[PATCHv2,03/11] Support for type-generic libm function implementations libm
diff mbox

Message ID 50e9ad99-870e-fbbe-89c5-02d9f4bc66bc@linux.vnet.ibm.com
State New
Headers show

Commit Message

Paul E. Murphy Aug. 5, 2016, 2:42 p.m. UTC
On 08/04/2016 11:42 AM, Joseph Myers wrote:
> On Wed, 3 Aug 2016, Paul E. Murphy wrote:
> 
>> +# Add special flags to control how the templated functions
>> +# are built.
>> +g_CPPFLAGS = -include math-type-macros.h
>> +g__fCPPFLAGS = $(g_CPPFLAGS) -DM_TYPE=M_FLOAT
>> +g__CPPFLAGS = $(g_CPPFLAGS) -DM_TYPE=M_DOUBLE
>> +g__lCPPFLAGS = $(g_CPPFLAGS) -DM_TYPE=M_LDOUBLE
> 
> Once you've got a separate header for each type, I wonder if you actually 
> need this M_FLOAT etc. enumeration, or if it would be better to -include a 
> per-type header (either before -include math-type-macros.h, or a header 
> which in turn #includes <math-type-macros.h>) and so avoid M_TYPE, M_FLOAT 
> etc.

The attached patch directly includes math-type-macros-<type>.h and removes
the M_<TYPE> macros.  Likewise, I've tried improve the math/Makefile
readability.

The only hiccup this caused was a simplification of b_clog10.c in a later
patch.  My original hope was this could be used to remove all ldbl-opt
overrides for generated functions.  Such is not feasible with the
current patches.  This is resolved by overriding and building s_clog10l.c
to maintain symbol and binary compatibility similar to what is done
now.

>> +#define CFLOAT __complex__ double
> 
> I think _Complex should be preferred over __complex__ for new code 
> (existing code is inconsistent).
> 
>> +   M_HUGE_VAL - Resolves to the type-specific variant of HUGE_VAL
> 
> You could just use (M_SUF (__builtin_huge_val) ()) rather than needing 
> type-specific headers to define this macro.
> 

Ok.

As requested earlier, I will hold off submission until we have
consensus on the entire patch series.  Likewise, once we reach
tentative agreement on this, I'll post updates to the remaining
patches.

Comments

Joseph Myers Aug. 10, 2016, 3:54 p.m. UTC | #1
On Fri, 5 Aug 2016, Paul E. Murphy wrote:

> +# Functions built through generic templates declared in gen-libm-calls
> +generated += $(foreach s,.c .S,$(call type-foreach,$(gen-libm-calls:%=%$(s))))

I don't see why you need to handle .S here (and so why you need the outer 
foreach at all).  These templates should only generate .c files.

> +/* Use a special epsilon value for IBM long double
> +   to avoid spurious overflows.  */

It's either overflows or underflows (or maybe both) depending on the 
function being built; the comment should reflect this rather than just 
saying overflows.

OK with those fixes unless anyone identifies any other issues within 48 
hours.
Paul E. Murphy Aug. 10, 2016, 4:31 p.m. UTC | #2
On 08/10/2016 10:54 AM, Joseph Myers wrote:
> On Fri, 5 Aug 2016, Paul E. Murphy wrote:
> 
>> +# Functions built through generic templates declared in gen-libm-calls
>> +generated += $(foreach s,.c .S,$(call type-foreach,$(gen-libm-calls:%=%$(s))))
> 
> I don't see why you need to handle .S here (and so why you need the outer 
> foreach at all).  These templates should only generate .c files.
> 
>> +/* Use a special epsilon value for IBM long double
>> +   to avoid spurious overflows.  */
> 
> It's either overflows or underflows (or maybe both) depending on the 
> function being built; the comment should reflect this rather than just 
> saying overflows.
> 
> OK with those fixes unless anyone identifies any other issues within 48 
> hours.
> 

Thinking on Carlos' comments regarding tst-wcstod-round, I am wondering if
maybe it wouldn't be better to add an extra make target which instead
generates a .c file based on existing naming conventions. E.g:

	/* This file is generated by 'make regen-libm-templates.  Do not edit.  */
	#include <math-type-macros-${type}>
	#include <${func_unsuffixed}_template.c>

This has the benefit of leaving the existing implicit override mechanism
in place, and no insane makefile hacks.  I would prefer them living in
their own directory maybe math/generated, ensuring they are always included
after sysdep targets.

This target would not run as part of the build process.  It would be invoked
periodically as libm supports new types, or changes are needed.

E.g placing them in math/:

regen-libm-templates:
       for gfunc in $(gen-libm-calls); do \
         gfunc_basefile=$${gfunc/F/}_template.c; \
         for type in $(foreach t,$(all-libm-types),$(t)__$(type-$(t)-suffix)); do \
           func=$${gfunc/F/$${type#*__}}; \
           type=$${type%__*}; \
           echo "/* This file is generated by 'make regen-libm-templates." >> $${func}.c; \
           echo "   Do not edit.  */" >> $${func}.c; \
           echo "#include <math-type-macros-${type}.h>" >> $${func}.c; \
           echo "#include \"$${gfunc_basefile}\"" >> $${func}.c; \
         done; \
       done;
Joseph Myers Aug. 10, 2016, 4:39 p.m. UTC | #3
On Wed, 10 Aug 2016, Paul E. Murphy wrote:

> Thinking on Carlos' comments regarding tst-wcstod-round, I am wondering if
> maybe it wouldn't be better to add an extra make target which instead
> generates a .c file based on existing naming conventions. E.g:
> 
> 	/* This file is generated by 'make regen-libm-templates.  Do not edit.  */
> 	#include <math-type-macros-${type}>
> 	#include <${func_unsuffixed}_template.c>
> 
> This has the benefit of leaving the existing implicit override mechanism
> in place, and no insane makefile hacks.  I would prefer them living in
> their own directory maybe math/generated, ensuring they are always included
> after sysdep targets.
> 
> This target would not run as part of the build process.  It would be invoked
> periodically as libm supports new types, or changes are needed.

