[llvm] r289001 - [InlineSpiller] Don't call TargetInstrInfo::foldMemoryOperand with an empty list.

Quentin Colombet via llvm-commits llvm-commits at lists.llvm.org
Wed Dec 7 16:06:51 PST 2016


Author: qcolombet
Date: Wed Dec  7 18:06:51 2016
New Revision: 289001

URL: http://llvm.org/viewvc/llvm-project?rev=289001&view=rev
Log:
[InlineSpiller] Don't call TargetInstrInfo::foldMemoryOperand with an empty list.

Since r287792 if we try to do that we will hit an assert.

Added:
    llvm/trunk/test/CodeGen/X86/implicit-use-spill.mir
Modified:
    llvm/trunk/lib/CodeGen/InlineSpiller.cpp

Modified: llvm/trunk/lib/CodeGen/InlineSpiller.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/InlineSpiller.cpp?rev=289001&r1=289000&r2=289001&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/InlineSpiller.cpp (original)
+++ llvm/trunk/lib/CodeGen/InlineSpiller.cpp Wed Dec  7 18:06:51 2016
@@ -768,6 +768,11 @@ foldMemoryOperand(ArrayRef<std::pair<Mac
       FoldOps.push_back(Idx);
   }
 
+  // If we only have implicit uses, we won't be able to fold that.
+  // Moreover, TargetInstrInfo::foldMemoryOperand will assert if we try!
+  if (FoldOps.empty())
+    return false;
+
   MachineInstrSpan MIS(MI);
 
   MachineInstr *FoldMI =

Added: llvm/trunk/test/CodeGen/X86/implicit-use-spill.mir
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/implicit-use-spill.mir?rev=289001&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/implicit-use-spill.mir (added)
+++ llvm/trunk/test/CodeGen/X86/implicit-use-spill.mir Wed Dec  7 18:06:51 2016
@@ -0,0 +1,22 @@
+# RUN: llc -run-pass=greedy -mtriple=x86_64-apple-macosx -o - %s 2>&1 | FileCheck %s
+
+# Make sure we don't assert when we try to reload a value that is just implicitly used.
+---
+# CHECK: name: foo
+# This test forces a spill of %0.
+name: foo
+registers:
+  - { id: 0, class: gr64 }
+body: |
+  bb.0:
+  ; CHECK: NOOP implicit-def [[VAL:%[0-9]+]]
+  ; VAL should be spilled before csr_noregs, i.e., before we clobber all the registers
+  ; CHECK-NEXT: MOV64mr [[SLOT:%stack.[0-9]+]], 1, _, 0, _, [[VAL]]
+  ; CHECK-NEXT: NOOP csr_noregs
+  ; We need to reload before the (implicit) use.
+  ; CHECK-NEXT: [[RELOADED_VAL:%[0-9]+]] = MOV64rm [[SLOT]], 1, _, 0, _
+  ; CHECK-NEXT: NOOP implicit [[RELOADED_VAL]]
+  NOOP implicit-def %0
+  NOOP csr_noregs
+  NOOP implicit %0
+...




More information about the llvm-commits mailing list