| author | |
| committer | |
| log | 96ecb402590df7a02526009f1630f27e14a0e77c |
| tree | 7c9816d48aec59f73228d474c67b682dfde9038b |
| parent | 4ac36d094c06b04c1c71d972d3f3e1187bccea95 |
3 files changed, 143 insertions(+), 13 deletions(-)
src/ir.cpp+5| ... | @@ -18184,6 +18184,11 @@ static TypeTableEntry *ir_analyze_instruction_atomic_rmw(IrAnalyze *ira, IrInstr | ... | @@ -18184,6 +18184,11 @@ static TypeTableEntry *ir_analyze_instruction_atomic_rmw(IrAnalyze *ira, IrInstr |
| 18184 | } else { | 18184 | } else { |
| 18185 | if (!ir_resolve_atomic_order(ira, instruction->ordering->other, &ordering)) | 18185 | if (!ir_resolve_atomic_order(ira, instruction->ordering->other, &ordering)) |
| 18186 | return ira->codegen->builtin_types.entry_invalid; | 18186 | return ira->codegen->builtin_types.entry_invalid; |
| 18187 | if (ordering == AtomicOrderUnordered) { | ||
| 18188 | ir_add_error(ira, instruction->ordering, | ||
| 18189 | buf_sprintf("@atomicRmw atomic ordering must not be Unordered")); | ||
| 18190 | return ira->codegen->builtin_types.entry_invalid; | ||
| 18191 | } | ||
| 18187 | } | 18192 | } |
| 18188 | 18193 | ||
| 18189 | if (instr_is_comptime(casted_operand) && instr_is_comptime(casted_ptr) && casted_ptr->value.data.x_ptr.mut == ConstPtrMutComptimeVar) | 18194 | if (instr_is_comptime(casted_operand) && instr_is_comptime(casted_ptr) && casted_ptr->value.data.x_ptr.mut == ConstPtrMutComptimeVar) |
std/atomic/stack.zig+82-5| ... | @@ -1,3 +1,6 @@ | ... | @@ -1,3 +1,6 @@ |
| 1 | const builtin = @import("builtin"); | ||
| 2 | const AtomicOrder = builtin.AtomicOrder; | ||
| 3 | |||
| 1 | /// Many reader, many writer, non-allocating, thread-safe, lock-free | 4 | /// Many reader, many writer, non-allocating, thread-safe, lock-free |
| 2 | pub fn Stack(comptime T: type) type { | 5 | pub fn Stack(comptime T: type) type { |
| 3 | return struct { | 6 | return struct { |
| ... | @@ -20,26 +23,100 @@ pub fn Stack(comptime T: type) type { | ... | @@ -20,26 +23,100 @@ pub fn Stack(comptime T: type) type { |
| 20 | /// being the first item in the stack, returns the other item that was there. | 23 | /// being the first item in the stack, returns the other item that was there. |
| 21 | pub fn pushFirst(self: &Self, node: &Node) ?&Node { | 24 | pub fn pushFirst(self: &Self, node: &Node) ?&Node { |
| 22 | node.next = null; | 25 | node.next = null; |
| 23 | return @cmpxchgStrong(?&Node, &self.root, null, node, AtomicOrder.AcqRel, AtomicOrder.AcqRel); | 26 | return @cmpxchgStrong(?&Node, &self.root, null, node, AtomicOrder.SeqCst, AtomicOrder.SeqCst); |
| 24 | } | 27 | } |
| 25 | 28 | ||
| 26 | pub fn push(self: &Self, node: &Node) void { | 29 | pub fn push(self: &Self, node: &Node) void { |
| 27 | var root = @atomicLoad(?&Node, &self.root, AtomicOrder.Acquire); | 30 | var root = @atomicLoad(?&Node, &self.root, AtomicOrder.SeqCst); |
| 28 | while (true) { | 31 | while (true) { |
| 29 | node.next = root; | 32 | node.next = root; |
| 30 | root = @cmpxchgWeak(?&Node, &self.root, root, node, AtomicOrder.Release, AtomicOrder.Acquire) ?? break; | 33 | root = @cmpxchgWeak(?&Node, &self.root, root, node, AtomicOrder.SeqCst, AtomicOrder.SeqCst) ?? break; |
