Fix detection of unmappable root domains #1952

naoyam · 2022-09-02T04:55:04Z

ComputeAtRootDomainMap flags domains that should not be mapped due to
reductions. Previously, checking if a domain potentially causes an
invalid mapping is only done with one domain in each group of domains
that are found to be mappable so far. That's not actually sufficient as
the unmappable domain set is created just once with no root mapping
information. The fix is to check all consumer domains of a producer
tensor. A small other fix is also done to address a different problem
discovered after the first fix.

Fixes #1950

ComputeAtRootDomainMap flags domains that should not be mapped due to reductions. Previously, checking if a domain potentially causes an invalid mapping is only done with one domain in each group of domains that are found to be mappable so far. That's not actually sufficient as the unmappable domain set is created just once with no root mapping information. The fix is to check all consumer domains of a producer tensor. A small other fix is also done to address a different problem discovered after the first fix.

naoyam · 2022-09-02T04:55:49Z

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

+    const DomainKeySet& consumer_domains,
    const ComputeAtRootDomainMap& root_map) const {
+  // Check each reduction domain if any of the consumer domains
+  // conflicts with it
  for (const auto& kv : reduction_domains_) {
    const DomainKey& reduction_domain = kv.first;
+    // Domains that must not be mapped with the reduction domain
    const DomainKeySet& incompatible_domains = kv.second;
-    DomainKey consumer_domain_with_reduction;
-    bool reduction_found = false;
+    // Input domains to the reduction domain
    const auto& input_keys = reduction_domain_inputs_.at(reduction_domain);
-    for (const DomainKey& consumer_domain : consumer_domains) {
-      for (const auto& input_key : input_keys) {
-        if (input_key == consumer_domain) {
-          consumer_domain_with_reduction = consumer_domain;
-          reduction_found = true;
-          break;
-        }
-      }
-    }
-    if (!reduction_found) {
+    // Check if any of the consumer domains is an input to the
+    // reduction
+    auto it = std::find_if(
+        consumer_domains.begin(),
+        consumer_domains.end(),
+        [&](const auto& consumer_domain) {
+          return std::find(
+                     input_keys.begin(), input_keys.end(), consumer_domain) !=
+              input_keys.end();
+        });
+    // None of the consumer domains is used for the reduction
+    // domain. They should be safe with respect to this reduction
+    // domain
+    if (it == consumer_domains.end()) {
      continue;
    }
-    // Make sure no incompatible domains will be merged with the reduction
-    // domain.
+
+    // A consumer domain that is an input to the reduction domain
+    const DomainKey& input_to_reduction = *it;
+
+    // Check if mapping input_to_reduction with the other domains in
+    // consumer_domains. If there's a domain that is a consumer of the
+    // reduction, they must not be mapped together
    for (const auto& consumer_domain : consumer_domains) {
-      if (consumer_domain == consumer_domain_with_reduction) {
+      if (consumer_domain == input_to_reduction) {


Just additional comments and cleanup

naoyam · 2022-09-02T04:56:37Z

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

+
  // Can't map if reduction output domains would be mapped
-  if (incompatible_domains_.isReductionOutputMapped(
-          unique_domains, root_map_) &&
+  if (incompatible_domains_.isReductionOutputMapped(domains, root_map_) &&


This is the main fix. Just checking unique domains isn't sufficient.

naoyam · 2022-09-02T04:59:11Z

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

+      // Do not include the tensor itself in its consumers
+      if (tv == out_tv) {
+        continue;
+      }


This is another related fix. Here, we are gathering IterDomains of all consumer tensors of a tensor with a reduction IterDomain. For this logic, we don't want to have the same tensor as its consumer.

shmsong

LGTM. Thanks for fixing this!

