[PATCH] R600/SI: Fix assertion on infinite loops.

Matt Arsenault Matthew.Arsenault at amd.com
Mon Feb 3 17:30:04 PST 2014


This isn't the most useful case to fix in the real world, but bugpoint runs into this.

http://llvm-reviews.chandlerc.com/D2682

Files:
  lib/Target/R600/SILowerControlFlow.cpp
  test/CodeGen/R600/infinite-loop.ll

Index: lib/Target/R600/SILowerControlFlow.cpp
===================================================================
--- lib/Target/R600/SILowerControlFlow.cpp
+++ lib/Target/R600/SILowerControlFlow.cpp
@@ -283,8 +283,10 @@
 }
 
 void SILowerControlFlowPass::Branch(MachineInstr &MI) {
-  assert(MI.getOperand(0).getMBB() == MI.getParent()->getNextNode());
-  MI.eraseFromParent();
+  if (MI.getOperand(0).getMBB() == MI.getParent()->getNextNode())
+    MI.eraseFromParent();
+
+  // If these aren't equal, this is probably an infinite loop.
 }
 
 void SILowerControlFlowPass::Kill(MachineInstr &MI) {
Index: test/CodeGen/R600/infinite-loop.ll
===================================================================
--- /dev/null
+++ test/CodeGen/R600/infinite-loop.ll
@@ -0,0 +1,17 @@
+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
+
+; SI-LABEL: @infinite_loop:
+; SI: V_MOV_B32_e32 [[REG:v[0-9]+]], 999
+; SI: BB0_1:
+; SI: BUFFER_STORE_DWORD [[REG]]
+; SI: S_WAITCNT vmcnt(0) expcnt(0)
+; SI: S_BRANCH BB0_1
+define void @infinite_loop(i32 addrspace(1)* %out) {
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  store i32 999, i32 addrspace(1)* %out, align 4
+  br label %for.body
+}
+
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D2682.1.patch
Type: text/x-patch
Size: 1292 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140203/1dc947db/attachment.bin>


More information about the llvm-commits mailing list