aside from the documented differences, which are the contents of this
patch, GCC's -Os also has hard-coded unwanted behaviors which are
impossible to override, like refusing to strength-reduce division by a
constant to multiplication, presumably because the div saves a couple
bytes of code. for this reason, getting rid of -Os and switching to an
equivalent default optimization profile based on -O2 has been a
long-term goal.
as follow-ups, it may make sense to evaluate which of these variations
from -O2 actually do anything useful, and eliminate the ones which are
not helpful or which throw away performance for insignificant size
savings. but for now, I've replicated -Os as closely as possible to
provide a baseline for such evaluation.
unlike most projects that use -fno-strict-aliasing, we aim to have all
sources respect the C language rules for effective type that make
type-based alias analysis optimizations possible. unfortunately, it
turns out that there are deep, and likely very difficult to fix, flaws
in the TBAA performed by GCC and likely other compilers, whereby this
kind of optimization can transform code that follows the rules
strictly in ways that will make it malfunction. see for example GCC
bugs 107107 and 107115, the latter of which also affects clang.
there are not presently any known instances of breakage due to wrong
type-based aliasing optimizations in our codebase. nonetheless, since
the transformations are unsound and could introduce breakage,
configure CFLAGS to build with -fno-strict-aliasing.
some casual analysis of the effects on codegen suggest that this is
unlikely to affect performance except possibly in the regex engine. in
general, we should probably prefer making better use of the restrict
keyword over relying on types to imply non-aliasing for optimization
purposes; doing so should be able to get back any performance that was
lost and more, should it turn out to matter (unlikely).
When the soft-float ABI for PowerPC was added in commit
5a92dd95c7, with Freescale cpus using
the alternative SPE FPU as the main use case, it was noted that we
could probably support hard float on them, but that it would involve
determining some difficult ABI constraints. This commit is the
completion of that work.
The Power-Arch-32 ABI supplement defines the ABI profiles, and indeed
ATR-SPE is built on ATR-SOFT-FLOAT. But setjmp/longjmp compatibility
are problematic for the same reason they're problematic on ARM, where
optional float-related parts of the register file are "call-saved if
present". This requires testing __hwcap, which is now done.
In keeping with the existing powerpc-sf subarch definition, which did
not have fenv, the fenv macros are not defined for SPE and the SPEFSCR
control register is left (and assumed to start in) the default mode.
now that -Wall is not used and we control which warnings are enabled,
it makes sense to have the wanted ones on by default. hopefully this
will also discourage manually adding -Wall to CFLAGS and making
incorrect changes or bug reports based on the compiler's output.
-Wall varies too much by compiler and version. rather than trying to
track all the unwanted style warnings that need to be subtracted, just
enable wanted warnings.
also, move -Wno-pointer-to-int-cast outside --enable-warnings
conditional so that it always applies, since it's turning off a
nuisance warning that's on-by-default with most compilers.
these four warning options were overlooked previously, likely because
they're not part of GCC's -Wall. they all detect constraint violations
(invalid C at the source level) and should always be on in -Werror
form.
the intent here is to keep oldmalloc as an option, at least for the
short term, in case any users are negatively impacted in some way by
mallocng and need to fallback until their issues are resolved.
coding style warnings enabled by default in clang have long been a
source of spurious questions/bug-reports. since clang provides a -w
that behaves differently from gcc's, and that lets us enable any
warnings we may actually want after turning them all off to start with
a clean slate, use it at configure time if clang is detected.
Author: Alex Suykov <alex.suykov@gmail.com>
Author: Aric Belsito <lluixhi@gmail.com>
Author: Drew DeVault <sir@cmpwn.com>
Author: Michael Clark <mjc@sifive.com>
Author: Michael Forney <mforney@mforney.org>
Author: Stefan O'Rear <sorear2@gmail.com>
This port has involved the work of many people over several years. I
have tried to ensure that everyone with substantial contributions has
been credited above; if any omissions are found they will be noted
later in an update to the authors/contributors list in the COPYRIGHT
file.
The version committed here comes from the riscv/riscv-musl repo's
commit 3fe7e2c75df78eef42dcdc352a55757729f451e2, with minor changes by
me for issues found during final review:
- a_ll/a_sc atomics are removed (according to the ISA spec, lr/sc
are not safe to use in separate inline asm fragments)
- a_cas[_p] is fixed to be a memory barrier
- the call from the _start assembly into the C part of crt1/ldso is
changed to allow for the possibility that the linker does not place
them nearby each other.
- DTP_OFFSET is defined correctly so that local-dynamic TLS works
- reloc.h LDSO_ARCH logic is simplified and made explicit.
