The previous fix clashed with the rewrite to emit SLP invariants
during the SLP walk. Thus the following adjusts the SLP tree
hacking vectorizable_shift does appropriately.
Still resisting the attempt of a rewrite of vectorizable_shift ...
2020-05-29 Richard Biener <rguenther@suse.de>
PR tree-optimization/95356
* tree-vect-stmts.c (vectorizable_shift): Do in-place SLP
node hacking during analysis.
it turns out I lost one hunk in the patch disabling extra streaming
which causes streamer to go out of sync in the case non-trivial scc
containing the node being streamed appears in local stream (which seems
quite rare since it does not happen during bootstrap).
2020-05-29 Jan Hubicka <hubicka@ucw.cz>
PR lto/95362
* lto-streamer-out.c (lto_output_tree): Disable redundant streaming.
Simple documentation update based on usage of GIT by both
LLVM and GCC.
libsanitizer/ChangeLog:
* HOWTO_MERGE: Do not mention not existing argument.
* README.gcc: Update LLVM repository location.
I've just tested the script and I'm going to install the patch
to all active branches.
contrib/ChangeLog:
* git-backport.py: The script did 'git co HEAD~' when
there was no modified ChangeLog file in a successful
git cherry pick.
Run cherry-pick --continue without editor.
As noticed by Arseny, I got the condition when to call the add removal hook
wrong wrong. Fixed thusly.
2020-05-28 Jakub Jelinek <jakub@redhat.com>
PR middle-end/95315
* omp-general.c (omp_resolve_declare_variant): Fix up addition of
declare variant cgraph node removal callback.
* gcc.dg/gomp/pr95315-2.c: New test.
In the following testcase, store_expr of e.g. 97 bytes long string literal
into 1MB long array is implemented by copying the 97 bytes from .rodata
section, followed by clearing the remaining bytes. But, as the STRING_CST
has type char[1024*1024], we actually allocate whole 1MB in .rodata section
for it, even when we only use the first 97 bytes from that.
The following patch tweaks it so that if we are going to initialize only the
small part from it, we don't emit all the zeros that we never use after it.
2020-05-29 Jakub Jelinek <jakub@redhat.com>
PR middle-end/95052
* expr.c (store_expr): If expr_size is constant and significantly
larger than TREE_STRING_LENGTH, set temp to just the
TREE_STRING_LENGTH portion of the STRING_CST.
* gcc.target/i386/pr95052.c: New test.
This makes sure to fold generated stmts so they do not survive
until RTL expansion and cause awkward code generation.
2020-05-29 Richard Biener <rguenther@suse.de>
PR tree-optimization/95393
* tree-ssa-phiopt.c (minmax_replacement): Use gimple_build
to build the min/max expression so we simplify cases like
MAX(0, s) immediately.
* gcc.dg/tree-ssa/phi-opt-21.c: New testcase.
* g++.dg/vect/slp-pr87105.cc: Adjust.
Extended patterns for these instructions to support unpacked vectors.
BIC will have to wait, as there is not currently support for unpacked
NOT.
2020-05-29 Joe Ramsay <joe.ramsay@arm.com>
gcc/
* config/aarch64/aarch64-sve.md (<LOGICAL:optab><mode>3): Add support
for unpacked EOR, ORR, AND.
gcc/testsuite/
* gcc.target/aarch64/sve/load_const_offset_2.c: Force using packed
vectors.
* gcc.target/aarch64/sve/logical_unpacked_and_1.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_2.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_3.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_4.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_5.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_6.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_and_7.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_1.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_2.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_3.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_4.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_5.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_6.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_eor_7.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_1.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_2.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_3.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_4.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_5.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_6.c: New test.
* gcc.target/aarch64/sve/logical_unpacked_orr_7.c: New test.
* gcc.target/aarch64/sve/scatter_store_6.c: Force using packed vectors.
* gcc.target/aarch64/sve/scatter_store_7.c: Force using packed vectors.
* gcc.target/aarch64/sve/strided_load_3.c: Force using packed vectors.
* gcc.target/aarch64/sve/strided_store_3.c: Force using packed vectors.
