mirror of
https://github.com/tile-ai/tilelang.git
synced 2026-10-27 07:32:58 +08:00
Lei Wang
8ad538557f
* [Refactor] Improve flash attention example and layout comparison logic - Removed unnecessary annotation for `lse_local_split` in the flash attention example to streamline the code. - Updated the handling of `lse_local_split` to utilize parallel processing for better performance. - Refactored kernel compilation and profiling logic to enhance clarity and maintainability in the flash attention example. - Added a condition in `FragmentNode::IsEqual` to handle broadcast cases, improving the robustness of layout comparisons. * lint fix * [Enhancement] Add support for shared memory scope in Fill operation - Introduced handling for `shared.dyn` and `shared` memory scopes in the Fill operation. - Implemented parallel operation and layout inference for improved performance in shared memory scenarios. - Updated thread loop partitioning and vectorization logic to accommodate new memory scope handling. * [Refactor] Remove deprecated decorator and enhance Cython kernel handling - Removed the deprecated decorator from the main module and added a new implementation in the utils module for better organization. - Introduced a pointer map in the Cython kernel adapter to manage pointer arguments, improving runtime shape resolution. - Updated the Cython kernel wrapper to utilize the new pointer map for handling kernel arguments. - Enhanced error checking in the tensor utility functions to ensure static shapes are enforced. - Added a new proxy module for buffer and tensor handling, streamlining the interface for TIR programs. * [Feature] Add matrix multiplication test and kernel implementation - Introduced a new test file `test_tilelang_language_ptr.py` that implements a matrix multiplication function using TileLang's primitives. - The `matmul_test` function defines a kernel for performing tile-level GEMM operations with customizable block sizes and data types. - Added a `run_matmul` function to compile and execute the kernel, along with a test function to validate the implementation. - Updated the `proxy.py` file to enhance type handling for buffer and tensor proxies, ensuring compatibility with TIR programs. - Minor formatting improvements in `deprecated.py` for better readability. * lint fix
68 lines
2.1 KiB
Python
68 lines
2.1 KiB
Python
import torch
|
|
from tilelang import tvm as tvm
|
|
import tilelang.testing
|
|
import tilelang as tl
|
|
import tilelang.language as T
|
|
from tilelang.utils import map_torch_type
|
|
|
|
|
|
def matmul_test(M, N, K, block_M, block_N, block_K, dtype="float16", accum_dtype="float"):
|
|
|
|
@T.prim_func
|
|
def main(
|
|
a_ptr: T.ptr,
|
|
b_ptr: T.ptr,
|
|
c_ptr: T.ptr,
|
|
m: T.int32,
|
|
n: T.int32,
|
|
k: T.int32,
|
|
):
|
|
A = T.Tensor.from_ptr(a_ptr, (m, k), dtype)
|
|
B = T.Tensor.from_ptr(b_ptr, (k, n), dtype)
|
|
C = T.Tensor.from_ptr(c_ptr, (m, n), accum_dtype)
|
|
|
|
# Initialize Kernel Context
|
|
with T.Kernel(T.ceildiv(N, block_N), T.ceildiv(M, block_M), threads=128) as (bx, by):
|
|
A_shared = T.alloc_shared((block_M, block_K), dtype)
|
|
B_shared = T.alloc_shared((block_N, block_K), dtype)
|
|
C_local = T.alloc_fragment((block_M, block_N), accum_dtype)
|
|
|
|
T.clear(C_local)
|
|
|
|
for ko in T.Pipelined(T.ceildiv(k, block_K), num_stages=3):
|
|
# Copy tile of A
|
|
T.copy(A[by * block_M, ko * block_K], A_shared)
|
|
T.copy(B[bx * block_N, ko * block_K], B_shared)
|
|
T.gemm(A_shared, B_shared, C_local, transpose_B=True)
|
|
|
|
T.copy(C_local, C[by * block_M, bx * block_N])
|
|
|
|
return main
|
|
|
|
|
|
def run_matmul(M, N, K, block_M, block_N, block_K, dtype="float16", accum_dtype="float"):
|
|
program = matmul_test(M, N, K, block_M, block_N, block_K, dtype, accum_dtype)
|
|
jit_kernel = tl.compile(program, target="cuda", execution_backend="cython")
|
|
|
|
def ref_program(a, b):
|
|
return (a @ b.T).to(torch.float32)
|
|
|
|
a = torch.randn(M, K, device="cuda", dtype=map_torch_type(dtype))
|
|
b = torch.randn(N, K, device="cuda", dtype=map_torch_type(dtype))
|
|
|
|
c = torch.zeros(M, N, device="cuda", dtype=map_torch_type(accum_dtype))
|
|
|
|
jit_kernel(a.data_ptr(), b.data_ptr(), c.data_ptr(), M, N, K)
|
|
|
|
ref_c = (a @ b.T).to(map_torch_type(accum_dtype))
|
|
|
|
torch.testing.assert_close(c, ref_c, atol=1e-2, rtol=1e-2)
|
|
|
|
|
|
def test_matmul():
|
|
run_matmul(1024, 1024, 1024, 128, 128, 32)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
tilelang.testing.main()
|