[llvm] 0a0c23b - [OpenMPOpt][FIX] Properly track changes to NestedParallelism

[Johannes Doerfert via llvm-commits]

Author: Johannes Doerfert
Date: 2023-10-20T19:28:09-07:00
New Revision: 0a0c23b9ce004664f8481d1ebd15727e736ba5cd

URL: https://github.com/llvm/llvm-project/commit/0a0c23b9ce004664f8481d1ebd15727e736ba5cd
DIFF: https://github.com/llvm/llvm-project/commit/0a0c23b9ce004664f8481d1ebd15727e736ba5cd.diff

LOG: [OpenMPOpt][FIX] Properly track changes to NestedParallelism

If we update the state, or indicate a pessimistic fixpoint, we need to
consider NestedParallelism too.

Fixes part of https://github.com/llvm/llvm-project/issues/66708

That said, the reproducer still needs malloc which we don't support on
AMD GPU. Will be added later.

Added: 
    

Modified: 
    llvm/lib/Transforms/IPO/OpenMPOpt.cpp
    llvm/test/Transforms/OpenMP/parallel_level_fold.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
index 473b56b80b60be9..e50e74ea6c0d5aa 100644
--- a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
+++ b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -733,6 +733,7 @@ struct KernelInfoState : AbstractState {
     SPMDCompatibilityTracker.indicatePessimisticFixpoint();
     ReachedKnownParallelRegions.indicatePessimisticFixpoint();
     ReachedUnknownParallelRegions.indicatePessimisticFixpoint();
+    NestedParallelism = true;
     return ChangeStatus::CHANGED;
   }
 
@@ -762,6 +763,8 @@ struct KernelInfoState : AbstractState {
       return false;
     if (ParallelLevels != RHS.ParallelLevels)
       return false;
+    if (NestedParallelism != RHS.NestedParallelism)
+      return false;
     return true;
   }
 
@@ -3580,7 +3583,8 @@ struct AAKernelInfo : public StateWrapper<KernelInfoState, AbstractAttribute> {
            ", #ParLevels: " +
            (ParallelLevels.isValidState()
                 ? std::to_string(ParallelLevels.size())
-                : "<invalid>");
+                : "<invalid>") +
+           ", NestedPar: " + (NestedParallelism ? "yes" : "no");
   }
 
   /// Create an abstract attribute biew for the position \p IRP.

diff  --git a/llvm/test/Transforms/OpenMP/parallel_level_fold.ll b/llvm/test/Transforms/OpenMP/parallel_level_fold.ll
index ac8a49ddfbfd26f..eb75d030a05f017 100644
--- a/llvm/test/Transforms/OpenMP/parallel_level_fold.ll
+++ b/llvm/test/Transforms/OpenMP/parallel_level_fold.ll
@@ -14,7 +14,7 @@ target triple = "nvptx64"
 ; CHECK: @[[G:[a-zA-Z0-9_$"\\.-]+]] = external global i16
 ; CHECK: @[[NONE_SPMD_KERNEL_ENVIRONMENT:[a-zA-Z0-9_$"\\.-]+]] = local_unnamed_addr constant [[STRUCT_KERNELENVIRONMENTTY:%.*]] { [[STRUCT_CONFIGURATIONENVIRONMENTTY:%.*]] { i8 0, i8 0, i8 1 }, ptr null, ptr null }
 ; CHECK: @[[SPMD_KERNEL_ENVIRONMENT:[a-zA-Z0-9_$"\\.-]+]] = local_unnamed_addr constant [[STRUCT_KERNELENVIRONMENTTY:%.*]] { [[STRUCT_CONFIGURATIONENVIRONMENTTY:%.*]] { i8 0, i8 0, i8 2 }, ptr null, ptr null }
-; CHECK: @[[PARALLEL_KERNEL_ENVIRONMENT:[a-zA-Z0-9_$"\\.-]+]] = local_unnamed_addr constant [[STRUCT_KERNELENVIRONMENTTY:%.*]] { [[STRUCT_CONFIGURATIONENVIRONMENTTY:%.*]] { i8 0, i8 0, i8 2 }, ptr null, ptr null }
+; CHECK: @[[PARALLEL_KERNEL_ENVIRONMENT:[a-zA-Z0-9_$"\\.-]+]] = local_unnamed_addr constant [[STRUCT_KERNELENVIRONMENTTY:%.*]] { [[STRUCT_CONFIGURATIONENVIRONMENTTY:%.*]] { i8 0, i8 1, i8 2 }, ptr null, ptr null }
 ;.
 define weak void @none_spmd() "kernel" {
 ; CHECK-LABEL: define {{[^@]+}}@none_spmd