[mips][msa] Fix generation of bm(n)zi and bins[lr]i instructions

[smaksimovic]

Summary

[mips][msa] Fix generation of bm(n)zi and bins[lr]i instructions

We have two cases here, the first one being the following instruction selection from the builtin function:
bm(n)zi builtin -> vselect node -> bins[lr]i machine instruction

In case of bm(n)zi having an immediate which has either its high or low bits set, a bins[lr] instruction can be selected through the selectVSplatMask[LR] function.
The function counts the number of bits set, and that value is being passed to the bins[lr]i instruction as its immediate, which in turn copies immediate modulo
the size of the element in bits plus 1 as per specs, where we get the off-by-one-error.

The other case is:
bins[lr]i -> vselect node -> bsel.v

In this case, a bsel.v instruction gets selected with a mask having one bit less set than required.

[smaksimovic]

Ping.

[sdardis]

Some comments inlined. rL297457 lifted the restriction on splat values being restricted to signed 10 bit values.

LGTM.

test/CodeGen/Mips/msa/bmzi_bmnzi.ll

20

This should be CHECK-LABEL:

43

This should be CHECK-LABEL:

test/CodeGen/Mips/msa/i5-b.ll

117

Unnecessary change. For the bins[lr]i.[hwd] cases, you don't need to modify the test cases. The .b cases need to be changed as otherwise the entire element gets selected and this gets optimized out.

test/CodeGen/Mips/msa/immediates.ll

619

This appears to the only line that requires changing in this test. The others are unnecessary.

[smaksimovic]

Made changes as per review comments.

lib/Target/Mips/MipsSEISelDAGToDAG.cpp

bool MipsSEDAGToDAGISel::selectVSplatMaskL(SDValue N, SDValue &Imm) const {
 
   if (selectVSplat(N.getNode(), ImmValue, EltTy.getSizeInBits()) &&
       ImmValue.getBitWidth() == EltTy.getSizeInBits()) {
     // Extract the run of set bits starting with bit zero from the bitwise
     // inverse of ImmValue, and test that the inverse of this is the same
     // as the original value.
     if (ImmValue == ~(~ImmValue & ~(~ImmValue + 1))) {
 
-      Imm = CurDAG->getTargetConstant(ImmValue.countPopulation(), SDLoc(N),
+      Imm = CurDAG->getTargetConstant(ImmValue.countPopulation() - 1, SDLoc(N),
                                       EltTy);
       return true;
     }
   }
 
   return false;
 }
 
bool MipsSEDAGToDAGISel::selectVSplatMaskR(SDValue N, SDValue &Imm) const {
   if (N->getOpcode() == ISD::BITCAST)
     N = N->getOperand(0);
 
   if (selectVSplat(N.getNode(), ImmValue, EltTy.getSizeInBits()) &&
       ImmValue.getBitWidth() == EltTy.getSizeInBits()) {
     // Extract the run of set bits starting with bit zero, and test that the
     // result is the same as the original value
     if (ImmValue == (ImmValue & ~(ImmValue + 1))) {
-      Imm = CurDAG->getTargetConstant(ImmValue.countPopulation(), SDLoc(N),
+      Imm = CurDAG->getTargetConstant(ImmValue.countPopulation() - 1, SDLoc(N),
                                       EltTy);
       return true;
     }
   }
 
   return false;
 }
 

lib/Target/Mips/MipsSEISelLowering.cpp

SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
   case Intrinsic::mips_binsli_w:
   case Intrinsic::mips_binsli_d: {
     // binsli_x(IfClear, IfSet, nbits) -> (vselect LBitsMask, IfSet, IfClear)
     EVT VecTy = Op->getValueType(0);
     EVT EltTy = VecTy.getVectorElementType();
     if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
       report_fatal_error("Immediate out of range");
     APInt Mask = APInt::getHighBitsSet(EltTy.getSizeInBits(),
-                                       Op->getConstantOperandVal(3));
+                                       Op->getConstantOperandVal(3) + 1);
     return DAG.getNode(ISD::VSELECT, DL, VecTy,
                        DAG.getConstant(Mask, DL, VecTy, true),
                        Op->getOperand(2), Op->getOperand(1));
   }
   case Intrinsic::mips_binsri_b:
   case Intrinsic::mips_binsri_h:
   case Intrinsic::mips_binsri_w:
   case Intrinsic::mips_binsri_d: {
     // binsri_x(IfClear, IfSet, nbits) -> (vselect RBitsMask, IfSet, IfClear)
     EVT VecTy = Op->getValueType(0);
     EVT EltTy = VecTy.getVectorElementType();
     if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
       report_fatal_error("Immediate out of range");
     APInt Mask = APInt::getLowBitsSet(EltTy.getSizeInBits(),
-                                      Op->getConstantOperandVal(3));
+                                      Op->getConstantOperandVal(3) + 1);
     return DAG.getNode(ISD::VSELECT, DL, VecTy,
                        DAG.getConstant(Mask, DL, VecTy, true),
                        Op->getOperand(2), Op->getOperand(1));
   }
   case Intrinsic::mips_bmnz_v:
     return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0), Op->getOperand(3),
                        Op->getOperand(2), Op->getOperand(1));
   case Intrinsic::mips_bmnzi_b:

test/CodeGen/Mips/msa/bitwise.ll

%3 = and <16 x i8> %1, <i8 192, i8 192, i8 192, i8 192,
                           i8 192, i8 192, i8 192, i8 192,
                           i8 192, i8 192, i8 192, i8 192,
                           i8 192, i8 192, i8 192, i8 192>
   %4 = and <16 x i8> %2, <i8 63, i8 63, i8 63, i8 63,
                           i8 63, i8 63, i8 63, i8 63,
                           i8 63, i8 63, i8 63, i8 63,
                           i8 63, i8 63, i8 63, i8 63>
   %5 = or <16 x i8> %3, %4
-  ; CHECK-DAG: binsli.b [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsli.b [[R2]], [[R1]], 1
   store <16 x i8> %5, <16 x i8>* %c
   ; CHECK-DAG: st.b [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsl_v16i8_i
 }
 
 define void @binsl_v8i16_i(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b) nounwind {
   ; CHECK: binsl_v8i16_i:
 
   %1 = load <8 x i16>, <8 x i16>* %a
   ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
   %2 = load <8 x i16>, <8 x i16>* %b
   ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
   %3 = and <8 x i16> %1, <i16 49152, i16 49152, i16 49152, i16 49152,
                           i16 49152, i16 49152, i16 49152, i16 49152>
   %4 = and <8 x i16> %2, <i16 16383, i16 16383, i16 16383, i16 16383,
                           i16 16383, i16 16383, i16 16383, i16 16383>
   %5 = or <8 x i16> %3, %4
-  ; CHECK-DAG: binsli.h [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsli.h [[R2]], [[R1]], 1
   store <8 x i16> %5, <8 x i16>* %c
   ; CHECK-DAG: st.h [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsl_v8i16_i
 }
 
 define void @binsl_v4i32_i(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b) nounwind {
   ; CHECK: binsl_v4i32_i:
 
   %1 = load <4 x i32>, <4 x i32>* %a
   ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
   %2 = load <4 x i32>, <4 x i32>* %b
   ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
   %3 = and <4 x i32> %1, <i32 3221225472, i32 3221225472, i32 3221225472, i32 3221225472>
   %4 = and <4 x i32> %2, <i32 1073741823, i32 1073741823, i32 1073741823, i32 1073741823>
   %5 = or <4 x i32> %3, %4
-  ; CHECK-DAG: binsli.w [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsli.w [[R2]], [[R1]], 1
   store <4 x i32> %5, <4 x i32>* %c
   ; CHECK-DAG: st.w [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsl_v4i32_i
 }
 
 define void @binsl_v2i64_i(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b) nounwind {
   ; CHECK: binsl_v2i64_i:
 
   %1 = load <2 x i64>, <2 x i64>* %a
   ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
   %2 = load <2 x i64>, <2 x i64>* %b
   ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
   %3 = and <2 x i64> %1, <i64 18446744073709551608, i64 18446744073709551608>
   %4 = and <2 x i64> %2, <i64 7, i64 7>
   %5 = or <2 x i64> %3, %4
   ; TODO: We use a particularly wide mask here to work around a legalization
   ;       issue. If the mask doesn't fit within a 10-bit immediate, it gets
   ;       legalized into a constant pool. We should add a test to cover the
   ;       other cases once they correctly select binsli.d.
-  ; CHECK-DAG: binsli.d [[R2]], [[R1]], 61
+  ; CHECK-DAG: binsli.d [[R2]], [[R1]], 60
   store <2 x i64> %5, <2 x i64>* %c
   ; CHECK-DAG: st.d [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsl_v2i64_i
 }
 
 define void @binsr_v16i8_i(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b) nounwind {
   ; CHECK: binsr_v16i8_i:
 
