mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-06-24 02:14:17 +00:00
9faf205601* move cifar into datasets * support for pathlib Tensors, tar_extract, and fetch gunzip * too early for Device.DEFAULT * simpler hlb_cifar + .to(None) is default * new compiler failure, start beautiful_cifar * beautiful cifar runs but is broken * jit train step * cleaner * std_mean, not mean_std * more correct * fast indexing * don't print that * torch load broken * add eval * nicer bar * decoraters are the way to do this * bounds check the load * a few ops * batchnorm bugfix, if track_running_stats is False, use online estimate * full timing * fix fusion * unneeded realize * master tensor
268 lines
16 KiB
Python
268 lines
16 KiB
Python
from typing import List, Dict, Optional, cast, Generator, Tuple, Union
|
|
import time, pprint
|
|
from collections import defaultdict
|
|
from dataclasses import dataclass, replace
|
|
from tinygrad.helpers import colored, getenv, DEBUG, GlobalCounters, ansilen, BEAM, NOOPT, all_int, CAPTURING, Metadata, Context, TRACEMETA, dedup
|
|
from tinygrad.ops import MetaOps, UOps, UOp
|
|
from tinygrad.dtype import dtypes
|
|
from tinygrad.device import Device, Buffer
|
|
from tinygrad.shape.symbolic import Variable, sym_infer, sint
|
|
from tinygrad.renderer import Renderer, Program
|
|
from tinygrad.codegen.kernel import Kernel
|
|
from tinygrad.engine.schedule import ScheduleItem
|
|
|
|
# **************** Program Creation ****************
|
|
|
|
logkerns, logkerns_level = open(getenv("LOGKERNS", ""), "a") if getenv("LOGKERNS", "") else None, getenv("LOGKERNS_LEVEL", 1)
|
|
def get_kernel(renderer:Renderer, ast:UOp) -> Kernel:
|
|
if DEBUG >= 5:
|
|
print(ast)
|
|
k = Kernel(ast, opts=renderer).required_optimizations()
|
|
if not NOOPT:
|
|
if not (used_tensor_cores:=k.apply_tensor_cores(getenv("TC", 1))): k.hand_coded_optimizations()
|
|
if BEAM >= 1:
|
|
from tinygrad.engine.search import beam_search, time_linearizer, bufs_from_lin
|
|
kb, k_opt = Kernel(ast, opts=renderer).required_optimizations(), k
|
|
rawbufs = bufs_from_lin(kb, allocate=False)
|
|
if BEAM.value >= 100:
|
|
from extra.mcts_search import mcts_search
|
|
k = mcts_search(kb, rawbufs, BEAM.value)
|
|
else:
|
|
k = beam_search(kb, rawbufs, BEAM.value, bool(getenv("BEAM_ESTIMATE", 1)))
|
|
if beam_compare:=getenv("BEAM_COMPARE", 1):
|
|
# TODO: move the HC/TC/BEAM compare to beam_search so it can be optionally cached which choice is better
|
|
lins: List[Tuple[str, Kernel]] = [(f"beam{BEAM.value}", k), (("tc" if used_tensor_cores else "hc"), k_opt)]
|
|
if used_tensor_cores: lins.append(("hc", Kernel(ast, opts=renderer).hand_coded_optimizations()))
|
|
timed = sorted([(nm, tk, time_linearizer(tk, rawbufs, allow_test_size=False, clear_l2=True)) for nm, tk in lins], key=lambda x: x[2])
|
|
if DEBUG >= 3: print(" < ".join(f"{nm:6s} : {lin.colored_shape(30, dense=True)} : {tm*1e6:8.2f} us" for nm, lin, tm in timed))
|
|
k = timed[0][1]
|
|
if logkerns is not None and logkerns_level > 1: logkerns.writelines([f"{(lin.ast, lin.applied_opts)}\n" for (_,lin,_) in timed[1:]])
|
|
if beam_compare == 2:
|
|
from tinygrad import Tensor
|
|
all_outs: List[List[Tensor]] = []
|
|
with Context(DEBUG=0, BEAM=0, CAPTURING=0):
|
|
rand_bufs = [Tensor.normal(buf.size, std=0.1, dtype=buf.dtype).data() if dtypes.is_float(buf.dtype) else \
|
|
(Tensor.randint(buf.size, low=0, high=2).cast(buf.dtype).data() if buf.dtype == dtypes.bool else \
|
|
Tensor.randint(buf.size, low=dtypes.min(buf.dtype), high=dtypes.max(buf.dtype), dtype=buf.dtype).data()) \
