SSE cmov support.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@25190 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Evan Cheng 2006-01-11 00:33:36 +00:00
parent f710062bfb
commit 4a46080fe0
3 changed files with 121 additions and 17 deletions

View File

@ -17,8 +17,9 @@
#include "X86TargetMachine.h"
#include "llvm/CallingConv.h"
#include "llvm/Function.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SSARegMap.h"
#include "llvm/Target/TargetOptions.h"
@ -1140,14 +1141,34 @@ LowerFrameReturnAddress(bool isFrameAddress, SDOperand Chain, unsigned Depth,
return std::make_pair(Result, Chain);
}
//===----------------------------------------------------------------------===//
// X86 Custom Lowering Hooks
//===----------------------------------------------------------------------===//
/// getCondBrOpcodeForX86CC - Returns the X86 conditional branch opcode
/// which corresponds to the condition code.
static unsigned getCondBrOpcodeForX86CC(unsigned X86CC) {
switch (X86CC) {
default: assert(0 && "Unknown X86 conditional code!");
case X86ISD::COND_A: return X86::JA;
case X86ISD::COND_AE: return X86::JAE;
case X86ISD::COND_B: return X86::JB;
case X86ISD::COND_BE: return X86::JBE;
case X86ISD::COND_E: return X86::JE;
case X86ISD::COND_G: return X86::JG;
case X86ISD::COND_GE: return X86::JGE;
case X86ISD::COND_L: return X86::JL;
case X86ISD::COND_LE: return X86::JLE;
case X86ISD::COND_NE: return X86::JNE;
case X86ISD::COND_NO: return X86::JNO;
case X86ISD::COND_NP: return X86::JNP;
case X86ISD::COND_NS: return X86::JNS;
case X86ISD::COND_O: return X86::JO;
case X86ISD::COND_P: return X86::JP;
case X86ISD::COND_S: return X86::JS;
}
}
/// SetCCToX86CondCode - do a one to one translation of a ISD::CondCode to
/// X86 specific CondCode. It returns a X86ISD::COND_INVALID if it cannot
/// getX86CC - do a one to one translation of a ISD::CondCode to the X86
/// specific condition code. It returns a X86ISD::COND_INVALID if it cannot
/// do a direct translation.
static unsigned CCToX86CondCode(SDOperand CC, bool isFP) {
static unsigned getX86CC(SDOperand CC, bool isFP) {
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
unsigned X86CC = X86ISD::COND_INVALID;
if (!isFP) {
@ -1192,11 +1213,10 @@ static unsigned CCToX86CondCode(SDOperand CC, bool isFP) {
return X86CC;
}
/// SupportedByFPCMOV - is there a floating point cmov for the specific
/// X86 condition code.
/// Current x86 isa includes the following FP cmov instructions:
/// hasFPCMov - is there a floating point cmov for the specific X86 condition
/// code. Current x86 isa includes the following FP cmov instructions:
/// fcmovb, fcomvbe, fcomve, fcmovu, fcmovae, fcmova, fcmovne, fcmovnu.
static bool SupportedByFPCMOV(unsigned X86CC) {
static bool hasFPCMov(unsigned X86CC) {
switch (X86CC) {
default:
return false;
@ -1212,6 +1232,64 @@ static bool SupportedByFPCMOV(unsigned X86CC) {
}
}
MachineBasicBlock *
X86TargetLowering::InsertAtEndOfBasicBlock(MachineInstr *MI,
MachineBasicBlock *BB) {
assert((MI->getOpcode() == X86::CMOV_FR32 ||
MI->getOpcode() == X86::CMOV_FR64) &&
"Unexpected instr type to insert");
// To "insert" a SELECT_CC instruction, we actually have to insert the diamond
// control-flow pattern. The incoming instruction knows the destination vreg
// to set, the condition code register to branch on, the true/false values to
// select between, and a branch opcode to use.
const BasicBlock *LLVM_BB = BB->getBasicBlock();
ilist<MachineBasicBlock>::iterator It = BB;
++It;
// thisMBB:
// ...
// TrueVal = ...
// cmpTY ccX, r1, r2
// bCC copy1MBB
// fallthrough --> copy0MBB
MachineBasicBlock *thisMBB = BB;
MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB);
unsigned Opc = getCondBrOpcodeForX86CC(MI->getOperand(3).getImmedValue());
BuildMI(BB, Opc, 1).addMBB(sinkMBB);
MachineFunction *F = BB->getParent();
F->getBasicBlockList().insert(It, copy0MBB);
F->getBasicBlockList().insert(It, sinkMBB);
// Update machine-CFG edges
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
// copy0MBB:
// %FalseValue = ...
// # fallthrough to sinkMBB
BB = copy0MBB;
// Update machine-CFG edges
BB->addSuccessor(sinkMBB);
// sinkMBB:
// %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
// ...
BB = sinkMBB;
BuildMI(BB, X86::PHI, 4, MI->getOperand(0).getReg())
.addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
.addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
delete MI; // The pseudo instruction is gone now.
return BB;
}
//===----------------------------------------------------------------------===//
// X86 Custom Lowering Hooks
//===----------------------------------------------------------------------===//
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDOperand X86TargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
@ -1383,7 +1461,7 @@ SDOperand X86TargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
Op.getOperand(0), Op.getOperand(1));
ISD::CondCode SetCCOpcode = cast<CondCodeSDNode>(CC)->get();
bool isFP = MVT::isFloatingPoint(Op.getOperand(1).getValueType());
unsigned X86CC = CCToX86CondCode(CC, isFP);
unsigned X86CC = getX86CC(CC, isFP);
if (X86CC != X86ISD::COND_INVALID) {
return DAG.getNode(X86ISD::SETCC, MVT::i8,
DAG.getConstant(X86CC, MVT::i8), Cond);
@ -1475,12 +1553,11 @@ SDOperand X86TargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
CC = Op0.getOperand(0);
Cond = Op0.getOperand(1);
isValid =
!(isFPStack &&
!SupportedByFPCMOV(cast<ConstantSDNode>(CC)->getSignExtended()));
!(isFPStack && !hasFPCMov(cast<ConstantSDNode>(CC)->getSignExtended()));
} else if (Op0.getOpcode() == ISD::SETCC) {
CC = Op0.getOperand(2);
bool isFP = MVT::isFloatingPoint(Op0.getOperand(1).getValueType());
unsigned X86CC = CCToX86CondCode(CC, isFP);
unsigned X86CC = getX86CC(CC, isFP);
CC = DAG.getConstant(X86CC, MVT::i8);
Cond = DAG.getNode(X86ISD::CMP, MVT::Flag,
Op0.getOperand(0), Op0.getOperand(1));
@ -1513,7 +1590,7 @@ SDOperand X86TargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
} else if (Cond.getOpcode() == ISD::SETCC) {
CC = Cond.getOperand(2);
bool isFP = MVT::isFloatingPoint(Cond.getOperand(1).getValueType());
unsigned X86CC = CCToX86CondCode(CC, isFP);
unsigned X86CC = getX86CC(CC, isFP);
CC = DAG.getConstant(X86CC, MVT::i8);
Cond = DAG.getNode(X86ISD::CMP, MVT::Flag,
Cond.getOperand(0), Cond.getOperand(1));

