Remove assert from MachineLoop::getLoopPredecessor()

[rampitec]

Summary

According to the documentation method returns predecessor
if the given loop's header has exactly one unique predecessor
outside the loop. Otherwise return null.

In reality it asserts if there is no predecessor outside of
the loop.

The testcase has the loop where predecessors outside of the
loop were not identified as analyzeBranch() was unable to
process the mask branch and returned true. That is also not
correct to assert for the truly dead loops.

[arsenm]

LGTM with test fix

llvm/test/CodeGen/AMDGPU/loop_header_nopred.mir

6–7 ↗

(On Diff #220388)

Can you generate full checks?

[rampitec]

Generated full checks.

llvm/trunk/include/llvm/Analysis/LoopInfoImpl.h

BlockT *LoopBase<BlockT, LoopT>::getLoopPredecessor() const {
   for (const auto Pred : children<Inverse<BlockT *>>(Header)) {
     if (!contains(Pred)) { // If the block is not in the loop...
       if (Out && Out != Pred)
         return nullptr; // Multiple predecessors outside the loop
       Out = Pred;
     }
   }
 
-  // Make sure there is only one exit out of the preheader.
-  assert(Out && "Header of loop has no predecessors from outside loop?");
   return Out;
 }
 
 /// getLoopLatch - If there is a single latch block for this loop, return it.
 /// A latch block is a block that contains a branch back to the header.
 template <class BlockT, class LoopT>
 BlockT *LoopBase<BlockT, LoopT>::getLoopLatch() const {
   assert(!isInvalid() && "Loop not in a valid state!");

llvm/trunk/test/CodeGen/AMDGPU/loop_header_nopred.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -o - -run-pass=block-placement -mcpu=gfx1010 -mattr=-inst-fwd-prefetch-bug -verify-machineinstrs %s | FileCheck -check-prefix=GCN %s
+
+# Used to fail with
+# Assertion `Out && "Header of loop has no predecessors from outside loop?"
+
+---
+name:            loop_header_nopred
+body:             |
+  ; GCN-LABEL: name: loop_header_nopred
+  ; GCN: bb.0:
+  ; GCN:   successors: %bb.2(0x40000000), %bb.3(0x40000000)
+  ; GCN:   S_CBRANCH_VCCZ %bb.3, implicit $vcc
+  ; GCN:   S_BRANCH %bb.2
+  ; GCN: bb.6 (align 64):
+  ; GCN:   successors: %bb.7(0x04000000), %bb.1(0x7c000000)
+  ; GCN:   S_CBRANCH_VCCNZ %bb.7, implicit $vcc
+  ; GCN: bb.1:
+  ; GCN:   successors: %bb.2(0x40000000), %bb.3(0x40000000)
+  ; GCN:   S_CBRANCH_VCCNZ %bb.2, implicit $vcc
+  ; GCN: bb.3:
+  ; GCN:   successors: %bb.4(0x40000000), %bb.6(0x40000000)
+  ; GCN:   SI_MASK_BRANCH %bb.6, implicit $exec
+  ; GCN:   S_BRANCH %bb.4
+  ; GCN: bb.2 (align 64):
+  ; GCN:   successors: %bb.4(0x40000000), %bb.6(0x40000000)
+  ; GCN:   SI_MASK_BRANCH %bb.6, implicit $exec
+  ; GCN:   S_BRANCH %bb.4
+  ; GCN: bb.4:
+  ; GCN:   successors: %bb.5(0x04000000), %bb.4(0x7c000000)
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_NOP 0
+  ; GCN:   S_CBRANCH_EXECZ %bb.4, implicit $exec
+  ; GCN: bb.5:
+  ; GCN:   successors: %bb.6(0x80000000)
+  ; GCN:   S_BRANCH %bb.6
+  ; GCN: bb.7:
+  ; GCN:   S_ENDPGM 0
+  bb.0:
+    successors: %bb.1(0x80000000)
+
+  bb.1:
+    successors: %bb.2(0x40000000), %bb.3(0x40000000)
+
+    S_CBRANCH_VCCZ %bb.3, implicit $vcc
+    S_BRANCH %bb.2
+
+  bb.2:
+    successors: %bb.3(0x80000000)
+
+  bb.3:
+    successors: %bb.4(0x40000000), %bb.6(0x40000000)
+
+    SI_MASK_BRANCH %bb.6, implicit $exec
+    S_BRANCH %bb.4
+
+  bb.4:
+    successors: %bb.5(0x04000000), %bb.4(0x7c000000)
+
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_NOP 0
+    S_CBRANCH_EXECZ %bb.4, implicit $exec
+
+  bb.5:
+    successors: %bb.6(0x80000000)
+
+  bb.6:
+    successors: %bb.7(0x04000000), %bb.1(0x7c000000)
+
+    S_CBRANCH_VCCZ %bb.1, implicit $vcc
+
+  bb.7:
+    S_ENDPGM 0
+...