[PATCH] D70138: [TargetLowering] Increase the storage size of NumRegistersForVT to allow the type break down for v256i1 and other types to be stored correctly

[Craig Topper via Phabricator via llvm-commits]

craig.topper created this revision.
craig.topper added reviewers: efriedma, spatel, arsenm, RKSimon.
Herald added subscribers: hiraditya, wdng.
Herald added a project: LLVM.

v256i1 on X86 without avx512 breaks down to 256 i8 values when passed between basic blocks. But the NumRegistersForVT was sized at a byte for each VT. This results in 256 being stored as 0.

This patch enlarges the type to 16 bits and adds an assert to ensure that no information is lost when the entry is stored.


https://reviews.llvm.org/D70138

Files:
  llvm/include/llvm/CodeGen/TargetLowering.h
  llvm/lib/CodeGen/TargetLoweringBase.cpp


Index: llvm/lib/CodeGen/TargetLoweringBase.cpp
===================================================================
--- llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -1332,8 +1332,11 @@
       MVT IntermediateVT;
       MVT RegisterVT;
       unsigned NumIntermediates;
-      NumRegistersForVT[i] = getVectorTypeBreakdownMVT(VT, IntermediateVT,
+      unsigned NumRegisters = getVectorTypeBreakdownMVT(VT, IntermediateVT,
           NumIntermediates, RegisterVT, this);
+      NumRegistersForVT[i] = NumRegisters;
+      assert(NumRegistersForVT[i] == NumRegisters &&
+             "NumRegistersForVT size cannot represent NumRegisters!");
       RegisterTypeForVT[i] = RegisterVT;
 
       MVT NVT = VT.getPow2VectorType();
Index: llvm/include/llvm/CodeGen/TargetLowering.h
===================================================================
--- llvm/include/llvm/CodeGen/TargetLowering.h
+++ llvm/include/llvm/CodeGen/TargetLowering.h
@@ -2741,7 +2741,7 @@
   /// This indicates the default register class to use for each ValueType the
   /// target supports natively.
   const TargetRegisterClass *RegClassForVT[MVT::LAST_VALUETYPE];
-  unsigned char NumRegistersForVT[MVT::LAST_VALUETYPE];
+  uint16_t NumRegistersForVT[MVT::LAST_VALUETYPE];
   MVT RegisterTypeForVT[MVT::LAST_VALUETYPE];
 
   /// This indicates the "representative" register class to use for each


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D70138.228931.patch
Type: text/x-patch
Size: 1410 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20191112/f0d58efe/attachment.bin>