|
|
for buf in rawbufs]
|
|
for _, tk in lins[::-1]:
|
|
for buf,data in zip(rawbufs, rand_bufs): buf.ensure_allocated().copyin(data)
|
|
time_linearizer(tk, rawbufs, allow_test_size=False, clear_l2=True, disable_cache=True)
|
|
all_outs.append([Tensor(bytes(buf.as_buffer()), dtype=buf.dtype) for buf in rawbufs[:len(ast.src)]])
|
|
with Context(DEBUG=0, BEAM=0, CAPTURING=0):
|
|
for bufs in zip(*all_outs):
|
|
for b in bufs[1:]:
|
|
if dtypes.is_float(bufs[0].dtype):
|
|
# we check both atol and rtol here
|
|
diff_count = (((b-bufs[0]).abs() > 1e-3) * (((b-bufs[0])/bufs[0]).abs() > 1e-3)).sum().item()
|
|
else:
|
|
diff_count = (b != bufs[0]).sum().item()
|
|
if diff_count != 0:
|
|
raise RuntimeError(f"mismatch of {diff_count}/{b.numel()} items with type {b.dtype}, max {(b-bufs[0]).abs().max().item()}")
|
|
if logkerns is not None: logkerns.writelines([f"{(k.ast, k.applied_opts)}\n"])
|
|
if DEBUG >= 5: print((k.ast, k.applied_opts)) # print here to show final applied_opts for all kernels instead of just in beam_search
|
|
return k
|
|
|
|
# **************** Runners ****************
|
|
|
|
class Runner:
|
|
def __init__(self, display_name:str, dname:str, op_estimate:sint=0, mem_estimate:sint=0, lds_estimate:Optional[sint]=None):
|
|
self.first_run, self.display_name, self.dname, self.op_estimate, self.mem_estimate, self.lds_estimate = \
|
|
True, display_name, dname, op_estimate, mem_estimate, mem_estimate if lds_estimate is None else lds_estimate
|
|
@property
|
|
def device(self): return Device[self.dname]
|
|
def exec(self, rawbufs:List[Buffer], var_vals:Optional[Dict[Variable, int]]=None) -> Optional[float]:
|
|
return self(rawbufs, {} if var_vals is None else var_vals)
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False) -> Optional[float]:
|
|
raise NotImplementedError("override this")
|
|
|
|
class CompiledRunner(Runner):
|
|
def __init__(self, p:Program, precompiled:Optional[bytes]=None):
|
|
if DEBUG >= 4: print(p.src)
|
|
self.p:Program = p
|
|
self.lib:bytes = precompiled if precompiled is not None else Device[p.dname].compiler.compile_cached(p.src)
|
|
self.clprg = Device[p.dname].runtime(p.function_name, self.lib)
|
|
super().__init__(p.name, p.dname, p.op_estimate, p.mem_estimate, p.lds_estimate)
|
|
|
|
def __reduce__(self): return self.__class__, (self.p, self.lib)
|
|
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False) -> Optional[float]:
|
|
global_size, local_size = self.p.launch_dims(var_vals)
|
|
if global_size is not None and local_size is None and all_int(self.p.global_size): # type: ignore[arg-type]
|
|
# TODO: this is copied from get_program
|
|
from tinygrad.engine.search import optimize_local_size
|
|
local_size = optimize_local_size(self.clprg, global_size, rawbufs)
|
|
global_size = [g//l if g%l == 0 else g/l for g,l in zip(global_size, local_size)]
|
|
self.p = replace(self.p, global_size=global_size, local_size=local_size)
|
|
lra = {}
|
|
if global_size:
|
|
lra['global_size'] = tuple(global_size)
|
|
assert len(global_size) == 3, "global size must have len 3"
|
|
if local_size:
|
|
lra['local_size'] = tuple(local_size)
|
|
assert len(local_size) == 3, "local size must have len 3"
|
|
return self.clprg(*[x._buf for x in rawbufs], **lra, vals=tuple(var_vals[k] for k in self.p.vars), wait=wait)
|
|
|
|
class CustomOp(Runner):
|
|
def __init__(self, fxn):
|
|
self.fxn = fxn
|
|
super().__init__(self.fxn.__name__, "CUSTOM", 0, 0)
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False): self.fxn(*rawbufs)
|
|
|
|
class EmptyOp(Runner):
|
|
def __init__(self, buf:Buffer): super().__init__(colored(f"empty {buf.size:10d} {buf.dtype}", "yellow"), buf.device)
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False): pass
|
|
|
|
class ViewOp(Runner):
|
|
def __init__(self, buf:Buffer): super().__init__(colored(f"view {buf.nbytes:8d} @ {buf.offset:<10d}", "yellow"), buf.device)
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False):
|
|
assert rawbufs[0]._base is not None and rawbufs[0]._base == rawbufs[1].base, f"must be base {rawbufs}"
|
|
|
|
class BufferCopy(Runner):
|
|
def __init__(self, total_sz, dest_device, src_device):
|
|
if total_sz >= 1e6: name = f"{type(self).__name__[6:].lower()} {total_sz/1e6:7.2f}M, {dest_device[:7]:>7s} <- {src_device[:7]:7s}"
|
|
else: name = f"{type(self).__name__[6:].lower()} {total_sz:8d}, {dest_device[:7]:>7s} <- {src_device[:7]:7s}"
|
|
super().__init__(colored(name, "yellow"), dest_device, 0, total_sz)
|
|
def copy(self, dest, src):
|
|
disk_supports_fast_copyout = src.device.startswith("DISK") and hasattr(src.allocator.device, 'io_uring') and hasattr(src.allocator.device, 'fd')
|
|
if src.device.startswith("DISK") and hasattr(dest.allocator, 'copy_from_disk') and disk_supports_fast_copyout and src.nbytes >= 4096:
|
|
dest.allocator.copy_from_disk(dest._buf, src._buf, src.nbytes)
|
|
elif src.device.startswith("DISK") and hasattr(dest.allocator, 'as_buffer'):
|
|
# fast(ish) path, uses readinto in diskbuffers
|
|
src.allocator.copyout(dest.allocator.as_buffer(dest._buf), src._buf)
|
|
else:
|
|
dest.copyin(src.as_buffer(allow_zero_copy=True)) # may allocate a CPU buffer depending on allow_zero_copy
|
|
def __call__(self, rawbufs:List[Buffer], var_vals:Dict[Variable, int], wait=False):
|
|
dest, src = rawbufs[0:2]
|
|
assert dest.size == src.size and dest.dtype == src.dtype, f"buffer copy mismatch, {dest.size} != {src.size}, {dest.dtype} != {src.dtype}"
|
|
st = time.perf_counter()
|
|
self.copy(dest, src)
|
|
if wait:
|
|
Device[dest.device].synchronize()
|
|
return time.perf_counter() - st
|
|
|
|
class BufferXfer(BufferCopy):
|
|
def copy(self, dest, src): dest.allocator.transfer(dest._buf, src._buf, dest.nbytes, src_dev=src.allocator.device, dest_dev=dest.allocator.device)
|
|
|
|
# **************** method cache ****************
|
|
|
|
method_cache: Dict[Tuple[str, bytes, int, int, bool], CompiledRunner] = {}
|
|
def get_runner(dname:str, ast:UOp) -> CompiledRunner:
|
|
ckey = (dname, ast.key, BEAM.value, NOOPT.value, False)
|
|
if cret:=method_cache.get(ckey): return cret
|
|
bkey = (dname.split(":")[0], ast.key, BEAM.value, NOOPT.value, True)
|
|
if bret:=method_cache.get(bkey):
|
|
method_cache[ckey] = ret = CompiledRunner(replace(bret.p, dname=dname), bret.lib)
|
|
else:
|
|
prg: Program = get_kernel(Device[dname].renderer, ast).to_program()
|
|
if getenv("FUZZ_UOPS"):
|
|
from test.external.fuzz_uops import UOpsFuzzerRunner
|
|
return UOpsFuzzerRunner(replace(prg, dname=dname))
|
|
method_cache[ckey] = method_cache[bkey] = ret = CompiledRunner(replace(prg, dname=dname))
|
|
return ret
|
|
|
|
# **************** lowering functions ****************
|
|
|
|
@dataclass(frozen=True)
|
|
class ExecItem:
|
|
prg: Runner
|
|
bufs: List[Optional[Buffer]]
|
|
metadata: Optional[List[Metadata]] = None
|
|
def run(self, var_vals:Optional[Dict[Variable, int]]=None, wait=False, jit=False, do_update_stats=True) -> Optional[float]:
|
|
bufs = [cast(Buffer, x) for x in self.bufs] if jit else [cast(Buffer, x).ensure_allocated() for x in self.bufs]
