[mips][msa] Test basic operations for the N32 ABI too.

[dsanders]

Summary

This required adding instruction aliases for dneg.

N64 will be enabled shortly but requires additional bugfixes.

Depends on D9340

[vkalintiris]

LGTM, with the typo fix and with the aliases marked appropriately.

lib/Target/Mips/Mips64InstrInfo.td

570–575

• I believe that we should mark these aliases as ISA_MIPS3.
• I can't find the second form of the dneg alias anywhere in binutils.
• dnegu isn't used in this patch.

Also, we should add some tests in test/MC for these aliases in a future patch.

test/CodeGen/Mips/msa/basic_operations.ll

5–6

Small typo: R1N -> RUN

[dsanders]

Thanks.

lib/Target/Mips/Mips64InstrInfo.td

570–575

I believe that we should mark these aliases as ISA_MIPS3.

Makes sense to me.

I can't find the second form of the dneg alias anywhere in binutils.

I mostly included this because the 1-operand form of neg is accepted by IAS. GAS accepts it as well with 'dneg $1' being equivalent to 'dneg $1, $1'.

dnegu isn't used in this patch.

I'll add some assembler/disassembler tests for it (and the others).

test/CodeGen/Mips/msa/basic_operations.ll

5–6

This isn't actually a typo. It's a placeholder that keeps the currently broken N64 disabled until the next patch. It shouldn't have been left in the patch though so I'll drop these two lines and merge them with the s/R!N/RUN/ from D9342 instead.

[dsanders]

dneg/dnegu are now available to MIPS-III and later 64-bit ISA's. They also
have assembly/disassembly tests.

Dropped 'R!N' lines.

Made basic_operations_float.ll test N32 too.

lib/Target/Mips/Mips64InstrInfo.td

def : MipsInstAlias<"daddu $rs, $imm",
                     (DADDiu GPR64Opnd:$rs, GPR64Opnd:$rs, simm16_64:$imm),
                     0>, ISA_MIPS3;
 def : MipsInstAlias<"dadd $rs, $imm",
                     (DADDi GPR64Opnd:$rs, GPR64Opnd:$rs, simm16_64:$imm),
                     0>, ISA_MIPS3_NOT_32R6_64R6;
 def : MipsInstAlias<"dsll $rd, $rt, $rs",
                     (DSLLV GPR64Opnd:$rd, GPR64Opnd:$rt, GPR32Opnd:$rs), 0>,
                     ISA_MIPS3;
+def : MipsInstAlias<"dneg $rt, $rs",
+                    (DSUB GPR64Opnd:$rt, ZERO_64, GPR64Opnd:$rs), 1>;
+def : MipsInstAlias<"dneg $rt",
+                    (DSUB GPR64Opnd:$rt, ZERO_64, GPR64Opnd:$rt), 0>;
+def : MipsInstAlias<"dnegu $rt, $rs",
+                    (DSUBu GPR64Opnd:$rt, ZERO_64, GPR64Opnd:$rs), 1>;
 def : MipsInstAlias<"dsubu $rt, $rs, $imm",
                     (DADDiu GPR64Opnd:$rt, GPR64Opnd:$rs,
                             InvertedImOperand64:$imm), 0>, ISA_MIPS3;
 def : MipsInstAlias<"dsubi $rs, $rt, $imm",
                     (DADDi GPR64Opnd:$rs, GPR64Opnd:$rt,
                            InvertedImOperand64:$imm),
                     0>, ISA_MIPS3_NOT_32R6_64R6;
 def : MipsInstAlias<"dsubi $rs, $imm",

lib/Target/Mips/MipsSEISelLowering.cpp

MipsSETargetLowering::emitINSERT_DF_VIDX(MachineInstr *MI,
   DebugLoc DL = MI->getDebugLoc();
   unsigned Wd = MI->getOperand(0).getReg();
   unsigned SrcVecReg = MI->getOperand(1).getReg();
   unsigned LaneReg = MI->getOperand(2).getReg();
   unsigned SrcValReg = MI->getOperand(3).getReg();
 
   const TargetRegisterClass *VecRC = nullptr;
   const TargetRegisterClass *GPRRC =
-      Subtarget.isGP64bit() ? &Mips::GPR64RegClass : &Mips::GPR32RegClass;
+      Subtarget.isABI_N64() ? &Mips::GPR64RegClass : &Mips::GPR32RegClass;
   unsigned EltLog2Size;
   unsigned InsertOp = 0;
   unsigned InsveOp = 0;
   switch (EltSizeInBytes) {
   default:
     llvm_unreachable("Unexpected size");
   case 1:
     EltLog2Size = 0;
BuildMI(*BB, MI, DL, TII->get(InsertOp), WdTmp2)
         .addReg(SrcValReg)
         .addImm(0);
   }
 
   // Rotate elements the rest of the way for a full rotation.
   // sld.df inteprets $rt modulo the number of columns so we only need to negate
   // the lane index to do this.
   unsigned LaneTmp2 = RegInfo.createVirtualRegister(GPRRC);
-  BuildMI(*BB, MI, DL, TII->get(Mips::SUB), LaneTmp2)
-      .addReg(Mips::ZERO)
+  BuildMI(*BB, MI, DL, TII->get(Subtarget.isABI_N64() ? Mips::DSUB : Mips::SUB),
+          LaneTmp2)
+      .addReg(Subtarget.isABI_N64() ? Mips::ZERO_64 : Mips::ZERO)
       .addReg(LaneReg);
   BuildMI(*BB, MI, DL, TII->get(Mips::SLD_B), Wd)
       .addReg(WdTmp2)
       .addReg(WdTmp2)
       .addReg(LaneTmp2);
 
   MI->eraseFromParent(); // The pseudo instruction is gone now.
   return BB;

test/CodeGen/Mips/msa/basic_operations.ll

-; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=MIPS32-BE %s
-; RUN: llc -march=mipsel -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=MIPS32-LE %s
+; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=O32 -check-prefix=MIPS32 -check-prefix=ALL-BE %s
+; RUN: llc -march=mipsel -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=O32 -check-prefix=MIPS32 -check-prefix=ALL-LE %s
+; RUN: llc -march=mips64 -target-abi n32 -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=N32 -check-prefix=MIPS64 -check-prefix=ALL-BE %s
+; RUN: llc -march=mips64el -target-abi n32 -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=N32 -check-prefix=MIPS64 -check-prefix=ALL-LE %s
+; R!N: llc -march=mips64 -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=N64 -check-prefix=MIPS64 -check-prefix=ALL-BE %s
+; R!N: llc -march=mips64el -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=ALL -check-prefix=N64 -check-prefix=MIPS64 -check-prefix=ALL-LE %s
 
 @v4i8 = global <4 x i8> <i8 0, i8 0, i8 0, i8 0>
 @v16i8 = 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>
 @v8i16 = global <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>
 @v4i32 = global <4 x i32> <i32 0, i32 0, i32 0, i32 0>
 @v2i64 = global <2 x i64> <i64 0, i64 0>
 @i32 = global i32 0
 @i64 = global i64 0
 
 define void @const_v16i8() nounwind {
   ; ALL-LABEL: const_v16i8:
 
   store volatile <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>, <16 x i8>*@v16i8
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 0
 
   store volatile <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, <16 x i8>*@v16i8
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 1
 
   store volatile <16 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 31>, <16 x i8>*@v16i8
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.b  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6>, <16 x i8>*@v16i8
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.b  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <16 x i8> <i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0, i8 1, i8 0>, <16 x i8>*@v16i8
-  ; MIPS32-BE: ldi.h [[R1:\$w[0-9]+]], 256
-  ; MIPS32-LE: ldi.h [[R1:\$w[0-9]+]], 1
+  ; ALL-BE: ldi.h [[R1:\$w[0-9]+]], 256
+  ; ALL-LE: ldi.h [[R1:\$w[0-9]+]], 1
 
   store volatile <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 1, i8 2, i8 3, i8 4, i8 1, i8 2, i8 3, i8 4, i8 1, i8 2, i8 3, i8 4>, <16 x i8>*@v16i8
-  ; MIPS32-BE-DAG: lui [[R2:\$[0-9]+]], 258
-  ; MIPS32-LE-DAG: lui [[R2:\$[0-9]+]], 1027
-  ; MIPS32-BE-DAG: ori [[R2]], [[R2]], 772
-  ; MIPS32-LE-DAG: ori [[R2]], [[R2]], 513
+  ; ALL-BE-DAG: lui [[R2:\$[0-9]+]], 258
+  ; ALL-LE-DAG: lui [[R2:\$[0-9]+]], 1027
+  ; ALL-BE-DAG: ori [[R2]], [[R2]], 772
+  ; ALL-LE-DAG: ori [[R2]], [[R2]], 513
   ; ALL-DAG: fill.w [[R1:\$w[0-9]+]], [[R2]]
 
   store volatile <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8>, <16 x i8>*@v16i8
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.b  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   ret void
 }
 
 define void @const_v8i16() nounwind {
   ; ALL-LABEL: const_v8i16:
 
   store volatile <8 x i16> <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>, <8 x i16>*@v8i16
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 0
 
   store volatile <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <8 x i16>*@v8i16
   ; ALL: ldi.h [[R1:\$w[0-9]+]], 1
 
