public inbox for newlib@sourceware.org
 help / color / mirror / Atom feed
* [PATCH v3] Add i386 and x86_64 fenv support
@ 2019-09-05 13:07 joel
  2019-09-05 13:07 ` [PATCH v4] Add i386 and x86_64 fenv support from Cygwin joel
  2019-09-05 13:10 ` [PATCH v3] Add i386 and x86_64 fenv support Joel Sherrill
  0 siblings, 2 replies; 3+ messages in thread
From: joel @ 2019-09-05 13:07 UTC (permalink / raw)
  To: newlib; +Cc: Joel Sherrill

From: Joel Sherrill <joel@rtems.org>

Hi

This patch adds i386 and x86_64 support by migrating the Cygwin
implementation in newlib/libm/machine directories. There were
some mechanical changes such as converting from C++ to C,
and converting use_sse to a function since _feinitialize() is not
guaranteed to be called.

Because of the way newlib builds, the non-functional files in
libm/fenv/ are all compiled and put in the library. For this
implementation, there is only an fenv.c, so every stub implementation
and fenv.c ended up in the final libm.a. Rather than split the file
and forcing all future implementations to do this, there is an fenv_stub.c

*** BLURB HERE ***

Joel Sherrill (1):
  Add i386 and x86_64 fenv support from Cygwin.

 newlib/libc/machine/i386/sys/fenv.h    |   1 +
 newlib/libc/machine/x86_64/sys/fenv.h  | 170 ++++++++++++
 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      | 487 +++++++++++++++++++++++++++++++++
 6 files changed, 678 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

-- 
1.8.3.1

^ permalink raw reply	[flat|nested] 3+ messages in thread

* [PATCH v4] Add i386 and x86_64 fenv support from Cygwin.
  2019-09-05 13:07 [PATCH v3] Add i386 and x86_64 fenv support joel
@ 2019-09-05 13:07 ` joel
  2019-09-05 13:10 ` [PATCH v3] Add i386 and x86_64 fenv support Joel Sherrill
  1 sibling, 0 replies; 3+ messages in thread
From: joel @ 2019-09-05 13:07 UTC (permalink / raw)
  To: newlib; +Cc: Joel Sherrill