* gcc.target/aarch64/sve/unpack_signed_1.c: Force using packed vectors.
The patch is about to handle situations like seen
in 3ea6977d0f.
contrib/ChangeLog:
* gcc-changelog/git_commit.py: Properly
handle duplicite authors.
* gcc-changelog/test_email.py: New test.
* gcc-changelog/test_patches.txt: New patch.
Two years ago Paolo has added the
else if (processing_template_decl && !COMPLETE_TYPE_P (type))
pedwarn (...);
lines into cp_finish_decomp. For type dependent decl we punt much earlier,
but even for types which aren't type dependent COMPLETE_TYPE_P might be
false as this testcase shows, so this patch tries to complete_type first
(the reason for writing it that way is that it is then followed by another
else if and if complete_type returns error_mark_node, we shouldn't report
anything, as a bug should have been reported already.
2020-05-28 Jakub Jelinek <jakub@redhat.com>
PR c++/95328
* decl.c (cp_finish_decomp): Call complete_type before checking
COMPLETE_TYPE_P.
* g++.dg/cpp1z/decomp53.C: New test.
The use of KIND, LEN, RE, and IM inquiry references for applicable intrinsic
types is valid only for suffienctly new Fortran standards. Add appropriate
check.
2020-05-28 Harald Anlauf <anlauf@gmx.de>
gcc/fortran/
PR fortran/95373
* primary.c (is_inquiry_ref): Check validity of inquiry
references against selected Fortran standard.
gcc/testsuite/
PR fortran/95373
* gfortran.dg/pr95373_1.f90: New test.
* gfortran.dg/pr95373_2.f90: New test.
The initial commit for this PR uncovered a latent issue with unit locking
in the Fortran run-time library. Add check for valid unit.
2020-05-28 Harald Anlauf <anlauf@gmx.de>
libgfortran/
PR libfortran/95104
* io/unit.c (unlock_unit): Guard by check for NULL pointer.
In r9-297 I was trying to be more flexible and treat static data members of
class templates more like variable templates, where the type need not be
determined until the variable is instantiated, but I suppose that in a class
the types of all the non-template members need to be determined at the time
of class instantiation.
gcc/cp/ChangeLog:
PR c++/94926
* decl.c (cp_finish_decl): Revert r9-297 change.
(check_static_variable_definition): Likewise.
* constexpr.c (ensure_literal_type_for_constexpr_object): Likewise.
* pt.c (instantiate_decl): Return early on type error.
gcc/testsuite/ChangeLog:
* g++.dg/cpp1z/pr86648.C: Expect error.
* g++.dg/cpp1z/static2.C: Expect error.
* g++.dg/cpp0x/nsdmi16.C: New test.
2020-05-28 Nicolas Bértolo <nicolasbertolo@gmail.com>
/ChangeLog:
* configure.ac: Don't require --enable-host-shared when building
for Mingw.
* configure: Regenerate.
2020-05-28 Nicolas Bértolo <nicolasbertolo@gmail.com>
gcc/ChangeLog:
* Makefile.in: don't look for libiberty in the "pic" subdirectory
when building for Mingw. Add dependency on xgcc with the proper
extension.
2020-05-28 Nicolas Bértolo <nicolasbertolo@gmail.com>
gcc/c/ChangeLog:
* Make-lang.in: Remove extra slash.
2020-05-28 Nicolas Bértolo <nicolasbertolo@gmail.com>
gcc/jit/ChangeLog:
* Make-lang.in: Remove extra slash. Build libgccjit.dll and its
import library in Windows.
* config-lang.in: Update comment about --enable-host-shared.
* jit-w32.h: New file.
* jit-w32.c: New file.
(print_last_error): New function that prints the error
string corresponding to GetLastError().
(get_TOKEN_USER_current_user): Helper function used for getting
the SID belonging to the current user.
(create_directory_for_current_user): Helper function to create
a directory with permissions such that only the current user can
access it.
(win_mkdtemp): Create a temporary directory using Windows APIs.
* jit-playback.c: Do not chmod files in Windows. Use LoadLibrary,
FreeLibrary and GetProcAddress instead of libdl.
