stage2 AArch64: add basic assertions to bits.zig for correct codegen

Includes many fixes of errors discovered by adding these assertions
This commit is contained in:
joachimschmidt557 2022-04-15 22:58:54 +02:00
parent 9c2cbe39c2
commit c78daeb642
No known key found for this signature in database
GPG Key ID: E0B575BE2884ACC5
3 changed files with 126 additions and 69 deletions

View File

@ -376,7 +376,7 @@ fn gen(self: *Self) !void {
// mov fp, sp
_ = try self.addInst(.{
.tag = .mov_to_from_sp,
.data = .{ .rr = .{ .rd = .x29, .rn = .xzr } },
.data = .{ .rr = .{ .rd = .x29, .rn = .sp } },
});
// sub sp, sp, #reloc
@ -421,7 +421,7 @@ fn gen(self: *Self) !void {
if (math.cast(u12, stack_size)) |size| {
self.mir_instructions.set(backpatch_reloc, .{
.tag = .sub_immediate,
.data = .{ .rr_imm12_sh = .{ .rd = .xzr, .rn = .xzr, .imm12 = size } },
.data = .{ .rr_imm12_sh = .{ .rd = .sp, .rn = .sp, .imm12 = size } },
});
} else |_| {
return self.failSymbol("TODO AArch64: allow larger stacks", .{});
@ -453,7 +453,7 @@ fn gen(self: *Self) !void {
// add sp, sp, #stack_size
_ = try self.addInst(.{
.tag = .add_immediate,
.data = .{ .rr_imm12_sh = .{ .rd = .xzr, .rn = .xzr, .imm12 = @intCast(u12, stack_size) } },
.data = .{ .rr_imm12_sh = .{ .rd = .sp, .rn = .sp, .imm12 = @intCast(u12, stack_size) } },
});
// <load other registers>
@ -882,7 +882,8 @@ fn spillCompareFlagsIfOccupied(self: *Self) !void {
/// allocated. A second call to `copyToTmpRegister` may return the same register.
/// This can have a side effect of spilling instructions to the stack to free up a register.
fn copyToTmpRegister(self: *Self, ty: Type, mcv: MCValue) !Register {
const reg = try self.register_manager.allocReg(null);
const raw_reg = try self.register_manager.allocReg(null);
const reg = registerAlias(raw_reg, ty.abiSize(self.target.*));
try self.genSetReg(ty, reg, mcv);
return reg;
}
@ -891,7 +892,9 @@ fn copyToTmpRegister(self: *Self, ty: Type, mcv: MCValue) !Register {
/// `reg_owner` is the instruction that gets associated with the register in the register table.
/// This can have a side effect of spilling instructions to the stack to free up a register.
fn copyToNewRegister(self: *Self, reg_owner: Air.Inst.Index, mcv: MCValue) !MCValue {
const reg = try self.register_manager.allocReg(reg_owner);
const raw_reg = try self.register_manager.allocReg(reg_owner);
const ty = self.air.typeOfIndex(reg_owner);
const reg = registerAlias(raw_reg, ty.abiSize(self.target.*));
try self.genSetReg(self.air.typeOfIndex(reg_owner), reg, mcv);
return MCValue{ .register = reg };
}
@ -1003,7 +1006,8 @@ fn airNot(self: *Self, inst: Air.Inst.Index) !void {
break :blk op_reg;
}
break :blk try self.register_manager.allocReg(null);
const raw_reg = try self.register_manager.allocReg(null);
break :blk raw_reg.to32();
};
_ = try self.addInst(.{
@ -1013,7 +1017,7 @@ fn airNot(self: *Self, inst: Air.Inst.Index) !void {
.rn = op_reg,
.imms = 0b000000,
.immr = 0b000000,
.n = 0b1,
.n = 0b0,
} },
});
@ -1035,7 +1039,8 @@ fn airNot(self: *Self, inst: Air.Inst.Index) !void {
break :blk op_reg;
}
break :blk try self.register_manager.allocReg(null);
const raw_reg = try self.register_manager.allocReg(null);
break :blk registerAlias(raw_reg, operand_ty.abiSize(self.target.*));
};
_ = try self.addInst(.{
@ -1124,7 +1129,8 @@ fn binOpRegister(
break :inst Air.refToIndex(bin_op.lhs).?;
} else null;
const reg = try self.register_manager.allocReg(track_inst);
const raw_reg = try self.register_manager.allocReg(track_inst);
const reg = registerAlias(raw_reg, lhs_ty.abiSize(self.target.*));
self.register_manager.freezeRegs(&.{reg});
if (track_inst) |inst| branch.inst_table.putAssumeCapacity(inst, .{ .register = reg });
@ -1139,7 +1145,8 @@ fn binOpRegister(
break :inst Air.refToIndex(bin_op.rhs).?;
} else null;
const reg = try self.register_manager.allocReg(track_inst);
const raw_reg = try self.register_manager.allocReg(track_inst);
const reg = registerAlias(raw_reg, rhs_ty.abiAlignment(self.target.*));
self.register_manager.freezeRegs(&.{reg});
if (track_inst) |inst| branch.inst_table.putAssumeCapacity(inst, .{ .register = reg });
@ -1156,7 +1163,8 @@ fn binOpRegister(
} else if (rhs_is_register and self.reuseOperand(inst, bin_op.rhs, 1, rhs)) {
break :blk rhs_reg;
} else {
break :blk try self.register_manager.allocReg(inst);
const raw_reg = try self.register_manager.allocReg(inst);
break :blk registerAlias(raw_reg, lhs_ty.abiSize(self.target.*));
}
} else try self.register_manager.allocReg(null);
@ -1276,7 +1284,8 @@ fn binOpImmediate(
).?;
} else null;
const reg = try self.register_manager.allocReg(track_inst);
const raw_reg = try self.register_manager.allocReg(track_inst);
const reg = registerAlias(raw_reg, lhs_ty.abiSize(self.target.*));
self.register_manager.freezeRegs(&.{reg});
if (track_inst) |inst| branch.inst_table.putAssumeCapacity(inst, .{ .register = reg });
@ -1298,7 +1307,8 @@ fn binOpImmediate(
)) {
break :blk lhs_reg;
} else {
break :blk try self.register_manager.allocReg(inst);
const raw_reg = try self.register_manager.allocReg(inst);
break :blk registerAlias(raw_reg, lhs_ty.abiSize(self.target.*));
}
} else try self.register_manager.allocReg(null),
};
@ -1965,7 +1975,8 @@ fn load(self: *Self, dst_mcv: MCValue, ptr: MCValue, ptr_ty: Type) InnerError!vo
},
.stack_offset => |off| {
if (elem_size <= 8) {
const tmp_reg = try self.register_manager.allocReg(null);
const raw_tmp_reg = try self.register_manager.allocReg(null);
const tmp_reg = registerAlias(raw_tmp_reg, elem_size);
self.register_manager.freezeRegs(&.{tmp_reg});
defer self.register_manager.unfreezeRegs(&.{tmp_reg});
@ -2001,12 +2012,8 @@ fn load(self: *Self, dst_mcv: MCValue, ptr: MCValue, ptr_ty: Type) InnerError!vo
.got_load,
.direct_load,
=> {
const reg = try self.register_manager.allocReg(null);
self.register_manager.freezeRegs(&.{reg});
defer self.register_manager.unfreezeRegs(&.{reg});
try self.genSetReg(ptr_ty, reg, ptr);
try self.load(dst_mcv, .{ .register = reg }, ptr_ty);
const addr_reg = try self.copyToTmpRegister(ptr_ty, ptr);
try self.load(dst_mcv, .{ .register = addr_reg }, ptr_ty);
},
}
}
@ -2091,6 +2098,7 @@ fn genInlineMemcpy(
fn airLoad(self: *Self, inst: Air.Inst.Index) !void {
const ty_op = self.air.instructions.items(.data)[inst].ty_op;
const elem_ty = self.air.typeOfIndex(inst);
const elem_size = elem_ty.abiSize(self.target.*);
const result: MCValue = result: {
if (!elem_ty.hasRuntimeBits())
break :result MCValue.none;
@ -2101,9 +2109,12 @@ fn airLoad(self: *Self, inst: Air.Inst.Index) !void {
break :result MCValue.dead;
const dst_mcv: MCValue = blk: {
if (self.reuseOperand(inst, ty_op.operand, 0, ptr)) {
if (elem_size <= 8 and self.reuseOperand(inst, ty_op.operand, 0, ptr)) {
// The MCValue that holds the pointer can be re-used as the value.
