Fixed a widening bug where we were not using the correct size for the load

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98920 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 8363c3af212188c36e33fa8cc7153cacb42504a0..ed5f24c1cc95048913cd5bbf8c2fa8d3cb1968b1 100644 (file)
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -2167,7 +2167,7 @@ SDValue DAGTypeLegalizer::GenWidenVectorLoads(SmallVector<SDValue, 16>& LdChain,
       } else
         return LdOp;
     } else {
-      unsigned NumElts = WidenWidth / LdWidth;
+      unsigned NumElts = WidenWidth / NewVTWidth;
       EVT NewVecVT = EVT::getVectorVT(*DAG.getContext(), NewVT, NumElts);
       SDValue VecOp = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, NewVecVT, LdOp);
       return DAG.getNode(ISD::BIT_CONVERT, dl, WidenVT, VecOp);

diff --git a/test/CodeGen/X86/widen_load-2.ll b/test/CodeGen/X86/widen_load-2.ll
index 11383fa3082980663fd8571f03690edfb75b01df..58b557a7915ccb893d5271fbc19118e161321a3f 100644 (file)
--- a/test/CodeGen/X86/widen_load-2.ll
+++ b/test/CodeGen/X86/widen_load-2.ll
@@ -152,4 +152,28 @@ define void @add31i8(%i8vec31* nocapture sret %ret, %i8vec31* %ap, %i8vec31* %bp
        %x = add %i8vec31 %a, %b
        store %i8vec31 %x, %i8vec31* %ret, align 16
        ret void
-}
\ No newline at end of file
+}
+
+
+%i8vec3pack = type { <3 x i8>, i8 }
+define %i8vec3pack  @rot() nounwind {
+; CHECK: shrb
+entry:
+  %X = alloca %i8vec3pack, align 4
+  %rot = alloca %i8vec3pack, align 4
+  %result = alloca %i8vec3pack, align 4
+  %storetmp = bitcast %i8vec3pack* %X to <3 x i8>*
+  store <3 x i8> <i8 -98, i8 -98, i8 -98>, <3 x i8>* %storetmp
+  %storetmp1 = bitcast %i8vec3pack* %rot to <3 x i8>*
+  store <3 x i8> <i8 1, i8 1, i8 1>, <3 x i8>* %storetmp1
+  %tmp = load %i8vec3pack* %X
+  %extractVec = extractvalue %i8vec3pack %tmp, 0
+  %tmp2 = load %i8vec3pack* %rot
+  %extractVec3 = extractvalue %i8vec3pack %tmp2, 0
+  %shr = lshr <3 x i8> %extractVec, %extractVec3
+  %storetmp4 = bitcast %i8vec3pack* %result to <3 x i8>*
+  store <3 x i8> %shr, <3 x i8>* %storetmp4
+  %tmp5 = load %i8vec3pack* %result
+  ret %i8vec3pack %tmp5
+}
+