tinygrad/tinygrad
2
0
Fork 3
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-08-29 10:56:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into real_pyrender

Browse Source
This commit is contained in:
George Hotz
2025-10-27 17:39:37 +08:00
committed by GitHub
parent 533f81bfc8 701a632907
commit ab7d8c24be
2 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/test_uops.py
+8 -2
View File
@@ -548,10 +548,16 @@ class TestUopsObject(unittest.TestCase):
class TestUOpRender(unittest.TestCase):
def test_render_vectorize_same(self):
u = UOp(Ops.VECTORIZE, dtype=dtypes.int.vec(3), src=(UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 0)))
self.assertEqual(u.render(), "{0, ...}")
self.assertEqual(u.render(simplify=False), "{0, ...}")
def test_render_vectorize_different(self):
u = UOp(Ops.VECTORIZE, dtype=dtypes.int.vec(3), src=(UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 1), UOp.const(dtypes.int, 2)))
self.assertEqual(u.render(), "{0,1,2}")
self.assertEqual(u.render(simplify=False), "{0,1,2}")
def test_render_vectorize_same_simplified(self):
u = UOp(Ops.VECTORIZE, dtype=dtypes.int.vec(3), src=(UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 0)))
self.assertEqual(u.render(), "0")
def test_render_vectorize_different_simplified(self):
u = UOp(Ops.VECTORIZE, dtype=dtypes.int.vec(3), src=(UOp.const(dtypes.int, 0), UOp.const(dtypes.int, 1), UOp.const(dtypes.int, 2)))
self.assertEqual(u.render(), "(0, 1, 2)")
if __name__ == '__main__':
unittest.main(verbosity=2)
tinygrad/runtime/support/nv/nvdev.py
+3 -1
View File
@@ -71,7 +71,7 @@ class NVMemoryManager(MemoryManager):
class NVDev(PCIDevImplBase):
def __init__(self, pci_dev:PCIDevice):
self.pci_dev, self.devfmt, self.vram, self.mmio = pci_dev, pci_dev.pcibus, pci_dev.map_bar(1), pci_dev.map_bar(0, fmt='I')
self.pci_dev, self.devfmt, self.mmio = pci_dev, pci_dev.pcibus, pci_dev.map_bar(0, fmt='I')
self.lock_fd = System.flock_acquire(f"nv_{self.devfmt}.lock")
@@ -134,6 +134,8 @@ class NVDev(PCIDevImplBase):
self.pte_t, self.pde_t, self.dual_pde_t = tuple([self.__dict__[name] for name in mmu_pd_names])
self.vram_size = self.reg("NV_PGC6_AON_SECURE_SCRATCH_GROUP_42").read() << 20
self.vram, self.mmio = self.pci_dev.map_bar(1), self.pci_dev.map_bar(0, fmt='I')
self.large_bar = self.vram.nbytes >= self.vram_size
def _alloc_boot_struct(self, struct:ctypes.Structure) -> tuple[ctypes.Structure, int]: