forked from OSchip/llvm-project
				
			
		
			
				
	
	
		
			376 lines
		
	
	
		
			10 KiB
		
	
	
	
		
			C++
		
	
	
	
			
		
		
	
	
			376 lines
		
	
	
		
			10 KiB
		
	
	
	
		
			C++
		
	
	
	
| //===-- ARMUtils.h ----------------------------------------------*- C++ -*-===//
 | |
| //
 | |
| //                     The LLVM Compiler Infrastructure
 | |
| //
 | |
| // This file is distributed under the University of Illinois Open Source
 | |
| // License. See LICENSE.TXT for details.
 | |
| //
 | |
| //===----------------------------------------------------------------------===//
 | |
| 
 | |
| #ifndef lldb_ARMUtils_h_
 | |
| #define lldb_ARMUtils_h_
 | |
| 
 | |
| #include "ARMDefines.h"
 | |
| #include "InstructionUtils.h"
 | |
| #include "llvm/Support/MathExtras.h" // for SignExtend64 template function
 | |
| 
 | |
| // Common utilities for the ARM/Thumb Instruction Set Architecture.
 | |
| 
 | |
| namespace lldb_private {
 | |
| 
 | |
| static inline uint32_t Align(uint32_t val, uint32_t alignment) {
 | |
|   return alignment * (val / alignment);
 | |
| }
 | |
| 
 | |
| static inline uint32_t DecodeImmShift(const uint32_t type, const uint32_t imm5,
 | |
|                                       ARM_ShifterType &shift_t) {
 | |
|   switch (type) {
 | |
|   default:
 | |
|   // assert(0 && "Invalid shift type");
 | |
|   case 0:
 | |
|     shift_t = SRType_LSL;
 | |
|     return imm5;
 | |
|   case 1:
 | |
|     shift_t = SRType_LSR;
 | |
|     return (imm5 == 0 ? 32 : imm5);
 | |
|   case 2:
 | |
|     shift_t = SRType_ASR;
 | |
|     return (imm5 == 0 ? 32 : imm5);
 | |
|   case 3:
 | |
|     if (imm5 == 0) {
 | |
|       shift_t = SRType_RRX;
 | |
|       return 1;
 | |
|     } else {
 | |
|       shift_t = SRType_ROR;
 | |
|       return imm5;
 | |
|     }
 | |
|   }
 | |
|   shift_t = SRType_Invalid;
 | |
|   return UINT32_MAX;
 | |
| }
 | |
| 
 | |
| // A8.6.35 CMP (register) -- Encoding T3
 | |
| // Convenience function.
 | |
| static inline uint32_t DecodeImmShiftThumb(const uint32_t opcode,
 | |
|                                            ARM_ShifterType &shift_t) {
 | |
|   return DecodeImmShift(Bits32(opcode, 5, 4),
 | |
|                         Bits32(opcode, 14, 12) << 2 | Bits32(opcode, 7, 6),
 | |
|                         shift_t);
 | |
| }
 | |
| 
 | |
| // A8.6.35 CMP (register) -- Encoding A1
 | |
| // Convenience function.
 | |
| static inline uint32_t DecodeImmShiftARM(const uint32_t opcode,
 | |
|                                          ARM_ShifterType &shift_t) {
 | |
|   return DecodeImmShift(Bits32(opcode, 6, 5), Bits32(opcode, 11, 7), shift_t);
 | |
| }
 | |
| 
 | |
| static inline uint32_t DecodeImmShift(const ARM_ShifterType shift_t,
 | |
|                                       const uint32_t imm5) {
 | |
|   ARM_ShifterType dont_care;
 | |
|   return DecodeImmShift(shift_t, imm5, dont_care);
 | |
| }
 | |
| 
 | |
| static inline ARM_ShifterType DecodeRegShift(const uint32_t type) {
 | |
|   switch (type) {
 | |
|   default:
 | |
|     // assert(0 && "Invalid shift type");
 | |
|     return SRType_Invalid;
 | |
|   case 0:
 | |
|     return SRType_LSL;
 | |
|   case 1:
 | |
|     return SRType_LSR;
 | |
