[mips] Fix lowering a signed immediate for *.d MSA instructions

[atanasyan]

Summary

The lowerMSASplatImm function zero-extends i32 immediates while building constant. If target type is i64, negative immediate loses the sign. As a result, for example __builtin_msa_ldi_d(-1) lowered to series of instruction loads incorrect value 0xffffffff to the $w0 register instead of single ldi.d $w0, -1 instruction.

The fix zero-extends unsigned immediates and signed-extend signed immediates.

[Petar.Avramovic]

LGTM.

llvm/lib/Target/Mips/MipsSEISelLowering.cpp

Result = DAG.getNode(ISD::BITCAST, DL, ResVecTy,
                          DAG.getNode(ISD::AND, DL, ViaVecTy, Result, One));
   }
 
   return Result;
 }
 
 static SDValue lowerMSASplatImm(SDValue Op, unsigned ImmOp, SelectionDAG &DAG,
                                 bool IsSigned = false) {
+  auto *CImm = cast<ConstantSDNode>(Op->getOperand(ImmOp));
   return DAG.getConstant(
       APInt(Op->getValueType(0).getScalarType().getSizeInBits(),
-            Op->getConstantOperandVal(ImmOp), IsSigned),
+            IsSigned ? CImm->getSExtValue() : CImm->getZExtValue(), IsSigned),
       SDLoc(Op), Op->getValueType(0));
 }
 
 static SDValue getBuildVectorSplat(EVT VecTy, SDValue SplatValue,
                                    bool BigEndian, SelectionDAG &DAG) {
   EVT ViaVecTy = VecTy;
   SDValue SplatValueA = SplatValue;
   SDValue SplatValueB = SplatValue;

llvm/test/CodeGen/Mips/msa/i10.ll

 
 declare i32 @llvm.mips.bnz.d(<2 x i64>) nounwind
 
 ; CHECK: llvm_mips_bnz_d_test:
 ; CHECK-DAG: ld.d [[R0:\$w[0-9]+]]
 ; CHECK-DAG: bnz.d [[R0]]
 ; CHECK: .size llvm_mips_bnz_d_test
 
+@llvm_mips_ldi_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_ldi_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+
+define void @llvm_mips_ldi_b_test() nounwind {
+entry:
+  %0 = call <16 x i8> @llvm.mips.ldi.b(i32 3)
+  store <16 x i8> %0, <16 x i8>* @llvm_mips_ldi_b_RES1
+  %1 = call <16 x i8> @llvm.mips.ldi.b(i32 -3)
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_ldi_b_RES2
+  ret void
+}
+
+declare <16 x i8> @llvm.mips.ldi.b(i32)
+
+; CHECK-LABEL: llvm_mips_ldi_b_test
+; CHECK-DAG: ldi.b {{\$w[0-9]}}, 3
+; CHECK-DAG: ldi.b {{\$w[0-9]}}, -3
+
+@llvm_mips_ldi_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_ldi_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+
+define void @llvm_mips_ldi_h_test() nounwind {
+entry:
+  %0 = call <8 x i16> @llvm.mips.ldi.h(i32 3)
+  store <8 x i16> %0, <8 x i16>* @llvm_mips_ldi_h_RES1
+  %1 = call <8 x i16> @llvm.mips.ldi.h(i32 -3)
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_ldi_h_RES2
+  ret void
+}
+
+declare <8 x i16> @llvm.mips.ldi.h(i32)
+
+; CHECK-LABEL: llvm_mips_ldi_h_test
+; CHECK-DAG: ldi.h {{\$w[0-9]}}, 3
+; CHECK-DAG: ldi.h {{\$w[0-9]}}, -3
+
+@llvm_mips_ldi_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_ldi_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+
+define void @llvm_mips_ldi_w_test() nounwind {
+entry:
+  %0 = call <4 x i32> @llvm.mips.ldi.w(i32 3)
+  store <4 x i32> %0, <4 x i32>* @llvm_mips_ldi_w_RES1
+  %1 = call <4 x i32> @llvm.mips.ldi.w(i32 -3)
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_ldi_w_RES2
+  ret void
+}
+
+declare <4 x i32> @llvm.mips.ldi.w(i32)
+
+; CHECK-LABEL: llvm_mips_ldi_w_test
+; CHECK-DAG: ldi.w {{\$w[0-9]}}, 3
+; CHECK-DAG: ldi.w {{\$w[0-9]}}, -3
+
+@llvm_mips_ldi_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_ldi_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
+
+define void @llvm_mips_ldi_d_test() nounwind {
+entry:
+  %0 = call <2 x i64> @llvm.mips.ldi.d(i32 3)
+  store <2 x i64> %0, <2 x i64>* @llvm_mips_ldi_d_RES1
+  %1 = call <2 x i64> @llvm.mips.ldi.d(i32 -3)
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_ldi_d_RES2
+  ret void
+}
+
+declare <2 x i64> @llvm.mips.ldi.d(i32)
+
+; CHECK-LABEL: llvm_mips_ldi_d_test
+; CHECK-DAG: ldi.d {{\$w[0-9]}}, 3
+; CHECK-DAG: ldi.d {{\$w[0-9]}}, -3

