riscemu/docs/assembly.md

Assembly

Assembly tokenization should be working completely. It knows what instructions the CPU implementation supports and parses based on them.

Instructions

• all loads/stores: lb, lh, lw, lbu, lhu, sw, sh, sb
  • supported arg format is either rd, imm(reg) or rd, reg, imm
• all branch statements: beq, bne, blt, bge, bltu, bgeu
• all jumps j, jal, jalr, ret
• basic arithmetic: add, addi, sub (not lui, auipc)
• shifts: sll, slli, srl, srli, sra, srai
• scall, ecall, sbreak, ebreak (both s and e version are the same instruction)
• compares (non immediate): slt, sltu (not slti, sltiu)
• logical (non immediate): and, or, xor (not andi, ori, xori)

All unimplemented instructions can be added quite easily

Pseudo-ops

The following pseudo-ops are implemented as of yet:

• .space <len> reverse bytes of zero
• .ascii 'text' put text into memory
• .asciiz 'text' put text into memory (null terminated)
• .section .<name> same as .<name>, see sections
• .set <name>, <value> to create a const symbol with a given value
• .global <name> mark symbol <name> as a global symbol. It is available from all loaded programs

Sections:

Currently only these three sections are supported:

• data read-write data (non-executable)
• rodata read-only data (non-executable)
• text executable data (read-only)

Allocating stack

another pseudo-op is recognized: .stack <len>. This marks the executable as requesting at least <len> bytes of stack. If the loader respects this wish, the sp is initialized pointing to the end of the stack.