|   case 2:
 | |
|     return SRType_ASR;
 | |
|   case 3:
 | |
|     return SRType_ROR;
 | |
|   }
 | |
| }
 | |
| 
 | |
| static inline uint32_t LSL_C(const uint32_t value, const uint32_t amount,
 | |
|                              uint32_t &carry_out, bool *success) {
 | |
|   if (amount == 0) {
 | |
|     *success = false;
 | |
|     return 0;
 | |
|   }
 | |
|   *success = true;
 | |
|   carry_out = amount <= 32 ? Bit32(value, 32 - amount) : 0;
 | |
|   return value << amount;
 | |
| }
 | |
| 
 | |
| static inline uint32_t LSL(const uint32_t value, const uint32_t amount,
 | |
|                            bool *success) {
 | |
|   *success = true;
 | |
|   if (amount == 0)
 | |
|     return value;
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = LSL_C(value, amount, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t LSR_C(const uint32_t value, const uint32_t amount,
 | |
|                              uint32_t &carry_out, bool *success) {
 | |
|   if (amount == 0) {
 | |
|     *success = false;
 | |
|     return 0;
 | |
|   }
 | |
|   *success = true;
 | |
|   carry_out = amount <= 32 ? Bit32(value, amount - 1) : 0;
 | |
|   return value >> amount;
 | |
| }
 | |
| 
 | |
| static inline uint32_t LSR(const uint32_t value, const uint32_t amount,
 | |
|                            bool *success) {
 | |
|   *success = true;
 | |
|   if (amount == 0)
 | |
|     return value;
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = LSR_C(value, amount, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t ASR_C(const uint32_t value, const uint32_t amount,
 | |
|                              uint32_t &carry_out, bool *success) {
 | |
|   if (amount == 0 || amount > 32) {
 | |
|     *success = false;
 | |
|     return 0;
 | |
|   }
 | |
|   *success = true;
 | |
|   bool negative = BitIsSet(value, 31);
 | |
|   if (amount <= 32) {
 | |
|     carry_out = Bit32(value, amount - 1);
 | |
|     int64_t extended = llvm::SignExtend64<32>(value);
 | |
|     return UnsignedBits(extended, amount + 31, amount);
 | |
|   } else {
 | |
|     carry_out = (negative ? 1 : 0);
 | |
|     return (negative ? 0xffffffff : 0);
 | |
|   }
 | |
| }
 | |
| 
 | |
| static inline uint32_t ASR(const uint32_t value, const uint32_t amount,
 | |
|                            bool *success) {
 | |
|   *success = true;
 | |
|   if (amount == 0)
 | |
|     return value;
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = ASR_C(value, amount, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t ROR_C(const uint32_t value, const uint32_t amount,
 | |
|                              uint32_t &carry_out, bool *success) {
 | |
|   if (amount == 0) {
 | |
|     *success = false;
 | |
|     return 0;
 | |
|   }
 | |
|   *success = true;
 | |
|   uint32_t amt = amount % 32;
 | |
|   uint32_t result = Rotr32(value, amt);
 | |
|   carry_out = Bit32(value, 31);
 | |
|   return result;
 | |
| }
 | |
| 
 | |
| static inline uint32_t ROR(const uint32_t value, const uint32_t amount,
 | |
|                            bool *success) {
 | |
|   *success = true;
 | |
|   if (amount == 0)
 | |
|     return value;
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = ROR_C(value, amount, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t RRX_C(const uint32_t value, const uint32_t carry_in,
 | |
|                              uint32_t &carry_out, bool *success) {
 | |
|   *success = true;
 | |
|   carry_out = Bit32(value, 0);
 | |
|   return Bit32(carry_in, 0) << 31 | Bits32(value, 31, 1);
 | |
| }
 | |
| 
 | |
| static inline uint32_t RRX(const uint32_t value, const uint32_t carry_in,
 | |
|                            bool *success) {
 | |
|   *success = true;
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = RRX_C(value, carry_in, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t Shift_C(const uint32_t value, ARM_ShifterType type,
 | |
|                                const uint32_t amount, const uint32_t carry_in,
 | |
|                                uint32_t &carry_out, bool *success) {
 | |
|   if (type == SRType_RRX && amount != 1) {
 | |
|     *success = false;
 | |
|     return 0;
 | |
|   }
 | |
|   *success = true;
 | |
| 
 | |
|   if (amount == 0) {
 | |
|     carry_out = carry_in;
 | |
|     return value;
 | |
|   }
 | |
|   uint32_t result;
 | |
|   switch (type) {
 | |
|   case SRType_LSL:
 | |
|     result = LSL_C(value, amount, carry_out, success);
 | |
|     break;
 | |
|   case SRType_LSR:
 | |
|     result = LSR_C(value, amount, carry_out, success);
 | |
|     break;
 | |
|   case SRType_ASR:
 | |
|     result = ASR_C(value, amount, carry_out, success);
 | |
|     break;
 | |
|   case SRType_ROR:
 | |
|     result = ROR_C(value, amount, carry_out, success);
 | |
|     break;
 | |
|   case SRType_RRX:
 | |
|     result = RRX_C(value, carry_in, carry_out, success);
 | |
|     break;
 | |
|   default:
 | |
|     *success = false;
 | |
|     break;
 | |
|   }
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t Shift(const uint32_t value, ARM_ShifterType type,
 | |
|                              const uint32_t amount, const uint32_t carry_in,
 | |
|                              bool *success) {
 | |
|   // Don't care about carry out in this case.
 | |
|   uint32_t dont_care;
 | |
|   uint32_t result = Shift_C(value, type, amount, carry_in, dont_care, success);
 | |
|   if (*success)
 | |
|     return result;
 | |
|   else
 | |
|     return 0;
 | |
| }
 | |
| 
 | |
| static inline uint32_t bits(const uint32_t val, const uint32_t msbit,
 | |
|                             const uint32_t lsbit) {
 | |
|   return Bits32(val, msbit, lsbit);
 | |
| }
 | |
| 
 | |
| static inline uint32_t bit(const uint32_t val, const uint32_t msbit) {
 | |
|   return bits(val, msbit, msbit);
 | |
| }
 | |
| 
 | |
| static uint32_t ror(uint32_t val, uint32_t N, uint32_t shift) {
 | |
|   uint32_t m = shift % N;
 | |
|   return (val >> m) | (val << (N - m));
 | |
| }
 | |
| 
 | |
| // (imm32, carry_out) = ARMExpandImm_C(imm12, carry_in)
 | |
| static inline uint32_t ARMExpandImm_C(uint32_t opcode, uint32_t carry_in,
 | |
|                                       uint32_t &carry_out) {
 | |
|   uint32_t imm32;                         // the expanded result
 | |
|   uint32_t imm = bits(opcode, 7, 0);      // immediate value
 | |
|   uint32_t amt = 2 * bits(opcode, 11, 8); // rotate amount
 | |
|   if (amt == 0) {
 | |
|     imm32 = imm;
 | |
|     carry_out = carry_in;
 | |
|   } else {
 | |
|     imm32 = ror(imm, 32, amt);
 | |
|     carry_out = Bit32(imm32, 31);
 | |
|   }
 | |
|   return imm32;
 | |
| }
 | |
| 
 | |
| static inline uint32_t ARMExpandImm(uint32_t opcode) {
 | |
|   // 'carry_in' argument to following function call does not affect the imm32
 | |
|   // result.
 | |
|   uint32_t carry_in = 0;
 | |
|   uint32_t carry_out;
 | |
|   return ARMExpandImm_C(opcode, carry_in, carry_out);
 | |
| }
 | |
| 
 | |
| // (imm32, carry_out) = ThumbExpandImm_C(imm12, carry_in)
 | |
| static inline uint32_t ThumbExpandImm_C(uint32_t opcode, uint32_t carry_in,
 | |
|                                         uint32_t &carry_out) {
 | |
|   uint32_t imm32; // the expanded result
 | |
|   const uint32_t i = bit(opcode, 26);
 | |
|   const uint32_t imm3 = bits(opcode, 14, 12);
 | |
|   const uint32_t abcdefgh = bits(opcode, 7, 0);
 | |
|   const uint32_t imm12 = i << 11 | imm3 << 8 | abcdefgh;
 | |
| 
 | |
|   if (bits(imm12, 11, 10) == 0) {
 | |
|     switch (bits(imm12, 9, 8)) {
 | |
|     default: // Keep static analyzer happy with a default case
 | |
|     case 0:
 | |
|       imm32 = abcdefgh;
 | |
|       break;
 | |
| 
 | |
|     case 1:
 | |
|       imm32 = abcdefgh << 16 | abcdefgh;
 | |
|       break;
 | |
| 
 | |
|     case 2:
 | |
|       imm32 = abcdefgh << 24 | abcdefgh << 8;
 | |
|       break;
 | |
| 
 | |
|     case 3:
 | |
|       imm32 = abcdefgh << 24 | abcdefgh << 16 | abcdefgh << 8 | abcdefgh;
 | |
|       break;
 | |
|     }
 | |
|     carry_out = carry_in;
 | |
|   } else {
 | |
|     const uint32_t unrotated_value = 0x80 | bits(imm12, 6, 0);
 | |
|     imm32 = ror(unrotated_value, 32, bits(imm12, 11, 7));
 | |
|     carry_out = Bit32(imm32, 31);
 | |
|   }
 | |
|   return imm32;
 | |
| }
 | |
| 
 | |
| static inline uint32_t ThumbExpandImm(uint32_t opcode) {
 | |
|   // 'carry_in' argument to following function call does not affect the imm32
 | |
|   // result.
 | |
|   uint32_t carry_in = 0;
 | |
|   uint32_t carry_out;
 | |
|   return ThumbExpandImm_C(opcode, carry_in, carry_out);
 | |
| }
 | |
| 
 | |
| // imm32 = ZeroExtend(i:imm3:imm8, 32)
 | |
| static inline uint32_t ThumbImm12(uint32_t opcode) {
 | |
|   const uint32_t i = bit(opcode, 26);
 | |
|   const uint32_t imm3 = bits(opcode, 14, 12);
 | |
|   const uint32_t imm8 = bits(opcode, 7, 0);
 | |
|   const uint32_t imm12 = i << 11 | imm3 << 8 | imm8;
 | |
|   return imm12;
 | |
| }
 | |
| 
 | |
| // imm32 = ZeroExtend(imm7:'00', 32)
 | |
| static inline uint32_t ThumbImm7Scaled(uint32_t opcode) {
 | |
|   const uint32_t imm7 = bits(opcode, 6, 0);
 | |
|   return imm7 * 4;
 | |
| }
 | |
| 
 | |
| // imm32 = ZeroExtend(imm8:'00', 32)
 | |
| static inline uint32_t ThumbImm8Scaled(uint32_t opcode) {
 | |
|   const uint32_t imm8 = bits(opcode, 7, 0);
 | |
|   return imm8 * 4;
 | |
| }
 | |
| 
 | |
| // This function performs the check for the register numbers 13 and 15 that are
 | |
| // not permitted for many Thumb register specifiers.
 | |
| static inline bool BadReg(uint32_t n) { return n == 13 || n == 15; }
 | |
| 
 | |
| } // namespace lldb_private
 | |
| 
 | |
| #endif // lldb_ARMUtils_h_
 |