break :blk ptr;
break :blk switch (ptr) {
.register => |r| MCValue{ .register = registerAlias(r, elem_size) },
else => ptr,
};
} else {
break :blk try self.allocRegOrMem(inst, true);
}
@ -2209,6 +2220,8 @@ fn genStrRegister(self: *Self, value_reg: Register, addr_reg: Register, abi_size
}
fn store(self: *Self, ptr: MCValue, value: MCValue, ptr_ty: Type, value_ty: Type) InnerError!void {
const abi_size = value_ty.abiSize(self.target.*);
switch (ptr) {
.none => unreachable,
.undef => unreachable,
@ -2226,14 +2239,14 @@ fn store(self: *Self, ptr: MCValue, value: MCValue, ptr_ty: Type, value_ty: Type
self.register_manager.freezeRegs(&.{addr_reg});
defer self.register_manager.unfreezeRegs(&.{addr_reg});
const abi_size = value_ty.abiSize(self.target.*);
switch (value) {
.register => |value_reg| {
try self.genStrRegister(value_reg, addr_reg, abi_size);
},
else => {
if (abi_size <= 8) {
const tmp_reg = try self.register_manager.allocReg(null);
const raw_tmp_reg = try self.register_manager.allocReg(null);
const tmp_reg = registerAlias(raw_tmp_reg, abi_size);
self.register_manager.freezeRegs(&.{tmp_reg});
defer self.register_manager.unfreezeRegs(&.{tmp_reg});
@ -3522,8 +3535,8 @@ fn genSetReg(self: *Self, ty: Type, reg: Register, mcv: MCValue) InnerError!void
.memory => |addr| {
// The value is in memory at a hard-coded address.
// If the type is a pointer, it means the pointer address is at this memory location.
try self.genSetReg(ty, reg, .{ .immediate = addr });
try self.genLdrRegister(reg, reg, ty.abiSize(self.target.*));
try self.genSetReg(ty, reg.to64(), .{ .immediate = addr });
try self.genLdrRegister(reg, reg.to64(), ty.abiSize(self.target.*));
},
.stack_offset => |off| {
const abi_size = ty.abiSize(self.target.*);
@ -3998,6 +4011,12 @@ fn resolveCallingConventionValues(self: *Self, fn_ty: Type) !CallMCValues {
var nsaa: u32 = 0; // Next stacked argument address
for (param_types) |ty, i| {
const param_size = @intCast(u32, ty.abiSize(self.target.*));
if (param_size == 0) {
result.args[i] = .{ .none = {} };
continue;
}
// We round up NCRN only for non-Apple platforms which allow the 16-byte aligned
// values to spread across odd-numbered registers.
if (ty.abiAlignment(self.target.*) == 16 and !self.target.isDarwin()) {
@ -4005,10 +4024,9 @@ fn resolveCallingConventionValues(self: *Self, fn_ty: Type) !CallMCValues {
ncrn += ncrn % 2;
}
const param_size = @intCast(u32, ty.abiSize(self.target.*));
if (std.math.divCeil(u32, param_size, 8) catch unreachable <= 8 - ncrn) {
if (param_size <= 8) {
result.args[i] = .{ .register = c_abi_int_param_regs[ncrn] };
result.args[i] = .{ .register = registerAlias(c_abi_int_param_regs[ncrn], param_size) };
ncrn += 1;
} else {
return self.fail("TODO MCValues with multiple registers", .{});
@ -4045,7 +4063,7 @@ fn resolveCallingConventionValues(self: *Self, fn_ty: Type) !CallMCValues {
.Unspecified, .C => {
const ret_ty_size = @intCast(u32, ret_ty.abiSize(self.target.*));
if (ret_ty_size <= 8) {
result.return_value = .{ .register = c_abi_int_return_regs[0] };
result.return_value = .{ .register = registerAlias(c_abi_int_return_regs[0], ret_ty_size) };
} else {
return self.fail("TODO support more return types for ARM backend", .{});
}

View File

@ -457,8 +457,13 @@ fn mirAddSubtractImmediate(emit: *Emit, inst: Mir.Inst.Index) !void {
const rn = r_imm12_sh.rn;
const imm12 = r_imm12_sh.imm12;
const sh = r_imm12_sh.sh == 1;
const zr: Register = switch (rn.size()) {
32 => .wzr,
64 => .xzr,
else => unreachable,
};
try emit.writeInstruction(Instruction.subs(.xzr, rn, imm12, sh));
try emit.writeInstruction(Instruction.subs(zr, rn, imm12, sh));
},
else => unreachable,
}
@ -674,8 +679,13 @@ fn mirAddSubtractShiftedRegister(emit: *Emit, inst: Mir.Inst.Index) !void {
const rm = rr_imm6_shift.rm;
const shift = rr_imm6_shift.shift;
const imm6 = rr_imm6_shift.imm6;
const zr: Register = switch (rn.size()) {
32 => .wzr,
64 => .xzr,
else => unreachable,
};
try emit.writeInstruction(Instruction.subsShiftedRegister(.xzr, rn, rm, shift, imm6));
try emit.writeInstruction(Instruction.subsShiftedRegister(zr, rn, rm, shift, imm6));
},
else => unreachable,
}
@ -686,7 +696,12 @@ fn mirConditionalSelect(emit: *Emit, inst: Mir.Inst.Index) !void {
switch (tag) {
.cset => {
const r_cond = emit.mir.instructions.items(.data)[inst].r_cond;
try emit.writeInstruction(Instruction.csinc(r_cond.rd, .xzr, .xzr, r_cond.cond));
const zr: Register = switch (r_cond.rd.size()) {
32 => .wzr,
64 => .xzr,
else => unreachable,
};
try emit.writeInstruction(Instruction.csinc(r_cond.rd, zr, zr, r_cond.cond));
},
else => unreachable,
}
@ -718,14 +733,14 @@ fn mirLoadMemoryPie(emit: *Emit, inst: Mir.Inst.Index) !void {
// PC-relative displacement to the entry in memory.
// adrp
const offset = @intCast(u32, emit.code.items.len);
try emit.writeInstruction(Instruction.adrp(reg, 0));
try emit.writeInstruction(Instruction.adrp(reg.to64(), 0));
switch (tag) {
.load_memory_got => {
// ldr reg, reg, offset
try emit.writeInstruction(Instruction.ldr(
reg,
reg,
reg.to64(),
Instruction.LoadStoreOffset.imm(0),
));
},
@ -739,11 +754,11 @@ fn mirLoadMemoryPie(emit: *Emit, inst: Mir.Inst.Index) !void {
// Note that this can potentially be optimised out by the codegen/linker if the
// target address is appropriately aligned.
// add reg, reg, offset
try emit.writeInstruction(Instruction.add(reg, reg, 0, false));
try emit.writeInstruction(Instruction.add(reg.to64(), reg.to64(), 0, false));
// ldr reg, reg, offset
try emit.writeInstruction(Instruction.ldr(
reg,
reg,
reg.to64(),
Instruction.LoadStoreOffset.imm(0),
));
},
@ -905,7 +920,13 @@ fn mirMoveRegister(emit: *Emit, inst: Mir.Inst.Index) !void {
switch (tag) {
.mov_register => {
const rr = emit.mir.instructions.items(.data)[inst].rr;
try emit.writeInstruction(Instruction.orrShiftedRegister(rr.rd, .xzr, rr.rn, .lsl, 0));
const zr: Register = switch (rr.rd.size()) {
32 => .wzr,
64 => .xzr,
else => unreachable,
};
try emit.writeInstruction(Instruction.orrShiftedRegister(rr.rd, zr, rr.rn, .lsl, 0));
},
.mov_to_from_sp => {
const rr = emit.mir.instructions.items(.data)[inst].rr;
@ -917,8 +938,13 @@ fn mirMoveRegister(emit: *Emit, inst: Mir.Inst.Index) !void {
const rm = rr_imm6_logical_shift.rm;
const shift = rr_imm6_logical_shift.shift;
const imm6 = rr_imm6_logical_shift.imm6;
const zr: Register = switch (rd.size()) {
32 => .wzr,
64 => .xzr,
else => unreachable,
};
try emit.writeInstruction(Instruction.ornShiftedRegister(rd, .xzr, rm, shift, imm6));
try emit.writeInstruction(Instruction.ornShiftedRegister(rd, zr, rm, shift, imm6));
},
else => unreachable,
}

View File

@ -695,6 +695,9 @@ pub const Instruction = union(enum) {
offset: LoadStoreOffset,
variant: LoadStoreVariant,
) Instruction {
assert(rn.size() == 64);
assert(rn.id() != Register.xzr.id());
const off = offset.toU12();
const op1: u2 = blk: {
switch (offset) {
@ -741,6 +744,9 @@ pub const Instruction = union(enum) {
encoding: u2,
load: bool,
) Instruction {
assert(rn.size() == 64);
assert(rn.id() != Register.xzr.id());
switch (rt1.size()) {
32 => {
assert(-256 <= offset and offset <= 252);
@ -849,38 +855,26 @@ pub const Instruction = union(enum) {
shift: LogicalShiftedRegisterShift,
amount: u6,
) Instruction {
switch (rd.size()) {
32 => {
assert(amount < 32);
return Instruction{
.logical_shifted_register = .{
.rd = rd.enc(),
.rn = rn.enc(),
.imm6 = amount,
.rm = rm.enc(),
.n = n,
.shift = @enumToInt(shift),
.opc = opc,
.sf = 0b0,
},
};
assert(rd.size() == rn.size());
assert(rd.size() == rm.size());
if (rd.size() == 32) assert(amount < 32);
return Instruction{
.logical_shifted_register = .{
.rd = rd.enc(),
.rn = rn.enc(),
.imm6 = amount,
.rm = rm.enc(),
.n = n,
.shift = @enumToInt(shift),
.opc = opc,
.sf = switch (rd.size()) {
32 => 0b0,
64 => 0b1,
else => unreachable,
},
},
64 => {
return Instruction{
.logical_shifted_register = .{
.rd = rd.enc(),
.rn = rn.enc(),
.imm6 = amount,
.rm = rm.enc(),
.n = n,
.shift = @enumToInt(shift),
.opc = opc,
.sf = 0b1,
},
};
},
else => unreachable, // unexpected register size
}
};
}
fn addSubtractImmediate(
@ -891,6 +885,9 @@ pub const Instruction = union(enum) {
imm12: u12,
shift: bool,
) Instruction {
assert(rd.size() == rn.size());
assert(rn.id() != Register.xzr.id());
return Instruction{
.add_subtract_immediate = .{
.rd = rd.enc(),
@ -916,6 +913,9 @@ pub const Instruction = union(enum) {
immr: u6,
n: u1,
) Instruction {
assert(rd.size() == rn.size());
assert(!(rd.size() == 32 and n == 1));
return Instruction{
.logical_immediate = .{
.rd = rd.enc(),
@ -941,6 +941,8 @@ pub const Instruction = union(enum) {
immr: u6,
imms: u6,
) Instruction {
assert(rd.size() == rn.size());
return Instruction{
.bitfield = .{
.rd = rd.enc(),
@ -969,6 +971,9 @@ pub const Instruction = union(enum) {
rm: Register,
imm6: u6,
) Instruction {
assert(rd.size() == rn.size());
assert(rd.size() == rm.size());
return Instruction{
.add_subtract_shifted_register = .{
.rd = rd.enc(),
@ -994,6 +999,7 @@ pub const Instruction = union(enum) {
offset: i21,
) Instruction {
assert(offset & 0b11 == 0b00);
return Instruction{
.conditional_branch = .{
.cond = @enumToInt(cond),
@ -1010,6 +1016,7 @@ pub const Instruction = union(enum) {
offset: i21,
) Instruction {
assert(offset & 0b11 == 0b00);
return Instruction{
.compare_and_branch = .{
.rt = rt.enc(),
@ -1033,6 +1040,9 @@ pub const Instruction = union(enum) {
rm: Register,
cond: Condition,
) Instruction {
assert(rd.size() == rn.size());
assert(rd.size() == rm.size());
return Instruction{
.conditional_select = .{
.rd = rd.enc(),
@ -1085,6 +1095,9 @@ pub const Instruction = union(enum) {
rn: Register,
rm: Register,
) Instruction {
assert(rd.size() == rn.size());
assert(rd.size() == rm.size());
return Instruction{
.data_processing_2_source = .{
.rd = rd.enc(),