Legalizer: Use the scalar bit width when promoting bit counting instrs on

vectors.

e.g. when promoting ctlz from <2 x i32> to <2 x i64> we have to fixup
the result by 32 bits, not 64. PR20917.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@217671 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 56 insertions, 6 deletions

diff --git a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 6feac0d..cffb0a1 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -343,9 +343,10 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTLZ(SDNode *N) {
   EVT NVT = Op.getValueType();
   Op = DAG.getNode(N->getOpcode(), dl, NVT, Op);
   // Subtract off the extra leading bits in the bigger type.
-  return DAG.getNode(ISD::SUB, dl, NVT, Op,
-                     DAG.getConstant(NVT.getSizeInBits() -
-                                     OVT.getSizeInBits(), NVT));
+  return DAG.getNode(
+      ISD::SUB, dl, NVT, Op,
+      DAG.getConstant(NVT.getScalarSizeInBits() - OVT.getScalarSizeInBits(),
+                      NVT));
 }
 
 SDValue DAGTypeLegalizer::PromoteIntRes_CTPOP(SDNode *N) {
@@ -363,8 +364,8 @@ SDValue DAGTypeLegalizer::PromoteIntRes_CTTZ(SDNode *N) {
     // The count is the same in the promoted type except if the original
     // value was zero.  This can be handled by setting the bit just off
     // the top of the original type.
-    APInt TopBit(NVT.getSizeInBits(), 0);
-    TopBit.setBit(OVT.getSizeInBits());
+    auto TopBit = APInt::getOneBitSet(NVT.getScalarSizeInBits(),
+                                      OVT.getScalarSizeInBits());
     Op = DAG.getNode(ISD::OR, dl, NVT, Op, DAG.getConstant(TopBit, NVT));
   }
   return DAG.getNode(N->getOpcode(), dl, NVT, Op);
diff --git a/test/CodeGen/X86/vec_ctbits.ll b/test/CodeGen/X86/vec_ctbits.ll
index bddd535..0aa72b1 100644
--- a/test/CodeGen/X86/vec_ctbits.ll
+++ b/test/CodeGen/X86/vec_ctbits.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -march=x86-64
+; RUN: llc < %s -march=x86-64 -mattr=+sse2 | FileCheck %s
 
 declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
 declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
@@ -7,12 +7,61 @@ declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>)
 define <2 x i64> @footz(<2 x i64> %a) nounwind {
   %c = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
   ret <2 x i64> %c
+
+; CHECK-LABEL: footz
+; CHECK: bsfq
+; CHECK: bsfq
 }
 define <2 x i64> @foolz(<2 x i64> %a) nounwind {
   %c = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
   ret <2 x i64> %c
+
+; CHECK-LABEL: foolz
+; CHECK: bsrq
+; CHECK: xorq $63
+; CHECK: bsrq
+; CHECK: xorq $63
 }
+
 define <2 x i64> @foopop(<2 x i64> %a) nounwind {
   %c = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a)
   ret <2 x i64> %c
 }
+
+declare <2 x i32> @llvm.cttz.v2i32(<2 x i32>, i1)
+declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1)
+declare <2 x i32> @llvm.ctpop.v2i32(<2 x i32>)
+
+define <2 x i32> @promtz(<2 x i32> %a) nounwind {
+  %c = call <2 x i32> @llvm.cttz.v2i32(<2 x i32> %a, i1 false)
+  ret <2 x i32> %c
+
+; CHECK: .quad 4294967296
+; CHECK: .quad 4294967296
+; CHECK-LABEL: promtz
+; CHECK: bsfq
+; CHECK: cmov
+; CHECK: bsfq
+; CHECK: cmov
+}
+define <2 x i32> @promlz(<2 x i32> %a) nounwind {
+  %c = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false)
+  ret <2 x i32> %c
+
+; CHECK: .quad 4294967295
+; CHECK: .quad 4294967295
+; CHECK: .quad 32
+; CHECK: .quad 32
+; CHECK-LABEL: promlz
+; CHECK: pand
+; CHECK: bsrq
+; CHECK: xorq $63
+; CHECK: bsrq
+; CHECK: xorq $63
+; CHECK: psub
+}
+
+define <2 x i32> @prompop(<2 x i32> %a) nounwind {
+  %c = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a)
+  ret <2 x i32> %c
+}