R600/SI: Refactor the VOP3_32 tablegen class

This will allow us to use a single MachineInstr to represent
instructions which behave the same but have different encodings
on some subtargets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209028 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 112 insertions, 34 deletions

diff --git a/lib/Target/R600/AMDGPUInstrInfo.cpp b/lib/Target/R600/AMDGPUInstrInfo.cpp
index 4f3ab5f544..1c3361a265 100644
--- a/lib/Target/R600/AMDGPUInstrInfo.cpp
+++ b/lib/Target/R600/AMDGPUInstrInfo.cpp
@@ -355,3 +355,14 @@ int AMDGPUInstrInfo::getMaskedMIMGOp(uint16_t Opcode, unsigned Channels) const {
   case 3: return AMDGPU::getMaskedMIMGOp(Opcode, AMDGPU::Channels_3);
   }
 }
+
+// Wrapper for Tablegen'd function.  enum Subtarget is not defined in any
+// header files, so we need to wrap it in a function that takes unsigned 
+// instead.
+namespace llvm {
+namespace AMDGPU {
+int getMCOpcode(uint16_t Opcode, unsigned Gen) {
+  return getMCOpcode(Opcode);
+}
+}
+}
diff --git a/lib/Target/R600/AMDGPUMCInstLower.cpp b/lib/Target/R600/AMDGPUMCInstLower.cpp
index d65b00f018..66d1074321 100644
--- a/lib/Target/R600/AMDGPUMCInstLower.cpp
+++ b/lib/Target/R600/AMDGPUMCInstLower.cpp
@@ -17,6 +17,7 @@
 #include "AMDGPUAsmPrinter.h"
 #include "InstPrinter/AMDGPUInstPrinter.h"
 #include "R600InstrInfo.h"
+#include "SIInstrInfo.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/IR/Constants.h"
@@ -31,12 +32,30 @@
 
 using namespace llvm;
 
-AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx):
-  Ctx(ctx)
+AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &st):
+  Ctx(ctx), ST(st)
 { }
 
+enum AMDGPUMCInstLower::SISubtarget
+AMDGPUMCInstLower::AMDGPUSubtargetToSISubtarget(unsigned Gen) const {
+  switch (Gen) {
+  default: return AMDGPUMCInstLower::SI;
+  }
+}
+
+unsigned AMDGPUMCInstLower::getMCOpcode(unsigned MIOpcode) const {
+
+  int MCOpcode = AMDGPU::getMCOpcode(MIOpcode,
+                              AMDGPUSubtargetToSISubtarget(ST.getGeneration()));
+  if (MCOpcode == -1)
+    MCOpcode = MIOpcode;
+
+  return MCOpcode;
+}
+
 void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
-  OutMI.setOpcode(MI->getOpcode());
+
+  OutMI.setOpcode(getMCOpcode(MI->getOpcode()));
 
   for (const MachineOperand &MO : MI->explicit_operands()) {
     MCOperand MCOp;
@@ -65,7 +84,8 @@ void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
 }
 
 void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
-  AMDGPUMCInstLower MCInstLowering(OutContext);
+  AMDGPUMCInstLower MCInstLowering(OutContext,
+                               MF->getTarget().getSubtarget<AMDGPUSubtarget>());
 
 #ifdef _DEBUG
   StringRef Err;
diff --git a/lib/Target/R600/AMDGPUMCInstLower.h b/lib/Target/R600/AMDGPUMCInstLower.h
index d7d538e925..2b7f1e3074 100644
--- a/lib/Target/R600/AMDGPUMCInstLower.h
+++ b/lib/Target/R600/AMDGPUMCInstLower.h
@@ -13,16 +13,30 @@
 
 namespace llvm {
 
+class AMDGPUSubtarget;
 class MCInst;
 class MCContext;
 class MachineInstr;
 
 class AMDGPUMCInstLower {
 
+  // This must be kept in sync with the SISubtarget class in SIInstrInfo.td
+  enum SISubtarget {
+    SI = 0
+  };
+
   MCContext &Ctx;
+  const AMDGPUSubtarget &ST;
+
+  /// Convert a member of the AMDGPUSubtarget::Generation enum to the
+  /// SISubtarget enum.
+  enum SISubtarget AMDGPUSubtargetToSISubtarget(unsigned Gen) const;
+
+  /// Get the MC opcode for this MachineInstr.
+  unsigned getMCOpcode(unsigned MIOpcode) const;
 
 public:
-  AMDGPUMCInstLower(MCContext &ctx);
+  AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &ST);
 