* jit-result.h, jit-result.c: Introduce result::handle_t to
abstract over the types used for dynamic library handles.
* jit-tempdir.c: Do not use mkdtemp() in Windows, use
win_mkdtemp().
* config/h8300/logical.md (HImode H8/SX bit-and splitter): Don't
make a nonzero adjustment to the memory offset.
(b<ior,xor>hi_msx): Turn into a splitter.
The bitregion_end field points to the next bit after the region.
gcc/ChangeLog
* gimple-ssa-store-merging.c (merged_store_group::can_be_merged_into):
Fix off-by-one error.
Check "! int128" instead of ilp32 since ILP32 targets can support int128.
gcc/testsuite/
* gcc.dg/builtin-bswap-10.c: Check "! int128" instead of ilp32
Based on a patch in the comments of the PR. That patch fixed this
problem but caused the test cases for PR93484 to fail. It has been
changed to reduce initialisation expressions if the expression is
not EXPR_VARIABLE and not EXPR_CONSTANT.
2020-05-28 Steven G. Kargl <kargl@gcc.gnu.org>
Mark Eggleston <markeggleston@gcc.gnu.org>
gcc/fortran/
PR fortran/94397
* match.c (gfc_match_type_spec): New variable ok initialised
to true. Set ok with the return value of gfc_reduce_init_expr
called only if the expression is not EXPR_CONSTANT and is not
EXPR_VARIABLE. Add !ok to the check for type not being integer
or the rank being greater than zero.
2020-05-28 Mark Eggleston <markeggleston@gcc.gnu.org>
gcc/testsuite/
PR fortran/94397
* gfortran.dg/pr94397.F90: New test.
wb_candidate1 and wb_candidate2 exist for two overlapping cases:
when we use an STR or STP with writeback to allocate the frame,
and when we set up a frame chain record (either using writeback
allocation or not).
However, aarch64_layout_frame was leaving these fields with
legitimate register numbers even if we decided to do neither
of those things. This prevented those registers from being
shrink-wrapped, even though we were otherwise treating them
as normal saves and restores.
The case this patch handles isn't the common case, so it might
not be worth going out of our way to optimise it. But I think
the patch actually makes the output of aarch64_layout_frame more
consistent.
2020-05-28 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/aarch64/aarch64.h (aarch64_frame): Add a comment above
wb_candidate1 and wb_candidate2.
* config/aarch64/aarch64.c (aarch64_layout_frame): Invalidate
wb_candidate1 and wb_candidate2 if we decided not to use them.
gcc/testsuite/
* gcc.target/aarch64/shrink_wrap_1.c: New test.
The stack frame for the function in the testcase consisted of two
SVE save slots. Both saves had been shrink-wrapped, but for different
blocks, meaning that the stack allocation and deallocation were
separate from the saves themselves. Before emitting the deallocation,
we tried to attach a REG_CFA_DEF_CFA note to the preceding instruction,
to redefine the CFA in terms of the stack pointer. But in this case
there was no preceding instruction.
This in practice only happens for SVE because:
(a) We don't try to shrink-wrap wb_candidate* registers even when
we've decided to treat them as normal saves and restores.
I have a fix for that.
(b) Even with (a) fixed, we're (almost?) guaranteed to emit
a stack tie for frames that are 64k or larger, so we end
up hanging the REG_CFA_DEF_CFA note on that instead.
We should only need to redefine the CFA if it was previously
defined in terms of the frame pointer. In other cases the CFA
should already be defined in terms of the stack pointer,
so redefining it is unnecessary but usually harmless.
2020-05-28 Richard Sandiford <richard.sandiford@arm.com>
gcc/
PR testsuite/95361
* config/aarch64/aarch64.c (aarch64_expand_epilogue): Assert that
we have at least some CFI operations when using a frame pointer.
Only redefine the CFA if we have CFI operations.
gcc/testsuite/
PR testsuite/95361
* gcc.target/aarch64/sve/pr95361.c: New test.
This removes handling of !SLP_TREE_VECTYPE from invariant costing.
