/fenv.c
> >  create mode 100644 newlib/libm/machine/x86_64/Makefile.am
> >  create mode 100644 newlib/libm/machine/x86_64/fenv.c
> >
> > diff --git a/newlib/libc/machine/i386/sys/fenv.h b/newlib/libc/machine/i386/sys/fenv.h
> > new file mode 120000
> > index 0000000..2180578
> > --- /dev/null
> > +++ b/newlib/libc/machine/i386/sys/fenv.h
> > @@ -0,0 +1 @@
> > +../../x86_64/sys/fenv.h
> > \ No newline at end of file
> > diff --git a/newlib/libc/machine/x86_64/sys/fenv.h b/newlib/libc/machine/x86_64/sys/fenv.h
> > new file mode 100644
> > index 0000000..69f7bef
> > --- /dev/null
> > +++ b/newlib/libc/machine/x86_64/sys/fenv.h
> > @@ -0,0 +1,150 @@
> > +/* fenv.h
> > +
> > +This file is part of Cygwin.
> > +
> > +This software is a copyrighted work licensed under the terms of the
> > +Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
> > +details. */
>
>   SPDX-License-Identifier: BSD-2-Clause
>
> not BSD-3-Clause as I wrote in other mail.

I added that using some other random file as an example.

>
> > +/*  The <fenv.h> header shall define the following constant, which
> > +   represents the default floating-point environment (that is, the one
> > +   installed at program startup) and has type pointer to const-qualified
> > +   fenv_t. It can be used as an argument to the functions within the
> > +   <fenv.h> header that manage the floating-point environment.  */
> > +
> > +extern const fenv_t *_fe_dfl_env;
> > +#define FE_DFL_ENV (_fe_dfl_env)
>
> These can go away, right?  They are already defined in
> newlib/libc/include/sys/fenv.h.

Each architecture overrides sys/fenv.h. There is no sharing of
libc/include/sys/fenv.h
with a functional implementation.

>
> > +#if __GNU_VISIBLE
> > +/*  If possible, the GNU C Library defines a macro FE_NOMASK_ENV which
> > +   represents an environment where every exception raised causes a trap
> > +   to occur. You can test for this macro using #ifdef. It is only defined
> > +   if _GNU_SOURCE is defined.  */
> > +extern const fenv_t *_fe_nomask_env;
> > +#define FE_NOMASK_ENV (_fe_nomask_env)
> > +#endif /* __GNU_VISIBLE */
>
> And those you just added to newlib/libc/include/sys/fenv.h in patch 2 of
> this set.

Ditto on architecture speciific sys/fenv.h

> > +/* These are writable so we can initialise them at startup.  */
> > +static fenv_t fe_nomask_env;
> > +
> > +/* These pointers provide the outside world with read-only access to them.  */
> > +const fenv_t *_fe_nomask_env = &fe_nomask_env;
>
> Given these are now declared in a shared header, shouldn't these be
> added to their own file, newlib/libm/fenv/fe_nomask_env.c parallel
> to newlib/libm/fenv/fe_dfl_env.c?

Sure. I need to add a stub for this.

>
> > +/*  Although Cygwin assumes i686 or above (hence SSE available) these
>
> Please drop Cygwin-specific comments.  They just don't make sense in
> common newlib code, except in rare cases to explain a difference to
> other targets.
>
> > [...]
> > +#if defined(__CYGWIN__)
>
> Great.
>
> > +/*  Returns the currently selected precision, represented by one of the
> > +   values of the defined precision macros.  */
> > +int
> > +fegetprec (void)
> > +{
> > [...]
> > +int
> > +fesetprec (int prec)
> > +{
> > [...]
> > +#endif
>
> Uh oh.  What about _feinitialise()?  Cygwin calls this function
> right from the initial code, but is it really the right thing
> to enforce this for all i386/x86_64 targets?
>
> Any idea how we can generate the default environment on the fly
> while maintaining backward compat on Cygwin?

Cygwin, libgloss, and RTEMS could call this I suppose. But each OS would have
to do their own thing.