   /// \brief Lower a MachineInstr to an MCInst
   void lower(const MachineInstr *MI, MCInst &OutMI) const;
diff --git a/lib/Target/R600/SIInstrInfo.h b/lib/Target/R600/SIInstrInfo.h
index d799274221..7b31a81f0a 100644
--- a/lib/Target/R600/SIInstrInfo.h
+++ b/lib/Target/R600/SIInstrInfo.h
@@ -176,6 +176,7 @@ namespace AMDGPU {
   int getVOPe64(uint16_t Opcode);
   int getCommuteRev(uint16_t Opcode);
   int getCommuteOrig(uint16_t Opcode);
+  int getMCOpcode(uint16_t Opcode, unsigned Gen);
 
   const uint64_t RSRC_DATA_FORMAT = 0xf00000000000LL;
 
diff --git a/lib/Target/R600/SIInstrInfo.td b/lib/Target/R600/SIInstrInfo.td
index 5f18d98195..2242e6d864 100644
--- a/lib/Target/R600/SIInstrInfo.td
+++ b/lib/Target/R600/SIInstrInfo.td
@@ -7,6 +7,13 @@
 //
 //===----------------------------------------------------------------------===//
 
+// Execpt for the NONE field, this must be kept in sync with the SISubtarget enum
+// in AMDGPUMCInstLower.h
+def SISubtarget {
+  int NONE = -1;
+  int SI = 0;
+}
+
 //===----------------------------------------------------------------------===//
 // SI DAG Nodes
 //===----------------------------------------------------------------------===//
@@ -245,6 +252,23 @@ class VOP2_REV <string revOp, bit isOrig> {
   bit IsOrig = isOrig;
 }
 
+class SIMCInstr <string pseudo, int subtarget> {
+  string PseudoInstr = pseudo;
+  int Subtarget = subtarget;
+}
+
+multiclass VOP3_m <bits<9> op, dag outs, dag ins, string asm, list<dag> pattern,
+                   string opName> {
+
+  def "" : InstSI <outs, ins, "", pattern>, VOP <opName>,
+           SIMCInstr<OpName, SISubtarget.NONE> {
+    let isPseudo = 1;
+  }
+
+  def _si : VOP3 <op, outs, ins, asm, []>, SIMCInstr<opName, SISubtarget.SI>;
+
+}
+
 // This must always be right before the operand being input modified.
 def InputMods : OperandWithDefaultOps <i32, (ops (i32 0))> {
   let PrintMethod = "printOperandAndMods";
@@ -364,13 +388,13 @@ multiclass VOPC_64 <bits<8> op, string opName,
   ValueType vt = untyped, PatLeaf cond = COND_NULL>
   : VOPC_Helper <op, VReg_64, VSrc_64, opName, vt, cond>;
 
-class VOP3_32 <bits<9> op, string opName, list<dag> pattern> : VOP3 <
+multiclass VOP3_32 <bits<9> op, string opName, list<dag> pattern> : VOP3_m <
   op, (outs VReg_32:$dst),
   (ins InputMods: $src0_modifiers, VSrc_32:$src0, InputMods:$src1_modifiers,
    VSrc_32:$src1, InputMods:$src2_modifiers, VSrc_32:$src2,
    InstFlag:$clamp, InstFlag:$omod),
-  opName#" $dst, $src0_modifiers, $src1, $src2, $clamp, $omod", pattern
->, VOP <opName>;
+  opName#" $dst, $src0_modifiers, $src1, $src2, $clamp, $omod", pattern, opName
+>;
 
 class VOP3_64_Shift <bits <9> op, string opName, list<dag> pattern> : VOP3 <
   op, (outs VReg_64:$dst),
@@ -681,4 +705,12 @@ def isDS : InstrMapping {
   let ValueCols = [["8"]];
 }
 