   %1 = load <16 x i8>, <16 x i8>* %a
   ; CHECK-DAG: ld.b [[R1:\$w[0-9]+]], 0($5)
   %2 = load <16 x i8>, <16 x i8>* %b
   ; CHECK-DAG: ld.b [[R2:\$w[0-9]+]], 0($6)
   %3 = and <16 x i8> %1, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3,
                           i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
   %4 = and <16 x i8> %2, <i8 252, i8 252, i8 252, i8 252,
                           i8 252, i8 252, i8 252, i8 252,
                           i8 252, i8 252, i8 252, i8 252,
                           i8 252, i8 252, i8 252, i8 252>
   %5 = or <16 x i8> %3, %4
-  ; CHECK-DAG: binsri.b [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsri.b [[R2]], [[R1]], 1
   store <16 x i8> %5, <16 x i8>* %c
   ; CHECK-DAG: st.b [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsr_v16i8_i
 }
 
 define void @binsr_v8i16_i(<8 x i16>* %c, <8 x i16>* %a, <8 x i16>* %b) nounwind {
   ; CHECK: binsr_v8i16_i:
 
   %1 = load <8 x i16>, <8 x i16>* %a
   ; CHECK-DAG: ld.h [[R1:\$w[0-9]+]], 0($5)
   %2 = load <8 x i16>, <8 x i16>* %b
   ; CHECK-DAG: ld.h [[R2:\$w[0-9]+]], 0($6)
   %3 = and <8 x i16> %1, <i16 3, i16 3, i16 3, i16 3,
                           i16 3, i16 3, i16 3, i16 3>
   %4 = and <8 x i16> %2, <i16 65532, i16 65532, i16 65532, i16 65532,
                           i16 65532, i16 65532, i16 65532, i16 65532>
   %5 = or <8 x i16> %3, %4
-  ; CHECK-DAG: binsri.h [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsri.h [[R2]], [[R1]], 1
   store <8 x i16> %5, <8 x i16>* %c
   ; CHECK-DAG: st.h [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsr_v8i16_i
 }
 
 define void @binsr_v4i32_i(<4 x i32>* %c, <4 x i32>* %a, <4 x i32>* %b) nounwind {
   ; CHECK: binsr_v4i32_i:
 
   %1 = load <4 x i32>, <4 x i32>* %a
   ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
   %2 = load <4 x i32>, <4 x i32>* %b
   ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
   %3 = and <4 x i32> %1, <i32 3, i32 3, i32 3, i32 3>
   %4 = and <4 x i32> %2, <i32 4294967292, i32 4294967292, i32 4294967292, i32 4294967292>
   %5 = or <4 x i32> %3, %4
-  ; CHECK-DAG: binsri.w [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsri.w [[R2]], [[R1]], 1
   store <4 x i32> %5, <4 x i32>* %c
   ; CHECK-DAG: st.w [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsr_v4i32_i
 }
 
 define void @binsr_v2i64_i(<2 x i64>* %c, <2 x i64>* %a, <2 x i64>* %b) nounwind {
   ; CHECK: binsr_v2i64_i:
 
   %1 = load <2 x i64>, <2 x i64>* %a
   ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
   %2 = load <2 x i64>, <2 x i64>* %b
   ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
   %3 = and <2 x i64> %1, <i64 3, i64 3>
   %4 = and <2 x i64> %2, <i64 18446744073709551612, i64 18446744073709551612>
   %5 = or <2 x i64> %3, %4
-  ; CHECK-DAG: binsri.d [[R2]], [[R1]], 2
+  ; CHECK-DAG: binsri.d [[R2]], [[R1]], 1
   store <2 x i64> %5, <2 x i64>* %c
   ; CHECK-DAG: st.d [[R2]], 0($4)
 
