github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-01-23 05:48:08 -05:00
Code Issues Packages Projects Releases Wiki Activity

fix op estimate

Browse Source
This commit is contained in:
George Hotz
2022-08-06 14:15:50 +00:00
parent f2847cb710
commit 94d526f8fc
2 changed files with 4 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
accel/MAPPING
View File

@@ -18,9 +18,6 @@ On M1 GPU, theoretical is 2.275 TFLOPS. https://www.notebookcheck.net/Apple-M1-G
We observe 2000ms for BS=8 (37 GFLOP). 37/2275 = 11.9 ms. tinygrad is over a factor of 100x off (similar on AMD GPU)
Making matters worse, flop estimates may be off by a factor of 10.
TPUs:

8
tinygrad/llops/ops_gpu.py
View File

@@ -50,14 +50,14 @@ class CLProgram:
self.clprg = self.clprogram.build(options=list(self.options)).__getattr__(self.name)
if self.argdtypes is not None: self.clprg.set_scalar_arg_dtypes(self.argdtypes)
CLProgram.kernel_cnt += 1
def __call__(self, *args):
def __call__(self, *args, op_estimate=0):
CL.kernel_count += 1
if CL.CACHE is not None: CL.CACHE.append((self, args))
else: e = self.clprg(CL().cl_queue, *args)
if DEBUG >= 2: CL.cl_queue.finish()
if DEBUG >= 1:
CL.ops_sum += max([x.size//4 for x in args[2:]])*(len(args)-2)
print(f"**CL** {CL.kernel_count:6d} {self.name:20s} args {len(args[2:]):5d} size {prod(args[0]):8d} kernels {str(args[0]):20s} {str(args[1]):14s} GOPs {CL.ops_sum/1e9:7.2f} " + \
CL.ops_sum += op_estimate
print(f"**CL** {CL.kernel_count:6d} {self.name:20s} args {len(args[2:]):5d} size {prod(args[0]):8d} kernels {str(args[0]):20s} {str(args[1]):14s} OPs {op_estimate/1e6:5.1f}M/{CL.ops_sum/1e9:7.2f}G " + \
("" if DEBUG <= 1 or CL.CACHE is not None else f"runtime {(e.profile.end - e.profile.start)/1e3:9.2f} us"))
if DEBUG >= 4: print(self.prg)
@@ -143,5 +143,5 @@ class GPUBuffer:
{chr(10).join([f' float {name} = ' + (f'get_{name}({name}_g, idx);' if views[name][1] else f'get_{name}(idx);') for name, _ in bufs if name not in earlybufs])}
output[gid] = {code};
}}""", argdtypes=tuple(None if i < 1+len(buf_types) else np.int32 for i in range(1+len(buf_types))))
conv_prg([prod(ret.shape), 1, 1], None, ret.cl, *[buf.cl for name, buf in bufs if name not in views or views[name][1]])
conv_prg([prod(ret.shape), 1, 1], None, ret.cl, *[buf.cl for name, buf in bufs if name not in views or views[name][1]], op_estimate=prod(reduce_shape[0])*len(earlybufs) + prod(reduce_shape[1])*len(bufs))
return ret