2006-09-04 04:16:09 +00:00
|
|
|
//===-- LLVMTargetMachine.cpp - Implement the LLVMTargetMachine class -----===//
|
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
2007-12-29 20:36:04 +00:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
2006-09-04 04:16:09 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// This file implements the LLVMTargetMachine class.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#include "llvm/Target/TargetMachine.h"
|
|
|
|
#include "llvm/PassManager.h"
|
|
|
|
#include "llvm/Pass.h"
|
2007-03-31 00:24:43 +00:00
|
|
|
#include "llvm/Assembly/PrintModulePass.h"
|
2007-03-06 21:14:09 +00:00
|
|
|
#include "llvm/Analysis/LoopPass.h"
|
2006-09-04 04:16:09 +00:00
|
|
|
#include "llvm/CodeGen/Passes.h"
|
|
|
|
#include "llvm/Target/TargetOptions.h"
|
|
|
|
#include "llvm/Transforms/Scalar.h"
|
2007-03-31 00:24:43 +00:00
|
|
|
#include "llvm/Support/CommandLine.h"
|
2006-09-04 04:16:09 +00:00
|
|
|
using namespace llvm;
|
|
|
|
|
2007-06-19 05:47:49 +00:00
|
|
|
static cl::opt<bool> PrintLSR("print-lsr-output", cl::Hidden,
|
|
|
|
cl::desc("Print LLVM IR produced by the loop-reduce pass"));
|
|
|
|
static cl::opt<bool> PrintISelInput("print-isel-input", cl::Hidden,
|
|
|
|
cl::desc("Print LLVM IR input to isel pass"));
|
2007-07-20 21:56:13 +00:00
|
|
|
static cl::opt<bool> PrintEmittedAsm("print-emitted-asm", cl::Hidden,
|
|
|
|
cl::desc("Dump emitter generated instructions as assembly"));
|
2007-06-19 05:47:49 +00:00
|
|
|
|
Add a really quick hack at a machine code sinking pass, enabled with --enable-sinking.
It is missing validity checks, so it is known broken. However, it is powerful enough
to compile this contrived code:
void test1(int C, double A, double B, double *P) {
double Tmp = A*A+B*B;
*P = C ? Tmp : A;
}
into:
_test1:
movsd 8(%esp), %xmm0
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movsd 16(%esp), %xmm1
mulsd %xmm1, %xmm1
mulsd %xmm0, %xmm0
addsd %xmm1, %xmm0
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm0, (%eax)
ret
instead of:
_test1:
movsd 16(%esp), %xmm0
mulsd %xmm0, %xmm0
movsd 8(%esp), %xmm1
movapd %xmm1, %xmm2
mulsd %xmm2, %xmm2
addsd %xmm0, %xmm2
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movapd %xmm2, %xmm1
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm1, (%eax)
ret
woo.
llvm-svn: 45570
2008-01-04 07:36:53 +00:00
|
|
|
// Hidden options to help debugging
|
|
|
|
static cl::opt<bool>
|
|
|
|
EnableSinking("enable-sinking", cl::init(false), cl::Hidden,
|
|
|
|
cl::desc("Perform sinking on machine code"));
|
2008-01-04 08:11:03 +00:00
|
|
|
static cl::opt<bool>
|
|
|
|
PerformLICM("machine-licm",
|
|
|
|
cl::init(false), cl::Hidden,
|
|
|
|
cl::desc("Perform loop-invariant code motion on machine code"));
|
Add a really quick hack at a machine code sinking pass, enabled with --enable-sinking.
It is missing validity checks, so it is known broken. However, it is powerful enough
to compile this contrived code:
void test1(int C, double A, double B, double *P) {
double Tmp = A*A+B*B;
*P = C ? Tmp : A;
}
into:
_test1:
movsd 8(%esp), %xmm0
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movsd 16(%esp), %xmm1
mulsd %xmm1, %xmm1
mulsd %xmm0, %xmm0
addsd %xmm1, %xmm0
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm0, (%eax)
ret
instead of:
_test1:
movsd 16(%esp), %xmm0
mulsd %xmm0, %xmm0
movsd 8(%esp), %xmm1
movapd %xmm1, %xmm2
mulsd %xmm2, %xmm2
addsd %xmm0, %xmm2
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movapd %xmm2, %xmm1
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm1, (%eax)
ret
woo.
