Update the Arm Optimized Routine library to v25.01

Sponsored by:	Arm Ltd

contrib/arm-optimized-routines/MAINTAINERS

@@ -1,12 +1,9 @@
 /
-	Szabolcs Nagy <szabolcs.nagy@arm.com>
+	Tamar Christina <tamar.christina@arm.com>
 math/
-	Szabolcs Nagy <szabolcs.nagy@arm.com>
-networking/
-	Szabolcs Nagy <szabolcs.nagy@arm.com>
-pl/
 	Pierre Blanchard <pierre.blanchard@arm.com>
 	Joe Ramsay <joe.ramsay@arm.com>
+networking/
+	Ola Liljedahl <ola.liljedahl@arm.com>
 string/
-	Szabolcs Nagy <szabolcs.nagy@arm.com>
 	Wilco Dijkstra <wilco.dijkstra@arm.com>

contrib/arm-optimized-routines/Makefile

@@ -1,6 +1,6 @@
 # Makefile - requires GNU make
 #
-# Copyright (c) 2018-2022, Arm Limited.
+# Copyright (c) 2018-2024, Arm Limited.
 # SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
 
 srcdir = .
@@ -11,7 +11,6 @@ includedir = $(prefix)/include
 
 # Configure these in config.mk, do not make changes in this file.
 SUBS = math string networking
-PLSUBS = math
 HOST_CC = cc
 HOST_CFLAGS = -std=c99 -O2
 HOST_LDFLAGS =
@@ -21,12 +20,22 @@ CPPFLAGS =
 CFLAGS = -std=c99 -O2
 CFLAGS_SHARED = -fPIC
 CFLAGS_ALL = -Ibuild/include $(CPPFLAGS) $(CFLAGS)
-CFLAGS_PL = -Ibuild/pl/include $(CPPFLAGS) $(CFLAGS) -DPL
 LDFLAGS =
 LDLIBS =
 AR = $(CROSS_COMPILE)ar
 RANLIB = $(CROSS_COMPILE)ranlib
 INSTALL = install
+# Detect OS.
+# Assume Unix environment: Linux, Darwin, or Msys.
+OS := $(shell uname -s)
+OS := $(patsubst MSYS%,Msys,$(OS))
+# Following math dependencies can be adjusted in config file
+# if necessary, e.g. for Msys.
+libm-libs = -lm
+libc-libs = -lc
+mpfr-libs = -lmpfr
+gmp-libs = -lgmp
+mpc-libs = -lmpc
 
 all:
 
@@ -53,7 +62,6 @@ $(DIRS):
 	mkdir -p $@
 
 $(filter %.os,$(ALL_FILES)): CFLAGS_ALL += $(CFLAGS_SHARED)
-$(filter %.os,$(ALL_FILES)): CFLAGS_PL += $(CFLAGS_SHARED)
 
 build/%.o: $(srcdir)/%.S
 	$(CC) $(CFLAGS_ALL) -c -o $@ $<

contrib/arm-optimized-routines/README

@@ -12,22 +12,42 @@ contribution requirements are documented in README.contributors of
 the appropriate subdirectory.
 
 Regular quarterly releases are tagged as vYY.MM, the latest
-release is v24.01.
+release is v25.01.
 
 Source code layout:
 
