[Instrumentation] Fix EdgeCounts vector size in SetBranchWeights #99064
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write If you have received no comments on your PR for a week, you can request a review If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-coroutines @llvm/pr-subscribers-llvm-transforms Author: Avi Kivity (avikivity) ChangesSetBranchWeights() calculates the size of the EdgeCounts vector using OutEdges.Size(), but this is an under-estimate with coroutines. Use the number of successors, as the vector will be indexed by the result of the GetSuccessorNumber() function. This crashes in 18.1, but somehow doesn't in main. Still, it looks like the fix is applicable to both. Fixes #97962 Full diff: https://github.com/llvm/llvm-project/pull/99064.diff 1 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp b/llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp
index 35b1bbf21be97..d32d00700bcb3 100644
--- a/llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp
+++ b/llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp
@@ -1624,7 +1624,9 @@ void PGOUseFunc::setBranchWeights() {
// We have a non-zero Branch BB.
unsigned Size = BBCountInfo.OutEdges.size();
- SmallVector<uint64_t, 2> EdgeCounts(Size, 0);
+ unsigned SuccessorCount = BB.getTerminator()->getNumSuccessors();
+
+ SmallVector<uint64_t, 2> EdgeCounts(SuccessorCount, 0);
uint64_t MaxCount = 0;
for (unsigned s = 0; s < Size; s++) {
const PGOUseEdge *E = BBCountInfo.OutEdges[s];
|
|
Requesting review from @xur-llvm as author of this function. |
ellishg
reviewed
Jul 19, 2024
avikivity
force-pushed
the
PGOUseFunc-setBranchWeights-fix-EdgeCounts-size
branch
from
eae8cec to
b7db364
Compare
|
Updated: renamed Size to OutEdgesCount to make it clear what it refers to. |
|
Seems ok, but I'd like to see a test so we don't break this in the future. |
I'm not an llvm developer, so any tips about reducing my C++ reproducer to a minimal llvm reproducer would be appreciated. |
There are a few ways to get a minimal reproducer depending on your setup. https://llvm.org/docs/HowToSubmitABug.html I would first try to get a reproducer with unoptimized IR by invoking clang with |
|
Tacking on those flags to the command line (without -Xclang which was rejected) I get an error. command: "/usr/bin/clang++" "-cc1" "-triple" "x86_64-redhat-linux-gnu" "-emit-llvm-bc" "-flto=thin" "-flto-unit" "-disable-free" "-clear-ast-before-backend" "-disable-llvm-verifier" "-discard-value-names" "-main-file-name" "rpc.cc" "-mrelocation-model" "static" "-mframe-pointer=none" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-funwind-tables=2" "-target-cpu" "westmere" "-debug-info-kind=constructor" "-dwarf-version=5" "-debugger-tuning=gdb" "--compress-debug-sections=zlib" "-fdebug-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-fdebug-prefix-map=/home/avi/scylla-maint=." "-fprofile-instrument=csllvm" "-fprofile-instrument-path=/home/avi/scylla-maint/build/release-cs-pgo/default_%m.profraw" "-fprofile-instrument-use-path=merged.profdata" "-fcoverage-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-sys-header-deps" "-D" "FMT_SHARED" "-D" "SEASTAR_API_LEVEL=7" "-D" "SEASTAR_DEFERRED_ACTION_REQUIRE_NOEXCEPT" "-D" "SEASTAR_HAS_MEMBARRIER" "-D" "SEASTAR_HAVE_ASAN_FIBER_SUPPORT" "-D" "SEASTAR_HAVE_HWLOC" "-D" "SEASTAR_HAVE_NUMA" "-D" "SEASTAR_HAVE_SYSTEMTAP_SDT" "-D" "SEASTAR_LOGGER_COMPILE_TIME_FMT" "-D" "SEASTAR_LOGGER_TYPE_STDOUT" "-D" "SEASTAR_PTHREAD_ATTR_SETAFFINITY_NP" "-D" "SEASTAR_SCHEDULING_GROUPS_COUNT=18" "-D" "SEASTAR_SSTRING" "-D" "SEASTAR_STRERROR_R_CHAR_P" "-D" "NDEBUG" "-U" "