AMDGPU/GlobalISel: Implement select for G_INSERT

Re-commit r344310.

Reviewers: arsenm

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, kristof.beyls, dstuttard, tpr, t-tye, llvm-commits

Differential Revision: https://reviews.llvm.org/D53116

llvm-svn: 355159

3 files changed, 80 insertions, 0 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 0619f1fc3db..f35c2a1a259 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -199,6 +199,34 @@ bool AMDGPUInstructionSelector::selectG_IMPLICIT_DEF(MachineInstr &I) const {
   return true;
 }
 
+bool AMDGPUInstructionSelector::selectG_INSERT(MachineInstr &I) const {
+  MachineBasicBlock *BB = I.getParent();
+  MachineFunction *MF = BB->getParent();
+  MachineRegisterInfo &MRI = MF->getRegInfo();
+  unsigned SubReg = TRI.getSubRegFromChannel(I.getOperand(3).getImm() / 32);
+  DebugLoc DL = I.getDebugLoc();
+  MachineInstr *Ins = BuildMI(*BB, &I, DL, TII.get(TargetOpcode::INSERT_SUBREG))
+                               .addDef(I.getOperand(0).getReg())
+                               .addReg(I.getOperand(1).getReg())
+                               .addReg(I.getOperand(2).getReg())
+                               .addImm(SubReg);
+
+  for (const MachineOperand &MO : Ins->operands()) {
+    if (!MO.isReg())
+      continue;
+    if (TargetRegisterInfo::isPhysicalRegister(MO.getReg()))
+      continue;
+
+    const TargetRegisterClass *RC =
+            TRI.getConstrainedRegClassForOperand(MO, MRI);
+    if (!RC)
+      continue;
+    RBI.constrainGenericRegister(MO.getReg(), *RC, MRI);
+  }
+  I.eraseFromParent();
+  return true;
+}
+
 bool AMDGPUInstructionSelector::selectG_INTRINSIC(MachineInstr &I,
                                           CodeGenCoverage &CoverageInfo) const {
   unsigned IntrinsicID =  I.getOperand(1).getIntrinsicID();
@@ -537,6 +565,8 @@ bool AMDGPUInstructionSelector::select(MachineInstr &I,
     return selectG_GEP(I);
   case TargetOpcode::G_IMPLICIT_DEF:
     return selectG_IMPLICIT_DEF(I);
+  case TargetOpcode::G_INSERT:
+    return selectG_INSERT(I);
   case TargetOpcode::G_INTRINSIC:
     return selectG_INTRINSIC(I, CoverageInfo);
   case TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
index dea13b92b2a..b7af40ac916 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
@@ -68,6 +68,7 @@ private:
   bool selectG_EXTRACT(MachineInstr &I) const;
   bool selectG_GEP(MachineInstr &I) const;
   bool selectG_IMPLICIT_DEF(MachineInstr &I) const;
+  bool selectG_INSERT(MachineInstr &I) const;
   bool selectG_INTRINSIC(MachineInstr &I, CodeGenCoverage &CoverageInfo) const;
   bool selectG_INTRINSIC_W_SIDE_EFFECTS(MachineInstr &I,
                                         CodeGenCoverage &CoverageInfo) const;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-insert.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-insert.mir
new file mode 100644
index 00000000000..93e35ead4d4
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-insert.mir
@@ -0,0 +1,49 @@
+# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s
+---
+
+name:            insert512
+legalized:       true
+regBankSelected: true
+
+# CHECK-LABEL: insert512
+# CHECK: [[BASE:%[0-9]+]]:sreg_512 = IMPLICIT_DEF
+# CHECK: [[VAL:%[0-9]+]]:sreg_32_xm0 = IMPLICIT_DEF
+# CHECK: [[BASE0:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE]], [[VAL]], %subreg.sub0
+# CHECK: [[BASE1:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE0]], [[VAL]], %subreg.sub1
+# CHECK: [[BASE2:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE1]], [[VAL]], %subreg.sub2
+# CHECK: [[BASE3:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE2]], [[VAL]], %subreg.sub3
+# CHECK: [[BASE4:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE3]], [[VAL]], %subreg.sub4
+# CHECK: [[BASE5:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE4]], [[VAL]], %subreg.sub5
+# CHECK: [[BASE6:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE5]], [[VAL]], %subreg.sub6
+# CHECK: [[BASE7:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE6]], [[VAL]], %subreg.sub7
+# CHECK: [[BASE8:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE7]], [[VAL]], %subreg.sub8
+# CHECK: [[BASE9:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE8]], [[VAL]], %subreg.sub9
+# CHECK: [[BASE10:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE9]], [[VAL]], %subreg.sub10
+# CHECK: [[BASE11:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE10]], [[VAL]], %subreg.sub11
+# CHECK: [[BASE12:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE11]], [[VAL]], %subreg.sub12
+# CHECK: [[BASE13:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE12]], [[VAL]], %subreg.sub13
+# CHECK: [[BASE14:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE13]], [[VAL]], %subreg.sub14
+# CHECK: [[BASE15:%[0-9]+]]:sreg_512 = INSERT_SUBREG [[BASE14]], [[VAL]], %subreg.sub15
+
+body: |
+  bb.0:
+    %0:sgpr(s512) = G_IMPLICIT_DEF
+    %1:sgpr(s32) = G_IMPLICIT_DEF
+    %2:sgpr(s512) = G_INSERT %0:sgpr, %1:sgpr(s32), 0
+    %3:sgpr(s512) = G_INSERT %2:sgpr, %1:sgpr(s32), 32
+    %4:sgpr(s512) = G_INSERT %3:sgpr, %1:sgpr(s32), 64
+    %5:sgpr(s512) = G_INSERT %4:sgpr, %1:sgpr(s32), 96
+    %6:sgpr(s512) = G_INSERT %5:sgpr, %1:sgpr(s32), 128
+    %7:sgpr(s512) = G_INSERT %6:sgpr, %1:sgpr(s32), 160
+    %8:sgpr(s512) = G_INSERT %7:sgpr, %1:sgpr(s32), 192
+    %9:sgpr(s512) = G_INSERT %8:sgpr, %1:sgpr(s32), 224
+    %10:sgpr(s512) = G_INSERT %9:sgpr, %1:sgpr(s32), 256
+    %11:sgpr(s512) = G_INSERT %10:sgpr, %1:sgpr(s32), 288
+    %12:sgpr(s512) = G_INSERT %11:sgpr, %1:sgpr(s32), 320
+    %13:sgpr(s512) = G_INSERT %12:sgpr, %1:sgpr(s32), 352
+    %14:sgpr(s512) = G_INSERT %13:sgpr, %1:sgpr(s32), 384
+    %15:sgpr(s512) = G_INSERT %14:sgpr, %1:sgpr(s32), 416
+    %16:sgpr(s512) = G_INSERT %15:sgpr, %1:sgpr(s32), 448
+    %17:sgpr(s512) = G_INSERT %16:sgpr, %1:sgpr(s32), 480
+    $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 = COPY %17:sgpr(s512)
+    SI_RETURN_TO_EPILOG $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15