Should be really be called from the beginning of each thread? Otherwise,
things are inconsistent.

--joel

> Thanks,
> Corinna
>
> --
> Corinna Vinschen
> Cygwin Maintainer
> Red Hat

Re: [PATCH 3/3] Add i386 and x86_64 fenv support from Cygwin.

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 2002 bytes --]

On Aug 28 11:57, Joel Sherrill wrote:
> On Wed, Aug 28, 2019 at 10:41 AM Corinna Vinschen <vinschen@redhat.com> wrote:
> > > +/*  The <fenv.h> header shall define the following constant, which
> > > +   represents the default floating-point environment (that is, the one
> > > +   installed at program startup) and has type pointer to const-qualified
> > > +   fenv_t. It can be used as an argument to the functions within the
> > > +   <fenv.h> header that manage the floating-point environment.  */
> > > +
> > > +extern const fenv_t *_fe_dfl_env;
> > > +#define FE_DFL_ENV (_fe_dfl_env)
> >
> > These can go away, right?  They are already defined in
> > newlib/libc/include/sys/fenv.h.
> 
> Each architecture overrides sys/fenv.h. There is no sharing of
> libc/include/sys/fenv.h
> with a functional implementation.

Miscomprehension on my side, sorry.

> > > +/*  Returns the currently selected precision, represented by one of the
> > > +   values of the defined precision macros.  */
> > > +int
> > > +fegetprec (void)
> > > +{
> > > [...]
> > > +int
> > > +fesetprec (int prec)
> > > +{
> > > [...]
> > > +#endif
> >
> > Uh oh.  What about _feinitialise()?  Cygwin calls this function
> > right from the initial code, but is it really the right thing
> > to enforce this for all i386/x86_64 targets?
> >
> > Any idea how we can generate the default environment on the fly
> > while maintaining backward compat on Cygwin?
> 
> Cygwin, libgloss, and RTEMS could call this I suppose. But each OS would have
> to do their own thing.
> 
> Should be really be called from the beginning of each thread? Otherwise,
> things are inconsistent.

What about initializing on the fly at the start of each affected function?
If we find a nice way to do that I'd also push that chage into Cygwin.
There's no obvious reason that we actually *have* to initalize the
fp environment in each process.


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH 1/3] riscv/include/fenv.h: Use shared fenv.h.

[joel]

From: Joel Sherrill <joel@rtems.org>

	libc/include/fenv.h was a direct copy of this file.
---
 newlib/libc/machine/riscv/include/fenv.h | 42 --------------------------------
 1 file changed, 42 deletions(-)
 delete mode 100644 newlib/libc/machine/riscv/include/fenv.h

diff --git a/newlib/libc/machine/riscv/include/fenv.h b/newlib/libc/machine/riscv/include/fenv.h
deleted file mode 100644
index 4795cc9..0000000
--- a/newlib/libc/machine/riscv/include/fenv.h
+++ /dev/null
@@ -1,42 +0,0 @@
-/* Copyright (c) 2017  SiFive Inc. All rights reserved.
-
-   This copyrighted material is made available to anyone wishing to use,
-   modify, copy, or redistribute it subject to the terms and conditions
-   of the FreeBSD License.   This program is distributed in the hope that
-   it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
-   including the implied warranties of MERCHANTABILITY or FITNESS FOR
-   A PARTICULAR PURPOSE.  A copy of this license is available at
-   http://www.opensource.org/licenses.
-*/
-
-#ifndef _FENV_H
-#define _FENV_H
-
-#include <sys/fenv.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-/* Exception */
-int feclearexcept(int excepts);
-int fegetexceptflag(fexcept_t *flagp, int excepts);
-int feraiseexcept(int excepts);
-int fesetexceptflag(const fexcept_t *flagp, int excepts);
-int fetestexcept(int excepts);
-
-/* Rounding mode */
-int fegetround(void);
-int fesetround(int rounding_mode);
-
-/* Float environment */
-int fegetenv(fenv_t *envp);
-int feholdexcept(fenv_t *envp);
-int fesetenv(const fenv_t *envp);
-int feupdateenv(const fenv_t *envp);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif
-- 
1.8.3.1

Re: [PATCH 1/3] riscv/include/fenv.h: Use shared fenv.h.