   ret void
   ; CHECK: .size binsr_v2i64_i
 }
 
 define void @bclr_v16i8(<16 x i8>* %c, <16 x i8>* %a, <16 x i8>* %b) nounwind {

test/CodeGen/Mips/msa/bmzi_bmnzi.ll

+; RUN: llc -march=mipsel -mattr=+msa,+fp64 -relocation-model=pic < %s | FileCheck %s
+
+@llvm_mips_bmnzi_b_ARG1 = global <16 x i8> <i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15, i8 15>, align 16
+@llvm_mips_bmnzi_b_ARG2 = global <16 x i8> zeroinitializer, align 16
+@llvm_mips_bmnzi_b_RES = global <16 x i8> zeroinitializer, align 16
+
+define void @llvm_mips_bmnzi_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_bmnzi_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_bmnzi_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.bmnzi.b(<16 x i8> %0, <16 x i8> %1, i32 240)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  %3 = tail call <16 x i8> @llvm.mips.bmnzi.b(<16 x i8> %0, <16 x i8> %1, i32 15)
+  store <16 x i8> %3, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  %4 = tail call <16 x i8> @llvm.mips.bmnzi.b(<16 x i8> %0, <16 x i8> %1, i32 170)
+  store <16 x i8> %4, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  ret void
+}
+; CHECK-LABEL: llvm_mips_bmnzi_b_test:
+; CHECK: lw [[R0:\$[0-9]+]], %got(llvm_mips_bmnzi_b_RES)(
+; CHECK: lw [[R1:\$[0-9]+]], %got(llvm_mips_bmnzi_b_ARG1)(
+; CHECK: lw [[R2:\$[0-9]+]], %got(llvm_mips_bmnzi_b_ARG2)(
+; CHECK: ld.b [[R3:\$w[0-9]+]], 0([[R2]])
+; CHECK: ld.b [[R4:\$w[0-9]+]], 0([[R1]])
+; CHECK: move.v [[R5:\$w[0-9]+]], [[R4]]
+; CHECK: binsli.b [[R5]], [[R3]], 3
+; CHECK: binsri.b [[R5]], [[R3]], 3
+; CHECK: bmnzi.b [[R4]], [[R3]], 170
+
+define void @llvm_mips_bmzi_b_test() nounwind {
+entry:
+  %0 = load <16 x i8>, <16 x i8>* @llvm_mips_bmnzi_b_ARG1
+  %1 = load <16 x i8>, <16 x i8>* @llvm_mips_bmnzi_b_ARG2
+  %2 = tail call <16 x i8> @llvm.mips.bmzi.b(<16 x i8> %0, <16 x i8> %1, i32 240)
+  store <16 x i8> %2, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  %3 = tail call <16 x i8> @llvm.mips.bmzi.b(<16 x i8> %0, <16 x i8> %1, i32 15)
+  store <16 x i8> %3, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  %4 = tail call <16 x i8> @llvm.mips.bmzi.b(<16 x i8> %0, <16 x i8> %1, i32 170)
+  store <16 x i8> %4, <16 x i8>* @llvm_mips_bmnzi_b_RES
+  ret void
+}
+; CHECK-LABEL: llvm_mips_bmzi_b_test:
+; CHECK: lw [[R0:\$[0-9]+]], %got(llvm_mips_bmnzi_b_RES)(
+; CHECK: lw [[R1:\$[0-9]+]], %got(llvm_mips_bmnzi_b_ARG2)(
+; CHECK: lw [[R2:\$[0-9]+]], %got(llvm_mips_bmnzi_b_ARG1)(
+; CHECK: ld.b [[R3:\$w[0-9]+]], 0([[R2]])
+; CHECK: ld.b [[R4:\$w[0-9]+]], 0([[R1]])
+; CHECK: move.v [[R5:\$w[0-9]+]], [[R4]]
+; CHECK: binsli.b [[R5]], [[R3]], 3
+; CHECK: binsri.b [[R5]], [[R3]], 3
+; bmnzi.b is the same as bmzi.b with ws and wd_in swapped
+; CHECK: bmnzi.b [[R4]], [[R3]], 170
+
+declare <16 x i8> @llvm.mips.bmnzi.b(<16 x i8>, <16 x i8>, i32) nounwind
+declare <16 x i8> @llvm.mips.bmzi.b(<16 x i8>, <16 x i8>, i32) nounwind

test/CodeGen/Mips/msa/i5-b.ll

 @llvm_mips_binsli_b_ARG1 = global <16 x i8> zeroinitializer, align 16
 @llvm_mips_binsli_b_ARG2 = 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_binsli_b_RES  = global <16 x i8> zeroinitializer, align 16
 
 define void @llvm_mips_binsli_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_binsli_b_ARG1
   %1 = load <16 x i8>, <16 x i8>* @llvm_mips_binsli_b_ARG2
-  %2 = tail call <16 x i8> @llvm.mips.binsli.b(<16 x i8> %0, <16 x i8> %1, i32 7)
+  %2 = tail call <16 x i8> @llvm.mips.binsli.b(<16 x i8> %0, <16 x i8> %1, i32 6)
   store <16 x i8> %2, <16 x i8>* @llvm_mips_binsli_b_RES
   ret void
 }
 
 declare <16 x i8> @llvm.mips.binsli.b(<16 x i8>, <16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_binsli_b_test:
 ; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_binsli_b_ARG1)(
 ; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_binsli_b_ARG2)(
 ; CHECK-DAG: ld.b [[R3:\$w[0-9]+]], 0([[R1]])
 ; CHECK-DAG: ld.b [[R4:\$w[0-9]+]], 0([[R2]])
-; CHECK-DAG: binsli.b [[R3]], [[R4]], 7
+; CHECK-DAG: binsli.b [[R3]], [[R4]], 6
 ; CHECK-DAG: lw [[R5:\$[0-9]+]], %got(llvm_mips_binsli_b_RES)(
 ; CHECK-DAG: st.b [[R3]], 0([[R5]])
 ; CHECK: .size llvm_mips_binsli_b_test
 