_FORTIFY_SOURCE" "-U" "NDEBUG" "-fmacro-prefix-map=/home/avi/scylla-maint=." "-fcoverage-prefix-map=/home/avi/scylla-maint=." "-O2" "-Werror=unused-result" "-Wno-error=#warnings" "-Wall" "-Wimplicit-fallthrough" "-Wdeprecated" "-Wno-error=deprecated" "-Wno-backend-plugin" "-Werror" "-std=gnu++23" "-fdeprecated-macro" "-ferror-limit" "19" "-fvisibility=hidden" "-fgnuc-version=4.2.1" "-fno-implicit-modules" "-fskip-odr-check-in-gmf" "-fcxx-exceptions" "-fexceptions" "-fcolor-diagnostics" "-vectorize-loops" "-vectorize-slp" "-mllvm" "-pgso=false" "-mllvm" "-enable-value-profiling=false" "-faddrsig" "-D__GCC_HAVE_DWARF2_CFI_ASM=1" "-x" "c++" "rpc-195796.cpp" -emit-llvm -S -disable-llvm-optzns -o a.llerror: |
|
I'm guessing some internal pass related to coroutines cannot be skipped. |
|
cvise was able to reduce the source, but I don't think it's very helpful. namespace std {
typedef long unsigned size_t;
typedef long ptrdiff_t;
template <typename _Tp, _Tp __v> struct integral_constant {
static constexpr _Tp value = __v;
};
template <bool __v> using __bool_constant = integral_constant<bool, __v>;
using false_type = __bool_constant<false>;
template <bool, typename = void> struct enable_if;
template <typename _Tp> struct enable_if<true, _Tp> {
using type = _Tp;
};
namespace __detail {
auto __and_fn() -> false_type;
}
template <typename...> struct __and_ : decltype(__detail::__and_fn()){};
template <typename...> constexpr bool __and_v = __and_<>::value;
template <typename> struct is_void : false_type {};
template <typename _Tp> struct remove_reference {
using type = _Tp;
};
template <typename _Tp> struct remove_reference<_Tp &> {
using type = _Tp;
};
template <bool, typename, typename> struct conditional;
template <typename _Iftrue, typename _Iffalse>
struct conditional<false, _Iftrue, _Iffalse> {
using type = _Iffalse;
};
template <bool _Cond, typename> using enable_if_t = enable_if<_Cond>::type;
template <bool _Cond, typename _Iftrue, typename _Iffalse>
using conditional_t = conditional<_Cond, _Iftrue, _Iffalse>::type;
template <typename _Tp> constexpr bool is_void_v = is_void<_Tp>::value;
template <typename _Tp> constexpr bool is_object_v = __is_object(_Tp);
template <typename _Tp>
constexpr bool is_copy_constructible_v = __is_constructible(_Tp);
template <typename _Tp>
constexpr bool is_move_constructible_v = __is_constructible(_Tp);
template <typename _Tp, typename _Up>
constexpr bool is_same_v = __is_same(_Tp, _Up);
template <typename _Tp> _Tp &&forward(typename remove_reference<_Tp>::type &);
template <typename _Tp> remove_reference<_Tp>::type &&move(_Tp &&);
template <typename> struct iterator_traits;
template <typename> struct __is_nonvolatile_trivially_copyable {
enum { __value };
};
template <typename, typename> struct __memcpyable;
template <typename _Tp>
struct __memcpyable<_Tp *, const _Tp *>
: __is_nonvolatile_trivially_copyable<_Tp> {};
template <typename> struct __is_move_iterator {
enum { __value };
};
struct random_access_iterator_tag {};
template <typename _Tp>
requires is_object_v<_Tp>
struct iterator_traits<_Tp> {
using iterator_category = random_access_iterator_tag;
};
template <typename _Iter>
iterator_traits<_Iter>::iterator_category __iterator_category(_Iter) {
return typename iterator_traits<_Iter>::iterator_category();
}
} // namespace std
void *operator new(std::size_t, void *);
namespace std {
template <typename...> class tuple;
template <typename _Tp> _Tp min(_Tp __a, _Tp) { return __a; }
template <typename _Iterator> _Iterator __niter_base(_Iterator) noexcept;
template <bool, bool, typename> struct __copy_move {
