Revision 47378ee5bd7cb304be9d61e0a636982c8a2623d0 authored by Steven Johnson on 29 January 2024, 01:28:13 UTC, committed by GitHub on 29 January 2024, 01:28:13 UTC
* Upgrade clang-format and clang-tidy to use LLVM 17 * trigger buildbots * trigger buildbots * trigger buildbots * trigger buildbots * Enable `bugprone-switch-missing-default-case` ...and fix existing warnings. * Update .clang-tidy * Update Parameter.cpp * Update .clang-tidy * Update .clang-tidy * Update .clang-tidy * Update .clang-tidy * Update CPlusPlusMangle.cpp
1 parent 4b2d211
Schedule.cpp
#include "Schedule.h"
#include "Func.h"
#include "Function.h"
#include "IR.h"
#include "IRMutator.h"
#include "Var.h"
namespace {
const char *const root_looplevel_name = "__root";
const char *const inline_looplevel_name = "";
const char *const undefined_looplevel_name = "__undefined_loop_level_var_name";
} // namespace
namespace Halide {
namespace Internal {
struct LoopLevelContents {
mutable RefCount ref_count;
// Note: func_name is empty for inline or root.
std::string func_name;
// If set to -1, this loop level does not refer to a particular stage of the
// function. 0 refers to initial stage, 1 refers to the 1st update stage, etc.
int stage_index;
// TODO: these two fields should really be VarOrRVar,
// but cyclical include dependencies make this challenging.
std::string var_name;
bool is_rvar;
bool locked;
LoopLevelContents(const std::string &func_name,
const std::string &var_name,
bool is_rvar,
int stage_index,
bool locked)
: func_name(func_name), stage_index(stage_index), var_name(var_name),
is_rvar(is_rvar), locked(locked) {
}
};
template<>
RefCount &ref_count<LoopLevelContents>(const LoopLevelContents *p) noexcept {
return p->ref_count;
}
template<>
void destroy<LoopLevelContents>(const LoopLevelContents *p) {
delete p;
}
} // namespace Internal
LoopLevel::LoopLevel(const std::string &func_name, const std::string &var_name,
bool is_rvar, int stage_index, bool locked)
: contents(new Internal::LoopLevelContents(func_name, var_name, is_rvar, stage_index, locked)) {
}
LoopLevel::LoopLevel(const Internal::Function &f, const VarOrRVar &v, int stage_index)
: LoopLevel(f.name(), v.name(), v.is_rvar, stage_index, false) {
}
LoopLevel::LoopLevel(const Func &f, const VarOrRVar &v, int stage_index)
: LoopLevel(f.function().name(), v.name(), v.is_rvar, stage_index, false) {
}
// Note that even 'undefined' LoopLevels get a LoopLevelContents; this is deliberate,
// as we want to be able to create an undefined LoopLevel, pass it to another function
// to use, then mutate it afterwards via 'set()'.
LoopLevel::LoopLevel()
: LoopLevel("", undefined_looplevel_name, false, -1, false) {
}
void LoopLevel::check_defined() const {
internal_assert(defined());
}
void LoopLevel::check_locked() const {
// A LoopLevel can be in one of two states:
// - Unlocked (the default state): An unlocked LoopLevel can be mutated freely (via the set() method),
// but cannot be inspected (calls to func(), var(), is_inlined(), is_root(), etc.
// will assert-fail). This is the only sort of LoopLevel that most user code will ever encounter.
// - Locked: Once a LoopLevel is locked, it can be freely inspected, but no longer mutated.
// Halide locks all LoopLevels during the lowering process to ensure that no user
// code (e.g. custom passes) can interfere with invariants.
user_assert(contents->locked)
<< "Cannot inspect an unlocked LoopLevel: "
<< contents->func_name << "." << contents->var_name
<< "\n";
}
void LoopLevel::check_defined_and_locked() const {
check_defined();
check_locked();
}
void LoopLevel::set(const LoopLevel &other) {
// Don't check locked(), since we don't care if it's defined() or not
user_assert(!contents->locked)
<< "Cannot call set() on a locked LoopLevel: "
<< contents->func_name << "." << contents->var_name
<< "\n";
contents->func_name = other.contents->func_name;
contents->stage_index = other.contents->stage_index;
contents->var_name = other.contents->var_name;
contents->is_rvar = other.contents->is_rvar;
}
LoopLevel &LoopLevel::lock() {
contents->locked = true;
// If you have an undefined LoopLevel at the point we're
// locking it (i.e., start of lowering), you've done something wrong,
// so let's give a more useful error message.
user_assert(defined())
<< "There should be no undefined LoopLevels at the start of lowering. "
<< "(Did you mean to use LoopLevel::inlined() instead of LoopLevel() ?)";
return *this;
}
bool LoopLevel::defined() const {
check_locked();
return contents->var_name != undefined_looplevel_name;
}
std::string LoopLevel::func() const {
check_defined_and_locked();
return contents->func_name;
}
int LoopLevel::stage_index() const {
check_defined_and_locked();
internal_assert(contents->stage_index >= 0);
return contents->stage_index;
}
VarOrRVar LoopLevel::var() const {
check_defined_and_locked();
internal_assert(!is_inlined() && !is_root());
return VarOrRVar(contents->var_name, contents->is_rvar);
}
/*static*/
LoopLevel LoopLevel::inlined() {
return LoopLevel("", inline_looplevel_name, false, -1);
}
bool LoopLevel::is_inlined() const {
// It's OK to be undefined (just return false).
check_locked();
return contents->var_name == inline_looplevel_name;
}
/*static*/
LoopLevel LoopLevel::root() {
return LoopLevel("", root_looplevel_name, false, -1);
}
bool LoopLevel::is_root() const {
// It's OK to be undefined (just return false).
check_locked();
return contents->var_name == root_looplevel_name;
}
int LoopLevel::get_stage_index() const {
return contents->stage_index;
}
std::string LoopLevel::func_name() const {
return contents->func_name;
}
std::string LoopLevel::var_name() const {
return contents->var_name;
}
bool LoopLevel::is_rvar() const {
return contents->is_rvar;
}
bool LoopLevel::locked() const {
return contents->locked;
}
std::string LoopLevel::to_string() const {
check_defined_and_locked();
if (contents->stage_index == -1) {
return contents->func_name + "." + contents->var_name;
} else {
return contents->func_name + ".s" + std::to_string(contents->stage_index) + "." + contents->var_name;
}
}
bool LoopLevel::match(const std::string &loop) const {
check_defined_and_locked();
if (contents->stage_index == -1) {
return Internal::starts_with(loop, contents->func_name + ".") &&
Internal::ends_with(loop, "." + contents->var_name);
} else {
std::string prefix = contents->func_name + ".s" + std::to_string(contents->stage_index) + ".";
return Internal::starts_with(loop, prefix) &&
Internal::ends_with(loop, "." + contents->var_name);
}
}
bool LoopLevel::match(const LoopLevel &other) const {
check_defined_and_locked();
other.check_defined_and_locked();
return (contents->func_name == other.contents->func_name &&
(contents->var_name == other.contents->var_name ||
Internal::ends_with(contents->var_name, "." + other.contents->var_name) ||
Internal::ends_with(other.contents->var_name, "." + contents->var_name)) &&
(contents->stage_index == other.contents->stage_index));
}
bool LoopLevel::operator==(const LoopLevel &other) const {
check_defined_and_locked();
other.check_defined_and_locked();
return (contents->func_name == other.contents->func_name) &&
(contents->stage_index == other.contents->stage_index) &&
(contents->var_name == other.contents->var_name);
}
namespace Internal {
typedef std::map<FunctionPtr, FunctionPtr> DeepCopyMap;
/** A schedule for a halide function, which defines where, when, and
* how it should be evaluated. */
struct FuncScheduleContents {
mutable RefCount ref_count;
LoopLevel store_level, compute_level, hoist_storage_level;
std::vector<StorageDim> storage_dims;
std::vector<Bound> bounds;
std::vector<Bound> estimates;
std::map<std::string, Internal::FunctionPtr> wrappers;
MemoryType memory_type = MemoryType::Auto;
bool memoized = false;
bool async = false;
// This is an extent of the ring buffer and expected to be a positive integer.
Expr ring_buffer;
Expr memoize_eviction_key;
FuncScheduleContents()
: store_level(LoopLevel::inlined()), compute_level(LoopLevel::inlined()), hoist_storage_level(LoopLevel::inlined()) {
}
// Pass an IRMutator through to all Exprs referenced in the FuncScheduleContents
void mutate(IRMutator *mutator) {
for (Bound &b : bounds) {
if (b.min.defined()) {
b.min = mutator->mutate(b.min);
}
if (b.extent.defined()) {
b.extent = mutator->mutate(b.extent);
}
if (b.modulus.defined()) {
b.modulus = mutator->mutate(b.modulus);
}
if (b.remainder.defined()) {
b.remainder = mutator->mutate(b.remainder);
}
}
for (Bound &b : estimates) {
if (b.min.defined()) {
b.min = mutator->mutate(b.min);
}
if (b.extent.defined()) {
b.extent = mutator->mutate(b.extent);
}
if (b.modulus.defined()) {
b.modulus = mutator->mutate(b.modulus);
}
if (b.remainder.defined()) {
b.remainder = mutator->mutate(b.remainder);
}
}
}
};
template<>
RefCount &ref_count<FuncScheduleContents>(const FuncScheduleContents *p) noexcept {
return p->ref_count;
}
template<>
void destroy<FuncScheduleContents>(const FuncScheduleContents *p) {
delete p;
}
/** A schedule for a sigle halide stage_index, which defines where, when, and
* how it should be evaluated. */
struct StageScheduleContents {
mutable RefCount ref_count;
std::vector<ReductionVariable> rvars;
std::vector<Split> splits;
std::vector<Dim> dims;
std::vector<PrefetchDirective> prefetches;
FuseLoopLevel fuse_level;
std::vector<FusedPair> fused_pairs;
bool touched = false;
bool allow_race_conditions = false;
bool atomic = false;
bool override_atomic_associativity_test = false;
StageScheduleContents()
: fuse_level(FuseLoopLevel()) {
}
// Pass an IRMutator through to all Exprs referenced in the StageScheduleContents
void mutate(IRMutator *mutator) {
for (ReductionVariable &r : rvars) {
if (r.min.defined()) {
r.min = mutator->mutate(r.min);
}
if (r.extent.defined()) {
r.extent = mutator->mutate(r.extent);
}
}
for (Split &s : splits) {
if (s.factor.defined()) {
s.factor = mutator->mutate(s.factor);
}
}
for (PrefetchDirective &p : prefetches) {
if (p.offset.defined()) {
p.offset = mutator->mutate(p.offset);
}
}
}
};
template<>
RefCount &ref_count<StageScheduleContents>(const StageScheduleContents *p) noexcept {
return p->ref_count;
}
template<>
void destroy<StageScheduleContents>(const StageScheduleContents *p) {
delete p;
}
FuncSchedule::FuncSchedule()
: contents(new FuncScheduleContents) {
}
FuncSchedule FuncSchedule::deep_copy(
std::map<FunctionPtr, FunctionPtr> &copied_map) const {
internal_assert(contents.defined()) << "Cannot deep-copy undefined FuncSchedule\n";
FuncSchedule copy;
copy.contents->store_level = contents->store_level;
copy.contents->compute_level = contents->compute_level;
copy.contents->hoist_storage_level = contents->hoist_storage_level;
copy.contents->storage_dims = contents->storage_dims;
copy.contents->bounds = contents->bounds;
copy.contents->estimates = contents->estimates;
copy.contents->memory_type = contents->memory_type;
copy.contents->memoized = contents->memoized;
copy.contents->memoize_eviction_key = contents->memoize_eviction_key;
copy.contents->async = contents->async;
copy.contents->ring_buffer = contents->ring_buffer;
// Deep-copy wrapper functions.
for (const auto &iter : contents->wrappers) {
FunctionPtr &copied_func = copied_map[iter.second];
internal_assert(copied_func.defined()) << Function(iter.second).name() << "\n";
copy.contents->wrappers[iter.first] = copied_func;
}
internal_assert(copy.contents->wrappers.size() == contents->wrappers.size());
return copy;
}
MemoryType FuncSchedule::memory_type() const {
return contents->memory_type;
}
MemoryType &FuncSchedule::memory_type() {
return contents->memory_type;
}
bool &FuncSchedule::memoized() {
return contents->memoized;
}
bool FuncSchedule::memoized() const {
return contents->memoized;
}
Expr &FuncSchedule::memoize_eviction_key() {
return contents->memoize_eviction_key;
}
Expr FuncSchedule::memoize_eviction_key() const {
return contents->memoize_eviction_key;
}
bool &FuncSchedule::async() {
return contents->async;
}
bool FuncSchedule::async() const {
return contents->async;
}
Expr &FuncSchedule::ring_buffer() {
return contents->ring_buffer;
}
Expr &FuncSchedule::ring_buffer() const {
return contents->ring_buffer;
}
std::vector<StorageDim> &FuncSchedule::storage_dims() {
return contents->storage_dims;
}
const std::vector<StorageDim> &FuncSchedule::storage_dims() const {
return contents->storage_dims;
}
std::vector<Bound> &FuncSchedule::bounds() {
return contents->bounds;
}
const std::vector<Bound> &FuncSchedule::bounds() const {
return contents->bounds;
}
std::vector<Bound> &FuncSchedule::estimates() {
return contents->estimates;
}
const std::vector<Bound> &FuncSchedule::estimates() const {
return contents->estimates;
}
std::map<std::string, Internal::FunctionPtr> &FuncSchedule::wrappers() {
return contents->wrappers;
}
const std::map<std::string, Internal::FunctionPtr> &FuncSchedule::wrappers() const {
return contents->wrappers;
}
void FuncSchedule::add_wrapper(const std::string &f,
const Internal::FunctionPtr &wrapper) {
if (contents->wrappers.count(f)) {
if (f.empty()) {
user_warning << "Replacing previous definition of global wrapper in function \""
<< f << "\"\n";
} else {
internal_error << "Wrapper redefinition in function \"" << f << "\" is not allowed\n";
}
}
contents->wrappers[f] = wrapper;
}
LoopLevel &FuncSchedule::store_level() {
return contents->store_level;
}
LoopLevel &FuncSchedule::compute_level() {
return contents->compute_level;
}
LoopLevel &FuncSchedule::hoist_storage_level() {
return contents->hoist_storage_level;
}
const LoopLevel &FuncSchedule::store_level() const {
return contents->store_level;
}
const LoopLevel &FuncSchedule::compute_level() const {
return contents->compute_level;
}
const LoopLevel &FuncSchedule::hoist_storage_level() const {
return contents->hoist_storage_level;
}
void FuncSchedule::accept(IRVisitor *visitor) const {
for (const Bound &b : bounds()) {
if (b.min.defined()) {
b.min.accept(visitor);
}
if (b.extent.defined()) {
b.extent.accept(visitor);
}
if (b.modulus.defined()) {
b.modulus.accept(visitor);
}
if (b.remainder.defined()) {
b.remainder.accept(visitor);
}
}
for (const Bound &b : estimates()) {
if (b.min.defined()) {
b.min.accept(visitor);
}
if (b.extent.defined()) {
b.extent.accept(visitor);
}
if (b.modulus.defined()) {
b.modulus.accept(visitor);
}
if (b.remainder.defined()) {
b.remainder.accept(visitor);
}
}
if (memoize_eviction_key().defined()) {
memoize_eviction_key().accept(visitor);
}
}
void FuncSchedule::mutate(IRMutator *mutator) {
if (contents.defined()) {
contents->mutate(mutator);
}
}
StageSchedule::StageSchedule()
: contents(new StageScheduleContents) {
}
StageSchedule::StageSchedule(const std::vector<ReductionVariable> &rvars, const std::vector<Split> &splits,
const std::vector<Dim> &dims, const std::vector<PrefetchDirective> &prefetches,
const FuseLoopLevel &fuse_level, const std::vector<FusedPair> &fused_pairs,
bool touched, bool allow_race_conditions, bool atomic, bool override_atomic_associativity_test)
: contents(new StageScheduleContents) {
contents->rvars = rvars;
contents->splits = splits;
contents->dims = dims;
contents->prefetches = prefetches;
contents->fuse_level = fuse_level;
contents->fused_pairs = fused_pairs;
contents->touched = touched;
contents->allow_race_conditions = allow_race_conditions;
contents->atomic = atomic;
contents->override_atomic_associativity_test = override_atomic_associativity_test;
}
StageSchedule StageSchedule::get_copy() const {
internal_assert(contents.defined()) << "Cannot copy undefined Schedule\n";
StageSchedule copy;
copy.contents->rvars = contents->rvars;
copy.contents->splits = contents->splits;
copy.contents->dims = contents->dims;
copy.contents->prefetches = contents->prefetches;
copy.contents->fuse_level = contents->fuse_level;
copy.contents->fused_pairs = contents->fused_pairs;
copy.contents->touched = contents->touched;
copy.contents->allow_race_conditions = contents->allow_race_conditions;
copy.contents->atomic = contents->atomic;
copy.contents->override_atomic_associativity_test = contents->override_atomic_associativity_test;
return copy;
}
bool &StageSchedule::touched() {
return contents->touched;
}
bool StageSchedule::touched() const {
return contents->touched;
}
std::vector<ReductionVariable> &StageSchedule::rvars() {
return contents->rvars;
}
const std::vector<ReductionVariable> &StageSchedule::rvars() const {
return contents->rvars;
}
const std::vector<Split> &StageSchedule::splits() const {
return contents->splits;
}
std::vector<Split> &StageSchedule::splits() {
return contents->splits;
}
std::vector<Dim> &StageSchedule::dims() {
return contents->dims;
}
const std::vector<Dim> &StageSchedule::dims() const {
return contents->dims;
}
std::vector<PrefetchDirective> &StageSchedule::prefetches() {
return contents->prefetches;
}
const std::vector<PrefetchDirective> &StageSchedule::prefetches() const {
return contents->prefetches;
}
FuseLoopLevel &StageSchedule::fuse_level() {
return contents->fuse_level;
}
const FuseLoopLevel &StageSchedule::fuse_level() const {
return contents->fuse_level;
}
std::vector<FusedPair> &StageSchedule::fused_pairs() {
return contents->fused_pairs;
}
const std::vector<FusedPair> &StageSchedule::fused_pairs() const {
return contents->fused_pairs;
}
bool &StageSchedule::allow_race_conditions() {
return contents->allow_race_conditions;
}
bool StageSchedule::allow_race_conditions() const {
return contents->allow_race_conditions;
}
bool &StageSchedule::atomic() {
return contents->atomic;
}
bool StageSchedule::atomic() const {
return contents->atomic;
}
bool &StageSchedule::override_atomic_associativity_test() {
return contents->override_atomic_associativity_test;
}
bool StageSchedule::override_atomic_associativity_test() const {
return contents->override_atomic_associativity_test;
}
void StageSchedule::accept(IRVisitor *visitor) const {
for (const ReductionVariable &r : rvars()) {
if (r.min.defined()) {
r.min.accept(visitor);
}
if (r.extent.defined()) {
r.extent.accept(visitor);
}
}
for (const Split &s : splits()) {
if (s.factor.defined()) {
s.factor.accept(visitor);
}
}
for (const PrefetchDirective &p : prefetches()) {
if (p.offset.defined()) {
p.offset.accept(visitor);
}
}
}
void StageSchedule::mutate(IRMutator *mutator) {
if (contents.defined()) {
contents->mutate(mutator);
}
}
} // namespace Internal
} // namespace Halide
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...