   store volatile <8 x i16> <i16 1, i16 1, i16 1, i16 2, i16 1, i16 1, i16 1, i16 31>, <8 x i16>*@v8i16
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.h  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <8 x i16> <i16 1028, i16 1028, i16 1028, i16 1028, i16 1028, i16 1028, i16 1028, i16 1028>, <8 x i16>*@v8i16
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 4
 
   store volatile <8 x i16> <i16 1, i16 2, i16 1, i16 2, i16 1, i16 2, i16 1, i16 2>, <8 x i16>*@v8i16
-  ; MIPS32-BE-DAG: lui [[R2:\$[0-9]+]], 1
-  ; MIPS32-LE-DAG: lui [[R2:\$[0-9]+]], 2
-  ; MIPS32-BE-DAG: ori [[R2]], [[R2]], 2
-  ; MIPS32-LE-DAG: ori [[R2]], [[R2]], 1
+  ; ALL-BE-DAG: lui [[R2:\$[0-9]+]], 1
+  ; ALL-LE-DAG: lui [[R2:\$[0-9]+]], 2
+  ; ALL-BE-DAG: ori [[R2]], [[R2]], 2
+  ; ALL-LE-DAG: ori [[R2]], [[R2]], 1
   ; ALL-DAG: fill.w [[R1:\$w[0-9]+]], [[R2]]
 
   store volatile <8 x i16> <i16 1, i16 2, i16 3, i16 4, i16 1, i16 2, i16 3, i16 4>, <8 x i16>*@v8i16
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.h  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   ret void
 }
 
 define void @const_v4i32() nounwind {
   ; ALL-LABEL: const_v4i32:
 
   store volatile <4 x i32> <i32 0, i32 0, i32 0, i32 0>, <4 x i32>*@v4i32
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 0
 
   store volatile <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32>*@v4i32
   ; ALL: ldi.w [[R1:\$w[0-9]+]], 1
 
   store volatile <4 x i32> <i32 1, i32 1, i32 1, i32 31>, <4 x i32>*@v4i32
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <4 x i32> <i32 16843009, i32 16843009, i32 16843009, i32 16843009>, <4 x i32>*@v4i32
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 1
 
   store volatile <4 x i32> <i32 65537, i32 65537, i32 65537, i32 65537>, <4 x i32>*@v4i32
   ; ALL: ldi.h [[R1:\$w[0-9]+]], 1
 
   store volatile <4 x i32> <i32 1, i32 2, i32 1, i32 2>, <4 x i32>*@v4i32
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <4 x i32> <i32 3, i32 4, i32 5, i32 6>, <4 x i32>*@v4i32
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
   ; ALL: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   ret void
 }
 
 define void @const_v2i64() nounwind {
   ; ALL-LABEL: const_v2i64:
 
   store volatile <2 x i64> <i64 0, i64 0>, <2 x i64>*@v2i64
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 0
 
   store volatile <2 x i64> <i64 72340172838076673, i64 72340172838076673>, <2 x i64>*@v2i64
   ; ALL: ldi.b [[R1:\$w[0-9]+]], 1
 
   store volatile <2 x i64> <i64 281479271743489, i64 281479271743489>, <2 x i64>*@v2i64
   ; ALL: ldi.h [[R1:\$w[0-9]+]], 1
 
   store volatile <2 x i64> <i64 4294967297, i64 4294967297>, <2 x i64>*@v2i64
   ; ALL: ldi.w [[R1:\$w[0-9]+]], 1
 
   store volatile <2 x i64> <i64 1, i64 1>, <2 x i64>*@v2i64
   ; ALL: ldi.d [[R1:\$w[0-9]+]], 1
 
   store volatile <2 x i64> <i64 1, i64 31>, <2 x i64>*@v2i64
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
-  ; ALL: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; MIPS32: ld.w [[R1:\$w[0-9]+]], 0([[G_PTR]])
+  ; MIPS64: ld.d [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   store volatile <2 x i64> <i64 3, i64 4>, <2 x i64>*@v2i64
-  ; ALL: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
-  ; ALL: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
+  ; O32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %lo($
+  ; N32: addiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; N64: daddiu [[G_PTR:\$[0-9]+]], {{.*}}, %got_ofst($
+  ; MIPS32: ld.w  [[R1:\$w[0-9]+]], 0([[G_PTR]])
+  ; MIPS64: ld.d  [[R1:\$w[0-9]+]], 0([[G_PTR]])
 
   ret void
 }
 
 define void @nonconst_v16i8(i8 signext %a, i8 signext %b, i8 signext %c, i8 signext %d, i8 signext %e, i8 signext %f, i8 signext %g, i8 signext %h) nounwind {
   ; ALL-LABEL: nonconst_v16i8:
 
   %1 = insertelement <16 x i8> undef, i8 %a, i32 0
define void @nonconst_v16i8(i8 signext %a, i8 signext %b, i8 signext %c, i8 signext %d, i8 signext %e, i8 signext %f, i8 signext %g, i8 signext %h) nounwind {
   %13 = insertelement <16 x i8> %12, i8 %h, i32 12
   %14 = insertelement <16 x i8> %13, i8 %h, i32 13
   %15 = insertelement <16 x i8> %14, i8 %h, i32 14
   %16 = insertelement <16 x i8> %15, i8 %h, i32 15
   ; ALL-DAG: insert.b [[R1:\$w[0-9]+]][0], $4
   ; ALL-DAG: insert.b [[R1]][1], $5
   ; ALL-DAG: insert.b [[R1]][2], $6
   ; ALL-DAG: insert.b [[R1]][3], $7
-  ; ALL-DAG: lw [[R2:\$[0-9]+]], 16($sp)
-  ; ALL-DAG: insert.b [[R1]][4], [[R2]]
-  ; ALL-DAG: lw [[R3:\$[0-9]+]], 20($sp)
-  ; ALL-DAG: insert.b [[R1]][5], [[R3]]
-  ; ALL-DAG: lw [[R4:\$[0-9]+]], 24($sp)
-  ; ALL-DAG: insert.b [[R1]][6], [[R4]]
-  ; ALL-DAG: lw [[R5:\$[0-9]+]], 28($sp)
-  ; ALL-DAG: insert.b [[R1]][7], [[R5]]
+  ; MIPS32-DAG: lw [[R2:\$[0-9]+]], 16($sp)
+  ; MIPS32-DAG: insert.b [[R1]][4], [[R2]]
+  ; MIPS64-DAG: insert.b [[R1]][4], $8
+  ; MIPS32-DAG: lw [[R3:\$[0-9]+]], 20($sp)
+  ; MIPS32-DAG: insert.b [[R1]][5], [[R3]]
+  ; MIPS64-DAG: insert.b [[R1]][5], $9
+  ; MIPS32-DAG: lw [[R4:\$[0-9]+]], 24($sp)
+  ; MIPS32-DAG: insert.b [[R1]][6], [[R4]]
+  ; MIPS64-DAG: insert.b [[R1]][6], $10
+  ; MIPS32-DAG: lw [[R5:\$[0-9]+]], 28($sp)
+  ; MIPS32-DAG: insert.b [[R1]][7], [[R5]]
+  ; MIPS64-DAG: insert.b [[R1]][7], [[R5:\$11]]
   ; ALL-DAG: insert.b [[R1]][8], [[R5]]
   ; ALL-DAG: insert.b [[R1]][9], [[R5]]
   ; ALL-DAG: insert.b [[R1]][10], [[R5]]
   ; ALL-DAG: insert.b [[R1]][11], [[R5]]
   ; ALL-DAG: insert.b [[R1]][12], [[R5]]
   ; ALL-DAG: insert.b [[R1]][13], [[R5]]
   ; ALL-DAG: insert.b [[R1]][14], [[R5]]
   ; ALL-DAG: insert.b [[R1]][15], [[R5]]
define void @nonconst_v8i16(i16 signext %a, i16 signext %b, i16 signext %c, i16 signext %d, i16 signext %e, i16 signext %f, i16 signext %g, i16 signext %h) nounwind {
   %5 = insertelement <8 x i16> %4, i16 %e, i32 4
   %6 = insertelement <8 x i16> %5, i16 %f, i32 5
   %7 = insertelement <8 x i16> %6, i16 %g, i32 6
   %8 = insertelement <8 x i16> %7, i16 %h, i32 7
   ; ALL-DAG: insert.h [[R1:\$w[0-9]+]][0], $4
   ; ALL-DAG: insert.h [[R1]][1], $5
   ; ALL-DAG: insert.h [[R1]][2], $6
   ; ALL-DAG: insert.h [[R1]][3], $7
-  ; ALL-DAG: lw [[R2:\$[0-9]+]], 16($sp)
-  ; ALL-DAG: insert.h [[R1]][4], [[R2]]
-  ; ALL-DAG: lw [[R2:\$[0-9]+]], 20($sp)
-  ; ALL-DAG: insert.h [[R1]][5], [[R2]]
-  ; ALL-DAG: lw [[R2:\$[0-9]+]], 24($sp)
-  ; ALL-DAG: insert.h [[R1]][6], [[R2]]
-  ; ALL-DAG: lw [[R2:\$[0-9]+]], 28($sp)
-  ; ALL-DAG: insert.h [[R1]][7], [[R2]]
+  ; MIPS32-DAG: lw [[R2:\$[0-9]+]], 16($sp)
+  ; MIPS32-DAG: insert.h [[R1]][4], [[R2]]
+  ; MIPS64-DAG: insert.h [[R1]][4], $8
+  ; MIPS32-DAG: lw [[R2:\$[0-9]+]], 20($sp)
+  ; MIPS32-DAG: insert.h [[R1]][5], [[R2]]
+  ; MIPS64-DAG: insert.h [[R1]][5], $9
+  ; MIPS32-DAG: lw [[R2:\$[0-9]+]], 24($sp)
+  ; MIPS32-DAG: insert.h [[R1]][6], [[R2]]
+  ; MIPS64-DAG: insert.h [[R1]][6], $10
+  ; MIPS32-DAG: lw [[R2:\$[0-9]+]], 28($sp)
+  ; MIPS32-DAG: insert.h [[R1]][7], [[R2]]
+  ; MIPS64-DAG: insert.h [[R1]][7], $11
 
   store volatile <8 x i16> %8, <8 x i16>*@v8i16
 
   ret void
 }
 
 define void @nonconst_v4i32(i32 signext %a, i32 signext %b, i32 signext %c, i32 signext %d) nounwind {
   ; ALL-LABEL: nonconst_v4i32:
define void @nonconst_v4i32(i32 signext %a, i32 signext %b, i32 signext %c, i32 signext %d) nounwind {
   ret void
 }
 
 define void @nonconst_v2i64(i64 signext %a, i64 signext %b) nounwind {
   ; ALL-LABEL: nonconst_v2i64:
 
   %1 = insertelement <2 x i64> undef, i64 %a, i32 0
   %2 = insertelement <2 x i64> %1, i64 %b, i32 1
-  ; ALL: insert.w [[R1:\$w[0-9]+]][0], $4
-  ; ALL: insert.w [[R1]][1], $5
-  ; ALL: insert.w [[R1]][2], $6
-  ; ALL: insert.w [[R1]][3], $7
+  ; MIPS32: insert.w [[R1:\$w[0-9]+]][0], $4
+  ; MIPS32: insert.w [[R1]][1], $5
+  ; MIPS32: insert.w [[R1]][2], $6
+  ; MIPS32: insert.w [[R1]][3], $7
+  ; MIPS64: insert.d [[R1:\$w[0-9]+]][0], $4
+  ; MIPS64: insert.d [[R1]][1], $5
 
   store volatile <2 x i64> %2, <2 x i64>*@v2i64
 
   ret void
 }
 
 define i32 @extract_sext_v16i8() nounwind {
   ; ALL-LABEL: extract_sext_v16i8:
define i64 @extract_sext_v2i64() nounwind {
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
   ; ALL-DAG: ld.d [[R1:\$w[0-9]+]],
 
   %2 = add <2 x i64> %1, %1
   ; ALL-DAG: addv.d [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = extractelement <2 x i64> %2, i32 1
-  ; ALL-DAG: copy_s.w [[R3:\$[0-9]+]], [[R1]][2]
-  ; ALL-DAG: copy_s.w [[R4:\$[0-9]+]], [[R1]][3]
+  ; MIPS32-DAG: copy_s.w [[R3:\$[0-9]+]], [[R1]][2]
+  ; MIPS32-DAG: copy_s.w [[R4:\$[0-9]+]], [[R1]][3]
+  ; MIPS64-DAG: copy_s.d [[R3:\$[0-9]+]], [[R1]][1]
   ; ALL-NOT: sll
   ; ALL-NOT: sra
 
   ret i64 %3
 }
 
 define i32 @extract_zext_v16i8() nounwind {
   ; ALL-LABEL: extract_zext_v16i8:
define i64 @extract_zext_v2i64() nounwind {
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
   ; ALL-DAG: ld.d [[R1:\$w[0-9]+]],
 
   %2 = add <2 x i64> %1, %1
   ; ALL-DAG: addv.d [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = extractelement <2 x i64> %2, i32 1
-  ; ALL-DAG: copy_{{[su]}}.w [[R3:\$[0-9]+]], [[R1]][2]
-  ; ALL-DAG: copy_{{[su]}}.w [[R4:\$[0-9]+]], [[R1]][3]
+  ; MIPS32-DAG: copy_{{[su]}}.w [[R3:\$[0-9]+]], [[R1]][2]
+  ; MIPS32-DAG: copy_{{[su]}}.w [[R4:\$[0-9]+]], [[R1]][3]
+  ; MIPS64-DAG: copy_{{[su]}}.d [[R3:\$[0-9]+]], [[R1]][1]
   ; ALL-NOT: andi
 
   ret i64 %3
 }
 
 define i32 @extract_sext_v16i8_vidx() nounwind {
   ; ALL-LABEL: extract_sext_v16i8_vidx:
 
   %1 = load <16 x i8>, <16 x i8>* @v16i8
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v16i8)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v16i8)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v16i8)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v16i8)(
   ; ALL-DAG: ld.b [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <16 x i8> %1, %1
   ; ALL-DAG: addv.b [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <16 x i8> %2, i32 %3
   %5 = sext i8 %4 to i32
   ; ALL-DAG: splat.b $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-DAG: sra [[R6:\$[0-9]+]], [[R5]], 24
 
   ret i32 %5
 }
 
 define i32 @extract_sext_v8i16_vidx() nounwind {
   ; ALL-LABEL: extract_sext_v8i16_vidx:
 
   %1 = load <8 x i16>, <8 x i16>* @v8i16
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v8i16)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v8i16)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v8i16)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v8i16)(
   ; ALL-DAG: ld.h [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <8 x i16> %1, %1
   ; ALL-DAG: addv.h [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <8 x i16> %2, i32 %3
   %5 = sext i16 %4 to i32
   ; ALL-DAG: splat.h $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-DAG: sra [[R6:\$[0-9]+]], [[R5]], 16
 
   ret i32 %5
 }
 
 define i32 @extract_sext_v4i32_vidx() nounwind {
   ; ALL-LABEL: extract_sext_v4i32_vidx:
 
   %1 = load <4 x i32>, <4 x i32>* @v4i32
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v4i32)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v4i32)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v4i32)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v4i32)(
   ; ALL-DAG: ld.w [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <4 x i32> %1, %1
   ; ALL-DAG: addv.w [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <4 x i32> %2, i32 %3
   ; ALL-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-NOT: sra
 
   ret i32 %4
 }
 
 define i64 @extract_sext_v2i64_vidx() nounwind {
   ; ALL-LABEL: extract_sext_v2i64_vidx:
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v2i64)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v2i64)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v2i64)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v2i64)(
   ; ALL-DAG: ld.d [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <2 x i64> %1, %1
   ; ALL-DAG: addv.d [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <2 x i64> %2, i32 %3
-  ; ALL-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
-  ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
-  ; ALL-DAG: splat.w $w[[R4:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
-  ; ALL-DAG: mfc1 [[R6:\$[0-9]+]], $f[[R4]]
+  ; MIPS32-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS32-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
+  ; MIPS32-DAG: splat.w $w[[R4:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS32-DAG: mfc1 [[R6:\$[0-9]+]], $f[[R4]]
+  ; MIPS64-DAG: splat.d $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS64-DAG: dmfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-NOT: sra
 
   ret i64 %4
 }
 
 define i32 @extract_zext_v16i8_vidx() nounwind {
   ; ALL-LABEL: extract_zext_v16i8_vidx:
 
   %1 = load <16 x i8>, <16 x i8>* @v16i8
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v16i8)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v16i8)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v16i8)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v16i8)(
   ; ALL-DAG: ld.b [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <16 x i8> %1, %1
   ; ALL-DAG: addv.b [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <16 x i8> %2, i32 %3
   %5 = zext i8 %4 to i32
   ; ALL-DAG: splat.b $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-DAG: srl [[R6:\$[0-9]+]], [[R5]], 24
 
   ret i32 %5
 }
 
 define i32 @extract_zext_v8i16_vidx() nounwind {
   ; ALL-LABEL: extract_zext_v8i16_vidx:
 
   %1 = load <8 x i16>, <8 x i16>* @v8i16
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v8i16)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v8i16)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v8i16)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v8i16)(
   ; ALL-DAG: ld.h [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <8 x i16> %1, %1
   ; ALL-DAG: addv.h [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <8 x i16> %2, i32 %3
   %5 = zext i16 %4 to i32
   ; ALL-DAG: splat.h $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-DAG: srl [[R6:\$[0-9]+]], [[R5]], 16
 
   ret i32 %5
 }
 
 define i32 @extract_zext_v4i32_vidx() nounwind {
   ; ALL-LABEL: extract_zext_v4i32_vidx:
 
   %1 = load <4 x i32>, <4 x i32>* @v4i32
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v4i32)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v4i32)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v4i32)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v4i32)(
   ; ALL-DAG: ld.w [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <4 x i32> %1, %1
   ; ALL-DAG: addv.w [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <4 x i32> %2, i32 %3
   ; ALL-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-NOT: srl
 
   ret i32 %4
 }
 
 define i64 @extract_zext_v2i64_vidx() nounwind {
   ; ALL-LABEL: extract_zext_v2i64_vidx:
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
-  ; ALL-DAG: lw [[PTR_V:\$[0-9]+]], %got(v2i64)(
+  ; O32-DAG: lw [[PTR_V:\$[0-9]+]], %got(v2i64)(
+  ; N32-DAG: lw [[PTR_V:\$[0-9]+]], %got_disp(v2i64)(
+  ; N64-DAG: ld [[PTR_V:\$[0-9]+]], %got_disp(v2i64)(
   ; ALL-DAG: ld.d [[R1:\$w[0-9]+]], 0([[PTR_V]])
 
   %2 = add <2 x i64> %1, %1
   ; ALL-DAG: addv.d [[R2:\$w[0-9]+]], [[R1]], [[R1]]
 
   %3 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %4 = extractelement <2 x i64> %2, i32 %3
-  ; ALL-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
-  ; ALL-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
-  ; ALL-DAG: splat.w $w[[R4:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
-  ; ALL-DAG: mfc1 [[R6:\$[0-9]+]], $f[[R4]]
+  ; MIPS32-DAG: splat.w $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS32-DAG: mfc1 [[R5:\$[0-9]+]], $f[[R3]]
+  ; MIPS32-DAG: splat.w $w[[R4:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS32-DAG: mfc1 [[R6:\$[0-9]+]], $f[[R4]]
+  ; MIPS64-DAG: splat.d $w[[R3:[0-9]+]], [[R1]]{{\[}}[[IDX]]]
+  ; MIPS64-DAG: dmfc1 [[R5:\$[0-9]+]], $f[[R3]]
   ; ALL-NOT: srl
 
   ret i64 %4
 }
 
 define void @insert_v16i8(i32 signext %a) nounwind {
   ; ALL-LABEL: insert_v16i8:
 
define void @insert_v4i32(i32 signext %a) nounwind {
 
   ret void
 }
 
 define void @insert_v2i64(i64 signext %a) nounwind {
   ; ALL-LABEL: insert_v2i64:
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
-  ; ALL-DAG: ld.w [[R1:\$w[0-9]+]],
+  ; MIPS32-DAG: ld.w [[R1:\$w[0-9]+]],
+  ; MIPS64-DAG: ld.d [[R1:\$w[0-9]+]],
 
   ; ALL-NOT: andi
   ; ALL-NOT: sra
 
   %2 = insertelement <2 x i64> %1, i64 %a, i32 1
-  ; ALL-DAG: insert.w [[R1]][2], $4
-  ; ALL-DAG: insert.w [[R1]][3], $5
+  ; MIPS32-DAG: insert.w [[R1]][2], $4
+  ; MIPS32-DAG: insert.w [[R1]][3], $5
+  ; MIPS64-DAG: insert.d [[R1]][1], $4
 
   store <2 x i64> %2, <2 x i64>* @v2i64
-  ; ALL-DAG: st.w [[R1]]
+  ; MIPS32-DAG: st.w [[R1]]
+  ; MIPS64-DAG: st.d [[R1]]
 
   ret void
 }
 
 define void @insert_v16i8_vidx(i32 signext %a) nounwind {
   ; ALL-LABEL: insert_v16i8_vidx:
 
   %1 = load <16 x i8>, <16 x i8>* @v16i8
   ; ALL-DAG: ld.b [[R1:\$w[0-9]+]],
 
   %2 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %a2 = trunc i32 %a to i8
   %a3 = sext i8 %a2 to i32
   %a4 = trunc i32 %a3 to i8
   ; ALL-NOT: andi
   ; ALL-NOT: sra
 
   %3 = insertelement <16 x i8> %1, i8 %a4, i32 %2
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[IDX]]]
   ; ALL-DAG: insert.b [[R1]][0], $4
-  ; ALL-DAG: neg [[NIDX:\$[0-9]+]], [[IDX]]
+  ; O32-DAG: neg [[NIDX:\$[0-9]+]], [[IDX]]
+  ; N32-DAG: neg [[NIDX:\$[0-9]+]], [[IDX]]
+  ; N64-DAG: dneg [[NIDX:\$[0-9]+]], [[IDX]]
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
 
   store <16 x i8> %3, <16 x i8>* @v16i8
   ; ALL-DAG: st.b [[R1]]
 
   ret void
 }
 
-define void @insert_v8i16_vidx(i32 %a) nounwind {
+define void @insert_v8i16_vidx(i32 signext %a) nounwind {
   ; ALL-LABEL: insert_v8i16_vidx:
 
   %1 = load <8 x i16>, <8 x i16>* @v8i16
   ; ALL-DAG: ld.h [[R1:\$w[0-9]+]],
 
   %2 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   %a2 = trunc i32 %a to i16
   %a3 = sext i16 %a2 to i32
   %a4 = trunc i32 %a3 to i16
   ; ALL-NOT: andi
   ; ALL-NOT: sra
 