| 31 | } | 34 | } |
| 32 | } | 35 | } |
| 33 | 36 | ||
| 34 | pub fn pop(self: &Self) ?&Node { | 37 | pub fn pop(self: &Self) ?&Node { |
| 35 | var root = @atomicLoad(?&Node, &self.root, AtomicOrder.Acquire); | 38 | var root = @atomicLoad(?&Node, &self.root, AtomicOrder.Acquire); |
| 36 | while (true) { | 39 | while (true) { |
| 37 | root = @cmpxchgWeak(?&Node, &self.root, root, (root ?? return null).next, AtomicOrder.Release, AtomicOrder.Acquire) ?? return root; | 40 | root = @cmpxchgWeak(?&Node, &self.root, root, (root ?? return null).next, AtomicOrder.SeqCst, AtomicOrder.SeqCst) ?? return root; |
| 38 | } | 41 | } |
| 39 | } | 42 | } |
| 40 | 43 | ||
| 41 | pub fn isEmpty(self: &Self) bool { | 44 | pub fn isEmpty(self: &Self) bool { |
| 42 | return @atomicLoad(?&Node, &self.root, AtomicOrder.Relaxed) == null; | 45 | return @atomicLoad(?&Node, &self.root, AtomicOrder.SeqCst) == null; |
| 43 | } | 46 | } |
| 44 | }; | 47 | }; |
| 45 | } | 48 | } |
| 49 | |||
| 50 | const std = @import("std"); | ||
| 51 | const Context = struct { | ||
| 52 | allocator: &std.mem.Allocator, | ||
| 53 | stack: &Stack(i32), | ||
| 54 | put_sum: isize, | ||
| 55 | get_sum: isize, | ||
| 56 | puts_done: u8, // TODO make this a bool | ||
| 57 | }; | ||
| 58 | const puts_per_thread = 1000; | ||
| 59 | const put_thread_count = 3; | ||
| 60 | |||
| 61 | test "std.atomic.stack" { | ||
| 62 | var direct_allocator = std.heap.DirectAllocator.init(); | ||
| 63 | defer direct_allocator.deinit(); | ||
| 64 | |||
| 65 | var plenty_of_memory = try direct_allocator.allocator.alloc(u8, 64 * 1024 * 1024); | ||
| 66 | defer direct_allocator.allocator.free(plenty_of_memory); | ||
| 67 | |||
| 68 | var fixed_buffer_allocator = std.heap.ThreadSafeFixedBufferAllocator.init(plenty_of_memory); | ||
| 69 | var a = &fixed_buffer_allocator.allocator; | ||
| 70 | |||
| 71 | var stack = Stack(i32).init(); | ||
| 72 | var context = Context { | ||
| 73 | .allocator = a, | ||
| 74 | .stack = &stack, | ||
| 75 | .put_sum = 0, | ||
| 76 | .get_sum = 0, | ||
| 77 | .puts_done = 0, | ||
| 78 | }; | ||
| 79 | |||
| 80 | var putters: [put_thread_count]&std.os.Thread = undefined; | ||
| 81 | for (putters) |*t| { | ||
| 82 | *t = try std.os.spawnThreadAllocator(a, &context, startPuts); | ||
| 83 | } | ||
| 84 | var getters: [put_thread_count]&std.os.Thread = undefined; | ||
| 85 | for (getters) |*t| { | ||
| 86 | *t = try std.os.spawnThreadAllocator(a, &context, startGets); | ||
| 87 | } | ||
| 88 | |||
| 89 | for (putters) |t| t.wait(); | ||
| 90 | _ = @atomicRmw(u8, &context.puts_done, builtin.AtomicRmwOp.Xchg, 1, AtomicOrder.SeqCst); | ||
| 91 | for (getters) |t| t.wait(); | ||
| 92 | |||
| 93 | std.debug.assert(context.put_sum == context.get_sum); | ||
| 94 | } | ||
| 95 | |||
| 96 | fn startPuts(ctx: &Context) u8 { | ||
| 97 | var put_count: usize = puts_per_thread; | ||
| 98 | var r = std.rand.DefaultPrng.init(0xdeadbeef); | ||
| 99 | while (put_count != 0) : (put_count -= 1) { | ||
