AMDGPU/GlobalISel: Legality and RegBankInfo for G_{INSERT|EXTRACT}_VECTOR_ELT

llvm-svn: 327269

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 5c1359e..de8147d 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -107,6 +107,17 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const SISubtarget &ST,
   setAction({G_LOAD, 1, S64}, Legal);
   setAction({G_STORE, 1, S64}, Legal);
 
+  for (unsigned Op : {G_EXTRACT_VECTOR_ELT, G_INSERT_VECTOR_ELT}) {
+    getActionDefinitionsBuilder(Op)
+      .legalIf([=](const LegalityQuery &Query) {
+          const LLT &VecTy = Query.Types[1];
+          const LLT &IdxTy = Query.Types[2];
+          return VecTy.getSizeInBits() % 32 == 0 &&
+            VecTy.getSizeInBits() <= 512 &&
+            IdxTy.getSizeInBits() == 32;
+        });
+  }
+
   // FIXME: Doesn't handle extract of illegal sizes.
   getActionDefinitionsBuilder(G_EXTRACT)
     .unsupportedIf([=](const LegalityQuery &Query) {

diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 72d5d6d..af75aee 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -50,6 +50,24 @@ AMDGPURegisterBankInfo::AMDGPURegisterBankInfo(const TargetRegisterInfo &TRI)
 
 }
 
+static bool isConstant(const MachineOperand &MO, int64_t &C) {
+  const MachineFunction *MF = MO.getParent()->getParent()->getParent();
+  const MachineRegisterInfo &MRI = MF->getRegInfo();
+  const MachineInstr *Def = MRI.getVRegDef(MO.getReg());
+  if (!Def)
+    return false;
+
+  if (Def->getOpcode() == AMDGPU::G_CONSTANT) {
+    C = Def->getOperand(1).getCImm()->getSExtValue();
+    return true;
+  }
+
+  if (Def->getOpcode() == AMDGPU::COPY)
+    return isConstant(Def->getOperand(1), C);
+
+  return false;
+}
+
 unsigned AMDGPURegisterBankInfo::copyCost(const RegisterBank &Dst,
                                           const RegisterBank &Src,
                                           unsigned Size) const {
@@ -415,6 +433,47 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
     OpdsMapping[3] = AMDGPU::getValueMapping(Op3Bank, Size);
     break;
   }
+
+
+  case AMDGPU::G_EXTRACT_VECTOR_ELT: {
+    unsigned IdxOp = 2;
+    int64_t Imm;
+    // XXX - Do we really need to fully handle these? The constant case should
+    // be legalized away before RegBankSelect?
+
+    unsigned OutputBankID = isSALUMapping(MI) && isConstant(MI.getOperand(IdxOp), Imm) ?
+                            AMDGPU::SGPRRegBankID : AMDGPU::VGPRRegBankID;
+
+    unsigned IdxBank = getRegBankID(MI.getOperand(2).getReg(), MRI, *TRI);
+    OpdsMapping[0] = AMDGPU::getValueMapping(OutputBankID, MRI.getType(MI.getOperand(0).getReg()).getSizeInBits());
+    OpdsMapping[1] = AMDGPU::getValueMapping(OutputBankID, MRI.getType(MI.getOperand(1).getReg()).getSizeInBits());
+
+    // The index can be either if the source vector is VGPR.
+    OpdsMapping[2] = AMDGPU::getValueMapping(IdxBank, MRI.getType(MI.getOperand(2).getReg()).getSizeInBits());
+    break;
+  }
+  case AMDGPU::G_INSERT_VECTOR_ELT: {
+    // XXX - Do we really need to fully handle these? The constant case should
+    // be legalized away before RegBankSelect?
+
+    int64_t Imm;
+
+    unsigned IdxOp = MI.getOpcode() == AMDGPU::G_EXTRACT_VECTOR_ELT ? 2 : 3;
+    unsigned BankID = isSALUMapping(MI) && isConstant(MI.getOperand(IdxOp), Imm) ?
+                      AMDGPU::SGPRRegBankID : AMDGPU::VGPRRegBankID;
+
+
+
+    // TODO: Can do SGPR indexing, which would obviate the need for the
+    // isConstant check.
+    for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
+      unsigned Size = getSizeInBits(MI.getOperand(i).getReg(), MRI, *TRI);
+      OpdsMapping[i] = AMDGPU::getValueMapping(BankID, Size);
+    }
+
+
+    break;
+  }
   case AMDGPU::G_INTRINSIC: {
     switch(MI.getOperand(1).getIntrinsicID()) {
     default:

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-extract-vector-elt.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-extract-vector-elt.mir
new file mode 100644 (file)
index 0000000..5ed787b
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-extract-vector-elt.mir
@@ -0,0 +1,180 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel %s -o - | FileCheck %s
+
+---
+name: extract_vector_elt_0_v2i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: extract_vector_elt_0_v2i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<2 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(s32) = G_CONSTANT i32 0
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...
+---
+name: extract_vector_elt_0_v3i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2
+    ; CHECK-LABEL: name: extract_vector_elt_0_v3i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<3 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
+    %1:_(s32) = G_CONSTANT i32 0
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...
+---
+name: extract_vector_elt_0_v4i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: extract_vector_elt_0_v4i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<4 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(s32) = G_CONSTANT i32 0
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: extract_vector_elt_0_v5i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: extract_vector_elt_0_v5i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[MV:%[0-9]+]]:_(<5 x s32>) = G_MERGE_VALUES [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[MV]](<5 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(<5 x s32>) = G_MERGE_VALUES %0, %0, %0, %0, %0
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: extract_vector_elt_0_v6i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: extract_vector_elt_0_v6i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[MV:%[0-9]+]]:_(<6 x s32>) = G_MERGE_VALUES [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[MV]](<6 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(<6 x s32>) = G_MERGE_VALUES %0, %0, %0, %0, %0, %0
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: extract_vector_elt_0_v7i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: extract_vector_elt_0_v7i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[MV:%[0-9]+]]:_(<7 x s32>) = G_MERGE_VALUES [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[MV]](<7 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(<7 x s32>) = G_MERGE_VALUES %0, %0, %0, %0, %0, %0, %0
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: extract_vector_elt_0_v8i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: extract_vector_elt_0_v8i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[MV:%[0-9]+]]:_(<8 x s32>) = G_MERGE_VALUES [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[MV]](<8 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(<8 x s32>) = G_MERGE_VALUES %0, %0, %0, %0, %0, %0, %0, %0
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: extract_vector_elt_0_v16i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: extract_vector_elt_0_v16i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[MV:%[0-9]+]]:_(<16 x s32>) = G_MERGE_VALUES [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32), [[COPY]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[MV]](<16 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(<16 x s32>) = G_MERGE_VALUES %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0, %0
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(s32) = G_EXTRACT_VECTOR_ELT %1, %2
+    $vgpr0 = COPY %3
+...
+
+---
+name: extract_vector_elt_var_v2i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2
+    ; CHECK-LABEL: name: extract_vector_elt_var_v2i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<2 x s32>), [[COPY1]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(s32) = COPY $vgpr2
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: extract_vector_elt_var_v8i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
+    ; CHECK-LABEL: name: extract_vector_elt_var_v8i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<8 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; CHECK: [[EVEC:%[0-9]+]]:_(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<8 x s32>), [[COPY1]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<8 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7
+    %1:_(s32) = COPY $vgpr2
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-insert-vector-elt.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-insert-vector-elt.mir
new file mode 100644 (file)
index 0000000..cc8021e
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-insert-vector-elt.mir
@@ -0,0 +1,21 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel %s -o - | FileCheck %s
+
+---
+name: insert_vector_elt_0_v2i32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2
+    ; CHECK-LABEL: name: insert_vector_elt_0_v2i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr2
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+    ; CHECK: [[IVEC:%[0-9]+]]:_(<2 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[COPY1]](s32), [[C]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[IVEC]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(s32) = COPY $vgpr2
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(<2 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $vgpr0_vgpr1 = COPY %3
+...