The single caller guards against this case already.
2020-05-28 Richard Biener <rguenther@suse.de>
* tree-vect-slp.c (vect_prologue_cost_for_slp): Remove
case for !SLP_TREE_VECTYPE.
(vect_slp_analyze_node_operations): Adjust.
This generates vector defs for externals and invariants during the SLP
walk rather than as part of getting vectorized defs when vectorizing
the users. This is a requirement to make sharing of external/invariant
nodes be reflected in actual code generation.
This temporarily adds a SLP_TREE_VEC_DEFS vector alongside the
SLP_TREE_VEC_STMTS one. Eventually the latter can go away.
2020-05-27 Richard Biener <rguenther@suse.de>
* tree-vectorizer.h (_slp_tree::vec_defs): Add.
(SLP_TREE_VEC_DEFS): Likewise.
* tree-vect-slp.c (_slp_tree::_slp_tree): Adjust.
(_slp_tree::~_slp_tree): Likewise.
(vect_mask_constant_operand_p): Remove unused function.
(vect_get_constant_vectors): Rename to...
(vect_create_constant_vectors): ... this. Take the
invariant node as argument and code generate it. Remove
dead code, remove temporary asserts. Pass a NULL stmt_info
to vect_init_vector.
(vect_get_slp_defs): Simplify.
(vect_schedule_slp_instance): Code-generate externals and
invariants using vect_create_constant_vectors.
This makes the call chain below vec_init_vector happy with a NULL
stmt_vec_info which is used as "context".
2020-05-27 Richard Biener <rguenther@suse.de>
* tree-vect-stmts.c (vect_finish_stmt_generation_1):
Conditionalize stmt_info use, assert the new stmt cannot throw
when not specified.
(vect_finish_stmt_generation): Adjust assert.
Covering all bases in vectorizable_shift is hard - this makes sure
to appropriately handle the case of PR95356 without breaking others.
2020-05-28 Richard Biener <rguenther@suse.de>
PR tree-optimization/95273
PR tree-optimization/95356
* tree-vect-stmts.c (vectorizable_shift): Adjust when and to
what we set the vector type of the shift operand SLP node
again.
* gcc.target/i386/pr95356.c: New testcase.
So far, we expect from a commit that renames a file to contain a
changelog entry only for the new name. For example, after the following
commit:
$ git move foo bar
$ git commit
We expect the following changelog:
* bar: Renamed from foo.
Git does not keep track of renamings, only file deletions and additions.
The display of patches then uses heuristics (with config-dependent
parameters) to try to match deleted and added files in the same commit.
It is thus brittle to rely on this information.
This commit modifies changelog processing so that renames are considered
as a deletion of a file plus an addition of another file. The following
changelog is now expected for the above example:
* foo: Move...
* bar: Here.
contrib/
* gcc-changelog/git_email.py (GitEmail.__init__): Interpret file
renamings as a file deletion plus a file addition.
* gcc-changelog/git_repository.py (parse_git_revisions):
Likewise.
* gcc-changelog/test_email.py: New testcase.
* gcc-changelog/test_patches.txt: New testcase.
The patch fixes various issues spotted by check-params-in-docs.py
script. I'm going to install the patch.
gcc/ChangeLog:
PR web/95380
* doc/invoke.texi: Add missing params, remove max-once-peeled-insns and
rename ipcp-unit-growth to ipa-cp-unit-growth.
This fixes the testcase to avoid out of bound shifts on ilp32 targets.
2020-05-28 Richard Biener <rguenther@suse.de>
PR testsuite/95363
* gcc.dg/vect/bb-slp-pr95271.c: Fix on ilp32 targets.
This patch fixes various issues in the testsuite that came up after
the dump/aux output revamp, namely:
- many outputs.exp tests used -flto without checking that LTO was
supported, getting lots of failures. With this patch, we test for LTO
support, and skip -flto tests on platforms that do not support it.