| 100 | std.os.time.sleep(0, 1); // let the os scheduler be our fuzz | ||
| 101 | const x = @bitCast(i32, r.random.scalar(u32)); | ||
| 102 | const node = ctx.allocator.create(Stack(i32).Node) catch unreachable; | ||
| 103 | node.data = x; | ||
| 104 | ctx.stack.push(node); | ||
| 105 | _ = @atomicRmw(isize, &ctx.put_sum, builtin.AtomicRmwOp.Add, x, AtomicOrder.SeqCst); | ||
| 106 | } | ||
| 107 | return 0; | ||
| 108 | } | ||
| 109 | |||
| 110 | fn startGets(ctx: &Context) u8 { | ||
| 111 | while (true) { | ||
| 112 | while (ctx.stack.pop()) |node| { | ||
| 113 | std.os.time.sleep(0, 1); // let the os scheduler be our fuzz | ||
| 114 | _ = @atomicRmw(isize, &ctx.get_sum, builtin.AtomicRmwOp.Add, node.data, builtin.AtomicOrder.SeqCst); | ||
| 115 | } | ||
| 116 | |||
| 117 | if (@atomicLoad(u8, &ctx.puts_done, builtin.AtomicOrder.SeqCst) == 1) { | ||
| 118 | break; | ||
| 119 | } | ||
| 120 | } | ||
| 121 | return 0; | ||
| 122 | } |
std/heap.zig+56-8| ... | @@ -47,13 +47,6 @@ pub const DirectAllocator = struct { | ... | @@ -47,13 +47,6 @@ pub const DirectAllocator = struct { |
| 47 | 47 | ||
| 48 | const HeapHandle = if (builtin.os == Os.windows) os.windows.HANDLE else void; | 48 | const HeapHandle = if (builtin.os == Os.windows) os.windows.HANDLE else void; |
| 49 | 49 | ||
| 50 | //pub const canary_bytes = []u8 {48, 239, 128, 46, 18, 49, 147, 9, 195, 59, 203, 3, 245, 54, 9, 122}; | ||
| 51 | //pub const want_safety = switch (builtin.mode) { | ||
| 52 | // builtin.Mode.Debug => true, | ||
| 53 | // builtin.Mode.ReleaseSafe => true, | ||
| 54 | // else => false, | ||
| 55 | //}; | ||
| 56 | |||
| 57 | pub fn init() DirectAllocator { | 50 | pub fn init() DirectAllocator { |
| 58 | return DirectAllocator { | 51 | return DirectAllocator { |
| 59 | .allocator = Allocator { | 52 | .allocator = Allocator { |
| ... | @@ -298,7 +291,7 @@ pub const FixedBufferAllocator = struct { | ... | @@ -298,7 +291,7 @@ pub const FixedBufferAllocator = struct { |
| 298 | 291 | ||
| 299 | fn alloc(allocator: &Allocator, n: usize, alignment: u29) ![]u8 { | 292 | fn alloc(allocator: &Allocator, n: usize, alignment: u29) ![]u8 { |
| 300 | const self = @fieldParentPtr(FixedBufferAllocator, "allocator", allocator); | 293 | const self = @fieldParentPtr(FixedBufferAllocator, "allocator", allocator); |
| 301 | const addr = @ptrToInt(&self.buffer[self.end_index]); | 294 | const addr = @ptrToInt(self.buffer.ptr) + self.end_index; |
| 302 | const rem = @rem(addr, alignment); | 295 | const rem = @rem(addr, alignment); |
| 303 | const march_forward_bytes = if (rem == 0) 0 else (alignment - rem); | 296 | const march_forward_bytes = if (rem == 0) 0 else (alignment - rem); |
| 304 | const adjusted_index = self.end_index + march_forward_bytes; | 297 | const adjusted_index = self.end_index + march_forward_bytes; |
| ... | @@ -325,6 +318,54 @@ pub const FixedBufferAllocator = struct { | ... | @@ -325,6 +318,54 @@ pub const FixedBufferAllocator = struct { |
| 325 | fn free(allocator: &Allocator, bytes: []u8) void { } | 318 | fn free(allocator: &Allocator, bytes: []u8) void { } |
