2005-10-16 05:39:50 +00:00
|
|
|
//===-- PPCISelLowering.h - PPC32 DAG Lowering Interface --------*- C++ -*-===//
|
2005-08-16 17:14:42 +00:00
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
|
|
|
// This file was developed by Chris Lattner and is distributed under
|
|
|
|
// the University of Illinois Open Source License. See LICENSE.TXT for details.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// This file defines the interfaces that PPC uses to lower LLVM code into a
|
|
|
|
// selection DAG.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#ifndef LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
|
|
|
|
#define LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
|
|
|
|
|
|
|
|
#include "llvm/Target/TargetLowering.h"
|
2005-08-26 20:25:03 +00:00
|
|
|
#include "llvm/CodeGen/SelectionDAG.h"
|
2005-10-14 23:51:18 +00:00
|
|
|
#include "PPC.h"
|
2005-08-16 17:14:42 +00:00
|
|
|
|
|
|
|
namespace llvm {
|
2005-08-26 20:25:03 +00:00
|
|
|
namespace PPCISD {
|
|
|
|
enum NodeType {
|
|
|
|
// Start the numbering where the builting ops and target ops leave off.
|
|
|
|
FIRST_NUMBER = ISD::BUILTIN_OP_END+PPC::INSTRUCTION_LIST_END,
|
|
|
|
|
|
|
|
/// FSEL - Traditional three-operand fsel node.
|
|
|
|
///
|
|
|
|
FSEL,
|
2005-08-31 21:09:52 +00:00
|
|
|
|
2005-09-06 22:03:27 +00:00
|
|
|
/// FCFID - The FCFID instruction, taking an f64 operand and producing
|
|
|
|
/// and f64 value containing the FP representation of the integer that
|
|
|
|
/// was temporarily in the f64 operand.
|
|
|
|
FCFID,
|
|
|
|
|
|
|
|
/// FCTI[D,W]Z - The FCTIDZ and FCTIWZ instructions, taking an f32 or f64
|
|
|
|
/// operand, producing an f64 value containing the integer representation
|
|
|
|
/// of that FP value.
|
|
|
|
FCTIDZ, FCTIWZ,
|
2005-11-17 07:30:41 +00:00
|
|
|
|
2006-03-01 05:50:56 +00:00
|
|
|
/// STFIWX - The STFIWX instruction. The first operand is an input token
|
|
|
|
/// chain, then an f64 value to store, then an address to store it to,
|
|
|
|
/// then a SRCVALUE for the address.
|
|
|
|
STFIWX,
|
|
|
|
|
2005-12-13 22:55:22 +00:00
|
|
|
// VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
|
|
|
|
// three v4f32 operands and producing a v4f32 result.
|
|
|
|
VMADDFP, VNMSUBFP,
|
|
|
|
|
2006-03-20 01:53:53 +00:00
|
|
|
/// VPERM - The PPC VPERM Instruction.
|
|
|
|
///
|
|
|
|
VPERM,
|
|
|
|
|
2005-11-17 07:30:41 +00:00
|
|
|
/// Hi/Lo - These represent the high and low 16-bit parts of a global
|
|
|
|
/// address respectively. These nodes have two operands, the first of
|
|
|
|
/// which must be a TargetGlobalAddress, and the second of which must be a
|
|
|
|
/// Constant. Selected naively, these turn into 'lis G+C' and 'li G+C',
|
|
|
|
/// though these are usually folded into other nodes.
|
|
|
|
Hi, Lo,
|
|
|
|
|
|
|
|
/// GlobalBaseReg - On Darwin, this node represents the result of the mflr
|
|
|
|
/// at function entry, used for PIC code.
|
|
|
|
GlobalBaseReg,
|
2005-12-06 02:10:38 +00:00
|
|
|
|
|
|
|
/// These nodes represent the 32-bit PPC shifts that operate on 6-bit
|
|
|
|
/// shift amounts. These nodes are generated by the multi-precision shift
|
|
|
|
/// code.
|
|
|
|
SRL, SRA, SHL,
|
When possible, custom lower 32-bit SINT_TO_FP to this:
_foo2:
extsw r2, r3
std r2, -8(r1)
lfd f0, -8(r1)
fcfid f0, f0
frsp f1, f0
blr
instead of this:
_foo2:
lis r2, ha16(LCPI2_0)
lis r4, 17200
xoris r3, r3, 32768
stw r3, -4(r1)
stw r4, -8(r1)
lfs f0, lo16(LCPI2_0)(r2)
lfd f1, -8(r1)
fsub f0, f1, f0
frsp f1, f0
blr
This speeds up Misc/pi from 2.44s->2.09s with LLC and from 3.01->2.18s
with llcbeta (16.7% and 38.1% respectively).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26943 91177308-0d34-0410-b5e6-96231b3b80d8
2006-03-22 05:30:33 +00:00
|
|
|
|
|
|
|
/// EXTSW_32 - This is the EXTSW instruction for use with "32-bit"
|
|
|
|
/// registers.
|
|
|
|
EXTSW_32,
|
2005-12-20 00:26:01 +00:00
|
|
|
|
When possible, custom lower 32-bit SINT_TO_FP to this:
_foo2:
extsw r2, r3
std r2, -8(r1)
lfd f0, -8(r1)
fcfid f0, f0
frsp f1, f0
blr
instead of this:
_foo2:
lis r2, ha16(LCPI2_0)
lis r4, 17200
xoris r3, r3, 32768
stw r3, -4(r1)
stw r4, -8(r1)
lfs f0, lo16(LCPI2_0)(r2)
lfd f1, -8(r1)
fsub f0, f1, f0
frsp f1, f0
blr
This speeds up Misc/pi from 2.44s->2.09s with LLC and from 3.01->2.18s
with llcbeta (16.7% and 38.1% respectively).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26943 91177308-0d34-0410-b5e6-96231b3b80d8
2006-03-22 05:30:33 +00:00
|
|
|
/// STD_32 - This is the STD instruction for use with "32-bit" registers.
|
|
|
|
STD_32,
|
|
|
|
|
2006-05-17 19:00:46 +00:00
|
|
|
/// CALL - A direct function call.
|
2006-01-27 23:34:02 +00:00
|
|
|
CALL,
|
|
|
|
|
2006-05-17 19:00:46 +00:00
|
|
|
/// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
|
|
|
|
/// MTCTR instruction.
|
|
|
|
MTCTR,
|
|
|
|
|
|
|
|
/// CHAIN,FLAG = BCTRL(CHAIN, INFLAG) - Directly corresponds to a
|
|
|
|
/// BCTRL instruction.
|
|
|
|
BCTRL,
|
|
|
|
|
2005-12-20 00:26:01 +00:00
|
|
|
/// Return with a flag operand, matched by 'blr'
|
|
|
|
RET_FLAG,
|
2006-03-26 10:06:40 +00:00
|
|
|
|
|
|
|
/// R32 = MFCR(CRREG, INFLAG) - Represents the MFCR/MFOCRF instructions.
|
|
|
|
/// This copies the bits corresponding to the specified CRREG into the
|
|
|
|
/// resultant GPR. Bits corresponding to other CR regs are undefined.
|
|
|
|
MFCR,
|
2006-03-31 05:13:27 +00:00
|
|
|
|
|
|
|
/// RESVEC = VCMP(LHS, RHS, OPC) - Represents one of the altivec VCMP*
|
|
|
|
/// instructions. For lack of better number, we use the opcode number
|
|
|
|
/// encoding for the OPC field to identify the compare. For example, 838
|
|
|
|
/// is VCMPGTSH.
|
|
|
|
VCMP,
|
2006-03-26 10:06:40 +00:00
|
|
|
|
|
|
|
/// RESVEC, OUTFLAG = VCMPo(LHS, RHS, OPC) - Represents one of the
|
|
|
|
/// altivec VCMP*o instructions. For lack of better number, we use the
|
|
|
|
/// opcode number encoding for the OPC field to identify the compare. For
|
|
|
|
/// example, 838 is VCMPGTSH.
|
Implement an important entry from README_ALTIVEC:
If an altivec predicate compare is used immediately by a branch, don't
use a (serializing) MFCR instruction to read the CR6 register, which requires
a compare to get it back to CR's. Instead, just branch on CR6 directly. :)
For example, for:
void foo2(vector float *A, vector float *B) {
if (!vec_any_eq(*A, *B))
*B = (vector float){0,0,0,0};
}
We now generate:
_foo2:
mfspr r2, 256
oris r5, r2, 12288
mtspr 256, r5
lvx v2, 0, r4
lvx v3, 0, r3
vcmpeqfp. v2, v3, v2
bne cr6, LBB1_2 ; UnifiedReturnBlock
LBB1_1: ; cond_true
vxor v2, v2, v2
stvx v2, 0, r4
mtspr 256, r2
blr
LBB1_2: ; UnifiedReturnBlock
mtspr 256, r2
blr
instead of:
_foo2:
mfspr r2, 256
oris r5, r2, 12288
mtspr 256, r5
lvx v2, 0, r4
lvx v3, 0, r3
vcmpeqfp. v2, v3, v2
mfcr r3, 2
rlwinm r3, r3, 27, 31, 31
cmpwi cr0, r3, 0
beq cr0, LBB1_2 ; UnifiedReturnBlock
LBB1_1: ; cond_true
vxor v2, v2, v2
stvx v2, 0, r4
mtspr 256, r2
blr
LBB1_2: ; UnifiedReturnBlock
mtspr 256, r2
blr
This implements CodeGen/PowerPC/vec_br_cmp.ll.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27804 91177308-0d34-0410-b5e6-96231b3b80d8
2006-04-18 17:59:36 +00:00
|
|
|
VCMPo,
|
|
|
|
|
|
|
|
/// CHAIN = COND_BRANCH CHAIN, CRRC, OPC, DESTBB [, INFLAG] - This
|
|
|
|
/// corresponds to the COND_BRANCH pseudo instruction. CRRC is the
|
|
|
|
/// condition register to branch on, OPC is the branch opcode to use (e.g.
|
|
|
|
/// PPC::BLE), DESTBB is the destination block to branch to, and INFLAG is
|
|
|
|
/// an optional input flag argument.
|
2006-07-10 20:56:58 +00:00
|
|
|
COND_BRANCH,
|
|
|
|
|
|
|
|
/// CHAIN = STBRX CHAIN, GPRC, Ptr, SRCVALUE, Type - This is a
|
|
|
|
/// byte-swapping store instruction. It byte-swaps the low "Type" bits of
|
|
|
|
/// the GPRC input, then stores it through Ptr. Type can be either i16 or
|
|
|
|
/// i32.
|
|
|
|
STBRX,
|
|
|
|
|
|
|
|
/// GPRC, CHAIN = LBRX CHAIN, Ptr, SRCVALUE, Type - This is a
|
|
|
|
/// byte-swapping load instruction. It loads "Type" bits, byte swaps it,
|
|
|
|
/// then puts it in the bottom bits of the GPRC. TYPE can be either i16
|
|
|
|
/// or i32.
|
|
|
|
LBRX
|
2006-01-27 23:34:02 +00:00
|
|
|
};
|
2006-03-20 06:15:45 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/// Define some predicates that are used for node matching.
|
|
|
|
namespace PPC {
|
2006-04-06 17:23:16 +00:00
|
|
|
/// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
|
|
|
|
/// VPKUHUM instruction.
|
2006-04-06 22:28:36 +00:00
|
|
|
bool isVPKUHUMShuffleMask(SDNode *N, bool isUnary);
|
2006-04-06 17:23:16 +00:00
|
|
|
|
|
|
|
/// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
|
|
|
|
/// VPKUWUM instruction.
|
2006-04-06 22:28:36 +00:00
|
|
|
bool isVPKUWUMShuffleMask(SDNode *N, bool isUnary);
|
2006-04-06 21:11:54 +00:00
|
|
|
|
|
|
|
/// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
|
|
|
|
/// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
|
2006-04-06 22:02:42 +00:00
|
|
|
bool isVMRGLShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary);
|
2006-04-06 21:11:54 +00:00
|
|
|
|
|
|
|
/// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
|
|
|
|
/// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
|
2006-04-06 22:02:42 +00:00
|
|
|
bool isVMRGHShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary);
|
2006-04-06 17:23:16 +00:00
|
|
|
|
2006-04-06 18:26:28 +00:00
|
|
|
/// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
|
|
|
|
/// amount, otherwise return -1.
|
2006-04-06 22:28:36 +00:00
|
|
|
int isVSLDOIShuffleMask(SDNode *N, bool isUnary);
|
2006-04-06 18:26:28 +00:00
|
|
|
|
2006-03-20 06:15:45 +00:00
|
|
|
/// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
|
|
|
|
/// specifies a splat of a single element that is suitable for input to
|
|
|
|
/// VSPLTB/VSPLTH/VSPLTW.
