-//===-- PPC32JITInfo.cpp - Implement the JIT interfaces for the PowerPC ---===//
-//
+//===-- PPCJITInfo.cpp - Implement the JIT interfaces for the PowerPC -----===//
+//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
-//
+//
//===----------------------------------------------------------------------===//
//
// This file implements the JIT interfaces for the 32-bit PowerPC target.
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "jit"
-#include "PPC32JITInfo.h"
-#include "PPC32Relocations.h"
+#include "PPCJITInfo.h"
+#include "PPCRelocations.h"
#include "llvm/CodeGen/MachineCodeEmitter.h"
#include "llvm/Config/alloca.h"
+#include <set>
using namespace llvm;
static TargetJITInfo::JITCompilerFn JITCompilerFunction;
AtI[3] = BUILD_BCTR(isCall); // bctr/bctrl
}
-static void CompilationCallback() {
- // Save R3-R31, since we want to restore arguments and nonvolatile regs used
- // by the compiler. We also save and restore the FP regs, although this is
- // probably just paranoia (gcc is unlikely to emit code that uses them for
- // for this function.
+extern "C" void PPC32CompilationCallback();
+
#if defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)
- unsigned IntRegs[29];
- double FPRegs[13];
- __asm__ __volatile__ (
- "stmw r3, 0(%0)\n"
- "stfd f1, 0(%1)\n" "stfd f2, 8(%1)\n" "stfd f3, 16(%1)\n"
- "stfd f4, 24(%1)\n" "stfd f5, 32(%1)\n" "stfd f6, 40(%1)\n"
- "stfd f7, 48(%1)\n" "stfd f8, 56(%1)\n" "stfd f9, 64(%1)\n"
- "stfd f10, 72(%1)\n" "stfd f11, 80(%1)\n" "stfd f12, 88(%1)\n"
- "stfd f13, 96(%1)\n" :: "b" (IntRegs), "b" (FPRegs) );
- /// FIXME: Need to safe and restore the rest of the FP regs!
+// CompilationCallback stub - We can't use a C function with inline assembly in
+// it, because we the prolog/epilog inserted by GCC won't work for us. Instead,
+// write our own wrapper, which does things our way, so we have complete control
+// over register saving and restoring.
+asm(
+ ".text\n"
+ ".align 2\n"
+ ".globl _PPC32CompilationCallback\n"
+"_PPC32CompilationCallback:\n"
+ // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
+ // FIXME: need to save v[0-19] for altivec?
+ // Set up a proper stack frame
+ "stwu r1, -208(r1)\n"
+ "mflr r0\n"
+ "stw r0, 216(r1)\n"
+ // Save all int arg registers
+ "stw r10, 204(r1)\n" "stw r9, 200(r1)\n"
+ "stw r8, 196(r1)\n" "stw r7, 192(r1)\n"
+ "stw r6, 188(r1)\n" "stw r5, 184(r1)\n"
+ "stw r4, 180(r1)\n" "stw r3, 176(r1)\n"
+ // Save all call-clobbered FP regs.
+ "stfd f13, 168(r1)\n" "stfd f12, 160(r1)\n"
+ "stfd f11, 152(r1)\n" "stfd f10, 144(r1)\n"
+ "stfd f9, 136(r1)\n" "stfd f8, 128(r1)\n"
+ "stfd f7, 120(r1)\n" "stfd f6, 112(r1)\n"
+ "stfd f5, 104(r1)\n" "stfd f4, 96(r1)\n"
+ "stfd f3, 88(r1)\n" "stfd f2, 80(r1)\n"
+ "stfd f1, 72(r1)\n"
+ // Arguments to Compilation Callback:
+ // r3 - our lr (address of the call instruction in stub plus 4)
+ // r4 - stub's lr (address of instruction that called the stub plus 4)
+ "mr r3, r0\n"
+ "lwz r2, 208(r1)\n" // stub's frame
+ "lwz r4, 8(r2)\n" // stub's lr
+ "bl _PPC32CompilationCallbackC\n"
+ "mtctr r3\n"
+ // Restore all int arg registers
+ "lwz r10, 204(r1)\n" "lwz r9, 200(r1)\n"
+ "lwz r8, 196(r1)\n" "lwz r7, 192(r1)\n"
+ "lwz r6, 188(r1)\n" "lwz r5, 184(r1)\n"
+ "lwz r4, 180(r1)\n" "lwz r3, 176(r1)\n"
+ // Restore all FP arg registers
+ "lfd f13, 168(r1)\n" "lfd f12, 160(r1)\n"
+ "lfd f11, 152(r1)\n" "lfd f10, 144(r1)\n"
+ "lfd f9, 136(r1)\n" "lfd f8, 128(r1)\n"
+ "lfd f7, 120(r1)\n" "lfd f6, 112(r1)\n"
+ "lfd f5, 104(r1)\n" "lfd f4, 96(r1)\n"
+ "lfd f3, 88(r1)\n" "lfd f2, 80(r1)\n"
+ "lfd f1, 72(r1)\n"
+ // Pop 3 frames off the stack and branch to target
+ "lwz r1, 208(r1)\n"
+ "lwz r2, 8(r1)\n"
+ "mtlr r2\n"
+ "bctr\n"
+ );
+#else
+void PPC32CompilationCallback() {
+ assert(0 && "This is not a power pc, you can't execute this!");
+ abort();
+}
#endif
- unsigned *CameFromStub = (unsigned*)__builtin_return_address(0);
- unsigned *CameFromOrig = (unsigned*)__builtin_return_address(1);
- unsigned *CCStackPtr = (unsigned*)__builtin_frame_address(0);
-//unsigned *StubStackPtr = (unsigned*)__builtin_frame_address(1);
- unsigned *OrigStackPtr = (unsigned*)__builtin_frame_address(2);
-
- // Adjust pointer to the branch, not the return address.
