/*
- * Copyright (C) 2009 University of Szeged
+ * Copyright (C) 2009, 2010 University of Szeged
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
#ifndef ARMAssembler_h
#define ARMAssembler_h
-#include <wtf/Platform.h>
-
#if ENABLE(ASSEMBLER) && CPU(ARM_TRADITIONAL)
#include "AssemblerBufferWithConstantPool.h"
+#include "JITCompilationEffort.h"
#include <wtf/Assertions.h>
namespace JSC {
r0 = 0,
r1,
r2,
- r3,
- S0 = r3,
+ r3, S0 = r3,
r4,
r5,
r6,
r7,
- r8,
- S1 = r8,
+ r8, S1 = r8,
r9,
r10,
r11,
r12,
- r13,
- sp = r13,
- r14,
- lr = r14,
- r15,
- pc = r15
+ r13, sp = r13,
+ r14, lr = r14,
+ r15, pc = r15
} RegisterID;
typedef enum {
d0,
d1,
d2,
- d3,
- SD0 = d3
+ d3, SD0 = d3,
+ d4,
+ d5,
+ d6,
+ d7,
+ d8,
+ d9,
+ d10,
+ d11,
+ d12,
+ d13,
+ d14,
+ d15,
+ d16,
+ d17,
+ d18,
+ d19,
+ d20,
+ d21,
+ d22,
+ d23,
+ d24,
+ d25,
+ d26,
+ d27,
+ d28,
+ d29,
+ d30,
+ d31
} FPRegisterID;
} // namespace ARMRegisters
typedef ARMRegisters::RegisterID RegisterID;
typedef ARMRegisters::FPRegisterID FPRegisterID;
typedef AssemblerBufferWithConstantPool<2048, 4, 4, ARMAssembler> ARMBuffer;
- typedef SegmentedVector<int, 64> Jumps;
+ typedef SegmentedVector<AssemblerLabel, 64> Jumps;
ARMAssembler() { }
MVN = (0xf << 21),
MUL = 0x00000090,
MULL = 0x00c00090,
- FADDD = 0x0e300b00,
- FDIVD = 0x0e800b00,
- FSUBD = 0x0e300b40,
- FMULD = 0x0e200b00,
- FCMPD = 0x0eb40b40,
+ VADD_F64 = 0x0e300b00,
+ VDIV_F64 = 0x0e800b00,
+ VSUB_F64 = 0x0e300b40,
+ VMUL_F64 = 0x0e200b00,
+ VCMP_F64 = 0x0eb40b40,
+ VSQRT_F64 = 0x0eb10bc0,
DTR = 0x05000000,
LDRH = 0x00100090,
STRH = 0x00000090,
FDTR = 0x0d000b00,
B = 0x0a000000,
BL = 0x0b000000,
- FMSR = 0x0e000a10,
- FMRS = 0x0e100a10,
- FSITOD = 0x0eb80bc0,
- FTOSID = 0x0ebd0b40,
- FMSTAT = 0x0ef1fa10,
+#if WTF_ARM_ARCH_AT_LEAST(5) || defined(__ARM_ARCH_4T__)
+ BX = 0x012fff10,
+#endif
+ VMOV_VFP = 0x0e000a10,
+ VMOV_ARM = 0x0e100a10,
+ VCVT_F64_S32 = 0x0eb80bc0,
+ VCVT_S32_F64 = 0x0ebd0b40,
+ VMRS_APSR = 0x0ef1fa10,
#if WTF_ARM_ARCH_AT_LEAST(5)
CLZ = 0x016f0f10,
- BKPT = 0xe120070,
+ BKPT = 0xe1200070,
+ BLX = 0x012fff30,
+ NOP_T2 = 0xf3af8000,
#endif
#if WTF_ARM_ARCH_AT_LEAST(7)
MOVW = 0x03000000,
MOVT = 0x03400000,
#endif
+ NOP = 0xe1a00000,
};
enum {
SET_CC = (1 << 20),
OP2_OFSREG = (1 << 25),
DT_UP = (1 << 23),
+ DT_BYTE = (1 << 22),
DT_WB = (1 << 21),
// This flag is inlcuded in LDR and STR
DT_PRE = (1 << 24),
enum {
padForAlign8 = 0x00,
padForAlign16 = 0x0000,
- padForAlign32 = 0xee120070,
+ padForAlign32 = 0xe12fff7f // 'bkpt 0xffff' instruction.
};
static const ARMWord INVALID_IMM = 0xf0000000;
+ static const ARMWord InvalidBranchTarget = 0xffffffff;
static const int DefaultPrefetching = 2;
- class JmpSrc {
- friend class ARMAssembler;
- public:
- JmpSrc()
- : m_offset(-1)
- {
- }
-
- private:
- JmpSrc(int offset)
- : m_offset(offset)
- {
- }
-
- int m_offset;
- };
-
- class JmpDst {
- friend class ARMAssembler;
- public:
- JmpDst()
- : m_offset(-1)
- , m_used(false)
- {
- }
-
- bool isUsed() const { return m_used; }
- void used() { m_used = true; }
- private:
- JmpDst(int offset)
- : m_offset(offset)
- , m_used(false)
- {
- ASSERT(m_offset == offset);
- }
-
- int m_offset : 31;
- int m_used : 1;
- };
-
// Instruction formating
void emitInst(ARMWord op, int rd, int rn, ARMWord op2)
{
- ASSERT ( ((op2 & ~OP2_IMM) <= 0xfff) || (((op2 & ~OP2_IMMh) <= 0xfff)) );
+ ASSERT(((op2 & ~OP2_IMM) <= 0xfff) || (((op2 & ~OP2_IMMh) <= 0xfff)));
m_buffer.putInt(op | RN(rn) | RD(rd) | op2);
}
+ void emitDoublePrecisionInst(ARMWord op, int dd, int dn, int dm)
+ {
+ ASSERT((dd >= 0 && dd <= 31) && (dn >= 0 && dn <= 31) && (dm >= 0 && dm <= 31));
+ m_buffer.putInt(op | ((dd & 0xf) << 12) | ((dd & 0x10) << (22 - 4))
+ | ((dn & 0xf) << 16) | ((dn & 0x10) << (7 - 4))
+ | (dm & 0xf) | ((dm & 0x10) << (5 - 4)));
+ }
+
+ void emitSinglePrecisionInst(ARMWord op, int sd, int sn, int sm)
+ {
+ ASSERT((sd >= 0 && sd <= 31) && (sn >= 0 && sn <= 31) && (sm >= 0 && sm <= 31));
+ m_buffer.putInt(op | ((sd >> 1) << 12) | ((sd & 0x1) << 22)
+ | ((sn >> 1) << 16) | ((sn & 0x1) << 7)
+ | (sm >> 1) | ((sm & 0x1) << 5));
+ }
+
void and_r(int rd, int rn, ARMWord op2, Condition cc = AL)
{
emitInst(static_cast<ARMWord>(cc) | AND, rd, rn, op2);
emitInst(static_cast<ARMWord>(cc) | CMP | SET_CC, 0, rn, op2);
}
+ void cmn_r(int rn, ARMWord op2, Condition cc = AL)
+ {
+ emitInst(static_cast<ARMWord>(cc) | CMN | SET_CC, 0, rn, op2);
+ }
+
void orr_r(int rd, int rn, ARMWord op2, Condition cc = AL)
{
emitInst(static_cast<ARMWord>(cc) | ORR, rd, rn, op2);
m_buffer.putInt(static_cast<ARMWord>(cc) | MULL | RN(rdhi) | RD(rdlo) | RS(rn) | RM(rm));
}
- void faddd_r(int dd, int dn, int dm, Condition cc = AL)
+ void vadd_f64_r(int dd, int dn, int dm, Condition cc = AL)
+ {
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VADD_F64, dd, dn, dm);
+ }
+
+ void vdiv_f64_r(int dd, int dn, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FADDD, dd, dn, dm);
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VDIV_F64, dd, dn, dm);
}
- void fdivd_r(int dd, int dn, int dm, Condition cc = AL)
+ void vsub_f64_r(int dd, int dn, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FDIVD, dd, dn, dm);
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VSUB_F64, dd, dn, dm);
}
- void fsubd_r(int dd, int dn, int dm, Condition cc = AL)
+ void vmul_f64_r(int dd, int dn, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FSUBD, dd, dn, dm);
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VMUL_F64, dd, dn, dm);
}
- void fmuld_r(int dd, int dn, int dm, Condition cc = AL)
+ void vcmp_f64_r(int dd, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FMULD, dd, dn, dm);
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VCMP_F64, dd, 0, dm);
}
- void fcmpd_r(int dd, int dm, Condition cc = AL)
+ void vsqrt_f64_r(int dd, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FCMPD, dd, 0, dm);