+def getMCOpcode : InstrMapping {
+  let FilterClass = "SIMCInstr";
+  let RowFields = ["PseudoInstr"];
+  let ColFields = ["Subtarget"];
+  let KeyCol = [!cast<string>(SISubtarget.NONE)];
+  let ValueCols = [[!cast<string>(SISubtarget.SI)]];
+}
+
 include "SIInstructions.td"
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index 9c6c35286e..b216762564 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1246,41 +1246,41 @@ defm V_CVT_PKRTZ_F16_F32 : VOP2_32 <0x0000002f, "V_CVT_PKRTZ_F16_F32",
 
 let neverHasSideEffects = 1 in {
 
-def V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
-def V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
-def V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
+defm V_MAD_LEGACY_F32 : VOP3_32 <0x00000140, "V_MAD_LEGACY_F32", []>;
+defm V_MAD_F32 : VOP3_32 <0x00000141, "V_MAD_F32", []>;
+defm V_MAD_I32_I24 : VOP3_32 <0x00000142, "V_MAD_I32_I24",
   [(set i32:$dst, (add (AMDGPUmul_i24 i32:$src0, i32:$src1), i32:$src2))]
 >;
-def V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
+defm V_MAD_U32_U24 : VOP3_32 <0x00000143, "V_MAD_U32_U24",
   [(set i32:$dst, (add (AMDGPUmul_u24 i32:$src0, i32:$src1), i32:$src2))]
 >;
 
 } // End neverHasSideEffects
-def V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
-def V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
-def V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;
-def V_CUBEMA_F32 : VOP3_32 <0x00000147, "V_CUBEMA_F32", []>;
+defm V_CUBEID_F32 : VOP3_32 <0x00000144, "V_CUBEID_F32", []>;
+defm V_CUBESC_F32 : VOP3_32 <0x00000145, "V_CUBESC_F32", []>;
+defm V_CUBETC_F32 : VOP3_32 <0x00000146, "V_CUBETC_F32", []>;
+defm V_CUBEMA_F32 : VOP3_32 <0x00000147, "V_CUBEMA_F32", []>;
 
 let neverHasSideEffects = 1, mayLoad = 0, mayStore = 0 in {
-def V_BFE_U32 : VOP3_32 <0x00000148, "V_BFE_U32",
+defm V_BFE_U32 : VOP3_32 <0x00000148, "V_BFE_U32",
   [(set i32:$dst, (AMDGPUbfe_u32 i32:$src0, i32:$src1, i32:$src2))]>;
-def V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32",
+defm V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32",
   [(set i32:$dst, (AMDGPUbfe_i32 i32:$src0, i32:$src1, i32:$src2))]>;
 }
 
-def V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32",
+defm V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32",
   [(set i32:$dst, (AMDGPUbfi i32:$src0, i32:$src1, i32:$src2))]>;
-def V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32",
+defm V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32",
   [(set f32:$dst, (fma f32:$src0, f32:$src1, f32:$src2))]
 >;
 def V_FMA_F64 : VOP3_64 <0x0000014c, "V_FMA_F64",
   [(set f64:$dst, (fma f64:$src0, f64:$src1, f64:$src2))]
 >;
 //def V_LERP_U8 : VOP3_U8 <0x0000014d, "V_LERP_U8", []>;
-def V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
+defm V_ALIGNBIT_B32 : VOP3_32 <0x0000014e, "V_ALIGNBIT_B32", []>;
 
-def V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
-def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
+defm V_ALIGNBYTE_B32 : VOP3_32 <0x0000014f, "V_ALIGNBYTE_B32", []>;
+defm V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
 ////def V_MIN3_F32 : VOP3_MIN3 <0x00000151, "V_MIN3_F32", []>;
 ////def V_MIN3_I32 : VOP3_MIN3 <0x00000152, "V_MIN3_I32", []>;
 ////def V_MIN3_U32 : VOP3_MIN3 <0x00000153, "V_MIN3_U32", []>;
@@ -1293,9 +1293,9 @@ def V_MULLIT_F32 : VOP3_32 <0x00000150, "V_MULLIT_F32", []>;
 //def V_SAD_U8 : VOP3_U8 <0x0000015a, "V_SAD_U8", []>;
 //def V_SAD_HI_U8 : VOP3_U8 <0x0000015b, "V_SAD_HI_U8", []>;
 //def V_SAD_U16 : VOP3_U16 <0x0000015c, "V_SAD_U16", []>;
-def V_SAD_U32 : VOP3_32 <0x0000015d, "V_SAD_U32", []>;
+defm V_SAD_U32 : VOP3_32 <0x0000015d, "V_SAD_U32", []>;
 ////def V_CVT_PK_U8_F32 : VOP3_U8 <0x0000015e, "V_CVT_PK_U8_F32", []>;
-def V_DIV_FIXUP_F32 : VOP3_32 <0x0000015f, "V_DIV_FIXUP_F32", []>;
+defm V_DIV_FIXUP_F32 : VOP3_32 <0x0000015f, "V_DIV_FIXUP_F32", []>;
 def V_DIV_FIXUP_F64 : VOP3_64 <0x00000160, "V_DIV_FIXUP_F64", []>;
 