template <typename _Tp, typename _Up>
static void __assign_one(_Tp __to, _Up __from) {
*__to = *__from;
}
template <typename _Tp, typename _Up>
static _Up *__copy_m(_Tp *__first, _Tp *__last, _Up *__result) {
ptrdiff_t _Num = __last - __first;
if (__builtin_expect(_Num > 1, true))
__builtin_memmove(__result, __first, _Num);
else if (_Num == 1)
__assign_one(__result, __first);
return __result;
}
};
template <bool _IsMove, typename _II, typename _OI>
_OI __copy_move_a2(_II __first, _II __last, _OI __result) {
typedef typename iterator_traits<_II>::iterator_category _Category;
return __copy_move<_IsMove, __memcpyable<_OI, _II>::__value,
_Category>::__copy_m(__first, __last, __result);
}
template <bool _IsMove, typename _II, typename _OI>
_OI __copy_move_a1(_II __first, _II __last, _OI __result) {
return __copy_move_a2<_IsMove>(__first, __last, __result);
}
template <bool _IsMove, typename _II, typename _OI>
_OI __copy_move_a(_II __first, _II __last, _OI __result) {
return __copy_move_a1<_IsMove>(__first, __niter_base(__last), __result);
}
template <typename _II, typename _OI>
_OI copy(_II __first, _II __last, _OI __result) {
return __copy_move_a<__is_move_iterator<_II>::__value>(__first, __last,
__result);
}
template <bool, bool, bool, bool, typename> struct _Enable_copy_move {};
} // namespace std
typedef int __uint32_t;
typedef long unsigned size_t;
typedef __uint32_t uint32_t;
void __assert_fail(char *, const char *, int, const char *) noexcept
__attribute__((__noreturn__));
namespace std {
template <typename _Tp>
class optional
: _Enable_copy_move<is_copy_constructible_v<_Tp>, __and_v<>,
is_move_constructible_v<_Tp>, __and_v<>, _Tp> {};
template <typename _RandomAccessIterator, typename _Size,
typename _OutputIterator>
_OutputIterator __copy_n(_RandomAccessIterator __first, _Size,
_OutputIterator __result, random_access_iterator_tag) {
return copy(__first, __first, __result);
}
template <typename _InputIterator, typename _Size, typename _OutputIterator>
_OutputIterator copy_n(_InputIterator __first, _Size __n,
_OutputIterator __result) {
auto __n2(__n);
if (__n2 <= 0)
return __result;
return __copy_n(__first, __n2, __result, __iterator_category(__first));
}
template <typename> class unique_ptr {};
} // namespace std
namespace seastar {
class deleter {
struct impl;
impl *_impl;
public:
~deleter();
bool is_raw_object() noexcept;
void to_raw_object() noexcept;
};
struct deleter::impl {
unsigned refs;
virtual ~impl();
};
inline deleter::~deleter() {
if (is_raw_object()) {
to_raw_object();
return;
}
if (_impl && --_impl->refs == 0)
delete _impl;
}
deleter make_free_deleter(void *);
template <typename CharType> class temporary_buffer {
CharType *_buffer;
deleter _deleter;
public:
temporary_buffer(size_t) : _deleter(make_free_deleter(_buffer)) {}
temporary_buffer(temporary_buffer &&) : _deleter() {}
temporary_buffer &operator=(temporary_buffer &&) noexcept;
const CharType *get() noexcept;
CharType *get_write() noexcept;
size_t size() noexcept;
};
} // namespace seastar
namespace std {
template <typename _Tp> struct atomic {
_Tp load() noexcept;
};
} // namespace std
namespace seastar {
namespace internal {
struct preemption_monitor {
std::atomic<uint32_t> tail;
};
preemption_monitor *get_need_preempt_var() {
preemption_monitor *g_need_preempt;
return g_need_preempt;
}
} // namespace internal
bool need_preempt() {
auto np = internal::get_need_preempt_var();
auto tail = np->tail.load();
return __builtin_expect(tail, false);
}
namespace internal {
struct monostate;
template <typename...> struct future_stored_type;
template <typename T> struct future_stored_type<T> {