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VSQRT_F64, dd, 0, dm);
}
void ldr_imm(int rd, ARMWord imm, Condition cc = AL)
dtr_u(true, reg, ARMRegisters::sp, 0, cc);
}
- void fmsr_r(int dd, int rn, Condition cc = AL)
+ void vmov_vfp_r(int sn, int rt, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FMSR, rn, dd, 0);
+ ASSERT(rt <= 15);
+ emitSinglePrecisionInst(static_cast<ARMWord>(cc) | VMOV_VFP, rt << 1, sn, 0);
}
- void fmrs_r(int rd, int dn, Condition cc = AL)
+ void vmov_arm_r(int rt, int sn, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FMRS, rd, dn, 0);
+ ASSERT(rt <= 15);
+ emitSinglePrecisionInst(static_cast<ARMWord>(cc) | VMOV_ARM, rt << 1, sn, 0);
}
- void fsitod_r(int dd, int dm, Condition cc = AL)
+ void vcvt_f64_s32_r(int dd, int sm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FSITOD, dd, 0, dm);
+ ASSERT(!(sm & 0x1)); // sm must be divisible by 2
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VCVT_F64_S32, dd, 0, (sm >> 1));
}
- void ftosid_r(int fd, int dm, Condition cc = AL)
+ void vcvt_s32_f64_r(int sd, int dm, Condition cc = AL)
{
- emitInst(static_cast<ARMWord>(cc) | FTOSID, fd, 0, dm);
+ ASSERT(!(sd & 0x1)); // sd must be divisible by 2
+ emitDoublePrecisionInst(static_cast<ARMWord>(cc) | VCVT_S32_F64, (sd >> 1), 0, dm);
}
- void fmstat(Condition cc = AL)
+ void vmrs_apsr(Condition cc = AL)
{
- m_buffer.putInt(static_cast<ARMWord>(cc) | FMSTAT);
+ m_buffer.putInt(static_cast<ARMWord>(cc) | VMRS_APSR);
}
#if WTF_ARM_ARCH_AT_LEAST(5)
dtr_dr(true, ARMRegisters::S0, ARMRegisters::S0, ARMRegisters::S0);
#endif
}
+
+ void nop()
+ {
+ m_buffer.putInt(OP_NOP_T2);
+ }
+
+ void nop()
+ {
+ m_buffer.putInt(NOP);
+ }
+
+ void bx(int rm, Condition cc = AL)
+ {
+#if WTF_ARM_ARCH_AT_LEAST(5) || defined(__ARM_ARCH_4T__)
+ emitInst(static_cast<ARMWord>(cc) | BX, 0, 0, RM(rm));
+#else
+ mov_r(ARMRegisters::pc, RM(rm), cc);
+#endif
+ }
+
+ AssemblerLabel blx(int rm, Condition cc = AL)
+ {
+#if WTF_ARM_ARCH_AT_LEAST(5)
+ emitInst(static_cast<ARMWord>(cc) | BLX, 0, 0, RM(rm));
+#else
+ ASSERT(rm != 14);
+ ensureSpace(2 * sizeof(ARMWord), 0);
+ mov_r(ARMRegisters::lr, ARMRegisters::pc, cc);
+ bx(rm, cc);
+#endif
+ return m_buffer.label();
+ }
static ARMWord lsl(int reg, ARMWord value)
{
// General helpers
- int size()
+ size_t codeSize() const
{
- return m_buffer.size();
+ return m_buffer.codeSize();
}
void ensureSpace(int insnSpace, int constSpace)
return m_buffer.sizeOfConstantPool();
}
- JmpDst label()
+ AssemblerLabel label()
{
- return JmpDst(m_buffer.size());
+ m_buffer.ensureSpaceForAnyOneInstruction();
+ return m_buffer.label();
}
- JmpDst align(int alignment)
+ AssemblerLabel align(int alignment)
{
while (!m_buffer.isAligned(alignment))
mov_r(ARMRegisters::r0, ARMRegisters::r0);
return label();
}
- JmpSrc jmp(Condition cc = AL, int useConstantPool = 0)
+ AssemblerLabel loadBranchTarget(int rd, Condition cc = AL, int useConstantPool = 0)
{
ensureSpace(sizeof(ARMWord), sizeof(ARMWord));
- int s = m_buffer.uncheckedSize();
- ldr_un_imm(ARMRegisters::pc, 0xffffffff, cc);
- m_jumps.append(s | (useConstantPool & 0x1));
- return JmpSrc(s);
+ m_jumps.append(m_buffer.codeSize() | (useConstantPool & 0x1));
+ ldr_un_imm(rd, InvalidBranchTarget, cc);
+ return m_buffer.label();
+ }
+
