AMDGPU/GlobalISel: Add support for init.exec intrinsics

TThe existing wave32 behavior seems broken and incomplete, but this
reproduces it.

llvm-svn: 373296

6 files changed, 42 insertions, 20 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUGISel.td b/llvm/lib/Target/AMDGPU/AMDGPUGISel.td
index 944f782db24..6ee11686f48 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUGISel.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPUGISel.td
@@ -202,3 +202,6 @@ def : GISelVop2Pat <or, V_OR_B32_e32, i32>;
 foreach Ty = [i64, p0, p1, p4] in {
   defm : SMRD_Pattern <"S_LOAD_DWORDX2",  Ty>;
 }
+
+def gi_as_i32timm : GICustomOperandRenderer<"renderTruncImm32">,
+  GISDNodeXFormEquiv<as_i32timm>;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index d808cc0ea33..4b186c2c8ab 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -2094,3 +2094,12 @@ AMDGPUInstructionSelector::selectDS1Addr1Offset(MachineOperand &Root) const {
       [=](MachineInstrBuilder &MIB) { MIB.addImm(0); }
     }};
 }
+
+void AMDGPUInstructionSelector::renderTruncImm32(MachineInstrBuilder &MIB,
+                                                 const MachineInstr &MI) const {
+  const MachineRegisterInfo &MRI = MI.getParent()->getParent()->getRegInfo();
+  assert(MI.getOpcode() == TargetOpcode::G_CONSTANT && "Expected G_CONSTANT");
+  Optional<int64_t> CstVal = getConstantVRegVal(MI.getOperand(0).getReg(), MRI);
+  assert(CstVal && "Expected constant value");
+  MIB.addImm(CstVal.getValue());
+}
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
index 5bff93c8a00..4aadcd9dc20 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.h
@@ -163,6 +163,9 @@ private:
   InstructionSelector::ComplexRendererFns
   selectDS1Addr1Offset(MachineOperand &Root) const;
 
+  void renderTruncImm32(MachineInstrBuilder &MIB,
+                        const MachineInstr &MI) const;
+
   const SIInstrInfo &TII;
   const SIRegisterInfo &TRI;
   const AMDGPURegisterBankInfo &RBI;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index da690c37c56..49a4c7b26b7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -2605,7 +2605,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
       OpdsMapping[2] = AMDGPU::getValueMapping(Bank, 32);
       break;
     }
-    case Intrinsic::amdgcn_end_cf: {
+    case Intrinsic::amdgcn_end_cf:
+    case Intrinsic::amdgcn_init_exec: {
       unsigned Size = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI);
       OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
       break;
@@ -2658,6 +2659,12 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
       OpdsMapping[5] = getSGPROpMapping(MI.getOperand(5).getReg(), MRI, *TRI);
       break;
     }
+    case Intrinsic::amdgcn_init_exec_from_input: {
+      unsigned Size = getSizeInBits(MI.getOperand(1).getReg(), MRI, *TRI);
+      OpdsMapping[1] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
+      OpdsMapping[2] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
+      break;
+    }
     default:
       if (const AMDGPU::RsrcIntrinsic *RSrcIntrin =
               AMDGPU::lookupRsrcIntrinsic(IntrID)) {
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index 95bba738f1a..e1b32c4964c 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -689,6 +689,10 @@ def as_i32imm: SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i32);
 }]>;
 
+def as_i32timm: SDNodeXForm<timm, [{
+  return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i32);
+}]>;
+
 def as_i64imm: SDNodeXForm<imm, [{
   return CurDAG->getTargetConstant(N->getSExtValue(), SDLoc(N), MVT::i64);
 }]>;
@@ -728,6 +732,10 @@ def UIMM16bit : ImmLeaf <i32,
   [{return isUInt<16>(Imm);}]
 >;
 
+def i64imm_32bit : ImmLeaf<i64, [{
+  return (Imm & 0xffffffffULL) == static_cast<uint64_t>(Imm);
+}]>;
+
 class InlineImm <ValueType vt> : PatLeaf <(vt imm), [{
   return isInlineImmediate(N);
 }]>;
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 3c0cc0051c6..d56af297752 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -349,7 +349,8 @@ def SI_INIT_M0 : SPseudoInstSI <(outs), (ins SSrc_b32:$src)> {
 }
 
 def SI_INIT_EXEC : SPseudoInstSI <
-  (outs), (ins i64imm:$src), []> {
+  (outs), (ins i64imm:$src),
+  [(int_amdgcn_init_exec (i64 timm:$src))]> {
   let Defs = [EXEC];
   let usesCustomInserter = 1;
   let isAsCheapAsAMove = 1;
@@ -365,12 +366,20 @@ def SI_INIT_EXEC_LO : SPseudoInstSI <
   let WaveSizePredicate = isWave32;
 }
 
+// FIXME: Wave32 version
 def SI_INIT_EXEC_FROM_INPUT : SPseudoInstSI <
-  (outs), (ins SSrc_b32:$input, i32imm:$shift), []> {
+  (outs), (ins SSrc_b32:$input, i32imm:$shift),
+  [(int_amdgcn_init_exec_from_input i32:$input, (i32 timm:$shift))]> {
   let Defs = [EXEC];
   let usesCustomInserter = 1;
 }
 
+def : GCNPat <
+  (int_amdgcn_init_exec timm:$src),
+  (SI_INIT_EXEC_LO (as_i32imm imm:$src))> {
+  let WaveSizePredicate = isWave32;
+}
+
 // Return for returning shaders to a shader variant epilog.
 def SI_RETURN_TO_EPILOG : SPseudoInstSI <
   (outs), (ins variable_ops), [(AMDGPUreturn_to_epilog)]> {
@@ -609,23 +618,6 @@ def : GCNPat <
   (SI_PC_ADD_REL_OFFSET $ptr_lo, (i32 0))
 >;
 
-def : GCNPat <
-  (int_amdgcn_init_exec i64:$src),
-  (SI_INIT_EXEC (as_i64imm $src))> {
-  let WaveSizePredicate = isWave64;
-}
-
-def : GCNPat <
-  (int_amdgcn_init_exec i64:$src),
-  (SI_INIT_EXEC_LO (as_i32imm $src))> {
-  let WaveSizePredicate = isWave32;
-}
-
-def : GCNPat <
-  (int_amdgcn_init_exec_from_input i32:$input, i32:$shift),
-  (SI_INIT_EXEC_FROM_INPUT (i32 $input), (as_i32imm $shift))
->;
-
 def : GCNPat<
   (AMDGPUtrap timm:$trapid),
   (S_TRAP $trapid)