mirror/zig
1
0
Fork 0
mirror of https://github.com/ziglang/zig.git synced 2025-12-16 19:23:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

x86_64: fix rem/mod behavior and hazards

Browse Source
This commit is contained in:
Jacob Young 2023-04-25 16:47:44 -04:00
parent a8842b6cbf
commit eff0e6a726
1 changed files with 49 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
src/arch/x86_64/CodeGen.zig
View File

@ -2601,7 +2601,13 @@ fn genIntMulDivOpMir(
else => unreachable, else => unreachable,
.mul, .imul => {}, .mul, .imul => {},
.div => try self.asmRegisterRegister(.xor, .edx, .edx), .div => try self.asmRegisterRegister(.xor, .edx, .edx),
.idiv => try self.asmOpOnly(.cqo), .idiv => switch (self.regBitSize(ty)) {
8 => try self.asmOpOnly(.cbw),
16 => try self.asmOpOnly(.cwd),
32 => try self.asmOpOnly(.cdq),
64 => try self.asmOpOnly(.cqo),
else => unreachable,
},
} }
const mat_rhs: MCValue = switch (rhs) { const mat_rhs: MCValue = switch (rhs) {
@ -2631,7 +2637,8 @@ fn genIntMulDivOpMir(
/// Always returns a register. /// Always returns a register.
/// Clobbers .rax and .rdx registers. /// Clobbers .rax and .rdx registers.
fn genInlineIntDivFloor(self: *Self, ty: Type, lhs: MCValue, rhs: MCValue) !MCValue { fn genInlineIntDivFloor(self: *Self, ty: Type, lhs: MCValue, rhs: MCValue) !MCValue {
const signedness = ty.intInfo(self.target.*).signedness; const abi_size = @intCast(u32, ty.abiSize(self.target.*));
const int_info = ty.intInfo(self.target.*);
const dividend: Register = switch (lhs) { const dividend: Register = switch (lhs) {
.register => |reg| reg, .register => |reg| reg,
else => try self.copyToTmpRegister(ty, lhs), else => try self.copyToTmpRegister(ty, lhs),
@ -2646,16 +2653,32 @@ fn genInlineIntDivFloor(self: *Self, ty: Type, lhs: MCValue, rhs: MCValue) !MCVa
const divisor_lock = self.register_manager.lockReg(divisor); const divisor_lock = self.register_manager.lockReg(divisor);
defer if (divisor_lock) |lock| self.register_manager.unlockReg(lock); defer if (divisor_lock) |lock| self.register_manager.unlockReg(lock);
try self.genIntMulDivOpMir(switch (signedness) { try self.genIntMulDivOpMir(switch (int_info.signedness) {
.signed => .idiv, .signed => .idiv,
.unsigned => .div, .unsigned => .div,
}, Type.isize, .{ .register = dividend }, .{ .register = divisor }); }, ty, .{ .register = dividend }, .{ .register = divisor });
try self.asmRegisterRegister(.xor, divisor.to64(), dividend.to64()); try self.asmRegisterRegister(
try self.asmRegisterImmediate(.sar, divisor.to64(), Immediate.u(63)); .xor,
try self.asmRegisterRegister(.@"test", .rdx, .rdx); registerAlias(divisor, abi_size),
try self.asmCmovccRegisterRegister(divisor.to64(), .rdx, .e); registerAlias(dividend, abi_size),
try self.genBinOpMir(.add, Type.isize, .{ .register = divisor }, .{ .register = .rax }); );
try self.asmRegisterImmediate(
.sar,
registerAlias(divisor, abi_size),
Immediate.u(int_info.bits - 1),
);
try self.asmRegisterRegister(
.@"test",
registerAlias(.rdx, abi_size),
registerAlias(.rdx, abi_size),
);
try self.asmCmovccRegisterRegister(
registerAlias(divisor, abi_size),
registerAlias(.rdx, abi_size),
.z,
);
try self.genBinOpMir(.add, ty, .{ .register = divisor }, .{ .register = .rax });
return MCValue{ .register = divisor }; return MCValue{ .register = divisor };
} }
@ -4928,8 +4951,24 @@ fn genMulDivBinOp(
switch (signedness) { switch (signedness) {
.signed => { .signed => {
const lhs_lock = switch (lhs) {
.register => |reg| self.register_manager.lockReg(reg),
else => null,
};
defer if (lhs_lock) |lock| self.register_manager.unlockReg(lock);
const rhs_lock = switch (rhs) {
.register => |reg| self.register_manager.lockReg(reg),
else => null,
};
defer if (rhs_lock) |lock| self.register_manager.unlockReg(lock);
// hack around hazard between rhs and div_floor by copying rhs to another register
const rhs_copy = try self.copyToTmpRegister(ty, rhs);
const rhs_copy_lock = self.register_manager.lockRegAssumeUnused(rhs_copy);
defer self.register_manager.unlockReg(rhs_copy_lock);
const div_floor = try self.genInlineIntDivFloor(ty, lhs, rhs); const div_floor = try self.genInlineIntDivFloor(ty, lhs, rhs);
try self.genIntMulComplexOpMir(ty, div_floor, rhs); try self.genIntMulComplexOpMir(ty, div_floor, .{ .register = rhs_copy });
const div_floor_lock = self.register_manager.lockReg(div_floor.register); const div_floor_lock = self.register_manager.lockReg(div_floor.register);
defer if (div_floor_lock) |lock| self.register_manager.unlockReg(lock); defer if (div_floor_lock) |lock| self.register_manager.unlockReg(lock);