[Feature] Unify the variable names for the alloc buffer.

[ArmandAlbert]

Feature Description

Unify the variable names for the alloc buffer.

Proposed Solution

The alloc_ub flag is set to shared.ub, the alloc_l1 flag is set to shared.l1, and the alloc_shared flag is set to shared. Then, the pass flag is changed to shared.ub or shared.l1.

tilelang-ascend/tilelang/language/allocate.py

Lines 32 to 47 in 35732a5

	def alloc_shared(shape, dtype, scope="shared.dyn"):
	"""Allocate a shared memory buffer for inter-thread communication.
	Args:
	shape (tuple): The shape of the buffer to allocate
	dtype (str): The data type of the buffer (e.g., 'float32', 'int32')
	scope (str, optional): The memory scope. Defaults to "shared.dyn"
	Returns:
	T.Buffer: A TVM buffer object allocated in shared memory
	"""
	if dtype == "bool":
	# lei: This is a hack to handle bool type.
	# Because tilelang's merge smem pass cannot merge bool type currently.
	scope = "shared"
	return T.alloc_buffer(shape, dtype, scope=scope)

tilelang-ascend/tilelang/language/allocate.py

Lines 154 to 171 in 35732a5

	def alloc_L1(shape, dtype):
	return T.alloc_buffer(shape, dtype, scope="shared.l1")
	def alloc_L0A(shape, dtype):
	return T.alloc_buffer(shape, dtype, scope="wmma.matrix_a")
	def alloc_L0B(shape, dtype):
	return T.alloc_buffer(shape, dtype, scope="wmma.matrix_b")
	def alloc_L0C(shape, dtype):
	return T.alloc_buffer(shape, dtype, scope="wmma.accumulator")
	def alloc_ub(shape, dtype):
	return T.alloc_buffer(shape, dtype, scope="shared")

Feature Type

• New operator/kernel (e.g., new GEMM variant, new attention mechanism)
• New API/primitive (e.g., new T.xxx operation)
• Compiler optimization (e.g., new pass, performance improvement)
• New backend support (e.g., new hardware target)
• Developer tooling (e.g., debugging, profiling)
• Documentation improvement
• Other: [specify]