- unused, non-functional crti/n asm files are removed.
- an empty .sdata section is added to crt1 so that the
__global_pointer reference is resolvable.
- indentation style errors in some asm files are fixed.
we have to avoid using ebx unconditionally in asm constraints for
i386, because gcc 3 and 4 and possibly other simplistic compilers
(pcc?) implement PIC via making ebx a fixed-use register, and disallow
its use for anything else. rather than hard-coding knowledge of which
compilers work (at least gcc 5+ and clang), perform a configure test;
this should give us the good codegen on any new compilers we don't yet
know about.
swapping ebx and edx is kept for 1- and 2-arg syscalls because it
avoids having any spills/stack-frame at all in small functions. for
6-arg, if ebx is directly usable, the complex shuffling introduced in
commit c8798ef974 can be avoided, and
ebp can be loaded the same way ebx is in 5-arg syscalls for compilers
that don't support direct use of ebx.
the clang internal assembler does not accept assembler options passed
via the usual -Wa mechanism, but it does accept -mimplicit-it directly
as an option to the compiler driver.
other compilers don't need this option, but gcc 3 and perhaps others
accept it despite not understanding it, then print warnings about it
at build time.
omitting it when not needed will also help shorten the command lines.
since commit dc2f368e56 this has been
disabled by default, but was left available in case users unhappy with
the resulting size or performance regressions wanted to try to make it
work. now that we make widespread use of hidden visibility for
internal interfaces, this no longer makes sense. if any costly calls
remain they can be fixed with hidden aliases.
three ABIs are supported: the default with 68881 80-bit fpu format and
results returned in floating point registers, softfloat-only with the
same format, and coldfire fpu with IEEE single/double only. only the
first is tested at all, and only under qemu which has fpu emulation
bugs.
basic functionality smoke tests have been performed for the most
common arch-specific breakage via libc-test and qemu user-level
emulation. some sysvipc failures remain, but are shared with other big
endian archs and will be fixed separately.
we have always bound symbols at libc.so link time rather than runtime
to minimize startup-time relocations and overhead of calls through the
PLT, and possibly also to preclude interposition that would not work
correctly anyway if allowed. historically, binding at link-time was
also necessary for the dynamic linker to work, but the dynamic linker
bootstrap overhaul in commit f3ddd17380
made it unnecessary.
our use of -Bsymbolic-functions, rather than -Bsymbolic, was chosen
because the latter is incompatible with public global data; it makes
it incompatible with copy relocations in the main program. however,
not all global data needs to be public. by using --dynamic-list
instead with an explicit list, we can reduce the number of symbolic
relocations left for runtime.
this change will also allow us to permit interposition of specific
functions (e.g. the allocator) if/when we want to, by adding them to
the dynamic list.
Maintainer's note: at one point, -lcompiler_rt apparently worked, and
may still work and be preferable if one has manually installed the
library in a public lib directory. but with current versions of clang,
the full pathname to the library file is needed. the original patch
removed the -lcompiler_rt check; I have left it in place in case there
are users depending on it, and since, when it does work, it's
preferable so as not to code a dependency on the specific compiler
version and paths in config.mak.
neither current compilers nor linkers treat protected visibility the
way I expected, as having fixed source-level semantics rather than
being dependent on target-specific ABI details, and change seems
unlikely. while the use here does not actually depend on the specific
semantics, at least some versions of some linkers, especially lld,
refuse to allow linking to a libc.so where the symbols have protected
visibility. this cannot be detected at configure-time because linking
libc.so itself works fine, and because even if we could test linking
an application against libc.so successfully, we could not justifiably
assume that the same linker used to link libc.so would also be used
later to link applications.
disable the vis.h hack by default at the configure level, but add an
explicit "auto" option to request the old configure-time detection
rather than just removing it. this leaves it easy to evaluate whether
it actually resulted in significant size or performance benefits while
ensuring that out-of-the-box builds are not unlinkable for some users.
fortunately, preliminary evaluation suggests that at least x86_64,
arm, and aarch64 don't suffer at all from the change, and impact on
other archs is low. if low is not low enough, it should not be hard to
analyze where the significant PLT call ABI costs are present and
mitigate them without the hack.
all assembly is now thumb2-compatible. on existing targets this is at
best a size optimization, but it will also facilitate porting to
thumb2-isa-only arm variants.
The --build flag is listed in two case statement entries in configure,
which causes the second entry to be ignored. This patch removes it
from the first entry.
