Merge master code 2018-10-25 to amd-staging-rel1.3 branch

amd · Jan 1, 2019 · 10c9201 · 10c9201
2 parents f92e789 + 550cdcf
commit 10c9201
Show file tree

Hide file tree

Showing 10 changed files with 92 additions and 207 deletions.
diff --git a/Makefile b/Makefile
@@ -3,7 +3,7 @@
 # Makefile
 #
 # Field G. Van Zee
-# 
+#
 # Top-level makefile for libflame linear algebra library.
 #
 #
@@ -89,18 +89,18 @@ ALL_FLAMEC_DLL_NAME            := libflame.so
 MK_ALL_FLAMEC_LIB                     := $(BASE_LIB_DIR)/$(ALL_FLAMEC_LIB_NAME)
 MK_ALL_FLAMEC_DLL                     := $(BASE_LIB_DIR)/$(ALL_FLAMEC_DLL_NAME)
 
-MK_BASE_FLAMEC_SRC                    := 
-MK_BASE_FLAMEC_OBJS                   := 
+MK_BASE_FLAMEC_SRC                    :=
+MK_BASE_FLAMEC_OBJS                   :=
 
-MK_BLAS_FLAMEC_SRC                    := 
-MK_BLAS_FLAMEC_OBJS                   := 
+MK_BLAS_FLAMEC_SRC                    :=
+MK_BLAS_FLAMEC_OBJS                   :=
 
-MK_LAPACK_FLAMEC_SRC                  := 
-MK_LAPACK_FLAMEC_OBJS                 := 
+MK_LAPACK_FLAMEC_SRC                  :=
+MK_LAPACK_FLAMEC_OBJS                 :=
 
 
-MK_MAP_LAPACK2FLAMEC_SRC              := 
-MK_MAP_LAPACK2FLAMEC_OBJS             := 
+MK_MAP_LAPACK2FLAMEC_SRC              :=
+MK_MAP_LAPACK2FLAMEC_OBJS             :=
 
 MK_MAP_LAPACK2FLAMEC_F2C_SRC          :=
 MK_MAP_LAPACK2FLAMEC_F2C_OBJS         :=
@@ -111,8 +111,8 @@ MK_MAP_LAPACK2FLAMEC_F2C_FLAMEC_OBJS  :=
 MK_MAP_LAPACK2FLAMEC_F2C_INSTALL_SRC  :=
 MK_MAP_LAPACK2FLAMEC_F2C_INSTALL_OBJS :=
 
-MK_FLABLAS_F2C_SRC                    := 
-MK_FLABLAS_F2C_OBJS                   := 
+MK_FLABLAS_F2C_SRC                    :=
+MK_FLABLAS_F2C_OBJS                   :=
 
 # --- Define install target names for static libraries ---
 MK_FLAMEC_LIBS                    := $(MK_ALL_FLAMEC_LIB)
@@ -242,6 +242,7 @@ FFLAGS          := $(FFLAGS) $(INCLUDE_PATHS)
 # Convert source file paths to object file paths by replaying the base source
 # directory with the base object directory, and also replacing the source file
 # suffix (ie: '.c' or '.f') with '.o'.
+
 MK_FLABLAS_F2C_OBJS                   := $(patsubst $(SRC_DIR)/%.c, $(BASE_OBJ_DIR)/%.o, \
                                                     $(filter %.c, $(MK_FLABLAS_F2C_SRC)))
 
@@ -269,7 +270,7 @@ MK_MAP_LAPACK2FLAMEC_F2C_INSTALL_OBJS := $(patsubst $(SRC_DIR)/%.c, $(BASE_OBJ_D
 # Combine the base, blas, and lapack libraries.
 MK_ALL_FLAMEC_OBJS        := $(MK_BASE_FLAMEC_OBJS) \
                              $(MK_BLAS_FLAMEC_OBJS) \
-                             $(MK_LAPACK_FLAMEC_OBJS) 
+                             $(MK_LAPACK_FLAMEC_OBJS)
 
 # Prepend the flablas source code files, if requested
 
@@ -288,13 +289,13 @@ ifeq ($(FLA_ENABLE_LAPACK2FLAME),no)
 MK_FLABLAS_F2C_OBJS       := $(MK_MAP_LAPACK2FLAMEC_F2C_INSTALL_OBJS) \
                              $(MK_FLABLAS_F2C_OBJS)
 endif
-
 MK_ALL_FLAMEC_OBJS        := $(MK_FLABLAS_F2C_OBJS) \
                              $(MK_ALL_FLAMEC_OBJS)
 endif
 
