Refactor TransormPropagator to allow specifying a position and propagating to part of the DAG (#1775)

`MaxInfoPropagator` is renamed to `MaxInfoSpanningTree`, it now only does path-finding, and the propagation is in a separate class `MaxInfoSpanningTree::Propagator`. Same for `MaxRootDomainInfoPropagator`.

`MaxInfoSpanningTree` and `MaxRootDomainInfoSpanningTree`  now allow specifying a selector, which controls which subgraph should be included in path-finding.

`MaxRootDomainInfoSpanningTree` also gets a few new constructors for convenience to use.

`TransormPropagator` is now a subclass of `MaxInfoSpanningTree::Propagator`, so the way to use it has changed.

Now `MaxInfoSpanningTree` and `MaxRootDomainInfoSpanningTree` will store the path after generation so that the same path can be traversed multiple times. This will be useful to support use cases like new `computeAt`. Pseudo-code:
```C++
void TensorView::computeAt(TensorView tv, int pos) {
  auto ComputeAtSubgraphSelector selector(this, tv);
  MaxRootDomainInfoSpanningTree path(tv, pos, &selector);
  TransformPropagator propagator(tv, pos);
  path.traverse(&propagator);
  ComputeAtPosPropagator ca_propagator(tv, pos);
  path.traverse(&ca_propagator);
}
```

Author: zasdfgbnm

torch/csrc/jit/codegen/cuda/maxinfo_propagator.cpp

@@ -6,16 +6,24 @@ namespace jit {
 namespace fuser {
 namespace cuda {
 
-bool MaxInfoPropagator::Information::operator>(const Information& r) const {
+bool MaxInfoSpanningTree::Information::operator>(const Information& r) const {
   return r < *this;
 }
 
-bool MaxInfoPropagator::Information::operator==(const Information& r) const {
+bool MaxInfoSpanningTree::Information::operator==(const Information& r) const {
   return !(r < *this) && !(*this < r);
 }
 
-// Dijkstra
-void MaxInfoPropagator::run() {
+// Prim's algorithm
+MaxInfoSpanningTree::MaxInfoSpanningTree(
+    TensorView* reference,
+    std::shared_ptr<Information> reference_info,
+    Selector* selector)
+    : reference_(reference),
+      reference_info_(reference_info),
+      selector_(selector) {}
+
+void MaxInfoSpanningTree::compute_spanning_tree() {
   // A set that allows us to quickly tell if a tensor has been replayed. If yes,
   // then we will not bother computing if a new path to this tensor is worth
   // taking (because the answer is always not worth)
@@ -28,88 +36,114 @@ void MaxInfoPropagator::run() {
   // std::list instead of std::priority_queue because C++'s
   // std::priority_queue does not support increase-key, and might not be
   // deterministic either.
-  std::list<NextHopInfo> propagation(1);
-  propagation.back().from = nullptr;
-  propagation.back().to = reference;
-  propagation.back().info_to = reference_info;
+  std::list<NextHopWithInfo> candidates(1);
+  candidates.back().next_hop.from = nullptr;
+  candidates.back().next_hop.to = reference_;
+  candidates.back().info_to = reference_info_;
 
-  // Insert the given next hop the correct position in `propagation`. If there
+  // Insert the given next hop the correct position in `candidates`. If there
   // is an existing next hop that preserves more information, then we will just
   // discard `info`.
-  auto insertNextHopInfo = [&](const NextHopInfo& info) {
+  auto insertNextHop = [&](const NextHopWithInfo& info) {
     if (!*(info.info_from)) {
       // When there is no more information about the starting tensor,
-      // we are not interested in continuing the propagation.
+      // we are not interested in continuing the path-finding.
       return;
     }
     // Find if there is already a path to the dest tensor
     auto existing = std::find_if(
-        propagation.begin(), propagation.end(), [&](const NextHopInfo& i) {
-          return i.to == info.to;
+        candidates.begin(), candidates.end(), [&](const NextHopWithInfo& i) {
+          return i.next_hop.to == info.next_hop.to;
         });
     // Only insert if there is no existing path to the dest tensor, or the new
     // path preserves more information about the starting tensor.
-    if (existing == propagation.end() || *existing < info) {
-      if (existing != propagation.end()) {
-        propagation.erase(existing);
+    if (existing == candidates.end() || *existing < info) {
+      if (existing != candidates.end()) {
+        candidates.erase(existing);
       }
-      auto pos = std::upper_bound(propagation.begin(), propagation.end(), info);
-      propagation.insert(pos, info);
+      auto pos = std::upper_bound(candidates.begin(), candidates.end(), info);
+      candidates.insert(pos, info);
+    }
+  };
+
+  auto allowPasC = [this](TensorView* from, TensorView* to) {
+    if (selector_ == nullptr) {
+      return true;
     }
+    return selector_->allowPasC(from, to);
   };
 
-  while (!propagation.empty()) {
-    auto next_hop = propagation.back();
-    propagation.pop_back();
+  auto allowCasP = [this](TensorView* from, TensorView* to) {
+    if (selector_ == nullptr) {
+      return true;
+    }
+    return selector_->allowCasP(from, to);
+  };
+
+  while (!candidates.empty()) {
+    const auto next_hop_info = candidates.back();
+    const auto& next_hop = next_hop_info.next_hop;
+    candidates.pop_back();
 
     if (next_hop.from != nullptr) {
       // nullptr used to start from reference
-      switch (next_hop.type) {
-        case NextHopType::C_AS_P:
-          propagateTvCasP(next_hop.from, next_hop.to);
-          break;
-        case NextHopType::P_AS_C:
-          propagateTvPasC(next_hop.from, next_hop.to);
-          break;
-      }
+      path_.push_back(next_hop);
     }
     replayed.emplace(next_hop.to);
 
     for (auto consumer_tv : ir_utils::consumerTvsOf(next_hop.to)) {
-      if (replayed.count(consumer_tv)) {
+      if (replayed.count(consumer_tv) || !allowCasP(next_hop.to, consumer_tv)) {
         continue;
       }
-      insertNextHopInfo(
-          {.type = NextHopType::C_AS_P,
-           .from = next_hop.to,
-           .to = consumer_tv,
-           .info_from = next_hop.info_to,
-           .info_to =
-               computeInfoCasP(next_hop.to, consumer_tv, next_hop.info_to)});
+      insertNextHop(
+          {.next_hop =
+               {.type = NextHopType::C_AS_P,
+                .from = next_hop.to,
+                .to = consumer_tv},
+           .info_from = next_hop_info.info_to,
+           .info_to = computeInfoCasP(
+               next_hop.to, consumer_tv, next_hop_info.info_to)});
     }
 
     for (auto producer_tv : ir_utils::producerTvsOf(next_hop.to)) {
-      if (replayed.count(producer_tv)) {
+      if (replayed.count(producer_tv) || !allowPasC(next_hop.to, producer_tv)) {
         continue;
       }
-      insertNextHopInfo(
-          {.type = NextHopType::P_AS_C,
-           .from = next_hop.to,
-           .to = producer_tv,
-           .info_from = next_hop.info_to,
-           .info_to =
-               computeInfoPasC(next_hop.to, producer_tv, next_hop.info_to)});
+      insertNextHop(
+          {.next_hop =
+               {.type = NextHopType::P_AS_C,
+                .from = next_hop.to,
+                .to = producer_tv},
+           .info_from = next_hop_info.info_to,
+           .info_to = computeInfoPasC(
+               next_hop.to, producer_tv, next_hop_info.info_to)});
     }
   }
 }
 
-MaxRootDomainInfoPropagator::RootDomainInfo::operator bool() const {
+void MaxInfoSpanningTree::traverse(Propagator* propagator) {
+  if (path_.empty()) {
+    compute_spanning_tree();
+  }
+  for (const auto& next_hop : path_) {
+    switch (next_hop.type) {
+      case NextHopType::C_AS_P:
+        propagator->propagateTvCasP(next_hop.from, next_hop.to);
+        break;
+      case NextHopType::P_AS_C:
+        propagator->propagateTvPasC(next_hop.from, next_hop.to);
+        break;
+    }
+  }
+}
+
+MaxRootDomainInfoSpanningTree::RootDomainInfo::operator bool() const {
   return !info.empty();
 }
 
-bool MaxRootDomainInfoPropagator::RootDomainInfo::operator<(
-    const MaxInfoPropagator::Information& r) const {
-  auto rr = dynamic_cast<const MaxRootDomainInfoPropagator::RootDomainInfo&>(r);
+bool MaxRootDomainInfoSpanningTree::RootDomainInfo::operator<(
+    const Information& r) const {
+  auto rr = dynamic_cast<const RootDomainInfo&>(r);
   if (info.size() != rr.info.size()) {
     return info.size() < rr.info.size();
   }
@@ -174,17 +208,17 @@ std::unordered_set<IterDomain*> mapRFactorToRoot(
 // Given the preserved reference root ID info of a producer, compute
 // the corresponding info in consumer. The given info may be represented by
 // producer's root domain, or rfactor domain, depending on how we reached the
-// producer during propagation. If the given info is already represented with
+// producer during path-finding. If the given info is already represented with
 // producer's rfactor domain, then we directly map it to the consumer's root
 // domain. If the given info is represented with producer's root domain, we need
 // to first map it to the rfactor domain of the producer, then we can map it to
 // the consumer's root domain. The computed info will be represented by root
 // domain as root domain contains the raw information.
-std::shared_ptr<MaxInfoPropagator::Information> MaxRootDomainInfoPropagator::
+std::shared_ptr<MaxInfoSpanningTree::Information> MaxRootDomainInfoSpanningTree::
     computeInfoCasP(
         TensorView* from,
         TensorView* to,
-        std::shared_ptr<Information> from_info) {
+        std::shared_ptr<Information> from_info) const {
   RootDomainInfo result;
 
   TensorView* producer = from;
@@ -231,17 +265,17 @@ std::shared_ptr<MaxInfoPropagator::Information> MaxRootDomainInfoPropagator::
 // Given the preserved reference root ID info of a consumer, compute
 // the corresponding info in producer. The given info may be represented by
 // consumer's root domain, or rfactor domain, depending on how we reached the
-// consumer during propagation. If the given info is already represented with
+// consumer during path-finding. If the given info is already represented with
 // consumer's root domain, then we directly map it to the producer's rfactor
 // domain. If the given info is represented with consumer's rfactor domain, we
 // need to first map it to the root domain of the consumer, then we can map it
 // to the producer's rfactor domain. The computed info will be represented by
 // rfactor domain as rfactor domain contains the raw information.
-std::shared_ptr<MaxInfoPropagator::Information> MaxRootDomainInfoPropagator::
+std::shared_ptr<MaxInfoSpanningTree::Information> MaxRootDomainInfoSpanningTree::
     computeInfoPasC(
         TensorView* from,
         TensorView* to,
-        std::shared_ptr<Information> from_info) {
+        std::shared_ptr<Information> from_info) const {
   RootDomainInfo result;
 
   TensorView* producer = to;
@@ -279,9 +313,9 @@ std::shared_ptr<MaxInfoPropagator::Information> MaxRootDomainInfoPropagator::
     // We will stop at the rfactor ids in producer, and will not further map
     // them into root ids in producer. This means, we only keep the unprocessed
     // raw information of a tensor. This behavior is important to make sure that
-    // info is as accurate as possible throughout the propagation.
+    // info is as accurate as possible throughout the path-finding.
     //
-    // For example, if we do a C->P->C' propagation, we want to do
+    // For example, in a C->P->C' path, we want to do
     //   C(root) -> P(rfactor) -> C'(root)
     // instead of
     //   C(root) -> P(rfactor) -> P(root) -> P(rfactor) -> C'(root)
@@ -305,6 +339,47 @@ std::shared_ptr<MaxInfoPropagator::Information> MaxRootDomainInfoPropagator::
   return std::make_shared<RootDomainInfo>(std::move(result));
 }
 
+std::shared_ptr<MaxRootDomainInfoSpanningTree::RootDomainInfo>
+MaxRootDomainInfoSpanningTree::getReferenceRootIDInfo(TensorView* tv) {
+  RootDomainInfo result;
+  const auto& root_domain = tv->getRootDomain();
+  result.info.reserve(root_domain.size());
+  for (auto id : root_domain) {
+    result.info.emplace_back(RootIDInfo{{id}, true, false});
+  }
+  return std::make_shared<RootDomainInfo>(std::move(result));
+}
+
+std::shared_ptr<MaxRootDomainInfoSpanningTree::RootDomainInfo>
+MaxRootDomainInfoSpanningTree::getReferenceRootIDInfo(
+    TensorView* tv,
+    int64_t leaf_pos) {
+  if (leaf_pos < 0) {
+    leaf_pos += int64_t(tv->nDims()) + 1;
+  }
+  TORCH_CHECK(
+      leaf_pos >= 0 && leaf_pos <= tv->nDims(),
+      "MaxRootDomainInfoSpanningTree called on an leaf_pos outside valid range.");
+  RootDomainInfo result;
+  const auto& root_domain = tv->getMaybeRFactorDomain();
+  const auto& leaf_domain = tv->domain()->domain();
+  std::unordered_set<IterDomain*> selected_leaves(
+      leaf_domain.begin(), leaf_domain.begin() + leaf_pos);
+  for (auto id : root_domain) {
+    if (selected_leaves.count(id) > 0) {
+      result.info.emplace_back(RootIDInfo{{id}, true, tv->hasRFactor()});
+      continue;
+    }
+    for (auto selected_leaf_id : selected_leaves) {
+      if (DependencyCheck::isDependencyOf(id, selected_leaf_id)) {
+        result.info.emplace_back(RootIDInfo{{id}, true, tv->hasRFactor()});
+        break;
+      }
+    }
+  }
+  return std::make_shared<RootDomainInfo>(std::move(result));
+}
+
 } // namespace cuda
 } // namespace fuser
 } // namespace jit