+ AssemblerLabel jmp(Condition cc = AL, int useConstantPool = 0)
+ {
+ return loadBranchTarget(ARMRegisters::pc, cc, useConstantPool);
}
- void* executableCopy(ExecutablePool* allocator);
+ PassRefPtr<ExecutableMemoryHandle> executableCopy(JSGlobalData&, void* ownerUID, JITCompilationEffort);
+
+ unsigned debugOffset() { return m_buffer.debugOffset(); }
// Patching helpers
static ARMWord* getLdrImmAddress(ARMWord* insn)
{
+#if WTF_ARM_ARCH_AT_LEAST(5)
+ // Check for call
+ if ((*insn & 0x0f7f0000) != 0x051f0000) {
+ // Must be BLX
+ ASSERT((*insn & 0x012fff30) == 0x012fff30);
+ insn--;
+ }
+#endif
// Must be an ldr ..., [pc +/- imm]
ASSERT((*insn & 0x0f7f0000) == 0x051f0000);
static void patchConstantPoolLoad(void* loadAddr, void* constPoolAddr);
+ // Read pointers
+ static void* readPointer(void* from)
+ {
+ ARMWord* insn = reinterpret_cast<ARMWord*>(from);
+ ARMWord* addr = getLdrImmAddress(insn);
+ return *reinterpret_cast<void**>(addr);
+ }
+
// Patch pointers
- static void linkPointer(void* code, JmpDst from, void* to)
+ static void linkPointer(void* code, AssemblerLabel from, void* to)
{
patchPointerInternal(reinterpret_cast<intptr_t>(code) + from.m_offset, to);
}
{
patchPointerInternal(reinterpret_cast<intptr_t>(from), reinterpret_cast<void*>(to));
}
+
+ static void repatchCompact(void* where, int32_t value)
+ {
+ repatchInt32(where, value);
+ }
static void repatchPointer(void* from, void* to)
{
patchPointerInternal(reinterpret_cast<intptr_t>(from), to);
}
- static void repatchLoadPtrToLEA(void* from)
+ // Linkers
+ static intptr_t getAbsoluteJumpAddress(void* base, int offset = 0)
{
- // On arm, this is a patch from LDR to ADD. It is restricted conversion,
- // from special case to special case, altough enough for its purpose
- ARMWord* insn = reinterpret_cast<ARMWord*>(from);
- ASSERT((*insn & 0x0ff00f00) == 0x05900000);
-
- *insn = (*insn & 0xf00ff0ff) | 0x02800000;
- ExecutableAllocator::cacheFlush(insn, sizeof(ARMWord));
+ return reinterpret_cast<intptr_t>(base) + offset - sizeof(ARMWord);
}
- // Linkers
-
- void linkJump(JmpSrc from, JmpDst to)
+ void linkJump(AssemblerLabel from, AssemblerLabel to)
{
- ARMWord* insn = reinterpret_cast<ARMWord*>(m_buffer.data()) + (from.m_offset / sizeof(ARMWord));
+ ARMWord* insn = reinterpret_cast<ARMWord*>(getAbsoluteJumpAddress(m_buffer.data(), from.m_offset));
ARMWord* addr = getLdrImmAddressOnPool(insn, m_buffer.poolAddress());
*addr = static_cast<ARMWord>(to.m_offset);
}
- static void linkJump(void* code, JmpSrc from, void* to)
+ static void linkJump(void* code, AssemblerLabel from, void* to)
{
- patchPointerInternal(reinterpret_cast<intptr_t>(code) + from.m_offset, to);
+ patchPointerInternal(getAbsoluteJumpAddress(code, from.m_offset), to);
}
static void relinkJump(void* from, void* to)
{
- patchPointerInternal(reinterpret_cast<intptr_t>(from) - sizeof(ARMWord), to);
+ patchPointerInternal(getAbsoluteJumpAddress(from), to);
}
- static void linkCall(void* code, JmpSrc from, void* to)
+ static void linkCall(void* code, AssemblerLabel from, void* to)
{
- patchPointerInternal(reinterpret_cast<intptr_t>(code) + from.m_offset, to);
+ patchPointerInternal(getAbsoluteJumpAddress(code, from.m_offset), to);
}
static void relinkCall(void* from, void* to)
{
- patchPointerInternal(reinterpret_cast<intptr_t>(from) - sizeof(ARMWord), to);
+ patchPointerInternal(getAbsoluteJumpAddress(from), to);
}
- // Address operations
-
- static void* getRelocatedAddress(void* code, JmpSrc jump)
+ static void* readCallTarget(void* from)
{
- return reinterpret_cast<void*>(reinterpret_cast<ARMWord*>(code) + jump.m_offset / sizeof(ARMWord) + 1);
+ return reinterpret_cast<void*>(readPointer(reinterpret_cast<void*>(getAbsoluteJumpAddress(from))));
}
- static void* getRelocatedAddress(void* code, JmpDst label)
+ // Address operations
+
+ static void* getRelocatedAddress(void* code, AssemblerLabel label)
{
- return reinterpret_cast<void*>(reinterpret_cast<ARMWord*>(code) + label.m_offset / sizeof(ARMWord));
+ return reinterpret_cast<void*>(reinterpret_cast<char*>(code) + label.m_offset);
}
// Address differences
- static int getDifferenceBetweenLabels(JmpDst from, JmpSrc to)
- {
- return (to.m_offset + sizeof(ARMWord)) - from.m_offset;
- }
-
- static int getDifferenceBetweenLabels(JmpDst from, JmpDst to)
+ static int getDifferenceBetweenLabels(AssemblerLabel a, AssemblerLabel b)
{
- return to.m_offset - from.m_offset;
+ return b.m_offset - a.m_offset;
}
- static unsigned getCallReturnOffset(JmpSrc call)
+ static unsigned getCallReturnOffset(AssemblerLabel call)
{
- return call.m_offset + sizeof(ARMWord);
+ return call.m_offset;
}
// Handle immediates
void moveImm(ARMWord imm, int dest);
ARMWord encodeComplexImm(ARMWord imm, int dest);
+ ARMWord getOffsetForHalfwordDataTransfer(ARMWord imm, int tmpReg)
+ {
+ // Encode immediate data in the instruction if it is possible
+ if (imm <= 0xff)
+ return getOp2Byte(imm);
+ // Otherwise, store the data in a temporary register
+ return encodeComplexImm(imm, tmpReg);
+ }
+
// Memory load/store helpers
- void dataTransfer32(bool isLoad, RegisterID srcDst, RegisterID base, int32_t offset);
- void baseIndexTransfer32(bool isLoad, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset);
+ void dataTransfer32(bool isLoad, RegisterID srcDst, RegisterID base, int32_t offset, bool bytes = false);
+ void baseIndexTransfer32(bool isLoad, RegisterID srcDst, RegisterID base, RegisterID index, int scale, int32_t offset, bool bytes = false);
void doubleTransfer(bool isLoad, FPRegisterID srcDst, RegisterID base, int32_t offset);
// Constant pool hnadlers
return AL | B | (offset & BRANCH_MASK);
}
+#if OS(LINUX) && COMPILER(RVCT)
+ static __asm void cacheFlush(void* code, size_t);
+#else
+ static void cacheFlush(void* code, size_t size)
+ {
+#if OS(LINUX) && COMPILER(GCC)
+ uintptr_t currentPage = reinterpret_cast<uintptr_t>(code) & ~(pageSize() - 1);
+ uintptr_t lastPage = (reinterpret_cast<uintptr_t>(code) + size) & ~(pageSize() - 1);
+ do {
+ asm volatile(
+ "push {r7}\n"
+ "mov r0, %0\n"
+ "mov r1, %1\n"
+ "mov r7, #0xf0000\n"
+ "add r7, r7, #0x2\n"
+ "mov r2, #0x0\n"
+ "svc 0x0\n"
+ "pop {r7}\n"
+ :
+ : "r" (currentPage), "r" (currentPage + pageSize())
+ : "r0", "r1", "r2");
+ currentPage += pageSize();
+ } while (lastPage >= currentPage);
+#elif OS(WINCE)
+ CacheRangeFlush(code, size, CACHE_SYNC_ALL);
+#elif OS(QNX) && ENABLE(ASSEMBLER_WX_EXCLUSIVE)
+ UNUSED_PARAM(code);
+ UNUSED_PARAM(size);
+#elif OS(QNX)
+ msync(code, size, MS_INVALIDATE_ICACHE);
+#else
+#error "The cacheFlush support is missing on this platform."
+#endif
+ }
+#endif
+
private:
ARMWord RM(int reg)
{