From: Joel Sherrill <joel@rtems.org>

---
 newlib/configure.host                        |   1 +
 newlib/libc/machine/i386/sys/fenv.h          |   1 +
 newlib/libc/machine/x86_64/sys/fenv.h        | 143 ++++++++
 newlib/libm/fenv/fenv_stub.c                 |  23 ++
 newlib/libm/machine/i386/Makefile.am         |   6 +-
 newlib/libm/machine/i386/feclearexcept.c     |   1 +
 newlib/libm/machine/i386/fegetenv.c          |   1 +
 newlib/libm/machine/i386/fegetexceptflag.c   |   1 +
 newlib/libm/machine/i386/fegetround.c        |   1 +
 newlib/libm/machine/i386/feholdexcept.c      |   1 +
 newlib/libm/machine/i386/fenv.c              |   1 +
 newlib/libm/machine/i386/feraiseexcept.c     |   1 +
 newlib/libm/machine/i386/fesetenv.c          |   1 +
 newlib/libm/machine/i386/fesetexceptflag.c   |   1 +
 newlib/libm/machine/i386/fesetround.c        |   1 +
 newlib/libm/machine/i386/fetestexcept.c      |   1 +
 newlib/libm/machine/i386/feupdateenv.c       |   1 +
 newlib/libm/machine/x86_64/Makefile.am       |  20 ++
 newlib/libm/machine/x86_64/feclearexcept.c   |   1 +
 newlib/libm/machine/x86_64/fegetenv.c        |   1 +
 newlib/libm/machine/x86_64/fegetexceptflag.c |   1 +
 newlib/libm/machine/x86_64/fegetround.c      |   1 +
 newlib/libm/machine/x86_64/feholdexcept.c    |   1 +
 newlib/libm/machine/x86_64/fenv.c            | 477 +++++++++++++++++++++++++++
 newlib/libm/machine/x86_64/feraiseexcept.c   |   1 +
 newlib/libm/machine/x86_64/fesetenv.c        |   1 +
 newlib/libm/machine/x86_64/fesetexceptflag.c |   1 +
 newlib/libm/machine/x86_64/fesetround.c      |   1 +
 newlib/libm/machine/x86_64/fetestexcept.c    |   1 +
 newlib/libm/machine/x86_64/feupdateenv.c     |   1 +
 30 files changed, 692 insertions(+), 2 deletions(-)
 create mode 120000 newlib/libc/machine/i386/sys/fenv.h
 create mode 100644 newlib/libc/machine/x86_64/sys/fenv.h
 create mode 100644 newlib/libm/fenv/fenv_stub.c
 create mode 120000 newlib/libm/machine/i386/feclearexcept.c
 create mode 120000 newlib/libm/machine/i386/fegetenv.c
 create mode 120000 newlib/libm/machine/i386/fegetexceptflag.c
 create mode 120000 newlib/libm/machine/i386/fegetround.c
 create mode 120000 newlib/libm/machine/i386/feholdexcept.c
 create mode 120000 newlib/libm/machine/i386/fenv.c
 create mode 120000 newlib/libm/machine/i386/feraiseexcept.c
 create mode 120000 newlib/libm/machine/i386/fesetenv.c
 create mode 120000 newlib/libm/machine/i386/fesetexceptflag.c
 create mode 120000 newlib/libm/machine/i386/fesetround.c
 create mode 120000 newlib/libm/machine/i386/fetestexcept.c
 create mode 120000 newlib/libm/machine/i386/feupdateenv.c
 create mode 100644 newlib/libm/machine/x86_64/Makefile.am
 create mode 120000 newlib/libm/machine/x86_64/feclearexcept.c
 create mode 120000 newlib/libm/machine/x86_64/fegetenv.c
 create mode 120000 newlib/libm/machine/x86_64/fegetexceptflag.c
 create mode 120000 newlib/libm/machine/x86_64/fegetround.c
 create mode 120000 newlib/libm/machine/x86_64/feholdexcept.c
 create mode 100644 newlib/libm/machine/x86_64/fenv.c
 create mode 120000 newlib/libm/machine/x86_64/feraiseexcept.c
 create mode 120000 newlib/libm/machine/x86_64/fesetenv.c
 create mode 120000 newlib/libm/machine/x86_64/fesetexceptflag.c
 create mode 120000 newlib/libm/machine/x86_64/fesetround.c
 create mode 120000 newlib/libm/machine/x86_64/fetestexcept.c
 create mode 120000 newlib/libm/machine/x86_64/feupdateenv.c