llvm/test/CodeGen/Mips/msa/i5-c.ll

 ; Test the MSA intrinsics that are encoded with the I5 instruction format.
 ; There are lots of these so this covers those beginning with 'c'
 
 ; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck %s
 ; RUN: llc -march=mipsel -mattr=+msa,+fp64 < %s | FileCheck %s
 
 @llvm_mips_ceqi_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
-@llvm_mips_ceqi_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_ceqi_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_ceqi_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_ceqi_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_ceqi_b_ARG1
   %1 = tail call <16 x i8> @llvm.mips.ceqi.b(<16 x i8> %0, i32 14)
-  store <16 x i8> %1, <16 x i8>* @llvm_mips_ceqi_b_RES
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_ceqi_b_RES1
+  %2 = tail call <16 x i8> @llvm.mips.ceqi.b(<16 x i8> %0, i32 -14)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_ceqi_b_RES2
   ret void
 }
 
 declare <16 x i8> @llvm.mips.ceqi.b(<16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_ceqi_b_test:
-; CHECK: ld.b
-; CHECK: ceqi.b
-; CHECK: st.b
+; CHECK: ld.b [[RS:\$w[0-9]+]]
+; CHECK: ceqi.b [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.b [[RD1]]
+; CHECK: ceqi.b [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.b [[RD2]]
 ; CHECK: .size llvm_mips_ceqi_b_test
 ;
 @llvm_mips_ceqi_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
-@llvm_mips_ceqi_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_ceqi_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_ceqi_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
 
 define void @llvm_mips_ceqi_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_ceqi_h_ARG1
   %1 = tail call <8 x i16> @llvm.mips.ceqi.h(<8 x i16> %0, i32 14)
-  store <8 x i16> %1, <8 x i16>* @llvm_mips_ceqi_h_RES
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_ceqi_h_RES1
+  %2 = tail call <8 x i16> @llvm.mips.ceqi.h(<8 x i16> %0, i32 -14)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_ceqi_h_RES2
   ret void
 }
 
 declare <8 x i16> @llvm.mips.ceqi.h(<8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_ceqi_h_test:
-; CHECK: ld.h
-; CHECK: ceqi.h
-; CHECK: st.h
+; CHECK: ld.h [[RS:\$w[0-9]+]]
+; CHECK: ceqi.h [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.h [[RD1]]
+; CHECK: ceqi.h [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.h [[RD2]]
 ; CHECK: .size llvm_mips_ceqi_h_test
 ;
 @llvm_mips_ceqi_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
-@llvm_mips_ceqi_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_ceqi_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_ceqi_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
 