|
2006-04-04 17:25:31 +00:00
|
|
|
bool isSplatShuffleMask(SDNode *N, unsigned EltSize);
|
2006-03-20 06:15:45 +00:00
|
|
|
|
|
|
|
/// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
|
|
|
|
/// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
|
2006-04-04 17:25:31 +00:00
|
|
|
unsigned getVSPLTImmediate(SDNode *N, unsigned EltSize);
|
2006-03-24 07:48:08 +00:00
|
|
|
|
2006-04-12 17:37:20 +00:00
|
|
|
/// get_VSPLTI_elt - If this is a build_vector of constants which can be
|
2006-04-08 06:46:53 +00:00
|
|
|
/// formed by using a vspltis[bhw] instruction of the specified element
|
|
|
|
/// size, return the constant being splatted. The ByteSize field indicates
|
|
|
|
/// the number of bytes of each element [124] -> [bhw].
|
2006-04-12 17:37:20 +00:00
|
|
|
SDOperand get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG);
|
2006-03-20 06:15:45 +00:00
|
|
|
}
|
2005-08-26 20:25:03 +00:00
|
|
|
|
2005-10-16 05:39:50 +00:00
|
|
|
class PPCTargetLowering : public TargetLowering {
|
2005-08-16 17:14:42 +00:00
|
|
|
int VarArgsFrameIndex; // FrameIndex for start of varargs area.
|
|
|
|
int ReturnAddrIndex; // FrameIndex for return slot.
|
|
|
|
public:
|
2005-10-16 05:39:50 +00:00
|
|
|
PPCTargetLowering(TargetMachine &TM);
|
2005-08-16 17:14:42 +00:00
|
|
|
|
2006-01-09 23:52:17 +00:00
|
|
|
/// getTargetNodeName() - This method returns the name of a target specific
|
|
|
|
/// DAG node.
|
|
|
|
virtual const char *getTargetNodeName(unsigned Opcode) const;
|
|
|
|
|
2005-08-26 00:52:45 +00:00
|
|
|
/// LowerOperation - Provide custom lowering hooks for some operations.
|
|
|
|
///
|
|
|
|
virtual SDOperand LowerOperation(SDOperand Op, SelectionDAG &DAG);
|
|
|
|
|
2006-03-01 04:57:39 +00:00
|
|
|
virtual SDOperand PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
|
|
|
|
|
2006-04-02 06:26:07 +00:00
|
|
|
virtual void computeMaskedBitsForTargetNode(const SDOperand Op,
|
|
|
|
uint64_t Mask,
|
|
|
|
uint64_t &KnownZero,
|
|
|
|
uint64_t &KnownOne,
|
|
|
|
unsigned Depth = 0) const;
|
2005-10-18 23:23:37 +00:00
|
|
|
|
2005-08-26 21:23:58 +00:00
|
|
|
virtual MachineBasicBlock *InsertAtEndOfBasicBlock(MachineInstr *MI,
|
|
|
|
MachineBasicBlock *MBB);
|
2006-01-31 19:20:21 +00:00
|
|
|
|
2006-02-07 20:16:30 +00:00
|
|
|
ConstraintType getConstraintType(char ConstraintLetter) const;
|
2006-01-31 19:20:21 +00:00
|
|
|
std::vector<unsigned>
|
2006-02-22 00:56:39 +00:00
|
|
|
getRegClassForInlineAsmConstraint(const std::string &Constraint,
|
|
|
|
MVT::ValueType VT) const;
|
2006-10-31 19:40:43 +00:00
|
|
|
SDOperand isOperandValidForConstraint(SDOperand Op, char ConstraintLetter,
|
|
|
|
SelectionDAG &DAG);
|
2006-03-13 23:20:37 +00:00
|
|
|
|
|
|
|
/// isLegalAddressImmediate - Return true if the integer value can be used
|
|
|
|
/// as the offset of the target addressing mode.
|
|
|
|
virtual bool isLegalAddressImmediate(int64_t V) const;
|
2006-08-28 01:02:49 +00:00
|
|
|
virtual bool isLegalAddressImmediate(llvm::GlobalValue*) const;
|
2005-08-16 17:14:42 +00:00
|
|
|
};
|
|
|
|
}
|
|
|
|
|
|
|
|
#endif // LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
|