[Kito Cheng]

LGTM :)

On Wed, Aug 28, 2019 at 11:07 PM <joel@rtems.org> wrote:
>
> From: Joel Sherrill <joel@rtems.org>
>
>         libc/include/fenv.h was a direct copy of this file.
> ---
>  newlib/libc/machine/riscv/include/fenv.h | 42 --------------------------------
>  1 file changed, 42 deletions(-)
>  delete mode 100644 newlib/libc/machine/riscv/include/fenv.h
>
> diff --git a/newlib/libc/machine/riscv/include/fenv.h b/newlib/libc/machine/riscv/include/fenv.h
> deleted file mode 100644
> index 4795cc9..0000000
> --- a/newlib/libc/machine/riscv/include/fenv.h
> +++ /dev/null
> @@ -1,42 +0,0 @@
> -/* Copyright (c) 2017  SiFive Inc. All rights reserved.
> -
> -   This copyrighted material is made available to anyone wishing to use,
> -   modify, copy, or redistribute it subject to the terms and conditions
> -   of the FreeBSD License.   This program is distributed in the hope that
> -   it will be useful, but WITHOUT ANY WARRANTY expressed or implied,
> -   including the implied warranties of MERCHANTABILITY or FITNESS FOR
> -   A PARTICULAR PURPOSE.  A copy of this license is available at
> -   http://www.opensource.org/licenses.
> -*/
> -
> -#ifndef _FENV_H
> -#define _FENV_H
> -
> -#include <sys/fenv.h>
> -
> -#ifdef __cplusplus
> -extern "C" {
> -#endif
> -
> -/* Exception */
> -int feclearexcept(int excepts);
> -int fegetexceptflag(fexcept_t *flagp, int excepts);
> -int feraiseexcept(int excepts);
> -int fesetexceptflag(const fexcept_t *flagp, int excepts);
> -int fetestexcept(int excepts);
> -
> -/* Rounding mode */
> -int fegetround(void);
> -int fesetround(int rounding_mode);
> -
> -/* Float environment */
> -int fegetenv(fenv_t *envp);
> -int feholdexcept(fenv_t *envp);
> -int fesetenv(const fenv_t *envp);
> -int feupdateenv(const fenv_t *envp);
> -
> -#ifdef __cplusplus
> -}
> -#endif
> -
> -#endif
> --
> 1.8.3.1
>

[PATCH 2/3] libc/include/sys/fenv.h: Add glibc extension of FE_NOMASK_ENV.

[joel]

From: Joel Sherrill <joel@rtems.org>

---
 newlib/libc/include/sys/fenv.h | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/newlib/libc/include/sys/fenv.h b/newlib/libc/include/sys/fenv.h
index e0983fa..d2c1302 100644
--- a/newlib/libc/include/sys/fenv.h
+++ b/newlib/libc/include/sys/fenv.h
@@ -110,4 +110,13 @@ typedef int fexcept_t;
 extern const fenv_t *_fe_dfl_env;
 #define FE_DFL_ENV _fe_dfl_env
 
+#if __GNU_VISIBLE
+/*  If possible, the GNU C Library defines a macro FE_NOMASK_ENV which
+   represents an environment where every exception raised causes a trap
+   to occur. You can test for this macro using #ifdef. It is only defined
+   if _GNU_SOURCE is defined.  */
+extern const fenv_t *_fe_nomask_env;
+#define FE_NOMASK_ENV (_fe_nomask_env)
+#endif /* __GNU_VISIBLE */
+
 #endif /* _SYS_FENV_H_ */
-- 
1.8.3.1

Re: [PATCH 2/3] libc/include/sys/fenv.h: Add glibc extension of FE_NOMASK_ENV.