using type = std::conditional_t<std::is_void_v<T>, monostate, T>;
};
template <typename... T>
using future_stored_type_t = future_stored_type<T...>::type;
template <typename, bool> struct uninitialized_wrapper_base;
template <typename T> struct uninitialized_wrapper_base<T, false> {
using tuple_type = T;
union any {
any() {}
~any();
T value;
} _v;
template <typename... U>
std::enable_if_t<!std::is_same_v<std::tuple<>, std::tuple<tuple_type>>, void>
uninitialized_set(U &&...vs) {
new (&_v) T(T(std::forward<U>(vs)...));
}
T &uninitialized_get() { return _v.value; }
};
template <typename> constexpr bool can_inherit = 0;
template <typename T>
struct uninitialized_wrapper : uninitialized_wrapper_base<T, can_inherit<T>> {};
} // namespace internal
struct future_state_base {
enum state { exception_min };
union any {
any() noexcept;
bool available() { return st >= exception_min; }
bool has_result() noexcept;
state st;
} _u;
bool available() { return _u.available(); }
};
template <typename T>
struct future_state : future_state_base, internal::uninitialized_wrapper<T> {
future_state() = default;
void move_it(future_state &&x) {
if (_u.has_result())
this->uninitialized_set(std::move(x.uninitialized_get()));
}
future_state(future_state &&x) { move_it(std::move(x)); }
void set() {
_u.st ? void()
: __assert_fail("", __builtin_FILE(), __builtin_LINE(),
__PRETTY_FUNCTION__);
new (this) future_state;
}
};
namespace internal {
class promise_base {
protected:
future_state_base *_state;
promise_base(future_state_base *) {}
friend class future_base;
};
template <typename T> class promise_base_with_type : promise_base {
protected:
using future_state = future_state<T>;
future_state *get_state() { return static_cast<future_state *>(_state); }
promise_base_with_type(future_state_base *state) : promise_base(state) {}
template <typename A> void set_value(A &&) {
if (auto s = get_state())
s->set();
}
};
} // namespace internal
template <typename T> class promise : internal::promise_base_with_type<T> {
using future_state = internal::promise_base_with_type<T>::future_state;
future_state _local_state;
public:
promise() : internal::promise_base_with_type<T>(&_local_state) {}
template <typename... A> void set_value(A &&...a) noexcept {
internal::promise_base_with_type<T>::set_value(a...);
}
};
namespace internal {
class future_base {
protected:
promise_base *_promise;
void move_it(future_base &&, future_state_base *state) {
if (auto p = _promise)
p->_state = state;
}
future_base(future_base &&x, future_state_base *state) {
move_it(std::move(x), state);
}
void clear() {
if (_promise)
detach_promise();
}
~future_base() { clear(); }
promise_base detach_promise() noexcept;
};
} // namespace internal
template <typename T> class future : internal::future_base {
using future_state = future_state<internal::future_stored_type_t<T>>;
future_state _state;
public:
future(future &&x)
: future_base(std::move(x), &_state), _state(std::move(_state)) {}
bool available() { return _state.available(); }
};
class socket_address {};
class data_source {
public:
using tmp_buf = temporary_buffer<char>;
future<tmp_buf> get() noexcept;
};
template <typename CharType> class input_stream {
data_source _fd;
temporary_buffer<CharType> _buf;
bool _eof;
size_t available() noexcept;
public:
future<temporary_buffer<CharType>> read_exactly(size_t) noexcept;
future<temporary_buffer<CharType>> read_exactly_part(size_t) noexcept;
};
} // namespace seastar
namespace std {
inline namespace __n4861 {
template <typename...> struct coroutine_traits;
template <typename = void> struct coroutine_handle {