- some linkers error out if an output file is not named, and the
a.{out,exe} construct that we used throughout outputs.exp to match the
default linker output would trigger a bug in tcl globbing. With this
patch, we detect the default linker output early. If none is found,
we arrange to pass -o a.out explicitly in tests that used to test the
default linker output. We now look for the detected default, or for
explicitly-specified output.
- collect2 will leave <execname>.cdtor.* files behind in -save-temps
tests. Ignore them.
- The prepending of -Wl, to file names in ldflags et al was done in a
way that introduced empty arguments when consecutive blanks appeared
in these board configuration knobs. Skip the empty strings between
consecutive blanks to avoid this problem.
gcc/testsuite/ChangeLog:
* lib/gcc-defs.exp: Avoid introducing empty arguments between
consecutive blanks in board linking options.
* gcc.misc-tests/outputs.exp: Likewise. Document
-gsplit-dwarf testing, skip LTO tests if -flto is not
supported, detect the default linker output name, cope with
the need for an explicit executable output.
This fixes 'non-delegitimized UNSPEC 3 found in variable location' notes
issued when building libraries which interferes with running tests.
2020-05-27 Max Filippov <jcmvbkbc@gmail.com>
gcc/
* config/xtensa/xtensa.c (xtensa_delegitimize_address): New
function.
(TARGET_DELEGITIMIZE_ADDRESS): New macro.
This patch introduces a new builtin named __builtin_bswap128 on targets
where TImode is supported, i.e. 64-bit targets only in practice. The
implementation simply reuses the existing double word path in optab, so
no routine is added to libgcc (which means that you get two calls to
_bswapdi2 in the worst case).
gcc/ChangeLog:
* builtin-types.def (BT_UINT128): New primitive type.
(BT_FN_UINT128_UINT128): New function type.
* builtins.def (BUILT_IN_BSWAP128): New GCC builtin.
* doc/extend.texi (__builtin_bswap128): Document it.
* builtins.c (expand_builtin): Deal with BUILT_IN_BSWAP128.
(is_inexpensive_builtin): Likewise.
* fold-const-call.c (fold_const_call_ss): Likewise.
* fold-const.c (tree_call_nonnegative_warnv_p): Likewise.
* tree-ssa-ccp.c (evaluate_stmt): Likewise.
* tree-vect-stmts.c (vect_get_data_ptr_increment): Likewise.
(vectorizable_call): Likewise.
* optabs.c (expand_unop): Always use the double word path for it.
* tree-core.h (enum tree_index): Add TI_UINT128_TYPE.
* tree.h (uint128_type_node): New global type.
* tree.c (build_common_tree_nodes): Build it if TImode is supported.
gcc/testsuite/ChangeLog:
* gcc.dg/builtin-bswap-10.c: New test.
* gcc.dg/builtin-bswap-11.c: Likewise.
* gcc.dg/builtin-bswap-12.c: Likewise.
* gcc.target/i386/builtin-bswap-5.c: Likewise.
PR libstdc++/95282
* include/bits/atomic_base.h (__atomic_impl::load): Add
cv-qualifiers to parameter so that _Tp is deduced as the
unqualified type.
* testsuite/29_atomics/atomic_float/95282.cc: New test.
The bug report is that transform_view's sentinel<false> cannot be
compared to its iterator<true>. The comparison is supposed to use
operator==(iterator<Const>, sentinel<Const>) after converting
sentinel<false> to sentinel<true>. However, the operator== is a hidden
friend so is not a candidate when comparing iterator<true> with
sentinel<false>. The required conversion would only happen if we'd found
the operator, but we can't find the operator until after the conversion
happens.
A new LWG issue has been reported, but not yet assigned a number. The
solution suggested by Casey Carter is to make the hidden friends of the
sentinel types work with iterators of any const-ness, so that no
conversions are required.
Patrick Palka observed that join_view has a similar problem and a
similar fix is used for its sentinel.
PR libstdc++/95322
* include/std/ranges (transform_view::_Sentinel): Allow hidden
friends to work with _Iterator<true> and _Iterator<false>.
(join_view::_Sentinel): Likewise.
* testsuite/std/ranges/adaptors/95322.cc: New test.