 ### Kyungjoo 2015.10.21
-AR_CHUNK_SIZE=4096
+#AR_CHUNK_SIZE=4096
+AR_CHUNK_SIZE=1024
 
 #
 # --- Targets/rules ------------------------------------------------------------
@@ -373,63 +374,88 @@ endif
 
 # --- Static library archiver rules for libflame ---
 $(MK_ALL_FLAMEC_LIB): $(MK_ALL_FLAMEC_OBJS)
-
 ifeq ($(FLA_ENABLE_VERBOSE_MAKE_OUTPUT),yes)
 ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
 ### Kyungjoo 2015.10.21
 	$(CAT) $(AR_OBJ_LIST_FILE) | xargs -n$(AR_CHUNK_SIZE) $(AR) $(ARFLAGS) $@
-### Previous hack (works on linux, not on osx)
+### Previous hack (works on linux, not on osx; osx's ar does not support @file)
 #	echo $(ARFLAGS) $@ > $(AR_ARG_LIST_FILE)
 #	$(CAT) $(AR_OBJ_LIST_FILE) >> $(AR_ARG_LIST_FILE)
 #	$(AR) @$(AR_ARG_LIST_FILE)
-	$(RANLIB) $@
 else
-	$(AR) $(ARFLAGS) $@ $?
-	$(RANLIB) $@
+#	NOTE: Can't use $^ automatic variable as long as $(AR_OBJ_LIST_FILE) is in
+#	the list of prerequisites.
+	$(AR) $(ARFLAGS) $@ $^
 endif
+	$(RANLIB) $@
 	mkdir -p include_local
 	cp -f $(MK_HEADER_FILES) include_local
 else
 	@echo "Archiving $@"
 ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
 ### Kyungjoo 2015.10.21
 	@$(CAT) $(AR_OBJ_LIST_FILE) | xargs -n$(AR_CHUNK_SIZE) $(AR) $(ARFLAGS) $@
-### Previous hack (works on linux, not on osx)
+### Previous hack (works on linux, not on osx; osx's ar does not support @file)
 #	@echo $(ARFLAGS) $@ > $(AR_ARG_LIST_FILE)
 #	@$(CAT) $(AR_OBJ_LIST_FILE) >> $(AR_ARG_LIST_FILE)
 #	@$(AR) @$(AR_ARG_LIST_FILE)
-	@$(RANLIB) $@
 else
-	@$(AR) $(ARFLAGS) $@ $?
-	@$(RANLIB) $@
+#	NOTE: Can't use $^ automatic variable as long as $(AR_OBJ_LIST_FILE) is in
+#	the list of prerequisites.
+	@$(AR) $(ARFLAGS) $@ $^
 endif
+	@$(RANLIB) $@
 	@mkdir -p include_local
 	@cp -f $(MK_HEADER_FILES) include_local
 endif
 
 
-
 # --- Dynamic library linker rules for libflame ---
 $(MK_ALL_FLAMEC_DLL): $(MK_ALL_FLAMEC_OBJS)
 ifeq ($(FLA_ENABLE_VERBOSE_MAKE_OUTPUT),yes)
 ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
-	$(file > [email protected],$^)
-	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @[email protected]
-	$(RM) [email protected]
+	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @$(AR_OBJ_LIST_FILE)
 else
-	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $?
+#	NOTE: Can't use $^ automatic variable as long as $(AR_OBJ_LIST_FILE) is in
+#	the list of prerequisites.
+	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $^
 endif
 else
 	@echo "Dynamically linking $@"
 ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
-	@$(file > [email protected],$^)
-	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @[email protected]
-	@$(RM) [email protected]
+	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @$(AR_OBJ_LIST_FILE)
 else
-	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $?
+#	NOTE: Can't use $^ automatic variable as long as $(AR_OBJ_LIST_FILE) is in
+#	the list of prerequisites.
+	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $^
 endif
 endif
 