diff --git a/newlib/configure.host b/newlib/configure.host
index 87bf78a..06b7bad 100644
--- a/newlib/configure.host
+++ b/newlib/configure.host
@@ -339,6 +339,7 @@ case "${host_cpu}" in
 	;;
   x86_64)
 	machine_dir=x86_64
+	libm_machine_dir=x86_64
 	;;
   xc16x*)
         machine_dir=xc16x
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..70620d8
--- /dev/null
+++ b/newlib/libc/machine/x86_64/sys/fenv.h
@@ -0,0 +1,143 @@
+/*
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * Copyright (c) 2010-2019 Red Hat, Inc.
+ */
+
+#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 __cplusplus
+}
+#endif
+
+#endif /* _FENV_H */
diff --git a/newlib/libm/fenv/fenv_stub.c b/newlib/libm/fenv/fenv_stub.c
new file mode 100644
index 0000000..a4eb652
--- /dev/null
+++ b/newlib/libm/fenv/fenv_stub.c
@@ -0,0 +1,23 @@
+/*
+ * SPDX-License-Identifier: BSD-2-Clause
+ * 
+ * (c) Copyright 2019 Joel Sherrill <joel@rtems.org
+ */
+
+/*
+ * This file is intentionally empty. 
+ *
+ * Newlib's build infrastructure needs a machine specific fiel to override
+ * the generic implementation in the library.  When a target
+ * implementation of the fenv.h methods puts all methods in a single file
+ * (e.g. fenv.c) or some as inline methods in its <sys/fenv.h>, it will need
+ * to override the default implementation found in a file in this directory.
+ *
+ * For each file that the target's machine directory needs to override,
+ * this file should be symbolically linked to that specific file name
+ * in the target directory. For example, the target may use fe_dfl_env.c
+ * from the default implementation but need to override all others.
+ */
+
+/* deliberately empty */
+
diff --git a/newlib/libm/machine/i386/Makefile.am b/newlib/libm/machine/i386/Makefile.am
index 6fade2d..249f876 100644
--- a/newlib/libm/machine/i386/Makefile.am
+++ b/newlib/libm/machine/i386/Makefile.am
@@ -12,8 +12,10 @@ 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 \
-	i386mach.h 
+	f_tan.S f_tanf.S f_math.h i386mach.h \
+	fenv.c feclearexcept.c fegetenv.c fegetexceptflag.c \
+	fegetround.c feholdexcept.c feraiseexcept.c fesetenv.c \
+	fesetexceptflag.c fesetround.c fetestexcept.c feupdateenv.c
 
 libi386_la_LDFLAGS = -Xcompiler -nostdlib
 
diff --git a/newlib/libm/machine/i386/feclearexcept.c b/newlib/libm/machine/i386/feclearexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/feclearexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fegetenv.c b/newlib/libm/machine/i386/fegetenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fegetenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fegetexceptflag.c b/newlib/libm/machine/i386/fegetexceptflag.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fegetexceptflag.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fegetround.c b/newlib/libm/machine/i386/fegetround.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fegetround.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/feholdexcept.c b/newlib/libm/machine/i386/feholdexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/feholdexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
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/i386/feraiseexcept.c b/newlib/libm/machine/i386/feraiseexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/feraiseexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fesetenv.c b/newlib/libm/machine/i386/fesetenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fesetenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fesetexceptflag.c b/newlib/libm/machine/i386/fesetexceptflag.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fesetexceptflag.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fesetround.c b/newlib/libm/machine/i386/fesetround.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fesetround.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/fetestexcept.c b/newlib/libm/machine/i386/fetestexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/fetestexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/i386/feupdateenv.c b/newlib/libm/machine/i386/feupdateenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/i386/feupdateenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.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..f309440
--- /dev/null
+++ b/newlib/libm/machine/x86_64/Makefile.am
@@ -0,0 +1,20 @@
+## 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 feclearexcept.c fegetenv.c fegetexceptflag.c \
+	fegetround.c feholdexcept.c feraiseexcept.c fesetenv.c \
+	fesetexceptflag.c fesetround.c fetestexcept.c feupdateenv.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/feclearexcept.c b/newlib/libm/machine/x86_64/feclearexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/feclearexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fegetenv.c b/newlib/libm/machine/x86_64/fegetenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fegetenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fegetexceptflag.c b/newlib/libm/machine/x86_64/fegetexceptflag.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fegetexceptflag.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fegetround.c b/newlib/libm/machine/x86_64/fegetround.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fegetround.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/feholdexcept.c b/newlib/libm/machine/x86_64/feholdexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/feholdexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fenv.c b/newlib/libm/machine/x86_64/fenv.c
new file mode 100644
index 0000000..ccc08e2
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fenv.c
@@ -0,0 +1,477 @@
+/*
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * Copyright (c) 2010-2019 Red Hat, Inc.
+ */
+
+#define _GNU_SOURCE        // for FE_NOMASK_ENV
+
+#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;
+
+/* Assume i686 or above (hence SSE available) these days, with the
+   compiler feels free to use it (depending on compile- time flags of
+   course), but 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 inline bool use_sse(void)
+{
+  unsigned int edx, eax;
+
+  /* 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)
+    return true;
+
+  return false;
+}
+
+/* forward declaration */
+static void _feinitialise (void);
+
+/*  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)
+{
+   if ((envp == FE_DFL_ENV || envp == FE_NOMASK_ENV) &&
+       envp->_fpu._fpu_cw == 0)
+     _feinitialise ();
+
+  __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.  */
+static void
+_feinitialise (void)
+{
+  extern fenv_t __fe_dfl_env;
+
+  /* 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);
+}
diff --git a/newlib/libm/machine/x86_64/feraiseexcept.c b/newlib/libm/machine/x86_64/feraiseexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/feraiseexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fesetenv.c b/newlib/libm/machine/x86_64/fesetenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fesetenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fesetexceptflag.c b/newlib/libm/machine/x86_64/fesetexceptflag.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fesetexceptflag.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fesetround.c b/newlib/libm/machine/x86_64/fesetround.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fesetround.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/fetestexcept.c b/newlib/libm/machine/x86_64/fetestexcept.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/fetestexcept.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
diff --git a/newlib/libm/machine/x86_64/feupdateenv.c b/newlib/libm/machine/x86_64/feupdateenv.c
new file mode 120000
index 0000000..f97d27d
--- /dev/null
+++ b/newlib/libm/machine/x86_64/feupdateenv.c
@@ -0,0 +1 @@
+../../fenv/fenv_stub.c
\ No newline at end of file
-- 
1.8.3.1