View File

@ -199,6 +199,9 @@ namespace llvm {
LowerFrameReturnAddress(bool isFrameAddr, SDOperand Chain, unsigned Depth,
SelectionDAG &DAG);
virtual MachineBasicBlock *InsertAtEndOfBasicBlock(MachineInstr *MI,
MachineBasicBlock *MBB);
/// getTargetNodeName - This method returns the name of a target specific
/// DAG node.
virtual const char *getTargetNodeName(unsigned Opcode) const;

View File

@ -361,8 +361,8 @@ class Ii32<bits<8> o, Format f, dag ops, string asm, list<dag> pattern>
// Instruction list...
//
// Pseudo-instructions:
def PHI : I<0, Pseudo, (ops variable_ops), "PHINODE", []>; // PHI node.
def NOOP : I<0x90, RawFrm, (ops), "nop", []>; // nop
def ADJCALLSTACKDOWN : I<0, Pseudo, (ops i32imm:$amt), "#ADJCALLSTACKDOWN",
[(X86callseq_start imm:$amt)]>;
@ -388,10 +388,29 @@ def IMPLICIT_DEF_FR64 : I<0, Pseudo, (ops FR64:$dst),
[(set FR64:$dst, (undef))]>, Requires<[HasSSE2]>;
// CMOV* - Used to implement the SSE SELECT DAG operation. Expanded by the
// scheduler into a branch sequence.
let usesCustomDAGSchedInserter = 1 in { // Expanded by the scheduler.
def CMOV_FR32 : I<0, Pseudo,
(ops FR32:$dst, FR32:$t, FR32:$f, i8imm:$cond),
"#CMOV PSEUDO!",
[(set FR32:$dst, (X86cmov FR32:$t, FR32:$f, imm:$cond,
STATUS))]>;
def CMOV_FR64 : I<0, Pseudo,
(ops FR64:$dst, FR64:$t, FR64:$f, i8imm:$cond),
"#CMOV PSEUDO!",
[(set FR64:$dst, (X86cmov FR64:$t, FR64:$f, imm:$cond,
STATUS))]>;
}
let isTerminator = 1 in
let Defs = [FP0, FP1, FP2, FP3, FP4, FP5, FP6] in
def FP_REG_KILL : I<0, Pseudo, (ops), "#FP_REG_KILL", []>;
// Nop
def NOOP : I<0x90, RawFrm, (ops), "nop", []>;
//===----------------------------------------------------------------------===//
// Control Flow Instructions...
//
@ -409,6 +428,7 @@ let isBranch = 1, isTerminator = 1, noResults = 1 in
class IBr<bits<8> opcode, dag ops, string asm, list<dag> pattern> :
I<opcode, RawFrm, ops, asm, pattern>;
// Conditional branches
let isBarrier = 1 in
def JMP : IBr<0xE9, (ops brtarget:$dst), "jmp $dst", [(br bb:$dst)]>;
@ -442,6 +462,10 @@ def JP : IBr<0x8A, (ops brtarget:$dst), "jp $dst",
[(X86brcond bb:$dst, X86_COND_P, STATUS)]>, Imp<[STATUS],[]>, TB;
def JNP : IBr<0x8B, (ops brtarget:$dst), "jnp $dst",
[(X86brcond bb:$dst, X86_COND_NP, STATUS)]>, Imp<[STATUS],[]>, TB;
def JO : IBr<0x80, (ops brtarget:$dst), "jo $dst",
[(X86brcond bb:$dst, X86_COND_O, STATUS)]>, Imp<[STATUS],[]>, TB;
def JNO : IBr<0x81, (ops brtarget:$dst), "jno $dst",
[(X86brcond bb:$dst, X86_COND_NO, STATUS)]>, Imp<[STATUS],[]>, TB;
//===----------------------------------------------------------------------===//
// Call Instructions...