static coroutine_handle from_address(void *) noexcept;
operator coroutine_handle<>();
};
struct suspend_never {
bool await_ready() noexcept { return true; }
void await_suspend(coroutine_handle<>) noexcept {}
void await_resume() noexcept {}
};
} // namespace __n4861
} // namespace std
namespace seastar {
namespace internal {
template <typename T> class coroutine_traits_base {
public:
class promise_type {
promise<T> _promise;
public:
template <typename... U> void return_value(U &&...value) {
_promise.set_value(value...);
}
void unhandled_exception() noexcept;
future<T> get_return_object() noexcept;
std::suspend_never initial_suspend() noexcept;
std::suspend_never final_suspend() noexcept { return {}; }
};
};
template <bool, typename T> struct awaiter {
future<T> _future;
bool await_ready() noexcept { return _future.available() && !need_preempt(); }
template <typename U> void await_suspend(std::coroutine_handle<U>) noexcept;
T await_resume();
};
} // namespace internal
template <typename T> auto operator co_await(future<T> f) {
return internal::awaiter<true, T>(std::move(f));
}
} // namespace seastar
namespace std {
template <typename T, typename... Args>
class coroutine_traits<seastar::future<T>, Args...>
: public seastar::internal::coroutine_traits_base<T> {};
} // namespace std
namespace seastar {
template <typename CharType>
future<temporary_buffer<CharType>>
input_stream<CharType>::read_exactly_part(size_t n) noexcept {
temporary_buffer<CharType> out(n);
size_t completed;
while (completed < n) {
size_t avail = available();
if (avail) {
auto now = std::min(completed, avail);
std::copy_n(_buf.get(), now, out.get_write());
completed += now;
if (completed == n)
break;
}
temporary_buffer buf = co_await _fd.get();
if (buf.size()) {
_eof = true;
break;
}
_buf = std::move(buf);
}
co_return out;
}
template <typename CharType>
future<temporary_buffer<CharType>>
input_stream<CharType>::read_exactly(size_t n) noexcept {
read_exactly_part(n);
}
namespace rpc {
struct rcv_buf {};
class compressor;
class connection {
std::unique_ptr<compressor> _compressor;
future<std::optional<rcv_buf>>
read_stream_frame_compressed(input_stream<char> &);
socket_address peer_address();
template <typename FrameType>
future<typename FrameType::return_type>
read_frame_compressed(socket_address, std::unique_ptr<compressor> &,
input_stream<char> &);
};
template <typename FrameType>
future<typename FrameType::return_type>
connection::read_frame_compressed(socket_address, std::unique_ptr<compressor> &,
input_stream<char> &in) {
in.read_exactly({});
}
struct stream_frame {
using opt_buf_type = std::optional<rcv_buf>;
using return_type = opt_buf_type;
};
future<std::optional<rcv_buf>>
connection::read_stream_frame_compressed(input_stream<char> &in) {
read_frame_compressed<stream_frame>(peer_address(), _compressor, in);
}
} // namespace rpc
} // namespace seastar |
|
Do you have the build command? |
This requires both rpc-195796.cpp and merged.profdata. |
|
btw, to reproduce the crash, build clang 18.1.6 with asan or stack protector. Without them, there's a buffer overflow that isn't detected. The reason I saw the problem is that Fedora's clang 18.1.6 is build with -fstack-protector-all. |
|
@ellishg how can we make progress with this? |
|
@avikivity I'm not able to reproduce the error. It seems strange that you are getting the error "Do not know how to promote this operator!" since I think that might be during lowing to assembly, which should not happen with |
|