Files with an explicit #include are fine as an alternative to generating 
special -include options.  But if you're generating them from the glibc 
build system, it would be better to generate them in the build directory 
rather than checking them in.

Patch
diff mbox

From 34567fc42c53ab6b126d46b5a41520dd34d69180 Mon Sep 17 00:00:00 2001
From: "Paul E. Murphy" <murphyp@linux.vnet.ibm.com>
Date: Tue, 14 Jun 2016 11:44:14 -0500
Subject: [PATCH 1/9] Support for type-generic libm function implementations
 libm

This defines two new classes of libm objects.  g for
generated objects, and b for basic templates to
generate those objects.  The end result should be
much less duplication, and no function changes,
other than potential shuffling of the code within
libm.

As a simple example, the trivial complex functions
are converted to use the new infrastructure.

This uses some makefile hackery to generate g_*
objects from from the base b_*.c.  The intent is to
convert most or all of the complex function wrappers,
possibly even the wrappers for the e functions.

Hooks are added to enable a platform to override a
subset of generated files and replace them with a
platform supplied list of files via the
libm-gen-replacement-calls variable.  This is a
little kludgy as the set of files being moved
to this will likely include both unprefixed and
w_ prefixed files.  The intent is to minimize
awkward file renames.

Likewise, the alias for platforms which previously
didn't have long double support in glibc is handled
for trivial cases.  The more complex cases are left
to be handled individually.

I have verified the built objects for ppc64, x86_64,
alpha, arm, and m68k do not change in any meaningful
way with these changes using the Fedora cross toolchains.
I have verified the x86_64 and ppc64 changes still run.

Verification is done by checking the objdump of
the g_*.{o,os} files with their equivalent s_
object prior to the patch.  Additionally nm is
used to verify the same set of symbols are
exported.

	* math/Makefile: Add support for routines generated
	via a macroed template file.
	(gen-libm-calls): New variable.
	(libm-calls-compat): Likewise.
	(gen-suffixes): Likewise.
	(g__CPPFLAGS): Likewise.
	(g__fCPPFLAGS): Likewise.
	(g__lCPPFLAGS): Likewise.

	* sysdeps/generic/math-type-macros.h: New file.
	* sysdeps/generic/math-type-macros-float.h: Likewise.
	* sysdeps/generic/math-type-macros-double.h: Likewise.
	* sysdeps/generic/math-type-macros-ldouble.h: Likewise.
	* sysdeps/ieee754/ldbl-opt/math-type-macros-double.h: Likewise.
	* sysdeps/ieee754/ldbl-opt/math-type-macros-ldouble.h: Likewise.

	* math/cabs.c: Refactor into
	* math/b_cabs.c: new file.
	* math/cabsf.c: Removed.
	* math/cabsl.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cabsf.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cabsl.c: Removed.

	* math/carg.c: Refactor into
	* math/b_carg.c: new file.
	* math/cargf.c: Removed.
	* math/cargl.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cargf.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cargl.c: Removed.

	* math/conj.c: Refactor into
	* math/b_conj.c: new file.
	* math/conjf.c: Removed.
	* math/conjl.c: Removed.
	* sysdeps/ieee754/ldbl-opt/conjf.c: Removed.
	* sysdeps/ieee754/ldbl-opt/conjl.c: Removed.

	* math/cimag.c: Refactor into
	* math/b_cimag.c: new file.
	* math/cimagf.c: Removed.
	* math/cimagl.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cimagf.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cimagl.c: Removed.

	* math/cimag.c: Refactor into
	* math/b_cimag.c: new file.
	* math/cimagf.c: Removed.
	* math/cimagl.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cimagf.c: Removed.
	* sysdeps/ieee754/ldbl-opt/cimagl.c: Removed.

	* sysdeps/ieee754/Makeconfig: New file.
	* Makeconfig: Move m_% sysd pattern rule into above.

	* scripts/sysd-rules.awk: Handle multiple rules
	with same DEP length correctly.
---
 Makeconfig                                         |   2 +-
 math/Makefile                                      |  62 ++++++++++-
 math/{cabs.c => b_cabs.c}                          |  17 +--
 math/{carg.c => b_carg.c}                          |  17 +--
 math/{cimag.c => b_cimag.c}                        |  15 +--
 math/{conj.c => b_conj.c}                          |  15 +--
 math/{creal.c => b_creal.c}                        |  15 +--
 math/cabsf.c                                       |  28 -----
 math/cabsl.c                                       |  28 -----
 math/cargf.c                                       |  28 -----
 math/cargl.c                                       |  28 -----
 math/cimagf.c                                      |  27 -----
 math/cimagl.c                                      |  27 -----
 math/conjf.c                                       |  27 -----
 math/conjl.c                                       |  27 -----
 math/crealf.c                                      |  27 -----
 math/creall.c                                      |  27 -----
 scripts/sysd-rules.awk                             |   5 +-
 sysdeps/alpha/fpu/Makefile                         |   4 +
 sysdeps/generic/math-type-macros-double.h          |  45 ++++++++
 .../generic/math-type-macros-float.h               |  32 +++---
 .../generic/math-type-macros-ldouble.h             |  27 ++---
 sysdeps/generic/math-type-macros.h                 | 121 +++++++++++++++++++++
 sysdeps/ieee754/Makeconfig                         |   6 +
 sysdeps/ieee754/ldbl-opt/cabs.c                    |   6 -
 sysdeps/ieee754/ldbl-opt/cabsl.c                   |   6 -
 sysdeps/ieee754/ldbl-opt/carg.c                    |   6 -
 sysdeps/ieee754/ldbl-opt/cargl.c                   |   6 -
 sysdeps/ieee754/ldbl-opt/cimag.c                   |   6 -
 sysdeps/ieee754/ldbl-opt/cimagl.c                  |   6 -
 sysdeps/ieee754/ldbl-opt/conj.c                    |   6 -
 sysdeps/ieee754/ldbl-opt/conjl.c                   |   6 -
 sysdeps/ieee754/ldbl-opt/creal.c                   |   6 -
 sysdeps/ieee754/ldbl-opt/creall.c                  |   6 -
 sysdeps/ieee754/ldbl-opt/math-type-macros-double.h |  58 ++++++++++
 .../ieee754/ldbl-opt/math-type-macros-ldouble.h    |  24 ++--
 36 files changed, 383 insertions(+), 416 deletions(-)
 rename math/{cabs.c => b_cabs.c} (77%)
 rename math/{carg.c => b_carg.c} (79%)
 copy math/{cimag.c => b_cimag.c} (82%)
 rename math/{conj.c => b_conj.c} (81%)
 rename math/{creal.c => b_creal.c} (82%)
 delete mode 100644 math/cabsf.c
 delete mode 100644 math/cabsl.c
 delete mode 100644 math/cargf.c
 delete mode 100644 math/cargl.c
 delete mode 100644 math/cimagf.c
 delete mode 100644 math/cimagl.c
 delete mode 100644 math/conjf.c
 delete mode 100644 math/conjl.c
 delete mode 100644 math/crealf.c
 delete mode 100644 math/creall.c
 create mode 100644 sysdeps/alpha/fpu/Makefile
 create mode 100644 sysdeps/generic/math-type-macros-double.h
 copy math/cimag.c => sysdeps/generic/math-type-macros-float.h (59%)
 copy math/cimag.c => sysdeps/generic/math-type-macros-ldouble.h (63%)
 create mode 100644 sysdeps/generic/math-type-macros.h
 create mode 100644 sysdeps/ieee754/Makeconfig
 delete mode 100644 sysdeps/ieee754/ldbl-opt/cabs.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/cabsl.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/carg.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/cargl.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/cimag.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/cimagl.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/conj.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/conjl.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/creal.c
 delete mode 100644 sysdeps/ieee754/ldbl-opt/creall.c
 create mode 100644 sysdeps/ieee754/ldbl-opt/math-type-macros-double.h
 rename math/cimag.c => sysdeps/ieee754/ldbl-opt/math-type-macros-ldouble.h (63%)

diff --git a/Makeconfig b/Makeconfig
index 03fd89c..8957180 100644
--- a/Makeconfig
+++ b/Makeconfig
@@ -1125,7 +1125,7 @@  endif
 # emitted into sysd-rules.  A sysdeps Makeconfig fragment can
 # add its own special object file prefix to this list with e.g. foo-%:%
 # to have foo-*.? compiled from *.? using $(foo-CPPFLAGS).
-sysd-rules-patterns := %:% rtld-%:rtld-% rtld-%:% m_%:s_%
+sysd-rules-patterns := %:% rtld-%:rtld-% rtld-%:%
 
 # Let sysdeps/ subdirs contain a Makeconfig fragment for us to include here.
 sysdep-makeconfigs := $(wildcard $(+sysdep_dirs:=/Makeconfig))
diff --git a/math/Makefile b/math/Makefile
index 925b8d4..01586ff 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -43,6 +43,30 @@  libm-support = s_lib_version s_matherr s_signgam			\
 	       fesetenv feupdateenv t_exp fedisblxcpt feenablxcpt	\
 	       fegetexcept
 
+# These functions are generated from a common source file using a
+# header file math-type-macros-<type>.h to supply implementations
+# for the type-generic macros.  These build from a common file
+# matching the name b_<func>.<source suffix>.
+#
+# Note, there is an explicit underscore between the function name
+# and the type suffix to simplify matching the rule used to build
+# the type specific variant.
+#
+# A platform can override any of these generated functions.
+# Overrides are accounted for after type suffix substitution occurs,
+# by adding a makefile fragment which defines libm-gen-calls-override
+# and libm-gen-replacement-calls for each overriden function.
+#
+# i.e:
+#
+# libm-gen-calls-override += g_cimag_
+# libm-gen-replacement-calls += cimag
+#
+# Would inhibit building the double version of cimag from the the
+# template b_cimag.c, and instead look for a buildable file matching
+# the name cimag.
+gen-libm-calls = g_carg_F g_conj_F g_cimag_F g_creal_F g_cabs_F
+
 libm-calls =								  \
 	e_acosF e_acoshF e_asinF e_atan2F e_atanhF e_coshF e_expF e_fmodF \
 	e_hypotF e_j0F e_j1F e_jnF e_lgammaF_r e_logF e_log10F e_powF	  \
@@ -59,13 +83,13 @@  libm-calls =								  \
 	w_ilogbF							  \
 	s_fpclassifyF s_fmaxF s_fminF s_fdimF s_nanF s_truncF		  \
 	s_remquoF e_log2F e_exp2F s_roundF s_nearbyintF s_sincosF	  \
-	conjF cimagF crealF cabsF cargF s_cexpF s_csinhF s_ccoshF s_clogF \
+	s_cexpF s_csinhF s_ccoshF s_clogF				  \
 	s_catanF s_casinF s_ccosF s_csinF s_ctanF s_ctanhF s_cacosF	  \
 	s_casinhF s_cacoshF s_catanhF s_csqrtF s_cpowF s_cprojF s_clog10F \
 	s_fmaF s_lrintF s_llrintF s_lroundF s_llroundF e_exp10F w_log2F	  \
 	s_issignalingF $(calls:s_%=m_%) x2y2m1F k_casinhF		  \
 	gamma_productF lgamma_negF lgamma_productF			  \
-	s_nextupF s_nextdownF
+	s_nextupF s_nextdownF $(gen-libm-calls)
 
 libm-compat-calls-ldouble-yes = w_lgamma_compatl k_standardl
 libm-compat-calls = w_lgamma_compatf w_lgamma_compat k_standard k_standardf \
@@ -105,7 +129,9 @@  type-float-routines :=
 type-foreach = $(foreach t,$(types),$(subst F,$(type-$(t)-suffix),$(1)))
 
 libm-routines = $(strip $(libm-support) $(libm-compat-calls)		\
-			$(call type-foreach, $(libm-calls))		\
+			$(filter-out $(libm-gen-calls-override),	\
+				$(call type-foreach, $(libm-calls)))	\
+			$(libm-gen-replacement-calls)			\
 			$(foreach t, $(types), $(type-$(t)-routines))) 	\
 