 build/          - build directory (created by make).
-math/           - math subproject sources.
+math/           - math subproject sources for generic scalar
+                  subroutines and sources shared with
+                  subdirectories of math/.
+                  All math routines should meet the quality
+                  requirements stated in math/README.contributors,
+                  routines that fail to do so are located in an
+                  experimental/ directory.
+math/aarch64/   - math subproject AArch64-specific sources
+                  and sources shared with subdirectories.
+math/aarch64/advsimd      - AdvSIMD-specific math sources.
+math/aarch64/experimental - Experimental math sources do not
+                            meet quality requirements stated in
+                            math/README.contributors.
+math/aarch64/sve          - SVE-specific math sources.
 math/include/   - math library public headers.
 math/test/      - math test and benchmark related sources.
 math/tools/     - tools used for designing the algorithms.
 networking/     - networking subproject sources.
 networking/include/ - networking library public headers.
 networking/test/ - networking test and benchmark related sources.
 string/         - string routines subproject sources.
+                  All string routines should meet the quality
+                  requirements stated in string/README.contributors,
+                  routines that fail to do so are located in an
+                  experimental/ directory.
+string/<arch>   - <arch>-specific string routines sources for
+                  <arch>=aarch64, and arm.
+string/aarch64/experimental - Experimental string routines which
+                              may not be fully optimized yet.
 string/include/ - string library public headers.
 string/test/    - string test and benchmark related sources.
-pl/...          - separately maintained performance library code.
 
 The steps to build the target libraries and run the tests:
 
@@ -50,6 +70,13 @@ Or building and testing the math subproject only:
 make all-math
 make check-math
 
+Note on compiler compability/requirement:
+
+SVE routines are always built by default - this means that on AArch64
+GCC >= 10 or LLVM >= 5 are always required for SVE ACLE compatibility.
+There is no explicit check for compatible compiler, therefore the SVE
+routines will fail to build if CC is too old.
+
 The test system requires libmpfr and libmpc.
 For example on debian linux they can be installed as:
 

contrib/arm-optimized-routines/config.mk.dist

@@ -1,14 +1,11 @@
 # Example config.mk
 #
-# Copyright (c) 2018-2023, Arm Limited.
+# Copyright (c) 2018-2024, Arm Limited.
 # SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
 
 # Subprojects to build
 SUBS = math string networking
 
-# Subsubprojects to build if subproject pl is built
-PLSUBS = math
-
 # Target architecture: aarch64, arm or x86_64
 ARCH = aarch64
 
@@ -30,6 +27,27 @@ HOST_CFLAGS += -Wall -Wno-unused-function
 HOST_CFLAGS += -g
 CFLAGS += -g
 
+ifeq ($(OS),Msys)
+  # llvm is the only available/valid native compiler
+  CC = clang
+  AR = llvm-ar
+  RANLIB = llvm-ranlib
+  HOST_CC = clang
+  SYSROOT = /c/wenv/msys2/msys64/clangarm64
+  # Common windows flags
+  COMMON_WIN_CFLAGS = -D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE
+  COMMON_WIN_CFLAGS += -Wno-deprecated-declarations -Wno-unused-variable
+  # For mathtest
+  HOST_CFLAGS += -I$(SYSROOT)/include
+  HOST_CFLAGS += $(COMMON_WIN_CFLAGS) -Wno-ignored-attributes
+  # Clear the default flag -fPIC, as not supported on Windows
+  CFLAGS_SHARED =
+  # For ulp.h with MPFR
+  CFLAGS += -I$(SYSROOT)/include
+  # For clang on Windows
+  CFLAGS += $(COMMON_WIN_CFLAGS)
+endif
+
 # Optimize the shared libraries on aarch64 assuming they fit in 1M.
 #CFLAGS_SHARED = -fPIC -mcmodel=tiny
 
@@ -45,12 +63,33 @@ math-cflags =
 math-ldlibs =
 math-ulpflags =
 math-testflags =
-string-cflags =
+string-cflags = -falign-functions=64
 networking-cflags =
 
-# Use if mpfr is available on the target for ulp error checking.
-#math-ldlibs += -lmpfr -lgmp
-#math-cflags += -DUSE_MPFR
+ifeq ($(OS),Msys)
+  # Libraries can be installed with pacman
+  libm-libs = -lmsvcrt -lvcruntime -lucrt
+  libc-libs =
+  # Linker will look for .lib but some systems only have .dll.a,
+  # therefore we have to give absolute path to libraries.
+  # This is system dependent and might need adjusting.
+  mpfr-libs = $(SYSROOT)/lib/libmpfr.dll.a
+  gmp-libs = $(SYSROOT)/lib/libgmp.dll.a
+  mpc-libs = $(SYSROOT)/lib/libmpc.dll.a
+endif
+
+# Use if mpfr is available on the target for ulp error checking. If
+# enabling this, it is advised to disable fenv checks by uncommenting
+# the two lines at the bottom of this block.
+USE_MPFR=0
+math-cflags += -DUSE_MPFR=$(USE_MPFR)
+ifeq ($(USE_MPFR), 1)
+  math-ldlibs += $(mpfr-libs) $(gmp-libs)
+  math-ulpflags += -m
+endif
+# Disable fenv checks
+#math-ulpflags = -q -f
+#math-testflags = -nostatus
 