+# Original implementation of the rule above.
+# FGVZ: This rule has been observed to not work on at least one system, where
+# it appears the ".in" file is not fully written out, or written out at all,
+# prior to the shared library link command being executed.
+#$(MK_ALL_FLAMEC_DLL): $(MK_ALL_FLAMEC_OBJS)
+#ifeq ($(FLA_ENABLE_VERBOSE_MAKE_OUTPUT),yes)
+#ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
+#	$(file > [email protected],$^)
+#	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @[email protected]
+#	$(RM) [email protected]
+#else
+#	$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $^
+#endif
+#else
+#	@echo "Dynamically linking $@"
+#ifeq ($(FLA_ENABLE_MAX_ARG_LIST_HACK),yes)
+#	@$(file > [email protected],$^)
+#	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ @[email protected]
+#	@$(RM) [email protected]
+#else
+#	@$(LINKER) -shared -Wl,-soname,libflame.so $(LDFLAGS) -o $@ $^
+#endif
+#endif
+
+
 
 
 # --- Test suite rules ---
@@ -533,11 +559,10 @@ endif
 
 # --- Clean rules ---
 cleanmost: check-config
-	- $(FIND) $(BASE_OBJ_DIR) -name "*.o" | $(XARGS) $(RM_F) 
-	- $(FIND) $(BASE_LIB_DIR) -name "*.a" | $(XARGS) $(RM_F) 
-	- $(FIND) $(BASE_LIB_DIR) -name "*.so" | $(XARGS) $(RM_F) 
+	- $(FIND) $(BASE_OBJ_DIR) -name "*.o" | $(XARGS) $(RM_F)
+	- $(FIND) $(BASE_LIB_DIR) -name "*.a" | $(XARGS) $(RM_F)
+	- $(FIND) $(BASE_LIB_DIR) -name "*.so" | $(XARGS) $(RM_F)
 	- $(RM_F) $(AR_OBJ_LIST_FILE)
-#	- $(RM_F) $(AR_ARG_LIST_FILE)
 	- $(RM_F) $(INCLUDE_LOCAL)/*.h
 
 distclean: check-config cleanmost cleanmk
@@ -551,7 +576,7 @@ distclean: check-config cleanmost cleanmk
 	- $(RM_RF) config.sys_type
 
 cleanmk: check-config
-	- $(FIND) $(SRC_DIR) -name "$(FRAGMENT_MK)" | $(XARGS) $(RM_F) 
+	- $(FIND) $(SRC_DIR) -name "$(FRAGMENT_MK)" | $(XARGS) $(RM_F)
 
 cleanleaves: check-config
 	- $(FIND) $(SRC_DIR) -name "*.[osx]" | $(XARGS) $(RM_F)

diff --git a/README → README.md b/README → README.md
@@ -1,9 +1,4 @@
-
-libflame library
-README
----
-
-Thank you for deciding to try out the libflame library!
+# libflame library
 
 libflame is a portable library for dense matrix computations, providing
 much of the functionality present in LAPACK. In fact, libflame includes
@@ -30,18 +25,18 @@ source. You may also find a nightly build of the document here:
 You can keep in touch with developers and other users of the project by
 joining the following mailing list:
 
-  o libflame-discuss - http://groups.google.com/group/libflame-discuss
+ - libflame-discuss - http://groups.google.com/group/libflame-discuss
     Please join and post to this mailing list if you have general questions
     or feedback regarding libflame.
 
-  o (NOTE: We do not yet have a developer's mailing list. For now,
+ -  (NOTE: We do not yet have a developer's mailing list. For now,
     developers and those with developer-related questions can simply post
     to libflame-discuss.
 
 Also, please read the LICENSE file for information on copying and
 distributing this software.
 
-Thanks again for your interest in libflame!
+Thanks for your interest in libflame!
 
 Regards,
 

diff --git a/build/ac-macros/fla_check_with_extra_cflags.m4 b/build/ac-macros/fla_check_with_extra_cflags.m4
diff --git a/build/ac-macros/fla_set_c_sse_flags.m4 b/build/ac-macros/fla_set_c_sse_flags.m4
@@ -15,6 +15,10 @@ AC_DEFUN([FLA_SET_C_SSE_FLAGS],
 		gcc)
 			fla_c_sse_flags='-msse3'
 		;;
+		dnl Clang.
+		clang)
+			fla_c_sse_flags='-msse3'
+		;;
 		dnl PathScale pathcc
 		pathcc)
 			fla_c_sse_flags='unknown'

diff --git a/build/config.mk.in b/build/config.mk.in
@@ -38,6 +38,11 @@ FLA_ENABLE_BUILTIN_BLAS        := @fla_enable_builtin_blas@
 FLA_ENABLE_LAPACK2FLAME        := @fla_enable_lapack2flame@
 FLA_MULTITHREADING_MODEL       := @fla_multithreading_model@
 