|
|
et = self.prg(bufs, var_vals if var_vals is not None else {}, wait=wait or DEBUG >= 2)
|
|
if do_update_stats:
|
|
GlobalCounters.kernel_count += 1
|
|
GlobalCounters.global_ops += (op_est:=sym_infer(self.prg.op_estimate, var_vals))
|
|
GlobalCounters.global_mem += (mem_est:=sym_infer(self.prg.mem_estimate, var_vals))
|
|
if et is not None: GlobalCounters.time_sum_s += et
|
|
if DEBUG >= 2:
|
|
lds_est = sym_infer(self.prg.lds_estimate, var_vals)
|
|
mem_est = min(mem_est, lds_est) # there can't be more memory accessed than loads/stores. remove this when symbolic is fixed
|
|
ptm = (colored(f"{et*1e3:9.2f}ms", "yellow") if et > 0.01 else f"{et*1e6:9.2f}us") if et is not None else ""
|
|
print(f"{colored(f'*** {self.prg.dname[:7]:7s} {GlobalCounters.kernel_count:4d}', 'magenta' if jit else ('green' if self.prg.first_run else None))} {self.prg.display_name+' '*(40-ansilen(self.prg.display_name))} mem {GlobalCounters.mem_used/1e9:5.2f} GB " + # noqa: E501
|
|
(str() if et is None else f"tm {ptm}/{GlobalCounters.time_sum_s*1e3:9.2f}ms ({op_est/((et or 1e-20)*1e9):9.2f} GFLOPS {mem_est/((et or 1e-20)*1e9):6.1f}|{lds_est/((et or 1e-20)*1e9):<7.1f} GB/s)" + # noqa: E501
|
|
f" {[repr(m) if TRACEMETA >= 2 else str(m) for m in self.metadata] if self.metadata else ''}"))
|
|
self.prg.first_run = False
|
|
return et
|
|
|
|
def lower_schedule_item(si:ScheduleItem) -> ExecItem:
|
|
assert len(set(x.device for x in si.bufs)) == 1 or (si.ast.op is UOps.EXT and si.ast.arg[0] is MetaOps.COPY)
|
|
if si.ast.op is UOps.SINK:
|
|
runner = get_runner(si.outputs[0].device, si.ast)
|
|
return ExecItem(runner, [si.bufs[x] for x in runner.p.globals], si.metadata)
|
|
out, (op, arg) = si.outputs[0], si.ast.arg
|
|
if op is MetaOps.COPY:
|
|
kernel_type = BufferCopy
|
|
if hasattr(Device[out.device].allocator, 'transfer') and out.device.split(":")[0] == si.inputs[0].device.split(":")[0]:
|
|
kernel_type = BufferXfer
|
|
return ExecItem(kernel_type(arg, out.device, si.inputs[0].device), list(si.bufs))
|
|
if op is MetaOps.CUSTOM: return ExecItem(CustomOp(arg), list(si.bufs))
|
|
if op is MetaOps.EMPTY: return ExecItem(EmptyOp(out), list(si.bufs))
|
|
if op is MetaOps.VIEW: return ExecItem(ViewOp(out), list(si.bufs))
|
|
raise RuntimeError(f"don't know how to lower {si.ast}")
|
|
|
|
def lower_schedule(schedule:List[ScheduleItem]) -> Generator[ExecItem, None, None]:
|
|
while len(schedule):
|
|
si = schedule.pop(0)
|
|
try: yield lower_schedule_item(si)
|
|
except Exception as e:
|
|
if DEBUG >= 2:
|
|
print(f"error lowering {si.ast.op}")
|
|
print("tensor operations:")
|
|
pprint.pprint(si.metadata, indent=2)
|
|
raise e
|
|
|
|
# **************** main run function ****************
|
|
|
|
capturing: List = [] # put classes with an add method in here
|
|
|
|
def run_schedule(schedule:List[ScheduleItem], var_vals:Optional[Dict[Variable, int]]=None, do_update_stats=True):
|
|
for ei in lower_schedule(schedule):
|
|
if len(capturing) and CAPTURING: capturing[0].add(ei)
|
|
ei.run(var_vals, do_update_stats=do_update_stats)
|
|
|
|
# **************** memory planning ****************
|
|
|
|
def _internal_memory_planner(buffers:List[Union[List[Buffer], Tuple[Buffer, ...]]], noopt_buffers=None, debug_prefix="") -> Dict[Buffer, Buffer]:
|
|
if getenv("NO_MEMORY_PLANNER"): return {}
|
|
first_appearance, last_appearance = {}, {}
|
|
for i,u in enumerate(buffers):
|
|
for buf in u:
|
|
if buf.is_allocated() or buf.lb_refcount > 0 or (noopt_buffers is not None and buf.base in noopt_buffers): continue