 # Use with gcc.
 math-cflags += -frounding-math -fexcess-precision=standard -fno-stack-protector
@@ -59,30 +98,36 @@ math-cflags += -ffp-contract=fast -fno-math-errno
 # Use with clang.
 #math-cflags += -ffp-contract=fast
 
-# Disable/enable SVE vector math code and tests.
-# If WANT_SVE_MATH is enabled, math-sve-cflags is added for SVE
-# routines only so that SVE code does not leak into scalar
-# routines. It is also necessary to add it for tools (e.g. ulp,
-# mathbench)
-WANT_SVE_MATH = 0
-ifeq ($(WANT_SVE_MATH), 1)
-  math-sve-cflags = -march=armv8-a+sve
-endif
-math-cflags += -DWANT_SVE_MATH=$(WANT_SVE_MATH)
-
 # If defined to 1, set errno in math functions according to ISO C.  Many math
 # libraries do not set errno, so this is 0 by default.  It may need to be
 # set to 1 if math.h has (math_errhandling & MATH_ERRNO) != 0.
 WANT_ERRNO = 0
 math-cflags += -DWANT_ERRNO=$(WANT_ERRNO)
 
+# Disable/enable SVE vector math tests/tools.
+ifeq ($(ARCH),aarch64)
+  WANT_SVE_TESTS = 1
+else
+  WANT_SVE_TESTS = 0
+endif
+math-cflags += -DWANT_SVE_TESTS=$(WANT_SVE_TESTS)
+
 # If set to 1, set fenv in vector math routines.
 WANT_SIMD_EXCEPT = 0
 math-cflags += -DWANT_SIMD_EXCEPT=$(WANT_SIMD_EXCEPT)
 
-# Disable fenv checks
-#math-ulpflags = -q -f
-#math-testflags = -nostatus
+# If set to 1, enable tests for exp10.
+WANT_EXP10_TESTS = 1
+math-cflags += -DWANT_EXP10_TESTS=$(WANT_EXP10_TESTS)
+
+# If set to 1, enable tests for sinpi and cospi. These functions are
+# only supported on aarch64
+ifeq ($(ARCH),aarch64)
+  WANT_TRIGPI_TESTS = 1
+else
+  WANT_TRIGPI_TESTS = 0
+endif
+math-cflags += -DWANT_TRIGPI_TESTS=$(WANT_TRIGPI_TESTS)
 
 # Remove GNU Property Notes from asm files.
 #string-cflags += -DWANT_GNU_PROPERTY=0
@@ -92,3 +137,13 @@ math-cflags += -DWANT_SIMD_EXCEPT=$(WANT_SIMD_EXCEPT)
 
 # Avoid auto-vectorization of scalar code and unroll loops
 networking-cflags += -O2 -fno-tree-vectorize -funroll-loops
+
+# Provide *_finite symbols and some of the glibc hidden symbols
+# so libmathlib can be used with binaries compiled against glibc
+# to interpose math functions with both static and dynamic linking
+USE_GLIBC_ABI = 1
+math-cflags += -DUSE_GLIBC_ABI=$(USE_GLIBC_ABI)
+
+# Enable experimental math routines - non-C23 vector math and low-accuracy scalar
+WANT_EXPERIMENTAL_MATH = 0
+math-cflags += -DWANT_EXPERIMENTAL_MATH=$(WANT_EXPERIMENTAL_MATH)