 def V_LSHL_B64 : VOP3_64_Shift <0x00000161, "V_LSHL_B64",
@@ -1321,16 +1321,16 @@ def V_LDEXP_F64 : VOP3_64 <0x00000168, "V_LDEXP_F64", []>;
 
 let isCommutable = 1 in {
 
-def V_MUL_LO_U32 : VOP3_32 <0x00000169, "V_MUL_LO_U32", []>;
-def V_MUL_HI_U32 : VOP3_32 <0x0000016a, "V_MUL_HI_U32", []>;
-def V_MUL_LO_I32 : VOP3_32 <0x0000016b, "V_MUL_LO_I32", []>;
-def V_MUL_HI_I32 : VOP3_32 <0x0000016c, "V_MUL_HI_I32", []>;
+defm V_MUL_LO_U32 : VOP3_32 <0x00000169, "V_MUL_LO_U32", []>;
+defm V_MUL_HI_U32 : VOP3_32 <0x0000016a, "V_MUL_HI_U32", []>;
+defm V_MUL_LO_I32 : VOP3_32 <0x0000016b, "V_MUL_LO_I32", []>;
+defm V_MUL_HI_I32 : VOP3_32 <0x0000016c, "V_MUL_HI_I32", []>;
 
 } // isCommutable = 1
 
-def V_DIV_SCALE_F32 : VOP3_32 <0x0000016d, "V_DIV_SCALE_F32", []>;
+defm V_DIV_SCALE_F32 : VOP3_32 <0x0000016d, "V_DIV_SCALE_F32", []>;
 def V_DIV_SCALE_F64 : VOP3_64 <0x0000016e, "V_DIV_SCALE_F64", []>;
-def V_DIV_FMAS_F32 : VOP3_32 <0x0000016f, "V_DIV_FMAS_F32", []>;
+defm V_DIV_FMAS_F32 : VOP3_32 <0x0000016f, "V_DIV_FMAS_F32", []>;
 def V_DIV_FMAS_F64 : VOP3_64 <0x00000170, "V_DIV_FMAS_F64", []>;
 //def V_MSAD_U8 : VOP3_U8 <0x00000171, "V_MSAD_U8", []>;
 //def V_QSAD_U8 : VOP3_U8 <0x00000172, "V_QSAD_U8", []>;
@@ -2312,9 +2312,9 @@ defm V_RNDNE_F64 : VOP1_64 <0x00000019, "V_RNDNE_F64",
   [(set f64:$dst, (frint f64:$src0))]
 >;
 
-def V_QSAD_PK_U16_U8 : VOP3_32 <0x00000173, "V_QSAD_PK_U16_U8", []>;
-def V_MQSAD_U16_U8 : VOP3_32 <0x000000172, "V_MQSAD_U16_U8", []>;
-def V_MQSAD_U32_U8 : VOP3_32 <0x00000175, "V_MQSAD_U32_U8", []>;
+defm V_QSAD_PK_U16_U8 : VOP3_32 <0x00000173, "V_QSAD_PK_U16_U8", []>;
+defm V_MQSAD_U16_U8 : VOP3_32 <0x000000172, "V_MQSAD_U16_U8", []>;
+defm V_MQSAD_U32_U8 : VOP3_32 <0x00000175, "V_MQSAD_U32_U8", []>;
 def V_MAD_U64_U32 : VOP3_64 <0x00000176, "V_MAD_U64_U32", []>;
 
 // XXX - Does this set VCC?