 define void @llvm_mips_ceqi_w_test() nounwind {
 entry:
   %0 = load <4 x i32>, <4 x i32>* @llvm_mips_ceqi_w_ARG1
   %1 = tail call <4 x i32> @llvm.mips.ceqi.w(<4 x i32> %0, i32 14)
-  store <4 x i32> %1, <4 x i32>* @llvm_mips_ceqi_w_RES
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_ceqi_w_RES1
+  %2 = tail call <4 x i32> @llvm.mips.ceqi.w(<4 x i32> %0, i32 -14)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_ceqi_w_RES2
   ret void
 }
 
 declare <4 x i32> @llvm.mips.ceqi.w(<4 x i32>, i32) nounwind
 
 ; CHECK: llvm_mips_ceqi_w_test:
-; CHECK: ld.w
-; CHECK: ceqi.w
-; CHECK: st.w
+; CHECK: ld.w [[RS:\$w[0-9]+]]
+; CHECK: ceqi.w [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.w [[RD1]]
+; CHECK: ceqi.w [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.w [[RD2]]
 ; CHECK: .size llvm_mips_ceqi_w_test
 ;
 @llvm_mips_ceqi_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
-@llvm_mips_ceqi_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_ceqi_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_ceqi_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
 
 define void @llvm_mips_ceqi_d_test() nounwind {
 entry:
   %0 = load <2 x i64>, <2 x i64>* @llvm_mips_ceqi_d_ARG1
   %1 = tail call <2 x i64> @llvm.mips.ceqi.d(<2 x i64> %0, i32 14)
-  store <2 x i64> %1, <2 x i64>* @llvm_mips_ceqi_d_RES
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_ceqi_d_RES1
+  %2 = tail call <2 x i64> @llvm.mips.ceqi.d(<2 x i64> %0, i32 -14)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_ceqi_d_RES2
   ret void
 }
 
 declare <2 x i64> @llvm.mips.ceqi.d(<2 x i64>, i32) nounwind
 
 ; CHECK: llvm_mips_ceqi_d_test:
-; CHECK: ld.d
-; CHECK: ceqi.d
-; CHECK: st.d
+; CHECK: ld.d [[RS:\$w[0-9]+]]
+; CHECK: ceqi.d [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.d [[RD1]]
+; CHECK: ceqi.d [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.d [[RD2]]
 ; CHECK: .size llvm_mips_ceqi_d_test
 ;
 @llvm_mips_clei_s_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
-@llvm_mips_clei_s_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_clei_s_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_clei_s_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_clei_s_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_clei_s_b_ARG1
   %1 = tail call <16 x i8> @llvm.mips.clei.s.b(<16 x i8> %0, i32 14)
-  store <16 x i8> %1, <16 x i8>* @llvm_mips_clei_s_b_RES
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_clei_s_b_RES1
+  %2 = tail call <16 x i8> @llvm.mips.clei.s.b(<16 x i8> %0, i32 -14)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_clei_s_b_RES2
   ret void
 }
 
 declare <16 x i8> @llvm.mips.clei.s.b(<16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_clei_s_b_test:
-; CHECK: ld.b
-; CHECK: clei_s.b
-; CHECK: st.b
+; CHECK: ld.b [[RS:\$w[0-9]+]]
+; CHECK: clei_s.b [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.b [[RD1]]
+; CHECK: clei_s.b [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.b [[RD2]]
 ; CHECK: .size llvm_mips_clei_s_b_test
 ;
 @llvm_mips_clei_s_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
-@llvm_mips_clei_s_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_clei_s_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_clei_s_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
 
 define void @llvm_mips_clei_s_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_clei_s_h_ARG1
   %1 = tail call <8 x i16> @llvm.mips.clei.s.h(<8 x i16> %0, i32 14)
-  store <8 x i16> %1, <8 x i16>* @llvm_mips_clei_s_h_RES
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_clei_s_h_RES1
+  %2 = tail call <8 x i16> @llvm.mips.clei.s.h(<8 x i16> %0, i32 -14)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_clei_s_h_RES2
   ret void
 }
 
