[llvm] 22ebbc4 - LoopUnrollAndJam] Only allow loops with single exit(ing) blocks

[Sidharth Baveja via llvm-commits]

Author: Sidharth Baveja
Date: 2021-02-05T16:10:53Z
New Revision: 22ebbc47655737eb543c8641b2e8491846b3aefb

URL: https://github.com/llvm/llvm-project/commit/22ebbc47655737eb543c8641b2e8491846b3aefb
DIFF: https://github.com/llvm/llvm-project/commit/22ebbc47655737eb543c8641b2e8491846b3aefb.diff

LOG: LoopUnrollAndJam] Only allow loops with single exit(ing) blocks

Summary:
This resolves an issue posted on Bugzilla. https://bugs.llvm.org/show_bug.cgi?id=48764
In this issue, the loop had  multiple exit blocks, which resulted in the
function getExitBlock to return a nullptr, which resulted in hitting the assert.
This patch ensures that loops which only have one exit block as allowed to be
unrolled and jammed.

Reviewed By: Whitney, Meinersbur, dmgreen

Differential Revision: https://reviews.llvm.org/D95806

Added: 
    llvm/test/Transforms/LoopUnrollAndJam/multiple_exit_blocks.ll

Modified: 
    llvm/lib/Transforms/Utils/LoopUnrollAndJam.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Utils/LoopUnrollAndJam.cpp b/llvm/lib/Transforms/Utils/LoopUnrollAndJam.cpp
index 6e32a2b865aa..c62049c464b2 100644
--- a/llvm/lib/Transforms/Utils/LoopUnrollAndJam.cpp
+++ b/llvm/lib/Transforms/Utils/LoopUnrollAndJam.cpp
@@ -831,6 +831,23 @@ static bool isEligibleLoopForm(const Loop &Root) {
     if (SubLoopsSize != 1)
       return false;
 
+    // Only loops with a single exit block can be unrolled and jammed.
+    // The function getExitBlock() is used for this check, rather than
+    // getUniqueExitBlock() to ensure loops with mulitple exit edges are
+    // disallowed.
+    if (!L->getExitBlock()) {
+      LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; only loops with single exit "
+                           "blocks can be unrolled and jammed.\n");
+      return false;
+    }
+
+    // Only loops with a single exiting block can be unrolled and jammed.
+    if (!L->getExitingBlock()) {
+      LLVM_DEBUG(dbgs() << "Won't unroll-and-jam; only loops with single "
+                           "exiting blocks can be unrolled and jammed.\n");
+      return false;
+    }
+
     L = L->getSubLoops()[0];
   } while (L);
 

diff  --git a/llvm/test/Transforms/LoopUnrollAndJam/multiple_exit_blocks.ll b/llvm/test/Transforms/LoopUnrollAndJam/multiple_exit_blocks.ll
new file mode 100644
index 000000000000..5b93dfeb8dba
--- /dev/null
+++ b/llvm/test/Transforms/LoopUnrollAndJam/multiple_exit_blocks.ll
@@ -0,0 +1,49 @@
+; REQUIRES: asserts
+; RUN: opt -debug-only=loop-unroll-and-jam -passes="loop-unroll-and-jam" -enable-unroll-and-jam -allow-unroll-and-jam -unroll-and-jam-count=8 -disable-output %s 2>&1 | FileCheck %s
+
+; CHECK: Loop Unroll and Jam: F[h] Loop %bb8
+; CHECK: Won't unroll-and-jam; only loops with single exit blocks can be unrolled and jammed
+
+ at a = external global i16, align 2
+ at e = external global i32, align 4
+ at f = external global i16, align 2
+ at b = external global i16, align 2
+ at c = external global i64, align 8
+
+define void @h() {
+bb:
+  store i32 4, i32* @e, align 4
+  %i15 = load i16, i16* @b, align 2
+  %i17 = icmp slt i16 %i15, 1
+  br label %bb8
+
+bb8:                                              ; preds = %bb, %bb47
+  %storemerge15 = phi i32 [ 4, %bb ], [ %i49, %bb47 ]
+  br label %bb24
+
+bb24:                                             ; preds = %bb43, %bb8
+  %storemerge312 = phi i16 [ 0, %bb8 ], [ %i45, %bb43 ]
+  br i1 %i17, label %bb46.preheader, label %bb43
+
+bb46.preheader:                                   ; preds = %bb24
+  store i16 %storemerge312, i16* @f, align 2
+  br label %bb46
+
+bb43:                                             ; preds = %bb24
+  %i45 = add nuw nsw i16 %storemerge312, 1
+  %i13 = icmp ult i16 %storemerge312, 7
+  br i1 %i13, label %bb24, label %bb47
+
+bb46:                                             ; preds = %bb46.preheader, %bb46
+  br label %bb46
+
+bb47:                                             ; preds = %bb43
+  %i49 = add nsw i32 %storemerge15, -1
+  store i32 %i49, i32* @e, align 4
+  %i7.not = icmp eq i32 %i49, 0
+  br i1 %i7.not, label %bb50, label %bb8
+
+bb50:                                             ; preds = %bb47
+  store i16 %i45, i16* @f, align 2
+  ret void
+}