llvm-svn: 45570
2008-01-04 07:36:53 +00:00
|
|
|
|
2007-02-08 01:36:53 +00:00
|
|
|
FileModel::Model
|
|
|
|
LLVMTargetMachine::addPassesToEmitFile(FunctionPassManager &PM,
|
|
|
|
std::ostream &Out,
|
|
|
|
CodeGenFileType FileType,
|
|
|
|
bool Fast) {
|
2006-09-04 04:16:09 +00:00
|
|
|
// Standard LLVM-Level Passes.
|
|
|
|
|
|
|
|
// Run loop strength reduction before anything else.
|
2007-03-31 00:24:43 +00:00
|
|
|
if (!Fast) {
|
|
|
|
PM.add(createLoopStrengthReducePass(getTargetLowering()));
|
|
|
|
if (PrintLSR)
|
|
|
|
PM.add(new PrintFunctionPass("\n\n*** Code after LSR *** \n", &cerr));
|
|
|
|
}
|
2006-09-04 04:16:09 +00:00
|
|
|
|
|
|
|
// FIXME: Implement efficient support for garbage collection intrinsics.
|
|
|
|
PM.add(createLowerGCPass());
|
2007-07-11 16:59:20 +00:00
|
|
|
|
2007-02-22 16:22:15 +00:00
|
|
|
if (!ExceptionHandling)
|
|
|
|
PM.add(createLowerInvokePass(getTargetLowering()));
|
2007-07-11 16:59:20 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Make sure that no unreachable blocks are instruction selected.
|
|
|
|
PM.add(createUnreachableBlockEliminationPass());
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2007-03-31 04:18:03 +00:00
|
|
|
if (!Fast)
|
|
|
|
PM.add(createCodeGenPreparePass(getTargetLowering()));
|
|
|
|
|
|
|
|
if (PrintISelInput)
|
|
|
|
PM.add(new PrintFunctionPass("\n\n*** Final LLVM Code input to ISel *** \n",
|
|
|
|
&cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Ask the target for an isel.
|
|
|
|
if (addInstSelector(PM, Fast))
|
2007-02-08 01:36:53 +00:00
|
|
|
return FileModel::Error;
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Print the instruction selected machine code...
|
|
|
|
if (PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2008-01-04 08:11:03 +00:00
|
|
|
if (PerformLICM)
|
|
|
|
PM.add(createMachineLICMPass());
|
Add a really quick hack at a machine code sinking pass, enabled with --enable-sinking.
It is missing validity checks, so it is known broken. However, it is powerful enough
to compile this contrived code:
void test1(int C, double A, double B, double *P) {
double Tmp = A*A+B*B;
*P = C ? Tmp : A;
}
into:
_test1:
movsd 8(%esp), %xmm0
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movsd 16(%esp), %xmm1
mulsd %xmm1, %xmm1
mulsd %xmm0, %xmm0
addsd %xmm1, %xmm0
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm0, (%eax)
ret
instead of:
_test1:
movsd 16(%esp), %xmm0
mulsd %xmm0, %xmm0
movsd 8(%esp), %xmm1
movapd %xmm1, %xmm2
mulsd %xmm2, %xmm2
addsd %xmm0, %xmm2
cmpl $0, 4(%esp)
je LBB1_2 # entry
LBB1_1: # entry
movapd %xmm2, %xmm1
LBB1_2: # entry
movl 24(%esp), %eax
movsd %xmm1, (%eax)
ret
woo.
llvm-svn: 45570
2008-01-04 07:36:53 +00:00
|
|
|
|
|
|
|
if (EnableSinking)
|
|
|
|
PM.add(createMachineSinkingPass());
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Perform register allocation to convert to a concrete x86 representation
|
|
|
|
PM.add(createRegisterAllocator());
|
|
|
|
|
|
|
|
if (PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
2007-07-27 07:36:14 +00:00
|
|
|
|
|
|
|
PM.add(createLowerSubregsPass());
|
|
|
|
|
|
|
|
if (PrintMachineCode) // Print the subreg lowered code
|
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Run post-ra passes.
|
|
|
|
if (addPostRegAlloc(PM, Fast) && PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Insert prolog/epilog code. Eliminate abstract frame index references...
|
|
|
|
PM.add(createPrologEpilogCodeInserter());
|
|
|
|
|
2007-07-13 17:13:54 +00:00
|
|
|
// Second pass scheduler.
|
2007-07-13 17:31:29 +00:00
|
|
|
if (!Fast)
|
|
|
|
PM.add(createPostRAScheduler());
|
2007-07-13 17:13:54 +00:00
|
|
|
|
2006-10-13 20:45:56 +00:00
|
|
|
// Branch folding must be run after regalloc and prolog/epilog insertion.
|
2006-10-24 16:11:49 +00:00
|
|
|
if (!Fast)
|
2007-05-22 18:31:04 +00:00
|
|
|
PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2006-11-07 19:33:46 +00:00
|
|
|
// Fold redundant debug labels.
|
|
|
|
PM.add(createDebugLabelFoldingPass());
|
2006-10-13 20:45:56 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
if (PrintMachineCode) // Print the register-allocated code
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
if (addPreEmitPass(PM, Fast) && PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
switch (FileType) {
|
2007-02-08 01:36:53 +00:00
|
|
|
default:
|
|
|
|
break;
|
|
|
|
case TargetMachine::AssemblyFile:
|
|
|
|
if (addAssemblyEmitter(PM, Fast, Out))
|
|
|
|
return FileModel::Error;
|
|
|
|
return FileModel::AsmFile;
|
|
|
|
case TargetMachine::ObjectFile:
|
|
|
|
if (getMachOWriterInfo())
|
|
|
|
return FileModel::MachOFile;
|
|
|
|
else if (getELFWriterInfo())
|
|
|
|
return FileModel::ElfFile;
|
2006-09-04 04:16:09 +00:00
|
|
|
}
|
2007-02-08 01:36:53 +00:00
|
|
|
|
|
|
|
return FileModel::Error;
|
|
|
|
}
|
|
|
|
|
|
|
|
/// addPassesToEmitFileFinish - If the passes to emit the specified file had to
|
|
|
|
/// be split up (e.g., to add an object writer pass), this method can be used to
|
|
|
|
/// finish up adding passes to emit the file, if necessary.
|
|
|
|
bool LLVMTargetMachine::addPassesToEmitFileFinish(FunctionPassManager &PM,
|
|
|
|
MachineCodeEmitter *MCE,
|
|
|
|
bool Fast) {
|
|
|
|
if (MCE)
|
2007-07-20 21:56:13 +00:00
|
|
|
addSimpleCodeEmitter(PM, Fast, PrintEmittedAsm, *MCE);
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Delete machine code for this function
|
|
|
|
PM.add(createMachineCodeDeleter());
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
return false; // success!
|
|
|
|
}
|
|
|
|
|
|
|
|
/// addPassesToEmitMachineCode - Add passes to the specified pass manager to
|
|
|
|
/// get machine code emitted. This uses a MachineCodeEmitter object to handle
|
|
|
|
/// actually outputting the machine code and resolving things like the address
|
|
|
|
/// of functions. This method should returns true if machine code emission is
|
|
|
|
/// not supported.
|
|
|
|
///
|
|
|
|
bool LLVMTargetMachine::addPassesToEmitMachineCode(FunctionPassManager &PM,
|
|
|
|
MachineCodeEmitter &MCE,
|
|
|
|
bool Fast) {
|
|
|
|
// Standard LLVM-Level Passes.
|
|
|
|
|
|
|
|
// Run loop strength reduction before anything else.
|
2007-03-31 04:18:03 +00:00
|
|
|
if (!Fast) {
|
|
|
|
PM.add(createLoopStrengthReducePass(getTargetLowering()));
|
|
|
|
if (PrintLSR)
|
|
|
|
PM.add(new PrintFunctionPass("\n\n*** Code after LSR *** \n", &cerr));
|
|
|
|
}
|
2006-09-04 04:16:09 +00:00
|
|
|
|
|
|
|
// FIXME: Implement efficient support for garbage collection intrinsics.
