[GitHub] [tvm] csullivan commented on a diff in pull request #12587: [Hexagon] Initial support for meta schedule tuning

[GitBox <gi...@apache.org>]

csullivan commented on code in PR #12587:
URL: https://github.com/apache/tvm/pull/12587#discussion_r955391842


##########
tests/python/contrib/test_hexagon/test_meta_schedule.py:
##########
@@ -0,0 +1,211 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+
+""" Test rpc based launcher for hexagon """
+import pytest
+import numpy as np
+import tempfile
+
+import tvm.testing
+from tvm import te
+from tvm import meta_schedule as ms
+from tvm.meta_schedule.arg_info import TensorInfo
+from tvm.meta_schedule.builder import BuilderInput
+from tvm.script import tir as T
+from tvm.tir import FloatImm
+from tvm.tir.tensor_intrin.hexagon import VRMPY_u8u8i32_INTRIN
+from tvm.meta_schedule.runner import RunnerInput
+from tvm.contrib.hexagon.meta_schedule import get_hexagon_local_builder, get_hexagon_rpc_runner
+
+MATMUL_N = 16
+MATMUL_M = 32
+
+
+@tvm.script.ir_module
+class MatmulModule:
+    @T.prim_func
+    def main(a: T.handle, b: T.handle, c: T.handle) -> None:  # pylint: disable=no-self-argument
+        T.func_attr({"global_symbol": "main", "tir.noalias": True})
+        A = T.match_buffer(a, (16, 16), "float32")
+        B = T.match_buffer(b, (16, 16), "float32")
+        C = T.match_buffer(c, (16, 16), "float32")
+        for i, j, k in T.grid(16, 16, 16):
+            with T.block("matmul"):
+                vi, vj, vk = T.axis.remap("SSR", [i, j, k])
+                with T.init():
+                    C[vi, vj] = 0.0
+                C[vi, vj] = C[vi, vj] + A[vi, vk] * B[vk, vj]
+
+
+@tvm.testing.requires_hexagon
+def test_builder_runner(hexagon_launcher):
+    if hexagon_launcher._serial_number == "simulator":
+        pytest.skip(msg="Tuning on simulator not supported.")
+
+    target_hexagon = tvm.target.hexagon("v68", link_params=True)
+    target = tvm.target.Target(target_hexagon, host=target_hexagon)
+    mod = MatmulModule
+
+    builder = get_hexagon_local_builder()
+    runner = get_hexagon_rpc_runner(hexagon_launcher, number=1, repeat=1, min_repeat_ms=0)
+
+    (builder_result,) = builder.build([BuilderInput(mod, target)])
+    assert builder_result.artifact_path is not None
+    assert builder_result.error_msg is None
+
+    runner_input = RunnerInput(
+        builder_result.artifact_path,
+        "llvm",
+        [
+            TensorInfo("float32", (MATMUL_N, MATMUL_N)),
+            TensorInfo("float32", (MATMUL_N, MATMUL_N)),
+            TensorInfo("float32", (MATMUL_N, MATMUL_N)),
+        ],
+    )
+
+    # Run the module
+    (runner_future,) = runner.run([runner_input])
+    runner_result = runner_future.result()
+
+    assert runner_result.error_msg is None
+    for result in runner_result.run_secs:
+        if isinstance(result, FloatImm):
+            result = result.value
+        assert isinstance(result, float)
+        assert result >= 0.0
+
+
+def dense(m, n, k):
+    X = te.placeholder((m, k), name="X", dtype="uint8")
+    packedW = te.placeholder((n // 32, k // 4, 32, 4), name="packedW", dtype="uint8")
+
+    ak = te.reduce_axis((0, k), name="k")
+    out = te.compute(
+        (m, n),
+        lambda i, j: te.sum(
+            X[i, ak].astype("int32")
+            * packedW[tvm.tir.indexdiv(j, 32), tvm.tir.indexdiv(ak, 4), j % 32, ak % 4].astype(
+                "int32"
+            ),
+            axis=ak,
+        ),
+        name="compute",
+    )
+    return [X, packedW, out]
+
+
+def schedule_dense(sch, block, M, do_tune):
+    a_y, a_x, _ = sch.get_loops(block)[-3:]
+
+    if do_tune:
+        y_factors = sch.sample_perfect_tile(a_y, n=2, max_innermost_factor=128)
+        a_yo, a_yi = sch.split(a_y, factors=y_factors)
+    else:
+        a_yo, a_yi = sch.split(a_y, factors=[None, min(M, 32)])
+
+    a_xo, a_xi = sch.split(a_x, factors=[None, 32])
+    sch.reorder(a_yo, a_xo, a_yi, a_xi)
+
+    a_xi, a_k = sch.get_loops(block)[-2:]
+    a_ko, a_ki = sch.split(a_k, factors=[None, 4])
+    sch.reorder(a_ko, a_xi, a_ki)
+
+    fused = sch.fuse(a_yo, a_xo)
+
+    sch.parallel(fused)
+
+    dec = sch.decompose_reduction(block, a_ko)
+
+    init_loop = sch.get_loops(dec)[-1]
+    sch.vectorize(init_loop)
+
+    sch.tensorize(a_xi, VRMPY_u8u8i32_INTRIN)
+
+
+def verify_dense(sch, target, M, N, K, hexagon_session):
+    f = tvm.build(sch.mod["main"], target=target, name="dense")
+    mod = hexagon_session.load_module(f)
+    dev = hexagon_session.device
+
+    a_np = np.random.uniform(1, 10, size=(M, K)).astype("uint8")
+    b_np = np.random.uniform(1, 10, size=(N, K)).astype("uint8")
+    c_np = np.dot(a_np.astype("int32"), b_np.transpose().astype("int32"))
+
+    packW = np.random.uniform(1, 10, size=(N // 32, (K // 4), 32, 4)).astype("uint8")
+
+    for r_idx in range(N // 32):
+        for ko in range(K // 4):
+            for s_idx in range(32):
+                for t_idx in range(4):
+                    packW[r_idx][ko][s_idx][t_idx] = b_np[r_idx * 32 + s_idx][ko * 4 + t_idx]
+
+    a = tvm.nd.array(a_np, dev)
+    b = tvm.nd.array(packW, dev)
+    c = tvm.nd.array(np.zeros((M, N), dtype="int32"), dev)
+
+    mod(a, b, c)
+    np.testing.assert_equal(c.numpy(), c_np)
+
+    evaluator = mod.time_evaluator(mod.entry_name, dev, number=10)
+    gflops = (N * M * K) * 2 / 1e9
+    time_ms = evaluator(a, b, c).mean * 1e3
+    print("%f ms, %f GOPS" % (time_ms, gflops / (time_ms / 1e3)))
+
+
+@pytest.mark.skip(reason="xgboost not installed on CI")
+@tvm.testing.requires_hexagon
+def test_vrmpy_dense(hexagon_launcher):
+    if hexagon_launcher._serial_number == "simulator":

Review Comment:
   It would be really great for this to be tested pre-commit on the simulator so future changes don't regress on the ability to tune. @kparzysz-quic do you think we could remove the use of a local x86 rpc server for targeting the hexagon simulator? Goal would be to make HexagonLauncherSimulator pickleable.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org