The std::reverse_iterator comparisons have always been implemented only
in terms of equality and less than. In C++98 that made no difference for
reasonable code, because when the underlying operators are the same type
they are required to support all comparisons anyway.
But since LWG 280 it's possible to compare reverse_iterator<X> and
reverse_iterator<Y>, and comparisons between X and Y might not support
the full set of equality and relational operators. This means that it
matters whether we implement operator!= as x.base() != y.base() or
!(x.base() == y.base()), and the current implementation is
non-conforming.
This was already fixed in GCC 10.1 for C++20, this change also fixes it
for all other -std modes.
PR libstdc++/94354
* include/bits/stl_iterator.h (reverse_iterator): Fix comparison
operators to use the correct operations on the underlying
iterators.
* testsuite/24_iterators/reverse_iterator/rel_ops.cc: New test.
PFPNACC insn is incorrectly modelled to perform addition and subtraction
of two operands, but in reality it performs horizontal addition and
subtraction:
Instruction: PFPNACC dest,src
Description:
dest[31:0] <- dest[31:0] - dest[63:32];
dest[63:32] <- src[31:0] + src[63:32];
Also, it is not possible to directly replace PFACC with HADDPS and PFNACC
with HSUBPS, because operands in the second word do not match.
PFACC does:
dest[31..0] <- dest[31..0] + dest[63..32];
dest[63..32] <- src[31..0] + src [63..32];
while HADDPS does:
dest[31..0] <- dest[31..0] + dest[63..32];
dest[63..32] <- dest[127..96] + dest[95..64];
dest[95..64] <- src [31..0] + src [64..32];
dest[127:96] <- src [127..96] + src [95..64];
2020-05-27 Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog:
* config/i386/mmx.md (*mmx_haddv2sf3): Remove SSE alternatives.
(mmx_hsubv2sf3): Ditto.
(mmx_haddsubv2sf3): New expander.
(*mmx_haddsubv2sf3): Rename from mmx_addsubv2sf3. Correct
RTL template to model horizontal subtraction and addition.
* config/i386/i386-builtin.def (IX86_BUILTIN_PFPNACC):
Update for rename.
For long module name, derive type and component name, the generated
name-mangled symbol did not fit into a buffer when coarrays were
enabled. Provide sufficiently large temporary.
2020-05-27 Harald Anlauf <anlauf@gmx.de>
gcc/fortran/
PR fortran/95090
* iresolve.c (gfc_get_string): Enlarge temporary for
name-mangling.
gcc/testsuite/
PR fortran/95090
* gfortran.dg/pr95090.f90: New test.
PR jit/95314 reports a internal error inside verify_gimple, which
turned out to be due to reusing the result of
gcc_jit_lvalue_get_address in several functions, leading to tree nodes
shared between multiple function bodies.
This patch fixes the issue by adopting the "Deep unsharing" strategy
described in the comment in gimplify.c preceding mostly_copy_tree_r:
to mark all of the jit "frontend"'s expression tree nodes with
TREE_VISITED, and to set LANG_HOOKS_DEEP_UNSHARING, so that "they are
unshared on the first reference within functions when the regular
unsharing algorithm runs".
gcc/jit/ChangeLog:
PR jit/95314
* dummy-frontend.c (LANG_HOOKS_DEEP_UNSHARING): Define to be true.
* jit-playback.h (gcc::jit::playback::rvalue): Mark tree node with
TREE_VISITED.
gcc/testsuite/ChangeLog:
PR jit/95314
* jit.dg/all-non-failing-tests.h: Add test-pr95314-rvalue-reuse.c.
* jit.dg/test-pr95314-rvalue-reuse.c: New test.
Richard Biener
Jan Hubicka
Martin Liska
Jakub Jelinek
Joe Ramsay
GCC Administrator
Harald Anlauf
Jason Merrill
Nicolás Bértolo
Jeff Law
Eric Botcazou
Marek Polacek
H.J. Lu
Mark Eggleston
Richard Sandiford
Andrea Corallo
Pierre-Marie de Rodat
liuhongt
Alexandre Oliva
Max Filippov
Jonathan Wakely
Uros Bizjak
David Malcolm