Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[flang]Check for dominance in loop versioning #68797

Merged
merged 2 commits into from Oct 12, 2023

Conversation

Leporacanthicus
Copy link
Contributor

This avoids trying to version loops that can't be versioned, and thus avoids hitting an assert.

Co-authored with Slava Zakharin (who provided the test-code).

This avoids trying to version loops that can't be versioned,
and thus avoids hitting an assert.

Co-authored with Slava Zakharin (who provided the test-code).
@llvmbot llvmbot added flang Flang issues not falling into any other category flang:fir-hlfir labels Oct 11, 2023
@llvmbot
Copy link
Collaborator

llvmbot commented Oct 11, 2023

@llvm/pr-subscribers-flang-fir-hlfir

Author: Mats Petersson (Leporacanthicus)

Changes

This avoids trying to version loops that can't be versioned, and thus avoids hitting an assert.

Co-authored with Slava Zakharin (who provided the test-code).


Full diff: https://github.com/llvm/llvm-project/pull/68797.diff

2 Files Affected:

  • (modified) flang/lib/Optimizer/Transforms/LoopVersioning.cpp (+6-4)
  • (modified) flang/test/Transforms/loop-versioning.fir (+61)
diff --git a/flang/lib/Optimizer/Transforms/LoopVersioning.cpp b/flang/lib/Optimizer/Transforms/LoopVersioning.cpp
index 4d3ea51ae1a5f71..38b46f635a18c73 100644
--- a/flang/lib/Optimizer/Transforms/LoopVersioning.cpp
+++ b/flang/lib/Optimizer/Transforms/LoopVersioning.cpp
@@ -51,6 +51,7 @@
 #include "flang/Optimizer/Dialect/Support/KindMapping.h"
 #include "flang/Optimizer/Transforms/Passes.h"
 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/IR/Dominance.h"
 #include "mlir/IR/Matchers.h"
 #include "mlir/IR/TypeUtilities.h"
 #include "mlir/Pass/Pass.h"
@@ -292,6 +293,8 @@ void LoopVersioningPass::runOnOperation() {
   // immediately nested in a loop.
   llvm::DenseMap<fir::DoLoopOp, ArgsUsageInLoop> argsInLoops;
 
+  auto &domInfo = getAnalysis<mlir::DominanceInfo>();
+
   // Traverse the loops in post-order and see
   // if those arguments are used inside any loop.
   func.walk([&](fir::DoLoopOp loop) {
@@ -309,12 +312,11 @@ void LoopVersioningPass::runOnOperation() {
       for (auto a : argsOfInterest) {
         if (a.arg == normaliseVal(operand)) {
           // Use the reboxed value, not the block arg when re-creating the loop.
-          // TODO: should we check that the operand dominates the loop?
-          // If this might be a case, we should record such operands in
-          // argsInLoop.cannotTransform, so that they disable the transformation
-          // for the parent loops as well.
           a.arg = operand;
 
+          if (!domInfo.dominates(a.arg, loop))
+            argsInLoop.cannotTransform.insert(a.arg);
+
           // No support currently for sliced arrays.
           // This means that we cannot transform properly
           // instructions referencing a.arg in the whole loop
diff --git a/flang/test/Transforms/loop-versioning.fir b/flang/test/Transforms/loop-versioning.fir
index f2768d7325f7407..c277dda84b2e8da 100644
--- a/flang/test/Transforms/loop-versioning.fir
+++ b/flang/test/Transforms/loop-versioning.fir
@@ -1434,4 +1434,65 @@ func.func @_QPtest_loop_nest(%arg0: !fir.box<!fir.array<?xf32>> {fir.bindc_name
 // CHECK: fir.if
 // CHECK-NOT: fir.if
 
+
+//-----
+
+// Check that a non-dominating operand isn't causing a problem.
+// Just check it compiles, and doesn't version this loop.
+func.func @sum1drebox(%arg0: !fir.box<!fir.array<?xf64>> {fir.bindc_name = "a"}, %arg1: !fir.ref<i32> {fir.bindc_name = "n"}) {
+  %decl = fir.declare %arg0 {uniq_name = "a"} : (!fir.box<!fir.array<?xf64>>) -> !fir.box<!fir.array<?xf64>>
+  %0 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QMmoduleFsum1dEi"}
+  %1 = fir.alloca f64 {bindc_name = "sum", uniq_name = "_QMmoduleFsum1dEsum"}
+  %cst = arith.constant 0.000000e+00 : f64
+  fir.store %cst to %1 : !fir.ref<f64>
+  %c1_i32 = arith.constant 1 : i32
+  %2 = fir.convert %c1_i32 : (i32) -> index
+  %3 = fir.load %arg1 : !fir.ref<i32>
+  %4 = fir.convert %3 : (i32) -> index
+  %c1 = arith.constant 1 : index
+  %5 = fir.convert %2 : (index) -> i32
+  %6:2 = fir.do_loop %arg2 = %2 to %4 step %c1 iter_args(%arg3 = %5) -> (index, i32) {
+    %rebox = fir.rebox %decl : (!fir.box<!fir.array<?xf64>>) -> !fir.box<!fir.array<?xf64>>
+    fir.store %arg3 to %0 : !fir.ref<i32>
+    %7 = fir.load %1 : !fir.ref<f64>
+    %8 = fir.load %0 : !fir.ref<i32>
+    %9 = fir.convert %8 : (i32) -> i64
+    %c1_i64 = arith.constant 1 : i64
+    %10 = arith.subi %9, %c1_i64 : i64
+    %11 = fir.coordinate_of %rebox, %10 : (!fir.box<!fir.array<?xf64>>, i64) -> !fir.ref<f64>
+    %12 = fir.load %11 : !fir.ref<f64>
+    %13 = arith.addf %7, %12 fastmath<contract> : f64
+    fir.store %13 to %1 : !fir.ref<f64>
+    %14 = arith.addi %arg2, %c1 : index
+    %15 = fir.convert %c1 : (index) -> i32
+    %16 = fir.load %0 : !fir.ref<i32>
+    %17 = arith.addi %16, %15 : i32
+    fir.result %14, %17 : index, i32
+  }
+  fir.store %6#1 to %0 : !fir.ref<i32>
+  return
+}
+// CHECK-LABEL: func @sum1drebox
+// No versioning -> no if-operation.
+// CHECK-NOT: fir.if
+
+
 } // End module
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+

flang/test/Transforms/loop-versioning.fir Outdated Show resolved Hide resolved
Comment on lines -312 to -315
// TODO: should we check that the operand dominates the loop?
// If this might be a case, we should record such operands in
// argsInLoop.cannotTransform, so that they disable the transformation
// for the parent loops as well.
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copy this comment below describing the change, instead of removing?

flang/test/Transforms/loop-versioning.fir Show resolved Hide resolved
Copy link
Contributor

@tblah tblah left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Please address Kiran's comments

@Leporacanthicus Leporacanthicus merged commit 8dcee58 into llvm:main Oct 12, 2023
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
flang:fir-hlfir flang Flang issues not falling into any other category
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants