gussmith23 commented on a change in pull request #5812: URL: https://github.com/apache/incubator-tvm/pull/5812#discussion_r469616234
########## File path: tests/python/unittest/test_custom_datatypes_change_dtype.py ########## @@ -0,0 +1,481 @@ +# 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. +"""Utilities for changing datatypes of models.""" +import tvm +import topi.testing +import numpy as np +from numpy.random import MT19937, RandomState, SeedSequence +from tvm import relay +from tvm.relay.testing.inception_v3 import get_workload as get_inception +from tvm.relay.testing.resnet import get_workload as get_resnet +from tvm.relay.testing.mobilenet import get_workload as get_mobilenet +from tvm.target.datatype import register, register_min_func, register_op, create_lower_func, lower_ite +from nose.tools import nottest + +tgt = "llvm" +# we use a random seed to generate input_data +# to guarantee stable tests +rs = RandomState(MT19937(SeedSequence(123456789))) + +def convert_ndarray(dst_dtype, *arrays): + """Converts NDArray(s) into the specified datatype""" + def convert(array): + x = relay.var('x', shape=array.shape, dtype=str(array.dtype)) + cast = relay.Function([x], x.astype(dst_dtype)) + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + return relay.create_executor('graph').evaluate(cast)(array) + + return tuple([convert(x) for x in arrays]) + + +def change_dtype(src, dst, module, params): + module = relay.frontend.ChangeDatatype(src, dst)(module) + module = relay.transform.InferType()(module) + params = dict((p, convert_ndarray(dst, params[p])) for p in params) + return module, params + +def compare(module, input, src_dtype, dst_dtype, rtol, atol, params = {}): + ex = relay.create_executor("graph", mod=module) + + correct = ex.evaluate()(*input, **params) + + module, _ = change_dtype(src_dtype, dst_dtype, module, []) + ex = relay.create_executor("graph", mod=module) + # converts all inputs to dst_dtype + x_converted = convert_ndarray(dst_dtype, *input) + + # Vectorization is not implemented with custom datatypes + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + maybe_correct = ex.evaluate()(*x_converted, **params) + # TODO(andrew) this only works on single output + maybe_correct_converted = convert_ndarray(src_dtype, maybe_correct)[0] + np.testing.assert_allclose(maybe_correct_converted.asnumpy(), + correct.asnumpy(), + rtol=rtol, + atol=atol) + +def setup(): + """Set up tests + + Currently, this registers some custom datatypes using the Bring Your + Own Datatypes framework. + """ + + # To use datatype operations in an external library, you should first load + # the library containing the datatype implementation: + # CDLL("libposit.so", RTLD_GLOBAL) + # In this case, the datatype library we are using is built right into TVM, + # so we do not need to explicitly load any library. + + # You can pick a code for your datatype arbitrarily, as long as it is + # greater than 128 and has not already been chosen. + + register("posites2", 131) + + register_op(create_lower_func( + { + (32, 32): "FloatToPosit32es2", + (32, 16): "FloatToPosit16es2", + (32, 8): 'FloatToPosit8es2', + }), + "Cast", "llvm", "posites2", "float") + register_op(create_lower_func( + { + (32, 32): "Posit32es2ToFloat", + (16, 32): 'Posit16es2ToFloat', + (8, 32): 'Posit8es2ToFloat', + }), + "Cast", "llvm", "float", "posites2") + register_op(create_lower_func( + { + (4, 32): 'IntToPosit32es2', + (4, 16): 'IntToPosit16es2', + (4, 8): 'IntToPosit8es2' + }), + "Cast", "llvm", "posites2", "int") + register_op(create_lower_func({ + 32: 'Posit32es2Add', + 16: 'Posit16es2Add', + 8: 'Posit8es2Add' + }), "Add", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sub', + 16: 'Posit16es2Sub', + 8: 'Posit8es2Sub' + }), "Sub", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'FloatToPosit32es2', + 16: 'FloatToPosit16es2', + 8: 'FloatToPosit8es2' + }), "FloatImm", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Mul', + 16: 'Posit16es2Mul', + 8: 'Posit8es2Mul' + }), "Mul", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Div', + 16: 'Posit16es2Div', + 8: 'Posit8es2Div' + }), "Div", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Max', + 16: 'Posit16es2Max', + 8: 'Posit8es2Max' + }), "Max", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sqrt', + 