|
|
|
|
PM.add(createLowerGCPass());
|
|
|
|
|
|
|
|
// FIXME: Implement the invoke/unwind instructions!
|
2006-09-04 06:21:35 +00:00
|
|
|
PM.add(createLowerInvokePass(getTargetLowering()));
|
2006-09-04 04:16:09 +00:00
|
|
|
|
|
|
|
// Make sure that no unreachable blocks are instruction selected.
|
|
|
|
PM.add(createUnreachableBlockEliminationPass());
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2007-03-31 04:18:03 +00:00
|
|
|
if (!Fast)
|
|
|
|
PM.add(createCodeGenPreparePass(getTargetLowering()));
|
|
|
|
|
|
|
|
if (PrintISelInput)
|
|
|
|
PM.add(new PrintFunctionPass("\n\n*** Final LLVM Code input to ISel *** \n",
|
|
|
|
&cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Ask the target for an isel.
|
|
|
|
if (addInstSelector(PM, Fast))
|
|
|
|
return true;
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Print the instruction selected machine code...
|
|
|
|
if (PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2008-01-04 08:11:03 +00:00
|
|
|
if (PerformLICM)
|
|
|
|
PM.add(createMachineLICMPass());
|
2008-01-05 06:14:16 +00:00
|
|
|
|
|
|
|
if (EnableSinking)
|
|
|
|
PM.add(createMachineSinkingPass());
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Perform register allocation to convert to a concrete x86 representation
|
|
|
|
PM.add(createRegisterAllocator());
|
|
|
|
|
|
|
|
if (PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
2007-07-27 07:36:14 +00:00
|
|
|
|
|
|
|
PM.add(createLowerSubregsPass());
|
|
|
|
|
|
|
|
if (PrintMachineCode) // Print the subreg lowered code
|
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
2007-02-08 01:36:53 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Run post-ra passes.
|
|
|
|
if (addPostRegAlloc(PM, Fast) && PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
// Insert prolog/epilog code. Eliminate abstract frame index references...
|
|
|
|
PM.add(createPrologEpilogCodeInserter());
|
|
|
|
|
|
|
|
if (PrintMachineCode) // Print the register-allocated code
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
2006-09-04 04:16:09 +00:00
|
|
|
|
2007-07-13 17:13:54 +00:00
|
|
|
// Second pass scheduler.
|
2007-07-13 17:31:29 +00:00
|
|
|
if (!Fast)
|
|
|
|
PM.add(createPostRAScheduler());
|
2007-07-13 17:13:54 +00:00
|
|
|
|
2006-11-16 01:00:07 +00:00
|
|
|
// Branch folding must be run after regalloc and prolog/epilog insertion.
|
|
|
|
if (!Fast)
|
2007-05-22 18:31:04 +00:00
|
|
|
PM.add(createBranchFoldingPass(getEnableTailMergeDefault()));
|
Initial commit of the machine code LICM pass. It successfully hoists this:
_foo:
li r2, 0
LBB1_1: ; bb
li r5, 0
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
to:
_foo:
li r2, 0
li r5, 0
LBB1_1: ; bb
stw r5, 0(r3)
addi r2, r2, 1
addi r3, r3, 4
cmplw cr0, r2, r4
bne cr0, LBB1_1 ; bb
LBB1_2: ; return
blr
ZOMG!! :-)
Moar to come...
llvm-svn: 44687
2007-12-07 21:42:31 +00:00
|
|
|
|
2006-09-04 04:16:09 +00:00
|
|
|
if (addPreEmitPass(PM, Fast) && PrintMachineCode)
|
2007-02-08 01:36:53 +00:00
|
|
|
PM.add(createMachineFunctionPrinterPass(cerr));
|
|
|
|
|
2007-07-20 21:56:13 +00:00
|
|
|
addCodeEmitter(PM, Fast, PrintEmittedAsm, MCE);
|
2006-09-04 04:16:09 +00:00
|
|
|
|
|
|
|
// Delete machine code for this function
|
|
|
|
PM.add(createMachineCodeDeleter());
|
|
|
|
|
|
|
|
return false; // success!
|
|
|
|
}
|