 declare <8 x i16> @llvm.mips.clei.s.h(<8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_clei_s_h_test:
-; CHECK: ld.h
-; CHECK: clei_s.h
-; CHECK: st.h
+; CHECK: ld.h [[RS:\$w[0-9]+]]
+; CHECK: clei_s.h [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.h [[RD1]]
+; CHECK: clei_s.h [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.h [[RD2]]
 ; CHECK: .size llvm_mips_clei_s_h_test
 ;
 @llvm_mips_clei_s_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
-@llvm_mips_clei_s_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_clei_s_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_clei_s_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
 
 define void @llvm_mips_clei_s_w_test() nounwind {
 entry:
   %0 = load <4 x i32>, <4 x i32>* @llvm_mips_clei_s_w_ARG1
   %1 = tail call <4 x i32> @llvm.mips.clei.s.w(<4 x i32> %0, i32 14)
-  store <4 x i32> %1, <4 x i32>* @llvm_mips_clei_s_w_RES
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_clei_s_w_RES1
+  %2 = tail call <4 x i32> @llvm.mips.clei.s.w(<4 x i32> %0, i32 -14)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_clei_s_w_RES2
   ret void
 }
 
 declare <4 x i32> @llvm.mips.clei.s.w(<4 x i32>, i32) nounwind
 
 ; CHECK: llvm_mips_clei_s_w_test:
-; CHECK: ld.w
-; CHECK: clei_s.w
-; CHECK: st.w
+; CHECK: ld.w [[RS:\$w[0-9]+]]
+; CHECK: clei_s.w [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.w [[RD1]]
+; CHECK: clei_s.w [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.w [[RD2]]
 ; CHECK: .size llvm_mips_clei_s_w_test
 ;
 @llvm_mips_clei_s_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
-@llvm_mips_clei_s_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_clei_s_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_clei_s_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
 
 define void @llvm_mips_clei_s_d_test() nounwind {
 entry:
   %0 = load <2 x i64>, <2 x i64>* @llvm_mips_clei_s_d_ARG1
   %1 = tail call <2 x i64> @llvm.mips.clei.s.d(<2 x i64> %0, i32 14)
-  store <2 x i64> %1, <2 x i64>* @llvm_mips_clei_s_d_RES
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_clei_s_d_RES1
+  %2 = tail call <2 x i64> @llvm.mips.clei.s.d(<2 x i64> %0, i32 -14)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_clei_s_d_RES2
   ret void
 }
 
 declare <2 x i64> @llvm.mips.clei.s.d(<2 x i64>, i32) nounwind
 
 ; CHECK: llvm_mips_clei_s_d_test:
-; CHECK: ld.d
-; CHECK: clei_s.d
-; CHECK: st.d
+; CHECK: ld.d [[RS:\$w[0-9]+]]
+; CHECK: clei_s.d [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.d [[RD1]]
+; CHECK: clei_s.d [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.d [[RD2]]
 ; CHECK: .size llvm_mips_clei_s_d_test
 ;
 @llvm_mips_clei_u_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
 @llvm_mips_clei_u_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_clei_u_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_clei_u_b_ARG1

 
 ; CHECK: llvm_mips_clei_u_d_test:
 ; CHECK: ld.d
 ; CHECK: clei_u.d
 ; CHECK: st.d
 ; CHECK: .size llvm_mips_clei_u_d_test
 ;
 @llvm_mips_clti_s_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
-@llvm_mips_clti_s_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_clti_s_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_clti_s_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_clti_s_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_clti_s_b_ARG1
   %1 = tail call <16 x i8> @llvm.mips.clti.s.b(<16 x i8> %0, i32 14)
-  store <16 x i8> %1, <16 x i8>* @llvm_mips_clti_s_b_RES
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_clti_s_b_RES1
+  %2 = tail call <16 x i8> @llvm.mips.clti.s.b(<16 x i8> %0, i32 -14)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_clti_s_b_RES2
   ret void
 }
 