 # These functions are in libc instead of libm because __printf_fp
@@ -118,6 +144,9 @@  calls = s_isinfF s_isnanF s_finiteF s_copysignF s_modfF s_scalbnF s_frexpF \
 generated += $(foreach s,.c .S,$(call type-foreach, $(calls:s_%=m_%$(s))))
 routines = $(call type-foreach, $(calls))
 
+# Functions built through generic templates declared in gen-libm-calls
+generated += $(foreach s,.c .S,$(call type-foreach,$(gen-libm-calls:%=%$(s))))
+
 ifeq ($(build-mathvec),yes)
 # We need to install libm.so as linker script
 # for more comfortable use of vector math library.
@@ -303,6 +332,33 @@  endef
 object-suffixes-left := $(all-object-suffixes)
 include $(o-iterator)
 
+# Generated per-type files.
+#
+# Suffixes for each generated file.  This hackery makes it possible
+# to correctly identify name of the templated file.
+gen-suffixes := $(foreach s,$(all-object-suffixes), \
+                       $(foreach t,$(types),_$(type-$(t)-suffix)$s))
+
+# Add special flags to control how the templated functions
+# are built.
+g__fCPPFLAGS = -include math-type-macros-float.h
+g__CPPFLAGS = -include math-type-macros-double.h
+g__lCPPFLAGS = -include math-type-macros-ldouble.h
+
+# Like m_% above, this builds the generic sources.  Makeconfig
+# enumerates the sysd-rules for overriding.
+define o-iterator-doit
+$(objpfx)g_%$o: b_%.c $(before-compile); $$(compile-command.c)
+endef
+object-suffixes-left := $(gen-suffixes)
+include $(o-iterator)
+
+# Tack on the appropriate CPPFLAGS to match sysd-rules
+define o-iterator-doit
+$(objpfx)g_%$o: CPPFLAGS += $(g_$(basename $o)CPPFLAGS)
+endef
+object-suffixes-left := $(gen-suffixes)
+include $(o-iterator)
 
 # This file defines the default _LIB_VERSION variable that controls
 # the error return conventions for the math functions.
diff --git a/math/cabs.c b/math/b_cabs.c
similarity index 77%
rename from math/cabs.c
rename to math/b_cabs.c
index d7e0665..5eff1b2 100644
--- a/math/cabs.c
+++ b/math/b_cabs.c
@@ -1,4 +1,4 @@ 
-/* Return the complex absolute value of double complex value.
+/* Return the complex absolute value of complex float type.
    Copyright (C) 1997-2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -20,13 +20,14 @@ 
 #include <complex.h>
 #include <math.h>
 
-double
-__cabs (double _Complex z)
+FLOAT
+M_DECL_FUNC (__cabs) (CFLOAT z)
 {
-  return __hypot (__real__ z, __imag__ z);
+  return M_SUF (__hypot) (__real__ z, __imag__ z);
 }
-weak_alias (__cabs, cabs)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__cabs, __cabsl)
-weak_alias (__cabs, cabsl)
+
+declare_mgen_alias (__cabs, cabs)
+
+#if M_LIBM_NEED_COMPAT (cabs)
+declare_mgen_libm_compat (__cabs, cabs)
 #endif
diff --git a/math/carg.c b/math/b_carg.c
similarity index 79%
rename from math/carg.c
rename to math/b_carg.c
index 61f1e0d..6205be8 100644
--- a/math/carg.c
+++ b/math/b_carg.c
@@ -1,4 +1,4 @@ 
-/* Compute argument of complex double value.
+/* Compute argument of complex float type.
    Copyright (C) 1997-2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -20,13 +20,14 @@ 
 #include <complex.h>
 #include <math.h>
 
-double
-__carg (__complex__ double x)
+FLOAT
+M_DECL_FUNC (__carg) (CFLOAT x)
 {
-  return __atan2 (__imag__ x, __real__ x);
+  return M_SUF (__atan2) (__imag__ x, __real__ x);
 }
-weak_alias (__carg, carg)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__carg, __cargl)
-weak_alias (__carg, cargl)
+
+declare_mgen_alias (__carg, carg)
+
+#if M_LIBM_NEED_COMPAT (carg)
+declare_mgen_libm_compat (__carg, carg)
 #endif
diff --git a/math/cimag.c b/math/b_cimag.c
similarity index 82%
copy from math/cimag.c
copy to math/b_cimag.c
index 1807ac2..582147d 100644
--- a/math/cimag.c
+++ b/math/b_cimag.c
@@ -1,4 +1,4 @@ 
-/* Return imaginary part of complex double value.
+/* Return imaginary part of complex float type.
    Copyright (C) 1997-2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -19,13 +19,14 @@ 
 
 #include <complex.h>
 
-double
-__cimag (double _Complex z)
+FLOAT
+M_DECL_FUNC (__cimag) (CFLOAT z)
 {
   return __imag__ z;
 }
-weak_alias (__cimag, cimag)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__cimag, __cimagl)
-weak_alias (__cimag, cimagl)
+
+declare_mgen_alias (__cimag, cimag)
+
+#if M_LIBM_NEED_COMPAT (cimag)
+declare_mgen_libm_compat (__cimag, cimag)
 #endif
diff --git a/math/conj.c b/math/b_conj.c
similarity index 81%
rename from math/conj.c
rename to math/b_conj.c
index d282985..72d1236 100644
--- a/math/conj.c
+++ b/math/b_conj.c
@@ -1,4 +1,4 @@ 
-/* Return complex conjugate of complex double value.
+/* Return complex conjugate of complex float type.
    Copyright (C) 1997-2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -19,13 +19,14 @@ 
 
 #include <complex.h>
 
-double _Complex
-__conj (double _Complex z)
+CFLOAT
+M_DECL_FUNC (__conj) (CFLOAT z)
 {
   return ~z;
 }
-weak_alias (__conj, conj)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__conj, __conjl)
-weak_alias (__conj, conjl)
+
+declare_mgen_alias (__conj, conj)
+
+#if M_LIBM_NEED_COMPAT (conj)
+declare_mgen_libm_compat (__conj, conj)
 #endif
diff --git a/math/creal.c b/math/b_creal.c
similarity index 82%
rename from math/creal.c
rename to math/b_creal.c
index 231d3b0..f840f43 100644
--- a/math/creal.c
+++ b/math/b_creal.c
@@ -1,4 +1,4 @@ 
-/* Return real part of complex double value.
+/* Return real part of complex float type.
    Copyright (C) 1997-2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
    Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
@@ -19,13 +19,14 @@ 
 