   %3 = insertelement <8 x i16> %1, i16 %a4, i32 %2
   ; ALL-DAG: sll [[BIDX:\$[0-9]+]], [[IDX]], 1
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
   ; ALL-DAG: insert.h [[R1]][0], $4
-  ; ALL-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; O32-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; N32-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; N64-DAG: dneg [[NIDX:\$[0-9]+]], [[BIDX]]
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
 
   store <8 x i16> %3, <8 x i16>* @v8i16
   ; ALL-DAG: st.h [[R1]]
 
   ret void
 }
 
 define void @insert_v4i32_vidx(i32 signext %a) nounwind {
   ; ALL-LABEL: insert_v4i32_vidx:
 
   %1 = load <4 x i32>, <4 x i32>* @v4i32
   ; ALL-DAG: ld.w [[R1:\$w[0-9]+]],
 
   %2 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   ; ALL-NOT: andi
   ; ALL-NOT: sra
 
   %3 = insertelement <4 x i32> %1, i32 %a, i32 %2
   ; ALL-DAG: sll [[BIDX:\$[0-9]+]], [[IDX]], 2
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
   ; ALL-DAG: insert.w [[R1]][0], $4
   ; ALL-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
   ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
 
   store <4 x i32> %3, <4 x i32>* @v4i32
   ; ALL-DAG: st.w [[R1]]
 
   ret void
 }
 
 define void @insert_v2i64_vidx(i64 signext %a) nounwind {
   ; ALL-LABEL: insert_v2i64_vidx:
 
   %1 = load <2 x i64>, <2 x i64>* @v2i64
-  ; ALL-DAG: ld.w [[R1:\$w[0-9]+]],
+  ; MIPS32-DAG: ld.w [[R1:\$w[0-9]+]],
+  ; MIPS64-DAG: ld.d [[R1:\$w[0-9]+]],
 
   %2 = load i32, i32* @i32
-  ; ALL-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; O32-DAG: lw [[PTR_I:\$[0-9]+]], %got(i32)(
+  ; N32-DAG: lw [[PTR_I:\$[0-9]+]], %got_disp(i32)(
+  ; N64-DAG: ld [[PTR_I:\$[0-9]+]], %got_disp(i32)(
   ; ALL-DAG: lw [[IDX:\$[0-9]+]], 0([[PTR_I]])
 
   ; ALL-NOT: andi
   ; ALL-NOT: sra
 
   %3 = insertelement <2 x i64> %1, i64 %a, i32 %2
   ; TODO: This code could be a lot better but it works. The legalizer splits
   ; 64-bit inserts into two 32-bit inserts because there is no i64 type on
   ; MIPS32. The obvious optimisation is to perform both insert.w's at once while
   ; the vector is rotated.
-  ; ALL-DAG: sll [[BIDX:\$[0-9]+]], [[IDX]], 2
-  ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
-  ; ALL-DAG: insert.w [[R1]][0], $4
-  ; ALL-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
-  ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
-  ; ALL-DAG: addiu [[IDX2:\$[0-9]+]], [[IDX]], 1
-  ; ALL-DAG: sll [[BIDX:\$[0-9]+]], [[IDX2]], 2
-  ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
-  ; ALL-DAG: insert.w [[R1]][0], $5
-  ; ALL-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
-  ; ALL-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
+  ; MIPS32-DAG: sll [[BIDX:\$[0-9]+]], [[IDX]], 2
+  ; MIPS32-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
+  ; MIPS32-DAG: insert.w [[R1]][0], $4
+  ; MIPS32-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; MIPS32-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
+  ; MIPS32-DAG: addiu [[IDX2:\$[0-9]+]], [[IDX]], 1
+  ; MIPS32-DAG: sll [[BIDX:\$[0-9]+]], [[IDX2]], 2
+  ; MIPS32-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
+  ; MIPS32-DAG: insert.w [[R1]][0], $5
+  ; MIPS32-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; MIPS32-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
+
+  ; MIPS64-DAG: sll [[BIDX:\$[0-9]+]], [[IDX]], 3
+  ; MIPS64-DAG: sld.b [[R1]], [[R1]]{{\[}}[[BIDX]]]
+  ; MIPS64-DAG: insert.d [[R1]][0], $4
+  ; N32-DAG: neg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; N64-DAG: dneg [[NIDX:\$[0-9]+]], [[BIDX]]
+  ; MIPS64-DAG: sld.b [[R1]], [[R1]]{{\[}}[[NIDX]]]
 
   store <2 x i64> %3, <2 x i64>* @v2i64
-  ; ALL-DAG: st.w [[R1]]
+  ; MIPS32-DAG: st.w [[R1]]
+  ; MIPS64-DAG: st.d [[R1]]
 
   ret void
 }
 
 define void @truncstore() nounwind {
   ; ALL-LABEL: truncstore:
 
   store volatile <4 x i8> <i8 -1, i8 -1, i8 -1, i8 -1>, <4 x i8>*@v4i8
   ; TODO: What code should be emitted?
 
   ret void
 }