AMDGPU/GlobalISel: legalize and select 32-bit G_SITOFP

[arsenm]

LGTM

llvm/trunk/lib/Target/AMDGPU/AMDGPUGISel.td

 def gi_vop3mods0 :
     GIComplexOperandMatcher<s32, "selectVOP3Mods0">,
     GIComplexPatternEquiv<VOP3Mods0>;
 
 def gi_vop3mods :
     GIComplexOperandMatcher<s32, "selectVOP3Mods">,
     GIComplexPatternEquiv<VOP3Mods>;
 
+def gi_vop3omods :
+    GIComplexOperandMatcher<s32, "selectVOP3OMods">,
+    GIComplexPatternEquiv<VOP3OMods>;
+
 class GISelSop2Pat <
   SDPatternOperator node,
   Instruction inst,
   ValueType dst_vt,
   ValueType src0_vt = dst_vt, ValueType src1_vt = src0_vt>   : GCNPat <
 
   (dst_vt (node (src0_vt SReg_32:$src0), (src1_vt SReg_32:$src1))),
   (inst src0_vt:$src0, src1_vt:$src1)

llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.h

private:
   bool selectG_STORE(MachineInstr &I) const;
 
   InstructionSelector::ComplexRendererFns
   selectVSRC0(MachineOperand &Root) const;
 
   InstructionSelector::ComplexRendererFns
   selectVOP3Mods0(MachineOperand &Root) const;
   InstructionSelector::ComplexRendererFns
+  selectVOP3OMods(MachineOperand &Root) const;
+  InstructionSelector::ComplexRendererFns
   selectVOP3Mods(MachineOperand &Root) const;
 
   const SIInstrInfo &TII;
   const SIRegisterInfo &TRI;
   const AMDGPURegisterBankInfo &RBI;
   const AMDGPUTargetMachine &TM;
   const SISubtarget &STI;
   bool EnableLateStructurizeCFG;

llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

if (!isPreISelGenericOpcode(I.getOpcode())) {
     if (I.isCopy())
       return selectCOPY(I);
     return true;
   }
 
   switch (I.getOpcode()) {
   default:
     break;
+  case TargetOpcode::G_SITOFP:
   case TargetOpcode::G_FMUL:
   case TargetOpcode::G_FADD:
   case TargetOpcode::G_FPTOUI:
   case TargetOpcode::G_OR:
     return selectImpl(I, CoverageInfo);
   case TargetOpcode::G_ADD:
     return selectG_ADD(I);
   case TargetOpcode::G_BITCAST:

 AMDGPUInstructionSelector::selectVOP3Mods0(MachineOperand &Root) const {
   return {{
       [=](MachineInstrBuilder &MIB) { MIB.add(Root); },
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // src0_mods
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // clamp
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }  // omod
   }};
 }
+InstructionSelector::ComplexRendererFns
+AMDGPUInstructionSelector::selectVOP3OMods(MachineOperand &Root) const {
+  return {{
+      [=](MachineInstrBuilder &MIB) { MIB.add(Root); },
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }, // clamp
+      [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }  // omod
+  }};
+}
 
 InstructionSelector::ComplexRendererFns
 AMDGPUInstructionSelector::selectVOP3Mods(MachineOperand &Root) const {
   return {{
       [=](MachineInstrBuilder &MIB) { MIB.add(Root); },
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }  // src_mods
   }};
 }

llvm/trunk/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

AMDGPULegalizerInfo::AMDGPULegalizerInfo(const SISubtarget &ST,
   setAction({G_FMUL, S32}, Legal);
 
   setAction({G_ZEXT, S64}, Legal);
   setAction({G_ZEXT, 1, S32}, Legal);
 
   setAction({G_FPTOSI, S32}, Legal);
   setAction({G_FPTOSI, 1, S32}, Legal);
 
+  setAction({G_SITOFP, S32}, Legal);
+  setAction({G_SITOFP, 1, S32}, Legal);
+
   setAction({G_FPTOUI, S32}, Legal);
   setAction({G_FPTOUI, 1, S32}, Legal);
 
   for (LLT PtrTy : AddrSpaces) {
     LLT IdxTy = LLT::scalar(PtrTy.getSizeInBits());
     setAction({G_GEP, PtrTy}, Legal);
     setAction({G_GEP, 1, IdxTy}, Legal);
   }

llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/inst-select-sitofp.mir

+# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -global-isel %s -o - | FileCheck %s -check-prefixes=GCN
+
+--- |
+  define amdgpu_kernel void @sitofp(i32 addrspace(1)* %global0) {ret void}
+...
+---
+
+name:            sitofp
+legalized:       true
+regBankSelected: true
+
+# GCN-LABEL: name: sitofp
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0, $vgpr3_vgpr4
+
+    ; GCN: [[SGPR:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    %0:sgpr(s32) = COPY $sgpr0
+
+    ; GCN: [[VGPR:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    %1:vgpr(s32) = COPY $vgpr0
+
+    %2:vgpr(s64) = COPY $vgpr3_vgpr4
+
+    ; sitofp s
+    ; GCN: V_CVT_F32_I32_e64 [[SGPR]], 0, 0
+    %3:vgpr(s32) = G_SITOFP %0
+
+    ; sitofp v
+    ; GCN: V_CVT_F32_I32_e64 [[VGPR]], 0, 0
+    %4:vgpr(s32) = G_SITOFP %1
+
+    G_STORE %3, %2 :: (store 4 into %ir.global0)
+    G_STORE %4, %2 :: (store 4 into %ir.global0)
+...
+---

llvm/trunk/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir

+# 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: test_sitofp_f32_to_i32
+body: |
+  bb.0:
+    liveins: $vgpr0
+
+    ; CHECK-LABEL: name: test_sitofp_f32_to_i32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = G_SITOFP %0
+...