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: [PATCH v3] Add i386 and x86_64 fenv support
  2019-09-05 13:07 [PATCH v3] Add i386 and x86_64 fenv support joel
  2019-09-05 13:07 ` [PATCH v4] Add i386 and x86_64 fenv support from Cygwin joel
@ 2019-09-05 13:10 ` Joel Sherrill
  1 sibling, 0 replies; 3+ messages in thread
From: Joel Sherrill @ 2019-09-05 13:10 UTC (permalink / raw)
  To: Newlib

Ignore this cover. I forgot to save before sending. Another will arrive shortly.

On Thu, Sep 5, 2019 at 8:07 AM <joel@rtems.org> wrote:
>
> From: Joel Sherrill <joel@rtems.org>
>
> Hi
>
> This patch adds i386 and x86_64 support by migrating the Cygwin
> implementation in newlib/libm/machine directories. There were
> some mechanical changes such as converting from C++ to C,
> and converting use_sse to a function since _feinitialize() is not
> guaranteed to be called.
>
> Because of the way newlib builds, the non-functional files in
> libm/fenv/ are all compiled and put in the library. For this
> implementation, there is only an fenv.c, so every stub implementation
> and fenv.c ended up in the final libm.a. Rather than split the file
> and forcing all future implementations to do this, there is an fenv_stub.c
>
> *** BLURB HERE ***
>
> Joel Sherrill (1):
>   Add i386 and x86_64 fenv support from Cygwin.
>
>  newlib/libc/machine/i386/sys/fenv.h    |   1 +
>  newlib/libc/machine/x86_64/sys/fenv.h  | 170 ++++++++++++
>  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      | 487 +++++++++++++++++++++++++++++++++
>  6 files changed, 678 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
>
> --
> 1.8.3.1
>

^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2019-09-05 13:10 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-09-05 13:07 [PATCH v3] Add i386 and x86_64 fenv support joel
2019-09-05 13:07 ` [PATCH v4] Add i386 and x86_64 fenv support from Cygwin joel
2019-09-05 13:10 ` [PATCH v3] Add i386 and x86_64 fenv support Joel Sherrill

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).