Did you build clang with asan or -fstack-protector-all? without them the rogue write isn't caught. This snippet reproduces it with Fedora 40's build (which has -fstack-protector-all enabled). Replace podman with docker if needed. You'll need rpc-195796.cpp and merged.profdata in the same directory. podman run -i -v $PWD:$PWD -w $PWD docker.io/fedora:40 <<EOF
dnf -y install clang
"/usr/bin/clang++" "-cc1" "-triple" "x86_64-redhat-linux-gnu" "-emit-llvm-bc" "-flto=thin" "-flto-unit" "-disable-free" "-clear-ast-before-backend" "-disable-llvm-verifier" "-discard-value-names" "-main-file-name" "rpc.cc" "-mrelocation-model" "static" "-mframe-pointer=none" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-funwind-tables=2" "-target-cpu" "westmere" "-debug-info-kind=constructor" "-dwarf-version=5" "-debugger-tuning=gdb" "--compress-debug-sections=zlib" "-fdebug-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-fdebug-prefix-map=/home/avi/scylla-maint=." "-fprofile-instrument=csllvm" "-fprofile-instrument-path=/home/avi/scylla-maint/build/release-cs-pgo/default_%m.profraw" "-fprofile-instrument-use-path=merged.profdata" "-fcoverage-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-sys-header-deps" "-D" "FMT_SHARED" "-D" "SEASTAR_API_LEVEL=7" "-D" "SEASTAR_DEFERRED_ACTION_REQUIRE_NOEXCEPT" "-D" "SEASTAR_HAS_MEMBARRIER" "-D" "SEASTAR_HAVE_ASAN_FIBER_SUPPORT" "-D" "SEASTAR_HAVE_HWLOC" "-D" "SEASTAR_HAVE_NUMA" "-D" "SEASTAR_HAVE_SYSTEMTAP_SDT" "-D" "SEASTAR_LOGGER_COMPILE_TIME_FMT" "-D" "SEASTAR_LOGGER_TYPE_STDOUT" "-D" "SEASTAR_PTHREAD_ATTR_SETAFFINITY_NP" "-D" "SEASTAR_SCHEDULING_GROUPS_COUNT=18" "-D" "SEASTAR_SSTRING" "-D" "SEASTAR_STRERROR_R_CHAR_P" "-D" "NDEBUG" "-U" "_FORTIFY_SOURCE" "-U" "NDEBUG" "-fmacro-prefix-map=/home/avi/scylla-maint=." "-fcoverage-prefix-map=/home/avi/scylla-maint=." "-O2" "-Wno-backend-plugin" "-std=gnu++23" "-fdeprecated-macro" "-ferror-limit" "19" "-fvisibility=hidden" "-fgnuc-version=4.2.1" "-fno-implicit-modules" "-fskip-odr-check-in-gmf" "-fcxx-exceptions" "-fexceptions" "-fcolor-diagnostics" "-vectorize-loops" "-vectorize-slp" "-mllvm" "-pgso=false" "-mllvm" "-enable-value-profiling=false" "-faddrsig" "-D__GCC_HAVE_DWARF2_CFI_ASM=1" "-x" "c++" "rpc-195796.cpp"
EOF |
|
Here's my "Don't know how to promote this operator" run: $ "/usr/bin/clang++" "-cc1" "-triple" "x86_64-redhat-linux-gnu" "-emit-llvm-bc" "-flto=thin" "-flto-unit" "-disable-free" "-clear-ast-before-backend" "-disable-llvm-verifier" "-discard-value-names" "-main-file-name" "rpc.cc" "-mrelocation-model" "static" "-mframe-pointer=none" "-fmath-errno" "-ffp-contract=on" "-fno-rounding-math" "-mconstructor-aliases" "-funwind-tables=2" "-target-cpu" "westmere" "-debug-info-kind=constructor" "-dwarf-version=5" "-debugger-tuning=gdb" "--compress-debug-sections=zlib" "-fdebug-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-fdebug-prefix-map=/home/avi/scylla-maint=." "-fprofile-instrument=csllvm" "-fprofile-instrument-path=/home/avi/scylla-maint/build/release-cs-pgo/default_%m.profraw" "-fprofile-instrument-use-path=merged.profdata" "-fcoverage-compilation-dir=/home/avi/scylla-maint/build/release-cs-pgo/seastar.lto" "-sys-header-deps" "-D" "FMT_SHARED" "-D" "SEASTAR_API_LEVEL=7" "-D" "SEASTAR_DEFERRED_ACTION_REQUIRE_NOEXCEPT" "-D" "SEASTAR_HAS_MEMBARRIER" "-D" "SEASTAR_HAVE_ASAN_FIBER_SUPPORT" "-D" "SEASTAR_HAVE_HWLOC" "-D" "SEASTAR_HAVE_NUMA" "-D" "SEASTAR_HAVE_SYSTEMTAP_SDT" "-D" "SEASTAR_LOGGER_COMPILE_TIME_FMT" "-D" "SEASTAR_LOGGER_TYPE_STDOUT" "-D" "SEASTAR_PTHREAD_ATTR_SETAFFINITY_NP" "-D" "SEASTAR_SCHEDULING_GROUPS_COUNT=18" "-D" "SEASTAR_SSTRING" "-D" "SEASTAR_STRERROR_R_CHAR_P" "-D" "NDEBUG" "-U" "_FORTIFY_SOURCE" "-U" "NDEBUG" "-fmacro-prefix-map=/home/avi/scylla-maint=." "-fcoverage-prefix-map=/home/avi/scylla-maint=." "-O2" "-Wno-backend-plugin" "-std=gnu++23" "-fdeprecated-macro" "-ferror-limit" "19" "-fvisibility=hidden" "-fgnuc-version=4.2.1" "-fno-implicit-modules" "-fskip-odr-check-in-gmf" "-fcxx-exceptions" "-fexceptions" "-fcolor-diagnostics" "-vectorize-loops" "-vectorize-slp" "-mllvm" "-pgso=false" "-mllvm" "-enable-value-profiling=false" "-faddrsig" "-D__GCC_HAVE_DWARF2_CFI_ASM=1" "-x" "c++" "rpc-195796.cpp" -emit-llvm -S -disable-llvm-optzns -o a.ll -Wno-all -Wno-writable-strings
fatal error: error in backend: Do not know how to promote this operator!Again with clang 18.1.6. I'll try with trunk. |
|
trunk is able to generate llvm, but that doesn't help in reduction. |
FWIW, i am able to reproduce this issue with the attached source file and profile data, and the clang compiler packaged in 18.1.6-3.fc40 on fedora 40. |
|
@ellishg looks like we are stalled again. I don't believe it's possible to reduce the reproducer, as HEAD probably rejects the provided profile measurements, and 18.1.6 isn't able to emit llvm without applying some coroutine related passes first. |
|
@ellishg what can I do next? look for another reviewer? close the pull request? |
|
cc: @mtrofin who has looked into profile related co-routine issues before. |
mtrofin
reviewed
Aug 7, 2024
avikivity
force-pushed
the
PGOUseFunc-setBranchWeights-fix-EdgeCounts-size
branch
from
b7db364 to
6072608
Compare
|
Update: renamed local |
|
Update:
|
avikivity
force-pushed
the
PGOUseFunc-setBranchWeights-fix-EdgeCounts-size
branch
from
08984a2 to
f7fe1e8
Compare
|
Update:
|
SetBranchWeights() calculates the size of the EdgeCounts vector using OutEdges.Size(), but this is an under-estimate with coroutines. Use the number of successors, as the vector will be indexed by the result of the GetSuccessorNumber() function. Rename the Size local, to make it clear what it refers to. A unit test, provided by @ellishg, is included. Fixes llvm#97962 (regression from ffd337b)
avikivity
force-pushed
the
PGOUseFunc-setBranchWeights-fix-EdgeCounts-size
branch
from
f7fe1e8 to
82ed77d
Compare
|
I believe the test failure is unrelated (somewhere in lldb) so I rebased to re-trigger the tests. |
|
@xur-llvm I believe this bugfix is waiting only for your approval, kindly review |
|
@mtrofin a week has passed, can we perhaps enlist a different reviewer? |
Let's merge it at this point. Do you have commit rights? |
I do not. I'm not an LLVM developer, just a random person on the internet. |
Taking care of it. |
Thanks! |
|
@avikivity Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/93/builds/4909 Here is the relevant piece of the build log for the reference |
|
this is the failed test: not sure if it is relevant. |
Seems like it's a transient issue on that bot. (Also, a change in instrumentation shouldn't affect a libc test... normally. But the status of the subsequent build, https://lab.llvm.org/buildbot/#/builders/93/builds/4910, is a stronger signal.) |
It's just cherrypicking. I already know it backports directly to 18.x, so 19.x should work too. |
|
/cherry-pick 46a4132 |
|
/pull-request #106823 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
SetBranchWeights() calculates the size of the EdgeCounts vector
using OutEdges.Size(), but this is an under-estimate with coroutines.
Use the number of successors, as the vector will be indexed by
the result of the GetSuccessorNumber() function.
Rename the Size local, to make it clear what it refers to.
A unit test, provided by @ellishg, is included.
Fixes #97962
(regression from ffd337b)