+# Alias the variable 'V' to the flag for verbosity.
+ifeq ($(V),1)
+FLA_ENABLE_VERBOSE_MAKE_OUTPUT := yes
+endif
+
 
 
 
@@ -90,26 +95,15 @@ endif
 
 # Aggregate all of the flags into two groups: one for optimizable code, and
 # one for code that should not be optimized.
-CFLAGS       := $(CDBGFLAGS) $(COPTFLAGS) $(CVECFLAGS) $(CWARNFLAGS) $(CMISCFLAGS) $(CPPROCFLAGS)
-CFLAGS_NOOPT := $(CDBGFLAGS) $(CWARNFLAGS) $(CMISCFLAGS) $(CPPROCFLAGS)
-
-# If the user requested that he supplement configure's CFLAGS with this own,
-# add them in.
-# *** Notice that we ALSO modify the 'no optimization' set of flags.
-#     THIS IS DANGEROUS. THE USER MUST TAKE CARE NOT TO ADD "EXTRA FLAGS"
-#     THAT WOULD DISRUPT THE NUMERICAL PROPERTIES OF SLAMCH/DLAMCH. ***
-ifeq (@fla_with_extra_cflags@,yes)
-CFLAGS       += @fla_extra_cflags@
-CFLAGS_NOOPT += @fla_extra_cflags@
-endif
+CFLAGS       := $(strip $(CDBGFLAGS) $(COPTFLAGS) $(CVECFLAGS) $(CWARNFLAGS) $(CMISCFLAGS) $(CPPROCFLAGS))
+CFLAGS_NOOPT := $(strip $(CDBGFLAGS) $(CWARNFLAGS) $(CMISCFLAGS) $(CPPROCFLAGS))
 
 # If the user provided his own CFLAGS, allow them to override our own.
-# *** Notice that we ALSO modify the 'no optimization' set of flags.
-#     THIS IS DANGEROUS. THE USER MUST TAKE CARE NOT TO ADD "EXTRA FLAGS"
-#     THAT WOULD DISRUPT THE NUMERICAL PROPERTIES OF SLAMCH/DLAMCH. ***
+# *** Notice that wo do not also modify the 'no optimization' set of flags.
+#     We do this to avoid potential numerical disruption of routines such
+#     as slamch/dlamch.
 ifneq (@fla_userdef_cflags@,)
-CFLAGS       := @CFLAGS@
-CFLAGS_NOOPT := @CFLAGS@
+CFLAGS       := @CFLAGS@ $(CFLAGS)
 endif
 
 

diff --git a/build/gen-make-frags.sh b/build/gen-make-frags.sh
@@ -7,5 +7,5 @@
 #
 
 # Generate makefile fragments recursively, level 1 verbosity.
-./build/gen-make-frag.sh -rhv1 build/templates/fragment.mk src
+./build/gen-make-frag.sh -rhv0 build/templates/fragment.mk src
 
diff --git a/build/post-configure.sh.in b/build/post-configure.sh.in
@@ -90,6 +90,9 @@ echo "   C compiler debug flags....................... : @fla_c_debug_flags@"
 echo "Enable compiler profiling symbols............... : @fla_enable_compiler_profiling@"
 echo "   C compiler profiling flags................... : @fla_c_prof_flags@"
 
+echo ""
+echo "User-specified CFLAGS (prepended)............... : @fla_userdef_cflags@"
+
 echo ""
 echo "Enable internal error checking.................. : @fla_enable_internal_error_checking@"
 if   [ "@fla_enable_internal_error_checking@" = "yes" ] ; then
@@ -142,12 +145,6 @@ echo "   Unmangled name............................... : @fla_f77_foobar_unmangl
 echo "   Mangled name................................. : @fla_f77_foobar_mangled@"
 fi
 
-echo ""
-echo "Compile with extra C compiler flags............. : @fla_with_extra_cflags@"
-if [ "@fla_with_extra_cflags@" = "yes" ] ; then
-echo "   User-requested extra C compiler flags........ : @fla_extra_cflags@"
-fi
-
 echo ""
 echo "libflame install directory prefix............... : @prefix@"
 echo ""