[PATCH] D74805: [AMDGPU] Fix assumption about LaneBitmask content

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec created this revision.
rampitec added a reviewer: arsenm.
Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
Herald added a project: LLVM.

Yet another assumption about an actual LaneBitmask content
is fixed.


https://reviews.llvm.org/D74805

Files:
  llvm/lib/Target/AMDGPU/GCNRegPressure.cpp


Index: llvm/lib/Target/AMDGPU/GCNRegPressure.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/GCNRegPressure.cpp
+++ llvm/lib/Target/AMDGPU/GCNRegPressure.cpp
@@ -103,7 +103,8 @@
                          LaneBitmask PrevMask,
                          LaneBitmask NewMask,
                          const MachineRegisterInfo &MRI) {
-  if (NewMask == PrevMask)
+  if (SIRegisterInfo::getNumCoveredRegs(NewMask) ==
+      SIRegisterInfo::getNumCoveredRegs(PrevMask))
     return;
 
   int Sign = 1;
@@ -111,21 +112,17 @@
     std::swap(NewMask, PrevMask);
     Sign = -1;
   }
-#ifndef NDEBUG
-  const auto MaxMask = MRI.getMaxLaneMaskForVReg(Reg);
-#endif
+
   switch (auto Kind = getRegKind(Reg, MRI)) {
   case SGPR32:
   case VGPR32:
   case AGPR32:
-    assert(PrevMask.none() && NewMask == MaxMask);
     Value[Kind] += Sign;
     break;
 
   case SGPR_TUPLE:
   case VGPR_TUPLE:
   case AGPR_TUPLE:
-    assert(NewMask < MaxMask || NewMask == MaxMask);
     assert(PrevMask < NewMask);
 
     Value[Kind == SGPR_TUPLE ? SGPR32 : Kind == AGPR_TUPLE ? AGPR32 : VGPR32] +=


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D74805.245298.patch
Type: text/x-patch
Size: 1127 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20200219/2bc91e0e/attachment-0001.bin>