[Howland, Craig D. - US via newlib]

________________________________________
> From: newlib-owner@sourceware.org <newlib-owner@sourceware.org> on behalf of joel@rtems.org <joel@rtems.org>
> Sent: Wednesday, August 28, 2019 11:06 AM
> To: newlib@sourceware.org
> Cc: Joel Sherrill
> Subject: [PATCH 2/3] libc/include/sys/fenv.h: Add glibc extension of FE_NOMASK_ENV.
>
> From: Joel Sherrill <joel@rtems.org>
>
> ---
>  newlib/libc/include/sys/fenv.h | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/newlib/libc/include/sys/fenv.h b/newlib/libc/include/sys/fenv.h
> index e0983fa..d2c1302 100644
> --- a/newlib/libc/include/sys/fenv.h
> +++ b/newlib/libc/include/sys/fenv.h
> @@ -110,4 +110,13 @@ typedef int fexcept_t;
>  extern const fenv_t *_fe_dfl_env;
>  #define FE_DFL_ENV _fe_dfl_env
>
> +#if __GNU_VISIBLE
> +/*  If possible, the GNU C Library defines a macro FE_NOMASK_ENV which
> +   represents an environment where every exception raised causes a trap
> +   to occur. You can test for this macro using #ifdef. It is only defined
> +   if _GNU_SOURCE is defined.  */
> +extern const fenv_t *_fe_nomask_env;
> +#define FE_NOMASK_ENV (_fe_nomask_env)
> +#endif /* __GNU_VISIBLE */
> +
>  #endif /* _SYS_FENV_H_ */
> --
> 1.8.3.1
--------------------------------------------------------------------------------
It says define it when possible.  But this is unconditional in terms of 
whether an implementation can do so or not.  (It is conditional on the
__GNU_VISIBLE option, but that's a compilation choice for GNU_SOURCE.)
There needs to be a way to know if traps are possible are not.  This
seems more appropriate to be undefined by default, as many embedded
applications don't even have the concept of traps.
Craig

Re: [PATCH 0/3] fenv support round 2

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 1410 bytes --]

On Aug 28 10:06, joel@rtems.org wrote:
> From: Joel Sherrill <joel@rtems.org>
> 
> There are three patches in this set and hopefully two will be uncontroversial.
> 
> (1) The common include/fenv.h was copied from riscv, so delete the 
> copy in the libc/machine/riscv directory.
> 
> (2) Add glibc extension FE_NOMASK_ENV to template sys/fenv.h.
> Code was copied from Cygwin fenv.h.
> 
> (3) Add x86_64 and i386 fenv support by copying from Cygwin. No
> functional changes were made. This needs some help.
> 
>    (a) x86_64 isn't picking up fenv.c. Not sure what isn't bootstrapping right.
>        Help needed to get it integrated into libm build.

Not sure yet, still looking into the code.

>    (b) I added a copyright to fenv.c since it just said "part of Cygwin".
>        I hope it is right.

No, that's not how it should be licensed since that implies GPL3+.
We're moving the file into newlib so we should loosen the copyright
so that the code is usable for non-GPLed code.  And as the sole
copyright holder, Red Hat can move the file to another copyright.

So what you can do here is to remove the entire Cygwin header from
the file and just keep a BSD-3-Clause SPDX header.

> After (3) is merged, a Cygwin maintainer will need to clean up winsup
> and make sure all is good.

Will check.


Thanks,
Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH 0/3] fenv support round 2

[Joel Sherrill]

On Wed, Aug 28, 2019 at 10:27 AM Corinna Vinschen <vinschen@redhat.com> wrote:
>
> On Aug 28 10:06, joel@rtems.org wrote:
> > From: Joel Sherrill <joel@rtems.org>
> >
> > There are three patches in this set and hopefully two will be uncontroversial.
> >
> > (1) The common include/fenv.h was copied from riscv, so delete the
> > copy in the libc/machine/riscv directory.
> >
> > (2) Add glibc extension FE_NOMASK_ENV to template sys/fenv.h.
> > Code was copied from Cygwin fenv.h.
> >
> > (3) Add x86_64 and i386 fenv support by copying from Cygwin. No
> > functional changes were made. This needs some help.
> >
> >    (a) x86_64 isn't picking up fenv.c. Not sure what isn't bootstrapping right.
> >        Help needed to get it integrated into libm build.
>
> Not sure yet, still looking into the code.