 #include <complex.h>
 
-double
-__creal (double _Complex z)
+FLOAT
+M_DECL_FUNC (__creal) (CFLOAT z)
 {
   return __real__ z;
 }
-weak_alias (__creal, creal)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__creal, __creall)
-weak_alias (__creal, creall)
+
+declare_mgen_alias (__creal, creal)
+
+#if M_LIBM_NEED_COMPAT (creal)
+declare_mgen_libm_compat (__creal, creal)
 #endif
diff --git a/math/cabsf.c b/math/cabsf.c
deleted file mode 100644
index 431a244..0000000
--- a/math/cabsf.c
+++ /dev/null
@@ -1,28 +0,0 @@ 
-/* Return the complex absolute value of float complex value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-#include <math.h>
-
-float
-__cabsf (float _Complex z)
-{
-  return __hypotf (__real__ z, __imag__ z);
-}
-weak_alias (__cabsf, cabsf)
diff --git a/math/cabsl.c b/math/cabsl.c
deleted file mode 100644
index d87e3a2..0000000
--- a/math/cabsl.c
+++ /dev/null
@@ -1,28 +0,0 @@ 
-/* Return the complex absolute value of long double complex value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-#include <math.h>
-
-long double
-__cabsl (long double _Complex z)
-{
-  return __hypotl (__real__ z, __imag__ z);
-}
-weak_alias (__cabsl, cabsl)
diff --git a/math/cargf.c b/math/cargf.c
deleted file mode 100644
index 620db3e..0000000
--- a/math/cargf.c
+++ /dev/null
@@ -1,28 +0,0 @@ 
-/* Compute argument of complex float value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-#include <math.h>
-
-float
-__cargf (__complex__ float x)
-{
-  return __atan2f (__imag__ x, __real__ x);
-}
-weak_alias (__cargf, cargf)
diff --git a/math/cargl.c b/math/cargl.c
deleted file mode 100644
index 31b7292..0000000
--- a/math/cargl.c
+++ /dev/null
@@ -1,28 +0,0 @@ 
-/* Compute argument of complex long double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-#include <math.h>
-
-long double
-__cargl (__complex__ long double x)
-{
-  return __atan2l (__imag__ x, __real__ x);
-}
-weak_alias (__cargl, cargl)
diff --git a/math/cimagf.c b/math/cimagf.c
deleted file mode 100644
index 67c37f4..0000000
--- a/math/cimagf.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return imaginary part of complex float value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-float
-__cimagf (float _Complex z)
-{
-  return __imag__ z;
-}
-weak_alias (__cimagf, cimagf)
diff --git a/math/cimagl.c b/math/cimagl.c
deleted file mode 100644
index c1d0910..0000000
--- a/math/cimagl.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return imaginary part of complex long double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-long double
-__cimagl (long double _Complex z)
-{
-  return __imag__ z;
-}
-weak_alias (__cimagl, cimagl)
diff --git a/math/conjf.c b/math/conjf.c
deleted file mode 100644
index 3587c94..0000000
--- a/math/conjf.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return complex conjugate of complex float value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-float _Complex
-__conjf (float _Complex z)
-{
-  return ~z;
-}
-weak_alias (__conjf, conjf)
diff --git a/math/conjl.c b/math/conjl.c
deleted file mode 100644
index 55bb393..0000000
--- a/math/conjl.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return complex conjugate of complex long double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-long double _Complex
-__conjl (long double _Complex z)
-{
-  return ~z;
-}
-weak_alias (__conjl, conjl)
diff --git a/math/crealf.c b/math/crealf.c
deleted file mode 100644
index 5883849..0000000
--- a/math/crealf.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return real part of complex float value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-float
-__crealf (float _Complex z)
-{
-  return __real__ z;
-}
-weak_alias (__crealf, crealf)
diff --git a/math/creall.c b/math/creall.c
deleted file mode 100644
index 4a31557..0000000
--- a/math/creall.c
+++ /dev/null
@@ -1,27 +0,0 @@ 
-/* Return real part of complex long double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
-   This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
-
-   The GNU C Library is free software; you can redistribute it and/or
-   modify it under the terms of the GNU Lesser General Public
-   License as published by the Free Software Foundation; either
-   version 2.1 of the License, or (at your option) any later version.
-
-   The GNU C Library is distributed in the hope that it will be useful,
-   but WITHOUT ANY WARRANTY; without even the implied warranty of
-   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
-   Lesser General Public License for more details.
-
-   You should have received a copy of the GNU Lesser General Public
-   License along with the GNU C Library; if not, see