- --CameFromStub;
-
- void *Target = JITCompilerFunction(CameFromStub);
-
- // Check to see if CameFromOrig[-1] is a 'bl' instruction, and if we can
- // rewrite it to branch directly to the destination. If so, rewrite it so it
- // does not need to go through the stub anymore.
- unsigned CameFromOrigInst = CameFromOrig[-1];
- if ((CameFromOrigInst >> 26) == 18) { // Direct call.
- intptr_t Offset = ((intptr_t)Target-(intptr_t)CameFromOrig) >> 2;
+extern "C" unsigned *PPC32CompilationCallbackC(unsigned *StubCallAddrPlus4,
+ unsigned *OrigCallAddrPlus4) {
+ // Adjust the pointer to the address of the call instruction in the stub
+ // emitted by emitFunctionStub, rather than the instruction after it.
+ unsigned *StubCallAddr = StubCallAddrPlus4 - 1;
+ unsigned *OrigCallAddr = OrigCallAddrPlus4 - 1;
+
+ void *Target = JITCompilerFunction(StubCallAddr);
+
+ // Check to see if *OrigCallAddr is a 'bl' instruction, and if we can rewrite
+ // it to branch directly to the destination. If so, rewrite it so it does not
+ // need to go through the stub anymore.
+ unsigned OrigCallInst = *OrigCallAddr;
+ if ((OrigCallInst >> 26) == 18) { // Direct call.
+ intptr_t Offset = ((intptr_t)Target - (intptr_t)OrigCallAddr) >> 2;
+
if (Offset >= -(1 << 23) && Offset < (1 << 23)) { // In range?
- // FIXME: hasn't been tested at all.
- // Clear the original target out:
- CameFromOrigInst &= (63 << 26) | 3;
- CameFromOrigInst |= Offset << 2;
- CameFromOrig[-1] = CameFromOrigInst;
+ // Clear the original target out.
+ OrigCallInst &= (63 << 26) | 3;
+ // Fill in the new target.
+ OrigCallInst |= (Offset & ((1 << 24)-1)) << 2;
+ // Replace the call.
+ *OrigCallAddr = OrigCallInst;
}
}
-
- // Locate the start of the stub. If this is a short call, adjust backwards
- // the short amount, otherwise the full amount.
- bool isShortStub = (*CameFromStub >> 26) == 18;
- CameFromStub -= isShortStub ? 2 : 6;
+
+ // Assert that we are coming from a stub that was created with our
+ // emitFunctionStub.
+ assert((*StubCallAddr >> 26) == 19 && "Call in stub is not indirect!");
+ StubCallAddr -= 6;
// Rewrite the stub with an unconditional branch to the target, for any users
// who took the address of the stub.
- EmitBranchToAt(CameFromStub, Target, false);
-
- // Change the SP so that we pop two stack frames off when we return.
- *CCStackPtr = (intptr_t)OrigStackPtr;
+ EmitBranchToAt(StubCallAddr, Target, false);
- // Put the address of the stub and the LR value that originally came into the
- // stub in a place that is easy to get on the stack after we restore all regs.
- CCStackPtr[2] = (intptr_t)Target;
- CCStackPtr[1] = (intptr_t)CameFromOrig;
-
- // Note, this is not a standard epilog!