Jeff's regenerate script wasn't even generating a x86_64 Makefile.am. I managed
to force it by hand but it's not being picked up.

Ahhh.. configure.host needs to have this:

 x86_64)
        machine_dir=x86_64
        libm_machine_dir=x86_64

But there must be more to it. Now the build looks for the file but it
isn't in the build tree.

There must be another piece of magic I am still missing.

> >    (b) I added a copyright to fenv.c since it just said "part of Cygwin".
> >        I hope it is right.
>
> No, that's not how it should be licensed since that implies GPL3+.
> We're moving the file into newlib so we should loosen the copyright
> so that the code is usable for non-GPLed code.  And as the sole
> copyright holder, Red Hat can move the file to another copyright.
>
> So what you can do here is to remove the entire Cygwin header from
> the file and just keep a BSD-3-Clause SPDX header.

I must not have been clear. I replaced the Cygwin comment with a
copy of a Red Hat BSD-2-Clause I found in another file. It was a move
in the right direction but I wasn't sure.

Just tell me what you want.

>
> > After (3) is merged, a Cygwin maintainer will need to clean up winsup
> > and make sure all is good.
>
> Will check.
>
>
> Thanks,
> Corinna
>
> --
> Corinna Vinschen
> Cygwin Maintainer
> Red Hat

Re: [PATCH 0/3] fenv support round 2

[Corinna Vinschen]

[-- Attachment #1: Type: text/plain, Size: 2373 bytes --]

On Aug 28 10:45, Joel Sherrill wrote:
> On Wed, Aug 28, 2019 at 10:27 AM Corinna Vinschen <vinschen@redhat.com> wrote:
> >
> > On Aug 28 10:06, joel@rtems.org wrote:
> > > From: Joel Sherrill <joel@rtems.org>
> > >
> > > There are three patches in this set and hopefully two will be uncontroversial.
> > >
> > > (1) The common include/fenv.h was copied from riscv, so delete the
> > > copy in the libc/machine/riscv directory.
> > >
> > > (2) Add glibc extension FE_NOMASK_ENV to template sys/fenv.h.
> > > Code was copied from Cygwin fenv.h.
> > >
> > > (3) Add x86_64 and i386 fenv support by copying from Cygwin. No
> > > functional changes were made. This needs some help.
> > >
> > >    (a) x86_64 isn't picking up fenv.c. Not sure what isn't bootstrapping right.
> > >        Help needed to get it integrated into libm build.
> >
> > Not sure yet, still looking into the code.
> 
> Jeff's regenerate script wasn't even generating a x86_64 Makefile.am. I managed
> to force it by hand but it's not being picked up.
> 
> Ahhh.. configure.host needs to have this:
> 
>  x86_64)
>         machine_dir=x86_64
>         libm_machine_dir=x86_64
> 
> But there must be more to it. Now the build looks for the file but it
> isn't in the build tree.
> 
> There must be another piece of magic I am still missing.
> 
> > >    (b) I added a copyright to fenv.c since it just said "part of Cygwin".
> > >        I hope it is right.
> >
> > No, that's not how it should be licensed since that implies GPL3+.
> > We're moving the file into newlib so we should loosen the copyright
> > so that the code is usable for non-GPLed code.  And as the sole
> > copyright holder, Red Hat can move the file to another copyright.
> >
> > So what you can do here is to remove the entire Cygwin header from
> > the file and just keep a BSD-3-Clause SPDX header.
> 
> I must not have been clear. I replaced the Cygwin comment with a
> copy of a Red Hat BSD-2-Clause I found in another file. It was a move
> in the right direction but I wasn't sure.
> 
> Just tell me what you want.

You didn't touch the header.  It has still the Cygwin copyright notice.
BSD-3-Clause or BSD-2-Clause for both files is fine, just pick one.
SPDX header is simpler and, presumably, the future.


Corinna

-- 
Corinna Vinschen
Cygwin Maintainer
Red Hat

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]