-   <http://www.gnu.org/licenses/>.  */
-
-#include <complex.h>
-
-long double
-__creall (long double _Complex z)
-{
-  return __real__ z;
-}
-weak_alias (__creall, creall)
diff --git a/scripts/sysd-rules.awk b/scripts/sysd-rules.awk
index 69af400..ccf1dff 100644
--- a/scripts/sysd-rules.awk
+++ b/scripts/sysd-rules.awk
@@ -18,6 +18,9 @@  BEGIN {
   # They are no in particular order.  We need to sort them so that
   # the longest TARGET is first, and, among elements with the same
   # TARGET, the longest DEP is first.
+  #
+  # If two TARGETs have the same length DEP, the first TARGET is
+  # always listed first.
   for (i = 1; i <= npatterns; ++i) {
     if (split(patterns[i], td, ":") != 2) {
       msg = "bad sysd-rules-patterns element '" patterns[i] "'";
@@ -25,7 +28,7 @@  BEGIN {
       exit 2;
     }
     target_order = sprintf("%09d", npatterns + 1 - length(td[1]));
-    dep_order = sprintf("%09d", npatterns - length(td[2]));
+    dep_order = sprintf("%09d%03d", npatterns - length(td[2]), i);
     sort_patterns[target_order SUBSEP dep_order] = patterns[i];
   }
   asorti(sort_patterns, map_patterns);
diff --git a/sysdeps/alpha/fpu/Makefile b/sysdeps/alpha/fpu/Makefile
new file mode 100644
index 0000000..afdd620
--- /dev/null
+++ b/sysdeps/alpha/fpu/Makefile
@@ -0,0 +1,4 @@ 
+# Inhibit generation of these math files as they are supplied via
+# the respectic s_*.c versions here.
+libm-gen-calls-override += g_cabs_f g_carg_f g_cimag_f g_creal_f g_conj_f
+libm-gen-replacement-calls += cabsf cargf cimagf crealf conjf
diff --git a/sysdeps/generic/math-type-macros-double.h b/sysdeps/generic/math-type-macros-double.h
new file mode 100644
index 0000000..4149b39
--- /dev/null
+++ b/sysdeps/generic/math-type-macros-double.h
@@ -0,0 +1,45 @@ 
+/* Helper macros for double variants of type generic functions of libm.
+   Copyright (C) 2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _MATH_TYPE_MACROS_DOUBLE
+#define _MATH_TYPE_MACROS_DOUBLE
+
+#define M_LIT(c) c
+#define M_MLIT(c) c
+#define M_PFX DBL
+#define M_SUF(c) c
+#define FLOAT double
+#define CFLOAT _Complex double
+
+/* Machines without a distinct long double type
+   alias long double functions to their double
+   equivalent.  */
+#if defined NO_LONG_DOUBLE
+# define declare_mgen_alias(from, to)	    \
+   weak_alias (from, to)		    \
+   strong_alias (from, from ## l)	    \
+   weak_alias (from, to ## l)
+#else
+# define declare_mgen_alias(from, to)	    \
+   weak_alias (M_SUF (from), M_SUF (to))
+#endif
+
+/* Supply the generic macros.  */
+#include <math-type-macros.h>
+
+#endif
diff --git a/math/cimag.c b/sysdeps/generic/math-type-macros-float.h
similarity index 59%
copy from math/cimag.c
copy to sysdeps/generic/math-type-macros-float.h
index 1807ac2..aab88fd 100644
--- a/math/cimag.c
+++ b/sysdeps/generic/math-type-macros-float.h
@@ -1,7 +1,6 @@ 
-/* Return imaginary part of complex double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
+/* Helper macros for float variants of type generic functions of libm.
+   Copyright (C) 2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -17,15 +16,20 @@ 
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-#include <complex.h>
-
-double
-__cimag (double _Complex z)
-{
-  return __imag__ z;
-}
-weak_alias (__cimag, cimag)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__cimag, __cimagl)
-weak_alias (__cimag, cimagl)
+#ifndef _MATH_TYPE_MACROS_FLOAT
+#define _MATH_TYPE_MACROS_FLOAT
+
+#define M_LIT(c) c ## f
+#define M_PFX FLT
+#define M_SUF(c) c ## f
+#define FLOAT float
+#define CFLOAT _Complex float
+
+/* Standard/GNU macro literals do not exist for the float type.  Use
+   the double macro constants.  */
+#define M_MLIT(c) c
+
+/* Supply the generic macros.  */
+#include <math-type-macros.h>
+
 #endif
diff --git a/math/cimag.c b/sysdeps/generic/math-type-macros-ldouble.h
similarity index 63%
copy from math/cimag.c
copy to sysdeps/generic/math-type-macros-ldouble.h
index 1807ac2..90773fa 100644
--- a/math/cimag.c
+++ b/sysdeps/generic/math-type-macros-ldouble.h
@@ -1,7 +1,6 @@ 
-/* Return imaginary part of complex double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
+/* Helper macros for long double variants of type generic functions of libm.
+   Copyright (C) 2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -17,15 +16,17 @@ 
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-#include <complex.h>
+#ifndef _MATH_TYPE_MACROS_LDOUBLE
+#define _MATH_TYPE_MACROS_LDOUBLE
+
+#define M_LIT(c) c ## L
+#define M_MLIT(c) c ## l
+#define M_PFX LDBL
+#define M_SUF(c) c ## l
+#define FLOAT long double
+#define CFLOAT _Complex long double
+
+/* Supply the generic macros.  */
+#include <math-type-macros.h>
 
-double
-__cimag (double _Complex z)
-{
-  return __imag__ z;
-}
-weak_alias (__cimag, cimag)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__cimag, __cimagl)
-weak_alias (__cimag, cimagl)
 #endif
diff --git a/sysdeps/generic/math-type-macros.h b/sysdeps/generic/math-type-macros.h
new file mode 100644
index 0000000..91394ba
--- /dev/null
+++ b/sysdeps/generic/math-type-macros.h
@@ -0,0 +1,121 @@ 
+/* Helper macros for type generic function implementations within libm.
+   Copyright (C) 2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _MATH_TYPE_MACROS
+#define _MATH_TYPE_MACROS
+
+/* Each type imports a header which is expected to
+   define:
+
+   M_LIT(x)   - Paste the type specific suffix onto the constant x.
+   M_MLIT(x)  - Paste the type specific suffix used by the macro
+		constants in math.h, i.e M_PI or M_PIl.
+   M_PFX      - The prefixed used by float.h macros like FLT_MANT_DIG.
+   M_SUF(x)   - Paste the the type specific suffix used by functions
+		i.e expf expl exp.
+   FLOAT      - Resolves to the C typename of M_TYPE.
+   CFLOAT     - Resolves to the complex typename of M_TYPE.
+
+  Optionally, these headers may inject a non-standard
+  definition for the following:
+
+  declare_mgen_alias(from,to)
+      This exposes the appropriate symbol(s) for a
+      function f of type FLOAT.
+
+  M_LIBM_NEED_COMPAT(func)
+      This is utilized in macro context to indicate
+      whether func should declare compat symbols.
+
+  declare_mgen_libm_compat(from,to)
+      This is used in conjunction with the above macro
+      outside of macro context to paste whatever is
+      required to generate a compat symbol. */
+
+#ifndef M_PFX
+# error "M_PFX must be defined."
+#endif
+#ifndef M_LIT
+# error "M_LIT must be defined."
+#endif
+#ifndef M_MLIT
+# error "M_MLIT must be defined."
+#endif
+#ifndef M_SUF
+# error "M_SUF must be defined."
+#endif
+#ifndef FLOAT
+# error "FLOAT must be defined."
+#endif
+#ifndef CFLOAT
+# error "CFLOAT must be defined."
+#endif
+
+#define __M_CONCAT(a,b) a ## b
+#define __M_CONCATX(a,b) __M_CONCAT(a,b)
+
+#define M_NAN M_SUF (__nan) ("")
+#define M_MAX_EXP __M_CONCATX (M_PFX, _MAX_EXP)
+#define M_MIN __M_CONCATX (M_PFX, _MIN)
+#define M_MAX __M_CONCATX (M_PFX, _MAX)
+#define M_MANT_DIG __M_CONCATX (M_PFX, _MANT_DIG)
+#define M_HUGE_VAL (M_SUF (__builtin_huge_val) ())
+
+/* Helper macros for commonly used functions.  */
+#define M_COPYSIGN M_SUF (__copysign)
+#define M_FABS M_SUF (fabs)
+#define M_SINCOS M_SUF (__sincos)
+#define M_SCALBN M_SUF (__scalbn)
+#define M_LOG1P M_SUF (__log1p)
+
+#define M_ATAN2 M_SUF (__ieee754_atan2)
+#define M_COSH M_SUF (__ieee754_cosh)
+#define M_EXP M_SUF (__ieee754_exp)
+#define M_HYPOT M_SUF (__ieee754_hypot)
+#define M_LOG M_SUF (__ieee754_log)
+#define M_SINH M_SUF (__ieee754_sinh)
+#define M_SQRT M_SUF (__ieee754_sqrt)
+
+/* Needed to evaluate M_MANT_DIG below.  */
+#include <float.h>
+
+/* Use a special epsilon value for IBM long double
+   to avoid spurious overflows.  */
+#if M_MANT_DIG != 106
+# define M_EPSILON __M_CONCATX (M_PFX, _EPSILON)
+#else
+# define M_EPSILON M_LIT (0x1p-106)
+#endif
+
+/* Enable overloading of function name to assist reuse.  */
+#ifndef M_DECL_FUNC
+# define M_DECL_FUNC(f) M_SUF (f)
+#endif
+
+/* If the type does not declare special aliasing, use the default.  */
+#ifndef declare_mgen_alias
+# define declare_mgen_alias(from, to) weak_alias (M_SUF (from), M_SUF (to))
+#endif
+
+/* Do not generate anything for compat symbols by default.  */
+#ifndef M_LIBM_NEED_COMPAT
+# define M_LIBM_NEED_COMPAT(func) 0
+# define declare_mgen_libm_compat(from, to)
+#endif
+
+#endif /* _MATH_TYPE_MACROS */
diff --git a/sysdeps/ieee754/Makeconfig b/sysdeps/ieee754/Makeconfig
new file mode 100644
index 0000000..15fd238
--- /dev/null
+++ b/sysdeps/ieee754/Makeconfig
@@ -0,0 +1,6 @@ 
+# Add the rules for generated math files in sysdeps
+# Not the perfect place, but close enough.
+#
+# Each type supported by libm should have a g_%_SUFFIX
+# pattern to match its suffix.
+sysd-rules-patterns += m_%:s_% g_%_f:b_% g_%_l:b_% g_%_:b_%
diff --git a/sysdeps/ieee754/ldbl-opt/cabs.c b/sysdeps/ieee754/ldbl-opt/cabs.c
deleted file mode 100644
index a181de2..0000000
--- a/sysdeps/ieee754/ldbl-opt/cabs.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#include <math/cabs.c>
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
-compat_symbol (libm, __cabs, cabsl, GLIBC_2_1);
-#endif
diff --git a/sysdeps/ieee754/ldbl-opt/cabsl.c b/sysdeps/ieee754/ldbl-opt/cabsl.c
deleted file mode 100644
index b861633..0000000
--- a/sysdeps/ieee754/ldbl-opt/cabsl.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#undef weak_alias
-#define weak_alias(n,a)
-#include <math/cabsl.c>
-long_double_symbol (libm, __cabsl, cabsl);
diff --git a/sysdeps/ieee754/ldbl-opt/carg.c b/sysdeps/ieee754/ldbl-opt/carg.c
deleted file mode 100644
index 2ed3581..0000000
--- a/sysdeps/ieee754/ldbl-opt/carg.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#include <math/carg.c>
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
-compat_symbol (libm, __carg, cargl, GLIBC_2_1);
-#endif
diff --git a/sysdeps/ieee754/ldbl-opt/cargl.c b/sysdeps/ieee754/ldbl-opt/cargl.c
deleted file mode 100644
index 952dc60..0000000
--- a/sysdeps/ieee754/ldbl-opt/cargl.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#undef weak_alias
-#define weak_alias(n,a)
-#include <math/cargl.c>
-long_double_symbol (libm, __cargl, cargl);
diff --git a/sysdeps/ieee754/ldbl-opt/cimag.c b/sysdeps/ieee754/ldbl-opt/cimag.c
deleted file mode 100644
index f805258..0000000
--- a/sysdeps/ieee754/ldbl-opt/cimag.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#include <math/cimag.c>
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
-compat_symbol (libm, __cimag, cimagl, GLIBC_2_1);
-#endif
diff --git a/sysdeps/ieee754/ldbl-opt/cimagl.c b/sysdeps/ieee754/ldbl-opt/cimagl.c
deleted file mode 100644
index 112365e..0000000
--- a/sysdeps/ieee754/ldbl-opt/cimagl.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#undef weak_alias
-#define weak_alias(n,a)
-#include <math/cimagl.c>
-long_double_symbol (libm, __cimagl, cimagl);
diff --git a/sysdeps/ieee754/ldbl-opt/conj.c b/sysdeps/ieee754/ldbl-opt/conj.c
deleted file mode 100644
index e4edade..0000000
--- a/sysdeps/ieee754/ldbl-opt/conj.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#include <math/conj.c>
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
-compat_symbol (libm, __conj, conjl, GLIBC_2_1);
-#endif
diff --git a/sysdeps/ieee754/ldbl-opt/conjl.c b/sysdeps/ieee754/ldbl-opt/conjl.c
deleted file mode 100644
index c98e0ed..0000000
--- a/sysdeps/ieee754/ldbl-opt/conjl.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#undef weak_alias
-#define weak_alias(n,a)
-#include <math/conjl.c>
-long_double_symbol (libm, __conjl, conjl);
diff --git a/sysdeps/ieee754/ldbl-opt/creal.c b/sysdeps/ieee754/ldbl-opt/creal.c
deleted file mode 100644
index 0d1c93e..0000000
--- a/sysdeps/ieee754/ldbl-opt/creal.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#include <math/creal.c>
-#if LONG_DOUBLE_COMPAT(libm, GLIBC_2_1)
-compat_symbol (libm, __creal, creall, GLIBC_2_1);
-#endif
diff --git a/sysdeps/ieee754/ldbl-opt/creall.c b/sysdeps/ieee754/ldbl-opt/creall.c
deleted file mode 100644
index 68fedd4..0000000
--- a/sysdeps/ieee754/ldbl-opt/creall.c
+++ /dev/null
@@ -1,6 +0,0 @@ 
-#include <complex.h>
-#include <math_ldbl_opt.h>
-#undef weak_alias
-#define weak_alias(n,a)
-#include <math/creall.c>
-long_double_symbol (libm, __creall, creall);
diff --git a/sysdeps/ieee754/ldbl-opt/math-type-macros-double.h b/sysdeps/ieee754/ldbl-opt/math-type-macros-double.h
new file mode 100644
index 0000000..8cb5694
--- /dev/null
+++ b/sysdeps/ieee754/ldbl-opt/math-type-macros-double.h
@@ -0,0 +1,58 @@ 
+/* Overrides for ldbl-opt versioning for double types.
+   Copyright (C) 2016 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _MATH_TYPE_MACROS_DOUBLE
+
+#include <math_ldbl_opt.h>
+
+#define LDOUBLE_cabsl_libm_version GLIBC_2_1
+#define LDOUBLE_cargl_libm_version GLIBC_2_1
+#define LDOUBLE_cimagl_libm_version GLIBC_2_1
+#define LDOUBLE_conjl_libm_version GLIBC_2_1
+#define LDOUBLE_creall_libm_version GLIBC_2_1
+#define LDOUBLE_cacosl_libm_version GLIBC_2_1
+#define LDOUBLE_cacoshl_libm_version GLIBC_2_1
+#define LDOUBLE_ccosl_libm_version GLIBC_2_1
+#define LDOUBLE_ccoshl_libm_version GLIBC_2_1
+#define LDOUBLE_casinl_libm_version GLIBC_2_1
+#define LDOUBLE_csinl_libm_version GLIBC_2_1
+#define LDOUBLE_casinhl_libm_version GLIBC_2_1
+#define LDOUBLE_csinhl_libm_version GLIBC_2_1
+#define LDOUBLE_catanl_libm_version GLIBC_2_1
+#define LDOUBLE_catanhl_libm_version GLIBC_2_1
+#define LDOUBLE_ctanl_libm_version GLIBC_2_1
+#define LDOUBLE_ctanhl_libm_version GLIBC_2_1
+#define LDOUBLE_cexpl_libm_version GLIBC_2_1
+#define LDOUBLE_clogl_libm_version GLIBC_2_1
+#define LDOUBLE_cprojl_libm_version GLIBC_2_1
+#define LDOUBLE_csqrtl_libm_version GLIBC_2_1
+#define LDOUBLE_cpowl_libm_version GLIBC_2_1
+#define LDOUBLE_clog10l_libm_version GLIBC_2_1
+#define LDOUBLE___clog10l_libm_version GLIBC_2_1
+
+/* Define compat symbols for long double on platforms
+   where it was not always a distinct type.  */
+#define M_LIBM_NEED_COMPAT(f) \
+  LONG_DOUBLE_COMPAT (libm, LDOUBLE_ ## f ## l_libm_version)
+
+#define declare_mgen_libm_compat(from, to)	      \
+  compat_symbol (libm, from, to ## l,		      \
+		 LDOUBLE_ ## to ## l_libm_version);
+
+#include_next <math-type-macros-double.h>
+#endif
diff --git a/math/cimag.c b/sysdeps/ieee754/ldbl-opt/math-type-macros-ldouble.h
similarity index 63%
rename from math/cimag.c
rename to sysdeps/ieee754/ldbl-opt/math-type-macros-ldouble.h
index 1807ac2..d2af4bb 100644
--- a/math/cimag.c
+++ b/sysdeps/ieee754/ldbl-opt/math-type-macros-ldouble.h
@@ -1,7 +1,6 @@ 
-/* Return imaginary part of complex double value.
-   Copyright (C) 1997-2016 Free Software Foundation, Inc.
+/* Overrides for ldbl-opt versioning for long double types.
+   Copyright (C) 2016 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
-   Contributed by Ulrich Drepper <drepper@cygnus.com>, 1997.
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -17,15 +16,14 @@ 
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-#include <complex.h>
+#ifndef _MATH_TYPE_MACROS_LDOUBLE
 
-double
-__cimag (double _Complex z)
-{
-  return __imag__ z;
-}
-weak_alias (__cimag, cimag)
-#ifdef NO_LONG_DOUBLE
-strong_alias (__cimag, __cimagl)
-weak_alias (__cimag, cimagl)
+#include <math_ldbl_opt.h>
+
+/* Use properly versioned symbols for long double on platforms where
+   it was not always a distinct type.  */
+#define declare_mgen_alias(from, to) \
+  long_double_symbol (libm, from ## l, to ## l);
+
+#include_next <math-type-macros-ldouble.h>
 #endif
-- 
2.4.11