 declare <16 x i8> @llvm.mips.clti.s.b(<16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_clti_s_b_test:
-; CHECK: ld.b
-; CHECK: clti_s.b
-; CHECK: st.b
+; CHECK: ld.b [[RS:\$w[0-9]+]]
+; CHECK: clti_s.b [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.b [[RD1]]
+; CHECK: clti_s.b [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.b [[RD2]]
 ; CHECK: .size llvm_mips_clti_s_b_test
 ;
 @llvm_mips_clti_s_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
-@llvm_mips_clti_s_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_clti_s_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_clti_s_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
 
 define void @llvm_mips_clti_s_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_clti_s_h_ARG1
   %1 = tail call <8 x i16> @llvm.mips.clti.s.h(<8 x i16> %0, i32 14)
-  store <8 x i16> %1, <8 x i16>* @llvm_mips_clti_s_h_RES
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_clti_s_h_RES1
+  %2 = tail call <8 x i16> @llvm.mips.clti.s.h(<8 x i16> %0, i32 -14)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_clti_s_h_RES2
   ret void
 }
 
 declare <8 x i16> @llvm.mips.clti.s.h(<8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_clti_s_h_test:
-; CHECK: ld.h
-; CHECK: clti_s.h
-; CHECK: st.h
+; CHECK: ld.h [[RS:\$w[0-9]+]]
+; CHECK: clti_s.h [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.h [[RD1]]
+; CHECK: clti_s.h [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.h [[RD2]]
 ; CHECK: .size llvm_mips_clti_s_h_test
 ;
 @llvm_mips_clti_s_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
-@llvm_mips_clti_s_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_clti_s_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_clti_s_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
 
 define void @llvm_mips_clti_s_w_test() nounwind {
 entry:
   %0 = load <4 x i32>, <4 x i32>* @llvm_mips_clti_s_w_ARG1
   %1 = tail call <4 x i32> @llvm.mips.clti.s.w(<4 x i32> %0, i32 14)
-  store <4 x i32> %1, <4 x i32>* @llvm_mips_clti_s_w_RES
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_clti_s_w_RES1
+  %2 = tail call <4 x i32> @llvm.mips.clti.s.w(<4 x i32> %0, i32 -14)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_clti_s_w_RES2
   ret void
 }
 
 declare <4 x i32> @llvm.mips.clti.s.w(<4 x i32>, i32) nounwind
 
 ; CHECK: llvm_mips_clti_s_w_test:
-; CHECK: ld.w
-; CHECK: clti_s.w
-; CHECK: st.w
+; CHECK: ld.w [[RS:\$w[0-9]+]]
+; CHECK: clti_s.w [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.w [[RD1]]
+; CHECK: clti_s.w [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.w [[RD2]]
 ; CHECK: .size llvm_mips_clti_s_w_test
 ;
 @llvm_mips_clti_s_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
-@llvm_mips_clti_s_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_clti_s_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_clti_s_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
 
 define void @llvm_mips_clti_s_d_test() nounwind {
 entry:
   %0 = load <2 x i64>, <2 x i64>* @llvm_mips_clti_s_d_ARG1
   %1 = tail call <2 x i64> @llvm.mips.clti.s.d(<2 x i64> %0, i32 14)
-  store <2 x i64> %1, <2 x i64>* @llvm_mips_clti_s_d_RES
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_clti_s_d_RES1
+  %2 = tail call <2 x i64> @llvm.mips.clti.s.d(<2 x i64> %0, i32 -14)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_clti_s_d_RES2
   ret void
 }
 
 declare <2 x i64> @llvm.mips.clti.s.d(<2 x i64>, i32) nounwind
 
 ; CHECK: llvm_mips_clti_s_d_test:
-; CHECK: ld.d
-; CHECK: clti_s.d
-; CHECK: st.d
+; CHECK: ld.d [[RS:\$w[0-9]+]]
+; CHECK: clti_s.d [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.d [[RD1]]
+; CHECK: clti_s.d [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.d [[RD2]]
 ; CHECK: .size llvm_mips_clti_s_d_test
 ;
 @llvm_mips_clti_u_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
 @llvm_mips_clti_u_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_clti_u_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_clti_u_b_ARG1

llvm/test/CodeGen/Mips/msa/i5-m.ll

 ; Test the MSA intrinsics that are encoded with the I5 instruction format.
 ; There are lots of these so this covers those beginning with 'm'
 
 ; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck %s
 ; RUN: llc -march=mipsel -mattr=+msa,+fp64 < %s | FileCheck %s
 
 @llvm_mips_maxi_s_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
-@llvm_mips_maxi_s_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_maxi_s_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_maxi_s_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_maxi_s_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_maxi_s_b_ARG1
   %1 = tail call <16 x i8> @llvm.mips.maxi.s.b(<16 x i8> %0, i32 14)
-  store <16 x i8> %1, <16 x i8>* @llvm_mips_maxi_s_b_RES
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_maxi_s_b_RES1
+  %2 = tail call <16 x i8> @llvm.mips.maxi.s.b(<16 x i8> %0, i32 -14)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_maxi_s_b_RES2
   ret void
 }
 
 declare <16 x i8> @llvm.mips.maxi.s.b(<16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_maxi_s_b_test:
-; CHECK: ld.b
-; CHECK: maxi_s.b
-; CHECK: st.b
+; CHECK: ld.b [[RS:\$w[0-9]+]]
+; CHECK: maxi_s.b [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.b [[RD1]]
+; CHECK: maxi_s.b [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.b [[RD2]]
 ; CHECK: .size llvm_mips_maxi_s_b_test
 ;
 @llvm_mips_maxi_s_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
-@llvm_mips_maxi_s_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_maxi_s_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_maxi_s_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
 
 define void @llvm_mips_maxi_s_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_maxi_s_h_ARG1
   %1 = tail call <8 x i16> @llvm.mips.maxi.s.h(<8 x i16> %0, i32 14)
-  store <8 x i16> %1, <8 x i16>* @llvm_mips_maxi_s_h_RES
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_maxi_s_h_RES1
+  %2 = tail call <8 x i16> @llvm.mips.maxi.s.h(<8 x i16> %0, i32 -14)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_maxi_s_h_RES2
   ret void
 }
 
 declare <8 x i16> @llvm.mips.maxi.s.h(<8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_maxi_s_h_test:
-; CHECK: ld.h
-; CHECK: maxi_s.h
-; CHECK: st.h
+; CHECK: ld.h [[RS:\$w[0-9]+]]
+; CHECK: maxi_s.h [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.h [[RD1]]
+; CHECK: maxi_s.h [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.h [[RD2]]
 ; CHECK: .size llvm_mips_maxi_s_h_test
 ;
 @llvm_mips_maxi_s_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
-@llvm_mips_maxi_s_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_maxi_s_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_maxi_s_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
 
 define void @llvm_mips_maxi_s_w_test() nounwind {
 entry:
   %0 = load <4 x i32>, <4 x i32>* @llvm_mips_maxi_s_w_ARG1
   %1 = tail call <4 x i32> @llvm.mips.maxi.s.w(<4 x i32> %0, i32 14)
-  store <4 x i32> %1, <4 x i32>* @llvm_mips_maxi_s_w_RES
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_maxi_s_w_RES1
+  %2 = tail call <4 x i32> @llvm.mips.maxi.s.w(<4 x i32> %0, i32 -14)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_maxi_s_w_RES2
   ret void
 }
 
 declare <4 x i32> @llvm.mips.maxi.s.w(<4 x i32>, i32) nounwind
 
 ; CHECK: llvm_mips_maxi_s_w_test:
-; CHECK: ld.w
-; CHECK: maxi_s.w
-; CHECK: st.w
+; CHECK: ld.w [[RS:\$w[0-9]+]]
+; CHECK: maxi_s.w [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.w [[RD1]]
+; CHECK: maxi_s.w [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.w [[RD2]]
 ; CHECK: .size llvm_mips_maxi_s_w_test
 ;
 @llvm_mips_maxi_s_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
-@llvm_mips_maxi_s_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_maxi_s_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_maxi_s_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
 
 define void @llvm_mips_maxi_s_d_test() nounwind {
 entry:
   %0 = load <2 x i64>, <2 x i64>* @llvm_mips_maxi_s_d_ARG1
   %1 = tail call <2 x i64> @llvm.mips.maxi.s.d(<2 x i64> %0, i32 14)
-  store <2 x i64> %1, <2 x i64>* @llvm_mips_maxi_s_d_RES
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_maxi_s_d_RES1
+  %2 = tail call <2 x i64> @llvm.mips.maxi.s.d(<2 x i64> %0, i32 -14)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_maxi_s_d_RES2
   ret void
 }
 
 declare <2 x i64> @llvm.mips.maxi.s.d(<2 x i64>, i32) nounwind
 
 ; CHECK: llvm_mips_maxi_s_d_test:
-; CHECK: ld.d
-; CHECK: maxi_s.d
-; CHECK: st.d
+; CHECK: ld.d [[RS:\$w[0-9]+]]
+; CHECK: maxi_s.d [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.d [[RD1]]
+; CHECK: maxi_s.d [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.d [[RD2]]
 ; CHECK: .size llvm_mips_maxi_s_d_test
 ;
 @llvm_mips_maxi_u_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
 @llvm_mips_maxi_u_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_maxi_u_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_maxi_u_b_ARG1

 
 ; CHECK: llvm_mips_maxi_u_d_test:
 ; CHECK: ld.d
 ; CHECK: maxi_u.d
 ; CHECK: st.d
 ; CHECK: .size llvm_mips_maxi_u_d_test
 ;
 @llvm_mips_mini_s_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
-@llvm_mips_mini_s_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_mini_s_b_RES1 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
+@llvm_mips_mini_s_b_RES2 = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_mini_s_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_mini_s_b_ARG1
   %1 = tail call <16 x i8> @llvm.mips.mini.s.b(<16 x i8> %0, i32 14)
-  store <16 x i8> %1, <16 x i8>* @llvm_mips_mini_s_b_RES
+  store <16 x i8> %1, <16 x i8>* @llvm_mips_mini_s_b_RES1
+  %2 = tail call <16 x i8> @llvm.mips.mini.s.b(<16 x i8> %0, i32 -14)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_mini_s_b_RES2
   ret void
 }
 
 declare <16 x i8> @llvm.mips.mini.s.b(<16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_mini_s_b_test:
-; CHECK: ld.b
-; CHECK: mini_s.b
-; CHECK: st.b
+; CHECK: ld.b [[RS:\$w[0-9]+]]
+; CHECK: mini_s.b [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.b [[RD1]]
+; CHECK: mini_s.b [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.b [[RD2]]
 ; CHECK: .size llvm_mips_mini_s_b_test
 ;
 @llvm_mips_mini_s_h_ARG1 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
-@llvm_mips_mini_s_h_RES  = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_mini_s_h_RES1 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
+@llvm_mips_mini_s_h_RES2 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, align 16
 
 define void @llvm_mips_mini_s_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_mini_s_h_ARG1
   %1 = tail call <8 x i16> @llvm.mips.mini.s.h(<8 x i16> %0, i32 14)
-  store <8 x i16> %1, <8 x i16>* @llvm_mips_mini_s_h_RES
+  store <8 x i16> %1, <8 x i16>* @llvm_mips_mini_s_h_RES1
+  %2 = tail call <8 x i16> @llvm.mips.mini.s.h(<8 x i16> %0, i32 -14)
+  store <8 x i16> %2, <8 x i16>* @llvm_mips_mini_s_h_RES2
   ret void
 }
 