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-extract-vector-elt.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-extract-vector-elt.mir
new file mode 100644 (file)
index 0000000..25e0992
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-extract-vector-elt.mir
@@ -0,0 +1,39 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect -global-isel %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+
+---
+name: extract_vector_elt_0_v2i32_s
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+    ; CHECK-LABEL: name: extract_vector_elt_0_v2i32_s
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<2 x s32>) = COPY $sgpr0_sgpr1
+    ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:sgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<2 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<2 x s32>) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_CONSTANT i32 0
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...
+
+
+---
+name: extract_vector_elt_0_v4i32_s
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK-LABEL: name: extract_vector_elt_0_v4i32_s
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+    ; CHECK: [[EVEC:%[0-9]+]]:sgpr(s32) = G_EXTRACT_VECTOR_ELT [[COPY]](<4 x s32>), [[C]](s32)
+    ; CHECK: $vgpr0 = COPY [[EVEC]](s32)
+    %0:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %1:_(s32) = G_CONSTANT i32 0
+    %2:_(s32) = G_EXTRACT_VECTOR_ELT %0, %1
+    $vgpr0 = COPY %2
+...

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-insert-vector-elt.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-insert-vector-elt.mir
new file mode 100644 (file)
index 0000000..a829d8f
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-insert-vector-elt.mir
@@ -0,0 +1,152 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect -global-isel %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+
+---
+name: insert_vector_elt_v4i32_s_s_k
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr5
+    ; CHECK-LABEL: name: insert_vector_elt_v4i32_s_s_k
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+    ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+    ; CHECK: [[IVEC:%[0-9]+]]:sgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[COPY1]](s32), [[C]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %1:_(s32) = COPY $sgpr5
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_v4i32_v_s_k
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr5
+    ; CHECK-LABEL: name: insert_vector_elt_v4i32_v_s_k
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+    ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[COPY2]](s32), [[COPY3]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(s32) = COPY $sgpr5
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_v4i32_s_v_k
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $vgpr5
+    ; CHECK-LABEL: name: insert_vector_elt_v4i32_s_v_k
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr2
+    ; CHECK: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(<4 x s32>) = COPY [[COPY]](<4 x s32>)
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[C]](s32)
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY2]], [[COPY1]](s32), [[COPY3]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %1:_(s32) = COPY $vgpr2
+    %2:_(s32) = G_CONSTANT i32 0
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_var_v4i32_s_s_s
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr5, $sgpr6
+    ; CHECK-LABEL: name: insert_vector_elt_var_v4i32_s_s_s
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+    ; CHECK: [[COPY2:%[0-9]+]]:sgpr(s32) = COPY $sgpr6
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s32>) = COPY [[COPY]](<4 x s32>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[COPY5:%[0-9]+]]:vgpr(s32) = COPY [[COPY2]](s32)
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY3]], [[COPY4]](s32), [[COPY5]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %1:_(s32) = COPY $sgpr5
+    %2:_(s32) = COPY $sgpr6
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_var_v4i32_s_s_v
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr5, $vgpr6
+    ; CHECK-LABEL: name: insert_vector_elt_var_v4i32_s_s_v
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(<4 x s32>) = COPY [[COPY]](<4 x s32>)
+    ; CHECK: [[COPY4:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY3]], [[COPY4]](s32), [[COPY2]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
+    %1:_(s32) = COPY $sgpr5
+    %2:_(s32) = COPY $vgpr6
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_var_v4i32_v_s_v
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $sgpr5, $vgpr6
+    ; CHECK-LABEL: name: insert_vector_elt_var_v4i32_v_s_v
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr5
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+    ; CHECK: [[COPY3:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[COPY3]](s32), [[COPY2]](s32)
+    ; CHECK: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(s32) = COPY $sgpr5
+    %2:_(s32) = COPY $vgpr6
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %3
+...
+
+---
+name: insert_vector_elt_var_v4i32_v_v_v
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3, $vgpr5, $vgpr6
+    ; CHECK-LABEL: name: insert_vector_elt_var_v4i32_v_v_v
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr5
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY $vgpr6
+    ; CHECK: [[IVEC:%[0-9]+]]:vgpr(<4 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[COPY1]](s32), [[COPY2]](s32)
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[IVEC]](<4 x s32>)
+    %0:_(<4 x s32>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(s32) = COPY $vgpr5
+    %2:_(s32) = COPY $vgpr6
+    %3:_(<4 x s32>) = G_INSERT_VECTOR_ELT %0, %1, %2
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
+...