def element_wise_invalid_annotation(a: ty.handle, c: ty.handle) -> None:
C = tir.match_buffer(c, [128, 128],
elem_offset=0,
align=128,
offset_factor=1)
A = tir.match_buffer(a, [128, 128],
elem_offset=0,
align=128,
offset_factor=1)
# body
with tir.block([], "root"):
tir.reads([])
tir.writes([])
B = tir.alloc_buffer([128, 128],
elem_offset=0,
align=128,
offset_factor=1)
for i0 in tir.serial(0, 128):
for ax1 in tir.serial(0, 128):
with tir.block([128, 128], "B") as [vi, vj]:
tir.block_attr({"buffer_dim_align": [0]})
tir.bind(vi, i0)
tir.bind(vj, ax1)
tir.reads([A[vi, vj]])
tir.writes([B[vi, vj]])
B[vi, vj] = (A[vi, vj] * tir.float32(2))
for i1 in tir.serial(0, 128):
with tir.block([128, 128], "C") as [vi_1, vj_1]:
tir.bind(vi_1, i0)
tir.bind(vj_1, i1)
tir.reads([B[vi_1, vj_1]])
tir.writes([C[vi_1, vj_1]])
C[vi_1, vj_1] = (B[vi_1, vj_1] + tir.float32(1))
def compacted_storage_align_func(a: ty.handle, c: ty.handle) -> None:
A = tir.match_buffer(a, (16, 16), "float32")
C = tir.match_buffer(c, (16, 16), "float32")
for i in range(0, 16):
with tir.block([]):
tir.reads(A[i, 0:16])
tir.writes(C[i, 0:16])
B = tir.alloc_buffer((1, 16), strides=(31, 1), dtypes="float32")
for j in range(0, 16):
with tir.block() as []:
tir.reads(A[i, j])
tir.writes(B[0, j])
tir.block_attr({"buffer_dim_align": [[0, 0, 16, 15]]})
B[0, j] = A[i, j] + 1.0
for j in range(0, 16):
with tir.block() as []:
tir.reads(B[0, j])
tir.writes(C[i, j])
C[i, j] = B[0, j] * 2.0
def duplicate_annotations() -> None:
with tir.block([16, 16]) as [vi, vj]:
tir.block_attr({})
tir.block_attr({}) # error