| 326 | }; | 319 | }; |
| 327 | 320 | ||
| 321 | /// lock free | ||
| 322 | pub const ThreadSafeFixedBufferAllocator = struct { | ||
| 323 | allocator: Allocator, | ||
| 324 | end_index: usize, | ||
| 325 | buffer: []u8, | ||
| 326 | |||
| 327 | pub fn init(buffer: []u8) ThreadSafeFixedBufferAllocator { | ||
| 328 | return ThreadSafeFixedBufferAllocator { | ||
| 329 | .allocator = Allocator { | ||
| 330 | .allocFn = alloc, | ||
| 331 | .reallocFn = realloc, | ||
| 332 | .freeFn = free, | ||
| 333 | }, | ||
| 334 | .buffer = buffer, | ||
| 335 | .end_index = 0, | ||
| 336 | }; | ||
| 337 | } | ||
| 338 | |||
| 339 | fn alloc(allocator: &Allocator, n: usize, alignment: u29) ![]u8 { | ||
| 340 | const self = @fieldParentPtr(ThreadSafeFixedBufferAllocator, "allocator", allocator); | ||
| 341 | var end_index = @atomicLoad(usize, &self.end_index, builtin.AtomicOrder.SeqCst); | ||
| 342 | while (true) { | ||
| 343 | const addr = @ptrToInt(self.buffer.ptr) + end_index; | ||
| 344 | const rem = @rem(addr, alignment); | ||
| 345 | const march_forward_bytes = if (rem == 0) 0 else (alignment - rem); | ||
| 346 | const adjusted_index = end_index + march_forward_bytes; | ||
| 347 | const new_end_index = adjusted_index + n; | ||
| 348 | if (new_end_index > self.buffer.len) { | ||
| 349 | return error.OutOfMemory; | ||
| 350 | } | ||
| 351 | end_index = @cmpxchgWeak(usize, &self.end_index, end_index, new_end_index, | ||
| 352 | builtin.AtomicOrder.SeqCst, builtin.AtomicOrder.SeqCst) ?? return self.buffer[adjusted_index .. new_end_index]; | ||
| 353 | } | ||
| 354 | } | ||
| 355 | |||
| 356 | fn realloc(allocator: &Allocator, old_mem: []u8, new_size: usize, alignment: u29) ![]u8 { | ||
| 357 | if (new_size <= old_mem.len) { | ||
| 358 | return old_mem[0..new_size]; | ||
| 359 | } else { | ||
| 360 | const result = try alloc(allocator, new_size, alignment); | ||
| 361 | mem.copy(u8, result, old_mem); | ||
| 362 | return result; | ||
| 363 | } | ||
| 364 | } | ||
| 365 | |||
| 366 | fn free(allocator: &Allocator, bytes: []u8) void { } | ||
| 367 | }; | ||
| 368 | |||
| 328 | 369 | ||
| 329 | 370 | ||
| 330 | test "c_allocator" { | 371 | test "c_allocator" { |
| ... | @@ -363,6 +404,13 @@ test "FixedBufferAllocator" { | ... | @@ -363,6 +404,13 @@ test "FixedBufferAllocator" { |
| 363 | try testAllocatorLargeAlignment(&fixed_buffer_allocator.allocator); | 404 | try testAllocatorLargeAlignment(&fixed_buffer_allocator.allocator); |
| 364 | } | 405 | } |
| 365 | 406 | ||
| 407 | test "ThreadSafeFixedBufferAllocator" { | ||
| 408 | var fixed_buffer_allocator = ThreadSafeFixedBufferAllocator.init(test_fixed_buffer_allocator_memory[0..]); | ||
| 409 | |||
| 410 | try testAllocator(&fixed_buffer_allocator.allocator); | ||
| 411 | try testAllocatorLargeAlignment(&fixed_buffer_allocator.allocator); | ||
| 412 | } | ||
| 413 | |||
| 366 | fn testAllocator(allocator: &mem.Allocator) !void { | 414 | fn testAllocator(allocator: &mem.Allocator) !void { |
| 367 | var slice = try allocator.alloc(&i32, 100); | 415 | var slice = try allocator.alloc(&i32, 100); |
| 368 | 416 |