mirror of
https://github.com/tile-ai/tilelang.git
synced 2026-10-21 06:40:23 +08:00
Lei Wang
8c8d8ca23b
* [Refactor] Rename `jit` class to `_JitImplementation` and improve debug path handling * Refactored the `jit` class to `_JitImplementation` for clarity and encapsulation. * Enhanced handling of `debug_root_path` to ensure it is correctly set as an absolute path when provided. * Updated the public `jit` function to serve as a decorator interface, allowing for both default and configured usage. * Added validation to ensure input tensors are contiguous in the Cython wrapper, improving error handling. * [Refactor] Improve formatting and handling in `_JitImplementation` and `jit` function * Refactored the `_JitImplementation` class to enhance readability by adjusting comment formatting and consolidating conditions for setting `debug_root_path`. * Updated the `jit` function signature for better alignment and clarity in parameter definitions. * Ensured consistent spacing and comments throughout the code for improved maintainability. * [Refactor] Update GEMM test parameters for performance optimization * Set num_stages to 0 and adjusted matrix dimensions in the GEMM test function to enhance performance and consistency across tests in test_tilelang_jit_gemm.py. * Reduced the number of threads used in the test to align with the updated configuration, improving overall test efficiency. * [Refactor] Enhance buffer error logging in layout inference * Updated the warning message in layout inference to provide clearer context when a buffer cannot be inferred due to its absence in the use list. This change improves the clarity of error reporting during layout inference operations. * Refactored tensor handling in the Cython wrapper to ensure input tensors are checked for contiguity before processing, enhancing error handling and robustness in tensor management. * bugfix
128 lines
3.0 KiB
Python
128 lines
3.0 KiB
Python
from tilelang import tvm as tvm
|
|
import tilelang.testing
|
|
import tilelang
|
|
import torch
|
|
|
|
|
|
@tilelang.jit(
|
|
out_idx=-1, # create the output tensor during runtime
|
|
)
|
|
def matmul_kernel_jit(
|
|
M,
|
|
N,
|
|
K,
|
|
block_M,
|
|
block_N,
|
|
block_K,
|
|
trans_A,
|
|
trans_B,
|
|
in_dtype,
|
|
out_dtype,
|
|
accum_dtype,
|
|
num_stages,
|
|
threads,
|
|
):
|
|
A_shape = (K, M) if trans_A else (M, K)
|
|
B_shape = (N, K) if trans_B else (K, N)
|
|
A_shared_shape = (block_K, block_M) if trans_A else (block_M, block_K)
|
|
B_shared_shape = (block_N, block_K) if trans_B else (block_K, block_N)
|
|
|
|
import tilelang.language as T
|
|
|
|
@T.prim_func
|
|
def main(
|
|
A: T.Tensor(A_shape, in_dtype),
|
|
B: T.Tensor(B_shape, in_dtype),
|
|
C: T.Tensor((M, N), out_dtype),
|
|
):
|
|
with T.Kernel(T.ceildiv(N, block_N), T.ceildiv(M, block_M), threads=threads) as (bx, by):
|
|
A_shared = T.alloc_shared(A_shared_shape, in_dtype)
|
|
B_shared = T.alloc_shared(B_shared_shape, in_dtype)
|
|
C_local = T.alloc_fragment((block_M, block_N), accum_dtype)
|
|
T.clear(C_local)
|
|
for k in T.Pipelined(T.ceildiv(K, block_K), num_stages=num_stages):
|
|
if trans_A:
|
|
T.copy(A[k * block_K, by * block_M], A_shared)
|
|
else:
|
|
T.copy(A[by * block_M, k * block_K], A_shared)
|
|
if trans_B:
|
|
T.copy(B[bx * block_N, k * block_K], B_shared)
|
|
else:
|
|
T.copy(B[k * block_K, bx * block_N], B_shared)
|
|
T.gemm(A_shared, B_shared, C_local, trans_A, trans_B)
|
|
T.copy(C_local, C[by * block_M, bx * block_N])
|
|
|
|
return main
|
|
|
|
|
|
def run_gemm_kernel_jit(
|
|
M,
|
|
N,
|
|
K,
|
|
trans_A,
|
|
trans_B,
|
|
in_dtype,
|
|
out_dtype,
|
|
dtypeAccum,
|
|
block_M,
|
|
block_N,
|
|
block_K,
|
|
num_stages=0,
|
|
num_threads=128,
|
|
):
|
|
matmul_kernel = matmul_kernel_jit(
|
|
M,
|
|
N,
|
|
K,
|
|
block_M,
|
|
block_N,
|
|
block_K,
|
|
trans_A,
|
|
trans_B,
|
|
in_dtype,
|
|
out_dtype,
|
|
dtypeAccum,
|
|
num_stages,
|
|
num_threads,
|
|
)
|
|
|
|
A = torch.randn(M, K, dtype=torch.__getattribute__(in_dtype)).cuda()
|
|
B = torch.randn(K, N, dtype=torch.__getattribute__(in_dtype)).cuda()
|
|
|
|
if trans_A:
|
|
A = A.T
|
|
if trans_B:
|
|
B = B.T
|
|
|
|
def ref_program(A, B):
|
|
import torch
|
|
C = torch.matmul(A.to(torch.float), B.to(torch.float))
|
|
C = C.to(torch.__getattribute__(out_dtype))
|
|
return C
|
|
|
|
ref_C = ref_program(A, B)
|
|
C = matmul_kernel(A, B)
|
|
|
|
tilelang.testing.torch_assert_close(C, ref_C, atol=1e-2, rtol=1e-2, max_mismatched_ratio=0.05)
|
|
|
|
|
|
def test_gemm_f16f16f16_nn_kernel_jit():
|
|
run_gemm_kernel_jit(
|
|
512,
|
|
1024,
|
|
768,
|
|
False,
|
|
False,
|
|
"float16",
|
|
"float16",
|
|
"float16",
|
|
128,
|
|
128,
|
|
32,
|
|
0,
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
tilelang.testing.main()
|