 declare <8 x i16> @llvm.mips.mini.s.h(<8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_mini_s_h_test:
-; CHECK: ld.h
-; CHECK: mini_s.h
-; CHECK: st.h
+; CHECK: ld.h [[RS:\$w[0-9]+]]
+; CHECK: mini_s.h [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.h [[RD1]]
+; CHECK: mini_s.h [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.h [[RD2]]
 ; CHECK: .size llvm_mips_mini_s_h_test
 ;
 @llvm_mips_mini_s_w_ARG1 = global <4 x i32> <i32 0, i32 1, i32 2, i32 3>, align 16
-@llvm_mips_mini_s_w_RES  = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_mini_s_w_RES1 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
+@llvm_mips_mini_s_w_RES2 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 16
 
 define void @llvm_mips_mini_s_w_test() nounwind {
 entry:
   %0 = load <4 x i32>, <4 x i32>* @llvm_mips_mini_s_w_ARG1
   %1 = tail call <4 x i32> @llvm.mips.mini.s.w(<4 x i32> %0, i32 14)
-  store <4 x i32> %1, <4 x i32>* @llvm_mips_mini_s_w_RES
+  store <4 x i32> %1, <4 x i32>* @llvm_mips_mini_s_w_RES1
+  %2 = tail call <4 x i32> @llvm.mips.mini.s.w(<4 x i32> %0, i32 -14)
+  store <4 x i32> %2, <4 x i32>* @llvm_mips_mini_s_w_RES2
   ret void
 }
 
 declare <4 x i32> @llvm.mips.mini.s.w(<4 x i32>, i32) nounwind
 
 ; CHECK: llvm_mips_mini_s_w_test:
-; CHECK: ld.w
-; CHECK: mini_s.w
-; CHECK: st.w
+; CHECK: ld.w [[RS:\$w[0-9]+]]
+; CHECK: mini_s.w [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.w [[RD1]]
+; CHECK: mini_s.w [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.w [[RD2]]
 ; CHECK: .size llvm_mips_mini_s_w_test
 ;
 @llvm_mips_mini_s_d_ARG1 = global <2 x i64> <i64 0, i64 1>, align 16
-@llvm_mips_mini_s_d_RES  = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_mini_s_d_RES1 = global <2 x i64> <i64 0, i64 0>, align 16
+@llvm_mips_mini_s_d_RES2 = global <2 x i64> <i64 0, i64 0>, align 16
 
 define void @llvm_mips_mini_s_d_test() nounwind {
 entry:
   %0 = load <2 x i64>, <2 x i64>* @llvm_mips_mini_s_d_ARG1
   %1 = tail call <2 x i64> @llvm.mips.mini.s.d(<2 x i64> %0, i32 14)
-  store <2 x i64> %1, <2 x i64>* @llvm_mips_mini_s_d_RES
+  store <2 x i64> %1, <2 x i64>* @llvm_mips_mini_s_d_RES1
+  %2 = tail call <2 x i64> @llvm.mips.mini.s.d(<2 x i64> %0, i32 -14)
+  store <2 x i64> %2, <2 x i64>* @llvm_mips_mini_s_d_RES2
   ret void
 }
 
 declare <2 x i64> @llvm.mips.mini.s.d(<2 x i64>, i32) nounwind
 
 ; CHECK: llvm_mips_mini_s_d_test:
-; CHECK: ld.d
-; CHECK: mini_s.d
-; CHECK: st.d
+; CHECK: ld.d [[RS:\$w[0-9]+]]
+; CHECK: mini_s.d [[RD1:\$w[0-9]]], [[RS]], 14
+; CHECK: st.d [[RD1]]
+; CHECK: mini_s.d [[RD2:\$w[0-9]]], [[RS]], -14
+; CHECK: st.d [[RD2]]
 ; CHECK: .size llvm_mips_mini_s_d_test
 ;
 @llvm_mips_mini_u_b_ARG1 = global <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15>, align 16
 @llvm_mips_mini_u_b_RES  = global <16 x i8> <i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, align 16
 
 define void @llvm_mips_mini_u_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_mini_u_b_ARG1