 @llvm_mips_binsli_h_ARG1 = global <8 x i16> zeroinitializer, align 16
 @llvm_mips_binsli_h_ARG2 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
 @llvm_mips_binsli_h_RES  = global <8 x i16> zeroinitializer, align 16
 
 define void @llvm_mips_binsli_h_test() nounwind {
 entry:
   %0 = load <8 x i16>, <8 x i16>* @llvm_mips_binsli_h_ARG1
   %1 = load <8 x i16>, <8 x i16>* @llvm_mips_binsli_h_ARG2
   %2 = tail call <8 x i16> @llvm.mips.binsli.h(<8 x i16> %0, <8 x i16> %1, i32 7)
   store <8 x i16> %2, <8 x i16>* @llvm_mips_binsli_h_RES
   ret void
 }
 
 declare <8 x i16> @llvm.mips.binsli.h(<8 x i16>, <8 x i16>, i32) nounwind
 
 ; CHECK: llvm_mips_binsli_h_test:
 ; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_binsli_h_ARG1)(

 @llvm_mips_binsri_b_ARG1 = global <16 x i8> zeroinitializer, align 16
 @llvm_mips_binsri_b_ARG2 = 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_binsri_b_RES  = global <16 x i8> zeroinitializer, align 16
 
 define void @llvm_mips_binsri_b_test() nounwind {
 entry:
   %0 = load <16 x i8>, <16 x i8>* @llvm_mips_binsri_b_ARG1
   %1 = load <16 x i8>, <16 x i8>* @llvm_mips_binsri_b_ARG2
-  %2 = tail call <16 x i8> @llvm.mips.binsri.b(<16 x i8> %0, <16 x i8> %1, i32 7)
+  %2 = tail call <16 x i8> @llvm.mips.binsri.b(<16 x i8> %0, <16 x i8> %1, i32 6)
   store <16 x i8> %2, <16 x i8>* @llvm_mips_binsri_b_RES
   ret void
 }
 
 declare <16 x i8> @llvm.mips.binsri.b(<16 x i8>, <16 x i8>, i32) nounwind
 
 ; CHECK: llvm_mips_binsri_b_test:
 ; CHECK-DAG: lw [[R1:\$[0-9]+]], %got(llvm_mips_binsri_b_ARG1)(
 ; CHECK-DAG: lw [[R2:\$[0-9]+]], %got(llvm_mips_binsri_b_ARG2)(
 ; CHECK-DAG: ld.b [[R3:\$w[0-9]+]], 0([[R1]])
 ; CHECK-DAG: ld.b [[R4:\$w[0-9]+]], 0([[R2]])
-; CHECK-DAG: binsri.b [[R3]], [[R4]], 7
+; CHECK-DAG: binsri.b [[R3]], [[R4]], 6
 ; CHECK-DAG: lw [[R5:\$[0-9]+]], %got(llvm_mips_binsri_b_RES)(
 ; CHECK-DAG: st.b [[R3]], 0([[R5]])
 ; CHECK: .size llvm_mips_binsri_b_test
 
 @llvm_mips_binsri_h_ARG1 = global <8 x i16> zeroinitializer, align 16
 @llvm_mips_binsri_h_ARG2 = global <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>, align 16
 @llvm_mips_binsri_h_RES  = global <8 x i16> zeroinitializer, align 16
 

test/CodeGen/Mips/msa/immediates.ll

 }
 
 define void @binsri_h(<8 x i16> * %ptr, <8 x i16> * %ptr2) {
 entry:
 ; CHECK-LABEL: binsri_h:
 ; CHECK: binsri.h
   %a = load <8 x i16>, <8 x i16> * %ptr, align 16
   %b = load <8 x i16>, <8 x i16> * %ptr2, align 16
-  %r = call <8 x i16> @llvm.mips.binsri.h(<8 x i16> %a, <8 x i16> %b, i32 15)
+  %r = call <8 x i16> @llvm.mips.binsri.h(<8 x i16> %a, <8 x i16> %b, i32 14)
   store <8 x i16> %r, <8 x i16> * %ptr, align 16
   ret void
 }
 
 define void @bnegi_h(<8 x i16> * %ptr) {
 entry:
 ; CHECK-LABEL: bnegi_h:
 ; CHECK: bnegi.h