Crate asm_riscv

Crate asm_riscv 

Source
Expand description

RISC-V 32-Bit Base Integer Instructions to compile into, with optional extensions that can be enabled.

  • RV32M Multiply Extension
  • RV32A Atomic Extension
  • RV32F Single-Precision Floating Point Extension
  • RV32D Double-Precision Floating Point Extension

§Ignored For Now

  • RV32Q Quadruple-Precision Floating Point Extension
  • RV32C Compression 16-bit Instructions
  • All 64 bit extensions

May be ported to other platforms with assembly translators.

§RISC-V Compiler Notes

This is intended to help anyone who needs it, if they’re building a compiler to RISC-V.

  • For speed: Make sure all loads and stores are 32-bit aligned
  • For shifts: Shifting the width of the register is a no-op, to clear register $R use ADDI $R, $ZERO, 0.

§Psuedo-Instructions

  • nop: addi $zero, $zero, 0
  • mv $d, $s: addi $d, $s, 0
  • not $d, $s: ori $d, $s, -1
  • neg $d, $s: sub $d, $zero, $s
  • j offset: jal $zero, offset (unconditional jump)
  • jal offset: jal $ra, offset (near function call)
  • call offset: (far function call)
auipc $ra, offset[31:12] + offset[11]
jalr $ra, offset[11:0]($ra)
  • ret: jalr $ra, 0($ra)
  • beqz $r, offset: beq $r, $zero, offset
  • bnez $r, offset: bne $r, $zero, offset
  • bgez $r, offset: bge $r, $zero, offset
  • bltz $r, offset: blt $r, $zero, offset
  • bgt $r1, $r2, offset: blt $r2, $r1, offset
  • ble $r1, $r2, offset: bge $r2, $r1, offset
  • fence: fence IORW, IORW
  • li $d, im: addi $d, $zero, im (set immediate)
  • li $d, im:
lui $d, im[31:12] + im[11]
addi $d, $zero, im[11:0]
  • la $d, symbol:
auipc $d, delta[31:12] + delta[11]
addi $d, $d, delta[11:0]
  • lw $d, symbol:
auipc $d, delta[31:12] + delta[11]
lw $d, $d, delta[11:0](rd)
  • sw $d, symbol, $t:
auipc $t, delta[31:12] + delta[11]
sw $d, $d, delta[11:0]($t)
  • seqz $d, $s: sltiu $d, $s, 0
  • snez $d, $s: sltu $d, $zero, $s Custom Pseudo-Instructions
  • zero $d: addi $r, $zero, 0 (set register to zero)
  • slt $d, $a, $b, $s: (with multiply cpu feature enabled)
# Option 1:
slt $d, $a, $b
mul $d, $d, $s
  • slt $d, $a, $b, $s: (without multiply cpu feature enabled)
# Option 1 (branching, ugh):
blt $a, $b, 12 # 12: ④
addi $d, $zero, 0
jal $zero 8 # Skip next instruction
add $d, $zero, $s # ④

# Option 2 (might be faster for without AND with multiply extension):
slt $d, $a, $b
slli $d, $d, 31
srai $d, $d, 31
and $d, $d, $s
  • slt $d, $a, $b, $s, $e: ($d: if $a < $b { $s } else { $e })
sub $s, $s, $e
slt $d, $a, $b, $s
add $d, $d, $e
add $s, $s, $e # can be elimated if s is dropped
  • slt $d, $a, $b, $s, $e, $t: ($d: if $a < $b { $s } else { $e })
sub $t, $s, $e
slt $d, $a, $b, $t
add $d, $d, $e

Enums§

ConversionError
Error types when converting u32 to I
I
An assembly instruction (im is limited to 12 bits)
Reg
A RISC-V Register