Signed-off-by: Michael LeMay <michael.lemay@intel.com>
the standard configure interface, which our configure script tries to
implement, identifies cross compiling (build != host) and searches for
the properly-prefixed cross tools. our script was not doing that,
forcing users to explicitly provide either CC or a CROSS_COMPILE tool
prefix, and the more common choice, just providing CC, was incomplete
because the Makefile would still invoke the native ar and ranlib
programs. this happened to work when building on ELF-based systems
with GNU binutils, but could easily fail when cross-compiling from
dissimilar systems.
like before, and like the standard configure behavior, an explicit CC
or CROSS_COMPILE variable on the command line or in the environment
overrides the automatic prefixing.
mips32r6 and mips64r6 are actually new isas at both the asm source and
opcode levels (pre-r6 code cannot run on r6) and thus need to be
treated as a new subarch. the following changes are made, some of
which yield code generation improvements for non-r6 targets too:
- add subarch logic in configure script and reloc.h files for dynamic
linker name.
- suppress use of .set mips2 asm directives (used to allow mips2
atomic instructions on baseline mips1 builds; the kernel has to
emulate them on mips1) except when actually needed. they cause wrong
instruction encodings on r6, and pessimize inlining on at least some
compilers.
- only hard-code sync instruction encoding on mips1.
- use "ZC" constraint instead of "m" constraint for llsc memory
operands on r6, where the ll/sc instructions no longer accept full
16-bit offsets.
- only hard-code rdhwr instruction encoding with .word on targets
(pre-r2) where it may need trap-and-emulate by the kernel.
otherwise, just use the instruction mnemonic, and allow an arbitrary
destination register to be used.
the SPE ABI may be compatible with soft-float, but actually making it
work requires some additional work, so for now it's best to make sure
broken builds don't happen.
Some PowerPC CPUs (e.g. Freescale MPC85xx) have a completely different
instruction set for floating point operations (SPE).
Executing regular PowerPC floating point instructions results in
"Illegal instruction" errors.
Make it possible to run these devices in soft-float mode.
the nt32 and nt64 archs will be provided by the midipix project for
building musl on top of its posix-like syscall layer for windows. at
present the needed arch files are in a separate repository, but having
the tuple matching in the upstream configure script should make it
possible to overlay the arch files without needing any further
patching.
commit e4355bd6be moved the math asm
from external source files to inline asm, but unfortunately, all
current releases of clang use the wrong inline asm constraint codes
for float and double ("w" and "P" instead of "t" and "w",
respectively). this patch adds detection for the bug in configure,
and, for now, just disables the affected asm on broken clang versions.
commit 80fbaac4cd broke all soft-float
archs, where gcc defines __GCC_IEC_559==0 because rounding modes and
exception flags are not supported. for now, just check for
__FAST_MATH__ as an indication of broken float. this won't detect all
possible misconfigurations but it probably catches the most common
one.
previous work overhauling the dynamic linker made it so that linking
libc with -Bsymbolic-functions was no longer mandatory, but the
configure logic that forced --disable-shared when ld failed to accept
the option was left in place.
this commit removes the hard-coded -Bsymbolic-functions from the
Makefile and changes the configure test to one that simply adds it to
the auto-detected LDFLAGS on success.
this sets the stage for the first phase of the bits deduplication.
bits headers which are identical for "most" archs will be moved to
arch/generic/bits.
now that .lo and .o files differ only by whether -fPIC is passed (and
no longer at the source level based on the SHARED macro), it's
possible to use the same object files for both static and shared libc
when the compiler would produce PIC for the static files anyway. this
happens if the user has included -fPIC in their CFLAGS or if the
compiler has been configured to produce PIE by default.
we use the .lo files for both, and still append -fPIC to the CFLAGS,
rather than using the .o files so that libc.so does not break
catastrophically if the user later removes -fPIC from CFLAGS in
config.mak or on the make command line. this also ensures that we get
full -fPIC in case -fpic, -fPIE, or some other lesser-PIC option was
passed in CFLAGS.
commit 2f853dd6b9 failed to change the
test for -include vis.h support to use $srcdir, so vis.h was always
disabled by configure for out-of-tree builds.
this change adds support for building musl outside of the source
tree. the implementation is similar to autotools where running
configure in a different directory creates config.mak in the current
working directory and symlinks the makefile, which contains the
logic for creating all necessary directories and resolving paths
relative to the source directory.
to support both in-tree and out-of-tree builds with implicit make
rules, all object files are now placed into a separate directory.
Stefan O'Rear
Hongliang Wang
Rich Felker
Wesley Wiser
Issam E. Maghni
rofl0r
Fangrui Song
Matúš Olekšák
Szabolcs Nagy
Dmitry Golovin
Bobby Bingham
LeMay, Michael
Felix Fietkau
Petr Hosek