github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

fix DEBUG=2 output for copy runners [pr] (#8579)

Browse Source 
* fix DEBUG=2 output for copy runners [pr]

* itemsize is constant
This commit is contained in:
qazal
2025-01-12 12:03:01 -05:00
committed by GitHub
parent 867004fbeb
commit cde18fddce
2 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
test/test_uops_stats.py
View File

@@ -7,6 +7,7 @@ from tinygrad.codegen.linearize import linearize_uop
from tinygrad.ops import Ops, UOp
from tinygrad.dtype import dtypes
from tinygrad.codegen.kernel import Kernel, Opt, OptOps, KernelOptError
from tinygrad.device import Device
def flops_mem(uops, ignore_indexing=False):
est = Estimates.from_uops(uops, ignore_indexing)
@@ -64,6 +65,15 @@ class TestMemoryCount(unittest.TestCase):
_, mem = get_stats(a.assign(a+a))
self.assertEqual(mem, 1024*1024*2) # 1 read + 1 write
@unittest.skipIf(Device.DEFAULT == "CLANG", "test copy to CLANG from other device")
def test_copyout(self):
a = Tensor.empty(32, dtype=dtypes.uint8).to("CLANG")
_, mem = get_stats(a)
self.assertEqual(mem, 32*1)
a = Tensor.empty(32, dtype=dtypes.uint32).to("CLANG")
_, mem = get_stats(a)
self.assertEqual(mem, 32*4)
# NOTE: this still isn't testing unroll using the acc
@unittest.skipUnless(getenv("PYTHON"), "only run test on emulated tensor cores")
class TestUOpsStatsMatmulHalf(unittest.TestCase):