|
|
if buf.base not in first_appearance: first_appearance[buf.base] = i
|
|
last_appearance[buf.base] = i
|
|
|
|
# Sort buffers by size in descending order, prioritizing largest buffers for allocation first.
|
|
# Track free segments, each containing (start, stop, and buffer that could be reused on this segment).
|
|
free_segs: Dict[Tuple, List[Tuple[int, int, Buffer]]] = defaultdict(list) # Dict[buffer key, Tuple[start, end, buffer to reuse on the seg]]
|
|
def find_replace_buffer(buf, st, en):
|
|
key = (buf.device, buf.dtype, buf.options) + ((buf.nbytes,) if not hasattr(Device[buf.device].allocator, "offset") else tuple())
|
|
|
|
default_buf = (0, len(buffers) - 1, buf) # will return the buffer itself if the replace one is not found.
|
|
seg_st, seg_en, seg_buf = next((free_segs[key].pop(i) for i,(sst,sen,_) in enumerate(free_segs[key]) if sst <= st and en <= sen), default_buf)
|
|
|
|
free_segs[key] += [(seg_st, st - 1, seg_buf)] if st - 1 >= seg_st else []
|
|
free_segs[key] += [(en + 1, seg_en, seg_buf)] if seg_en >= en + 1 else []
|
|
|
|
return seg_buf if seg_buf.nbytes == buf.nbytes else Buffer(buf.device, buf.size, buf.dtype, base=seg_buf)
|
|
|
|
buffer_requests = sorted([(first_appearance[buf], last_appearance[buf], buf) for buf in first_appearance.keys()], key=lambda x: -x[2].nbytes)
|
|
assigned = {buf:find_replace_buffer(buf, st, en) for st, en, buf in buffer_requests}
|
|
|
|
for i,u in enumerate(buffers):
|
|
for buf in u:
|
|
if buf.is_allocated() or buf.lb_refcount > 0 or (noopt_buffers is not None and buf.base in noopt_buffers): continue
|
|
if buf._base is not None: assigned[buf] = Buffer(buf.device, buf.size, buf.dtype, base=assigned.get(buf.base, buf.base).base, offset=buf.offset)
|
|
else: assigned[buf] = assigned.get(buf, buf)
|
|
|
|
if DEBUG >= 1 and len(ak:=dedup(x for x in assigned.keys() if x._base is None)) != len(av:=dedup(x for x in assigned.values() if x._base is None)):
|
|
print(debug_prefix+f"memory reduced from {sum([x.nbytes for x in ak])/1e6:.2f} MB -> {sum([x.nbytes for x in av])/1e6:.2f} MB,",
|
|
f"{len(ak)} -> {len(av)} bufs")
|
|
return assigned
|
|
|
|
def memory_planner(schedule:List[ScheduleItem]) -> List[ScheduleItem]:
|
|
# Exclude buffers involved in load ops (e.g transfers) to preserve parallelism in graphs.
|
|
assigned = _internal_memory_planner([si.bufs for si in schedule],
|
|
noopt_buffers={b for si in schedule if si.ast.op is not UOps.SINK for b in si.bufs})
|
|
return [ScheduleItem(si.ast, tuple(assigned.get(x, x) for x in si.bufs), si.metadata) for si in schedule]
|