T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

aarch64: Migrate {s,u}{div,rem} to ISLE (#3572)

Browse Source 
* aarch64: Migrate {s,u}{div,rem} to ISLE

This commit migrates four different instructions at once to ISLE:

* `sdiv`
* `udiv`
* `srem`
* `urem`

These all share similar codegen and center around the `div` instruction
to use internally. The main feature of these was to model the manual
traps since the `div` instruction doesn't trap on overflow, instead
requiring manual checks to adhere to the semantics of the instruction
itself.

While I was here I went ahead and implemented an optimization for these
instructions when the right-hand-side is a constant with a known value.
For `udiv`, `srem`, and `urem` if the right-hand-side is a nonzero
constant then the checks for traps can be skipped entirely. For `sdiv`
if the constant is not 0 and not -1 then additionally all checks can be
elided. Finally if the right-hand-side of `sdiv` is -1 the zero-check is
elided, but it still needs a check for `i64::MIN` on the left-hand-side
and currently there's a TODO where `-1` is still checked too.

* Rebasing and review conflicts
This commit is contained in:
Alex Crichton
2021-12-13 17:27:11 -06:00
committed by GitHub
parent f1225dfd93
commit 20e090b114
12 changed files with 567 additions and 215 deletions
Download Patch File Download Diff File Expand all files Collapse all files

130
cranelift/codegen/src/isa/aarch64/lower_inst.rs
View File

@@ -75,135 +75,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
Opcode::Umulhi | Opcode::Smulhi => implemented_in_isle(ctx),
Opcode::Udiv | Opcode::Sdiv | Opcode::Urem | Opcode::Srem => {
let ty = ty.unwrap();
if ty.is_vector() || ty_bits(ty) > 64 {
return Err(CodegenError::Unsupported(format!(
"{}: Unsupported type: {:?}",
op, ty
)));
}
let is_signed = match op {
Opcode::Udiv | Opcode::Urem => false,
Opcode::Sdiv | Opcode::Srem => true,
_ => unreachable!(),
};
let is_rem = match op {
Opcode::Udiv | Opcode::Sdiv => false,
Opcode::Urem | Opcode::Srem => true,
_ => unreachable!(),
};
let narrow_mode = if is_signed {
NarrowValueMode::SignExtend64
} else {
NarrowValueMode::ZeroExtend64
};
// TODO: Add SDiv32 to implement 32-bit directly, rather
// than extending the input.
let div_op = if is_signed {
ALUOp::SDiv64
} else {
ALUOp::UDiv64
};
let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
let rn = put_input_in_reg(ctx, inputs[0], narrow_mode);
let rm = put_input_in_reg(ctx, inputs[1], narrow_mode);
// The div instruction does not trap on divide by zero or signed overflow
// so checks are inserted below.
//
// div rd, rn, rm
ctx.emit(Inst::AluRRR {
alu_op: div_op,
rd,
rn,
rm,
});
if is_rem {
// Remainder (rn % rm) is implemented as:
//
// tmp = rn / rm
// rd = rn - (tmp*rm)
//
// use 'rd' for tmp and you have:
//
// div rd, rn, rm ; rd = rn / rm
// cbnz rm, #8 ; branch over trap
// udf ; divide by zero
// msub rd, rd, rm, rn ; rd = rn - rd * rm
// Check for divide by 0.
let trap_code = TrapCode::IntegerDivisionByZero;
ctx.emit(Inst::TrapIf {
trap_code,
kind: CondBrKind::Zero(rm),
});
ctx.emit(Inst::AluRRRR {
alu_op: ALUOp3::MSub64,
rd,
rn: rd.to_reg(),
rm,
ra: rn,
});
} else {
if div_op == ALUOp::SDiv64 {
// cbnz rm, #8
// udf ; divide by zero
// cmn rm, 1
// ccmp rn, 1, #nzcv, eq
// b.vc #8
// udf ; signed overflow
// Check for divide by 0.
let trap_code = TrapCode::IntegerDivisionByZero;
ctx.emit(Inst::TrapIf {
trap_code,
kind: CondBrKind::Zero(rm),
});
// Check for signed overflow. The only case is min_value / -1.
// The following checks must be done in 32-bit or 64-bit, depending
// on the input type. Even though the initial div instruction is
// always done in 64-bit currently.
let size = OperandSize::from_ty(ty);
// Check RHS is -1.
ctx.emit(Inst::AluRRImm12 {
alu_op: choose_32_64(ty, ALUOp::AddS32, ALUOp::AddS64),
rd: writable_zero_reg(),
rn: rm,
imm12: Imm12::maybe_from_u64(1).unwrap(),
});
// Check LHS is min_value, by subtracting 1 and branching if
// there is overflow.
ctx.emit(Inst::CCmpImm {
size,
rn,
imm: UImm5::maybe_from_u8(1).unwrap(),
nzcv: NZCV::new(false, false, false, false),
cond: Cond::Eq,
});
let trap_code = TrapCode::IntegerOverflow;
ctx.emit(Inst::TrapIf {
trap_code,
kind: CondBrKind::Cond(Cond::Vs),
});
} else {
// cbnz rm, #8
// udf ; divide by zero
// Check for divide by 0.
let trap_code = TrapCode::IntegerDivisionByZero;
ctx.emit(Inst::TrapIf {
trap_code,
kind: CondBrKind::Zero(rm),
});
}
}
}
Opcode::Udiv | Opcode::Sdiv | Opcode::Urem | Opcode::Srem => implemented_in_isle(ctx),
Opcode::Uextend | Opcode::Sextend => {
let output_ty = ty.unwrap();