The writing in the book is really bad. Summarize a little bit.

Bytecode representation Bytecode

Bitwise description

The data layout used to represent the instruction binary code.

[A|G|op BBBB F|E|D|C]

• A letter represents 4bit data (a round of hexadecimal)

• op means 8bit operation code (one byte, two hexadecimal)

• Ø indicates that this bit must be zero in the instruction

• One space per 16 bits (one word)

• Vertical bars separate different data

Instruction format identification Format IDs

It is used to indicate the length and parameter type of the instruction.

N N a [s/i]

Additional data types:

grammar

A human-oriented instruction syntax defined with bit descriptions and format flags.

op vAA, string@BBBB

• comma separated parameters

• The instruction type op is the low byte of the first word. The upper eight bits can be parameters and can be empty

• vX is a register, and several Xs are the size of the register label (address)

• #+X constant

• +X is the relative (current instruction) address offset

• @X is the constant pool index. kind=[string, type, field, meth, site] site=call site

Register nomenclature

Let the function foo() use M registers and have N parameters.

command list

The table below contains a list of all directives.

original:https://source.android.com/devices/tech/dalvik/dalvik-bytecode.html#instructions

The second part of the instruction format ID in the first column of the above table, the corresponding (for people to see) instruction data layout list, see

original:https://source.android.com/devices/tech/dalvik/instruction-formats.html#formats

You don’t need to memorize the second form, just look it up when you use it.

accomplish

• The dalvik.bytecode.Opcodes interface defines a complete list of bytecodes.

  • The source code is located in libcore/dalvik/src/main/java/dalvik/bytecode/Opcodes.java

  • dalvik/opcode-gen/bytecode.txt is used to generate the instruction set of the above interface

  • The Android source code of kitkat version and below can be found at dalvik/vm/mterp/c Find the instruction implementation

  • lollipop and above, available at art/runtime/interpreter/mterp/ Find the corresponding instruction implementation (assembly) under [arch]

Take kitcat’s OP_MOVE.cpp as an example.

INST_A macro: Get the lower 4 bits of the upper eight bits of the instruction.

INST_B: Get the upper 4 bits of the upper eight bits.

GET_REGISTER: store value

fp[]: stack frame register, the local variable area of the function, storing all register values. Its index is the register number.

FINISH: Adjust the PC register and move to after this instruction.

underlying architecture

• Part of the registers are mapped to CPU registers, part is simulated by the call stack.

• The registers are all 32 bits, and the 64-bit data is represented by two adjacent registers.

• Register numbers v0-v65535 because operators are 16 bits (v0000-vFFFF)

  • Most functions use less than 16 registers, so registers below 16, numbered less than 1 byte, correspond to special instructions (faster?)

• The number of registers reserves the corresponding space when the function is called and allocates the call stack, and is specified by the .register command in the function header.