-#if defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)
- register unsigned *IRR asm ("r2") = IntRegs;
- register double *FRR asm ("r3") = FPRegs;
- __asm__ __volatile__ (
- "lfd f1, 0(%0)\n" "lfd f2, 8(%0)\n" "lfd f3, 16(%0)\n"
- "lfd f4, 24(%0)\n" "lfd f5, 32(%0)\n" "lfd f6, 40(%0)\n"
- "lfd f7, 48(%0)\n" "lfd f8, 56(%0)\n" "lfd f9, 64(%0)\n"
- "lfd f10, 72(%0)\n" "lfd f11, 80(%0)\n" "lfd f12, 88(%0)\n"
- "lfd f13, 96(%0)\n"
- "lmw r3, 0(%1)\n" // Load all integer regs
- "lwz r0,4(r1)\n" // Get CameFromOrig (LR into stub)
- "mtlr r0\n" // Put it in the LR register
- "lwz r0,8(r1)\n" // Get target function pointer
- "mtctr r0\n" // Put it into the CTR register
- "lwz r1,0(r1)\n" // Pop two frames off
- "bctr\n" :: // Return to stub!
- "b" (FRR), "b" (IRR));
-#endif
+ // Put the address of the target function to call and the address to return to
+ // after calling the target function in a place that is easy to get on the
+ // stack after we restore all regs.
+ return (unsigned *)Target;
}
-TargetJITInfo::LazyResolverFn
-PPC32JITInfo::getLazyResolverFunction(JITCompilerFn Fn) {
+TargetJITInfo::LazyResolverFn
+PPCJITInfo::getLazyResolverFunction(JITCompilerFn Fn) {
JITCompilerFunction = Fn;
- return CompilationCallback;
+ return PPC32CompilationCallback;
}
-void *PPC32JITInfo::emitFunctionStub(void *Fn, MachineCodeEmitter &MCE) {
+void *PPCJITInfo::emitFunctionStub(void *Fn, MachineCodeEmitter &MCE) {
// If this is just a call to an external function, emit a branch instead of a
// call. The code is the same except for one bit of the last instruction.
- if (Fn != CompilationCallback) {
+ if (Fn != PPC32CompilationCallback) {
MCE.startFunctionStub(4*4);
void *Addr = (void*)(intptr_t)MCE.getCurrentPCValue();
- MCE.emitWord(0);
- MCE.emitWord(0);
- MCE.emitWord(0);
- MCE.emitWord(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
EmitBranchToAt(Addr, Fn, false);
return MCE.finishFunctionStub(0);
}
MCE.startFunctionStub(4*7);
- MCE.emitWord(0x9421ffe0); // stwu r1,-32(r1)
- MCE.emitWord(0x7d6802a6); // mflr r11
- MCE.emitWord(0x91610028); // stw r11, 40(r1)
+ MCE.emitWordBE(0x9421ffe0); // stwu r1,-32(r1)
+ MCE.emitWordBE(0x7d6802a6); // mflr r11
+ MCE.emitWordBE(0x91610028); // stw r11, 40(r1)
void *Addr = (void*)(intptr_t)MCE.getCurrentPCValue();
- MCE.emitWord(0);
- MCE.emitWord(0);
- MCE.emitWord(0);
- MCE.emitWord(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
+ MCE.emitWordBE(0);
EmitBranchToAt(Addr, Fn, true/*is call*/);
return MCE.finishFunctionStub(0);
}
-void PPC32JITInfo::relocate(void *Function, MachineRelocation *MR,
- unsigned NumRelocs) {
+void PPCJITInfo::relocate(void *Function, MachineRelocation *MR,
+ unsigned NumRelocs, unsigned char* GOTBase) {
for (unsigned i = 0; i != NumRelocs; ++i, ++MR) {
unsigned *RelocPos = (unsigned*)Function + MR->getMachineCodeOffset()/4;
intptr_t ResultPtr = (intptr_t)MR->getResultPointer();
"Relocation out of range!");
*RelocPos |= (ResultPtr & ((1 << 24)-1)) << 2;
break;
- case PPC::reloc_absolute_loadhi: // Relocate high bits into addis
- case PPC::reloc_absolute_la: // Relocate low bits into addi
+ case PPC::reloc_absolute_ptr_high: // Pointer relocations.
+ case PPC::reloc_absolute_ptr_low:
+ case PPC::reloc_absolute_high: // high bits of ref -> low 16 of instr
+ case PPC::reloc_absolute_low: // low bits of ref -> low 16 of instr
ResultPtr += MR->getConstantVal();
- if (MR->getRelocationType() == PPC::reloc_absolute_loadhi) {
+ // If this is a high-part access, get the high-part.
+ if (MR->getRelocationType() == PPC::reloc_absolute_high ||
+ MR->getRelocationType() == PPC::reloc_absolute_ptr_high) {
// If the low part will have a carry (really a borrow) from the low
// 16-bits into the high 16, add a bit to borrow from.
if (((int)ResultPtr << 16) < 0)
}
}
-void PPC32JITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
+void PPCJITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
EmitBranchToAt(Old, New, false);
}