Syncing nvfuser devel branch to upstream master. https://github.com/csarofeen/pytorch/ Codegen changes include: * codegen improvement: i. allow non-root trivial reductions, allow empty/no-op fusion ii. fixes vectorization checks and size calculation iii. bank conflict handle improvement iv. enables transpose scheduler * misc: i. CI tests failure fixes ii. cpp tests file clean up iii. trivial forwarding supports added in codegen runtime iv. added factory methods support in codegen Commits that's in this PR from the devel branch: ``` 7117a7e patching nvfuser conv cudnn test numerics mismatch (#2048) 65af1a4 Inserting sync for redundant parallel types is already done at the (#2023) 6ac74d1 Fix sync map (#2047) f5bca33 Bank conflict checker improvements (#2032) d2ca7e3 Minor update on cp.async code generation. (#1901) d36cf61 Test file cleanup (#2040) 0b8e83f Allow non-root trivial reductions (#2037) a2dfe40 Fix vectorize size calculation (#2035) e040676 Use withPredicate to replace setPredicate to maintain Exprs immutable (#2025) 197221b removing ci workflow (#2034) 40e2703 Reduction rand like patch (#2031) bc77266 Add utility for checking bank conflict of shared memory (#2029) ddd1cf7 Add back FusionReductionWithTrivialReduction_CUDA (#2030) fbd97e5 Revert "Cleanup trivial reduction workarounds (#2006)" (#2024) bca20c1 Cleanup trivial reduction workarounds (#2006) e4b6585 Trivial forwarding (#1995) 1a0e355 Fix contiguity analysis of predicates to match updated contiguity. (#1991) a4effa6 Enable output allocation cache (#2010) 35440b7 Patching bn inference (#2016) 0f9f0b4 Add matmul benchmark (#2007) 45045cd Enable tests previously disabled due to an aliasing bug (#2005) 967aa77 Contiguous indexing for View operations (#1990) a43cb20 Make inlining even more modular (#2004) dc45835 Test util cleanup (#2003) 3ca21eb More strict validation (#2000) a7a7d57 Fix build problem (#1999) fc235b0 Just fixes comments (#1998) 482386c cleanup (#1997) 4cbe0db Improve divisible split detection (#1970) 42ccc52 Minor build fix. (#1996) fcf8c09 Cleanup of lower_utils.cpp: Isolate out GpuLower usage (#1989) 15f2f6d Move ConcretizedBroadcastDomains to shared_ptr in GpuLower. (#1988) 8f1c7f5 Minor cleanup lower_unroll.cpp (#1994) 1d9858c Minor cleanup (#1992) f262d9c Add support for uniform RNG (#1986) eb1dad1 Remove non-const functions, remove GpuLower instance on build, pass in ca_map. (#1987) 634820c Add support for some empty fusion (#1981) eabe8d8 Segment self mapping fusions (#1954) e96aacf Enable Transpose operation (#1882) 425dce2 Add a null scheduler that helps segmenting away no-op schedules (#1835) 306d4a6 Fix canScheduleCompileTime check of transpose scheduler (#1969) b1bd32c Minor fix (#1967) bd93578 Enable transpose scheduler (#1927) b7a206e Move scheduler vectorize utilities into their own file (#1959) d9420e4 View scheduling (#1928) c668e13 Upstream push ci fixes (#1965) c40202b Fix dump effective bandwidth (#1962) 93505bc WAR on index mapping when exact and permissive maps differ (#1960) 45e95fd Allow splitting inner-most ID to create virtual innermost ID in transpose scheduler (#1930) a3ecb33 Improve the comments at the beginning of index_compute.h (#1946) f7bc341 Remove unused variables (#1955) df3393a Some cleanup (#1957) 7d1d7c8 TVDomainGuard factory (#1953) 357ba22 Fill allocation with nan on tests (#1956) 8eafc54 Fix detection of unmappable root domains (#1952) 90a51f2 Some indexing cleanups, Add eye support (#1940) ddc01e4 Exclude unsupported data types (#1951) 992e17c test the groups the same order as they are merged (#1949) 208262b Move detection of self mapping IDs to IterDomainGraph from (#1941) ac4de38 Merge pull request #1945 from csarofeen/master_merge_0828 6310948 Add full, full_like, zeros, zeros_like, ones, ones_like (#1943) aab10bc Merge remote-tracking branch 'upstream/viable/strict' into HEAD 4c254c0 Fix arange when step is negative (#1942) 89330aa Tensor factories must set the output shape as its input (#1939) ``` RUN_TORCHBENCH: nvfuser Differential Revision: [D40869846](https://our.internmc.facebook.com/intern/diff/D40869846) Pull Request resolved: pytorch#87779 Approved by: https://github.com/davidberard98

naoyam commented Sep 2, 2022

View reviewed changes

naoyam requested a review from shmsong September 2, 2022 04:59

shmsong approved these changes Sep 2, 2022

View reviewed changes

naoyam merged commit 8eafc54 into devel Sep 2, 2022

naoyam deleted the fix_root_mapping branch September 2, 2022 19:57

csarofeen mentioned this pull request Sep 5, 2022

Fusion Segmenter asserts #1947

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix detection of unmappable root domains #1952

Fix detection of unmappable root domains #1952

Uh oh!

naoyam commented Sep 2, 2022

Uh oh!

naoyam Sep 2, 2022

Uh oh!

naoyam Sep 2, 2022

Uh oh!

naoyam Sep 2, 2022

Uh oh!

shmsong left a comment

Uh oh!

Uh oh!

Fix detection of unmappable root domains #1952

Fix detection of unmappable root domains #1952

Uh oh!

Conversation

naoyam commented Sep 2, 2022

Uh oh!

naoyam Sep 2, 2022

Choose a reason for hiding this comment

Uh oh!

naoyam Sep 2, 2022

Choose a reason for hiding this comment

Uh oh!

naoyam Sep 2, 2022

Choose a reason for hiding this comment

Uh oh!

shmsong left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!