Last active
November 22, 2022 13:28
-
-
Save Laurawly/5288b88a091ed4646a5840317ab59d84 to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import numpy as np | |
import tvm | |
from tvm import te, auto_scheduler, topi | |
from tvm.topi.testing import conv2d_nchw_python | |
from tvm.contrib import cublas | |
target = tvm.target.Target('cuda') | |
M = 8192 | |
N = 2304 | |
K = 768 | |
A = te.placeholder((M, K), name='data', dtype='float16') | |
B = te.placeholder((N, K), name='kernel', dtype='float16') | |
C = cublas.matmul(A, B, False, True, dtype='float32') | |
sch = te.create_schedule(C.op) | |
args = [A, B, C] | |
func = tvm.build(sch, args, target) | |
# Check correctness | |
data_np = np.random.uniform(size=(M, K)).astype(np.float16) | |
weight_np = np.random.uniform(size=(N, K)).astype(np.float16) | |
out_np = np.matmul(data_np, weight_np.T) | |
ctx = tvm.gpu() | |
data_tvm = tvm.nd.array(data_np, ctx=ctx) | |
weight_tvm = tvm.nd.array(weight_np, ctx=ctx) | |
out_tvm = tvm.nd.array(np.zeros((M, N), dtype=C.dtype), ctx=ctx) | |
func(data_tvm, weight_tvm, out_tvm) | |
# Check results | |
np.testing.assert_allclose(out_np, out_tvm.asnumpy(), rtol=1e-3) | |
# Evaluate execution time | |
evaluator = func.time_evaluator(func.entry_name, ctx, number=100, repeat=10) | |
time = np.median(evaluator(data_tvm, weight_tvm, out_tvm).results) | |
print("shape", data_np.shape, weight_np.shape) | |
print("Execution time of this operator: %.3f ms" % (time * 1000)) | |
print("Speed: %.3f TFLOPS" % (2 * (M*N*K) / time / 1e12)) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment