mirror of
https://github.com/tile-ai/tilelang.git
synced 2026-10-21 06:40:23 +08:00
Tong WUandLeiWang1999
228c7c0486
* Add experimental IKET CUDA backend hooks * Keep IKET APIs under namespace * Use real Perfetto screenshot for IKET example * Show IKET events in Perfetto timeline screenshot * Add IKET validation overhead example * Remove IKET validation overhead example * Fix IKET lint issues * Move IKET integration to CUDA tools * Add tools documentation and fix workflow regressions --------- Co-authored-by: LeiWang1999 <leiwang1999@outlook.com>
72 lines
2.0 KiB
Python
72 lines
2.0 KiB
Python
import tilelang
|
|
import tilelang.language as T
|
|
from tilelang.tools import Analyzer
|
|
|
|
|
|
class _Device:
|
|
compute_capability = "80"
|
|
bandwidth = [750, 12080]
|
|
|
|
|
|
@tilelang.jit
|
|
def _gemm(A, B):
|
|
M = N = K = 64
|
|
block_M = block_N = block_K = 32
|
|
|
|
A: T.Tensor((M, K), T.float16)
|
|
B: T.Tensor((N, K), T.float16)
|
|
C = T.empty((M, N), T.float16)
|
|
|
|
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), T.float16)
|
|
B_shared = T.alloc_shared((block_N, block_K), T.float16)
|
|
C_local = T.alloc_fragment((block_M, block_N), T.float32)
|
|
|
|
T.clear(C_local)
|
|
for k in T.serial(T.ceildiv(K, block_K)):
|
|
T.copy(A[by * block_M, k * block_K], A_shared)
|
|
T.copy(B[bx * block_N, k * 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 C
|
|
|
|
|
|
@tilelang.jit
|
|
def _single_block_gemm(A, B):
|
|
M = N = K = 32
|
|
|
|
A: T.Tensor((M, K), T.float16)
|
|
B: T.Tensor((N, K), T.float16)
|
|
C = T.empty((M, N), T.float16)
|
|
|
|
with T.Kernel(1, threads=128):
|
|
A_shared = T.alloc_shared((M, K), T.float16)
|
|
B_shared = T.alloc_shared((N, K), T.float16)
|
|
C_local = T.alloc_fragment((M, N), T.float32)
|
|
|
|
T.copy(A, A_shared)
|
|
T.copy(B, B_shared)
|
|
T.clear(C_local)
|
|
T.gemm(A_shared, B_shared, C_local, transpose_B=True)
|
|
T.copy(C_local, C)
|
|
|
|
return C
|
|
|
|
|
|
def test_analyzer_recognizes_tileop_names():
|
|
result = Analyzer.analysis(_gemm.get_tir(), _Device())
|
|
|
|
assert result.total_flops == 2 * 64 * 64 * 64
|
|
assert result.total_global_bytes == 40_960
|
|
|
|
|
|
def test_analyzer_resets_grid_extents_for_each_prim_func():
|
|
analyzer = Analyzer(_single_block_gemm.get_tir(), _Device())
|
|
analyzer.block_counts["blockIdx.y"] = 7
|
|
|
|
result = analyzer.ir_pass().calculate()
|
|
|
|
assert result.total_flops == 2 * 32 * 32 * 32
|
|
assert result.total_global_bytes == 6_144
|