16: 'Posit16es2Sqrt', + 8: 'Posit8es2Sqrt' + }), "Call", "llvm", "posites2", intrinsic_name="sqrt") + register_op(lower_ite, + "Call", + "llvm", + "posites2", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({ + 32: 'Posit32es2Exp', + 16: 'Posit16es2Exp', + 8: 'Posit8es2Exp' + }), "Call", "llvm", "posites2", intrinsic_name="exp") + register_op(create_lower_func({ + 32: 'Posit32es2Log', + 16: 'Posit16es2Log', + 8: 'Posit8es2Log' + }), "Call", "llvm", "posites2", intrinsic_name="log") + register_op(create_lower_func({ + 32: 'Posit32es2Sigmoid', + 16: 'Posit16es2Sigmoid', + 8: 'Posit8es2Sigmoid' + }), "Call", "llvm", "posites2", intrinsic_name="sigmoid") + register_op(create_lower_func({ + 32: 'Posit32es2Tanh', + 16: 'Posit16es2Tanh', + 8: 'Posit8es2Tanh' + }), "Call", "llvm", "posites2", intrinsic_name="tanh") + register_min_func(lambda num_bits: - (2 ** 2 ** 2) ** (num_bits - 2), "posites2") + + register("noptype", 132) + register_op(create_lower_func({ + (32, 32): "FloatToNop32" + }), "Cast", "llvm", "noptype", "float") + register_op(create_lower_func({ + (32, 32): 'Nop32ToFloat' + }), "Cast", "llvm", "float", "noptype") + register_op(create_lower_func({ + (4, 32): "IntToNop32" + }), "Cast", "llvm", "noptype", "int") + register_op(create_lower_func({32: 'IntToNop32'}), "Add", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sub'}), "Sub", "llvm", "noptype") + register_op(create_lower_func({32: 'FloatToNop32'}), "FloatImm", "llvm", + "noptype") + register_op(create_lower_func({32: 'Nop32Mul'}), "Mul", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Div'}), "Div", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Max'}), "Max", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sqrt'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sqrt") + # TODO(gus) not sure if this will work... + register_op(lower_ite, + "Call", + "llvm", + "noptype", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({32: 'Nop32Exp'}), + "Call", + "llvm", + "noptype", + intrinsic_name="exp") + register_op(create_lower_func({32: 'Nop32Log'}), + "Call", + "llvm", + "noptype", + intrinsic_name="log") + register_op(create_lower_func({32: 'Nop32Sigmoid'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sigmoid") + register_op(create_lower_func({32: 'Nop32Tanh'}), + "Call", + "llvm", + "noptype", + intrinsic_name="tanh") + # This can be anything, considering the type isn't functionally correct. + register_min_func(lambda num_bits: 0, "noptype") + + +def run_ops(src_dtype, dst_dtype, rtol=1e-7, atol=1e-7): + """Run the same op, but with two different datatypes""" + def check_unary_op(op, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + x = relay.var("x", t1) + z = op(x) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + + module = tvm.IRModule.from_expr(relay.Function([x], z)) + + compare(module, (x_data, ), src_dtype, dst_dtype, rtol, atol) + # print(maybe_correct_converted) + # print(correct) + + for op in [ + relay.nn.softmax, + tvm.relay.log, + tvm.relay.exp, + tvm.relay.sqrt, + tvm.relay.rsqrt, + tvm.relay.sigmoid, + tvm.relay.tanh, + relay.nn.relu, + ]: + check_unary_op(op, src_dtype, dst_dtype) + + def check_binary_op(opfunc, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + t2 = relay.TensorType((5, ), src_dtype) + x = relay.var("x", t1) + y = relay.var("y", t2) + z = opfunc(x, y) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + y_data = rs.rand(5).astype(t2.dtype) + module = tvm.IRModule.from_expr(relay.Function([x, y], z)) + + compare(module, (x_data, y_data), src_dtype, dst_dtype, rtol, atol) + + for op in [ + relay.add, + relay.subtract, + relay.divide, + relay.multiply, + ]: + check_binary_op(op, src_dtype, dst_dtype) + + # we would like to test tvm_if_then_else + # but Relay.IfNode is not lowered to this intrinsic, + # so to keep our tests consistent with relay, we decide to not unit test + # Note: tvm_if_then_else is tested as part of the mobile_net model + + +def run_model(get_workload, + input_shape, + src_dtype, + dst_dtype, + num_classes, + rtol=0.0001, + atol=0.0001): + module, params = get_workload(image_shape=input_shape, + num_classes=num_classes) + + # Convert the input into the correct format. + input = tvm.nd.array(rs.rand(*input_shape).astype(src_dtype)) + + compare(module, (input, ), src_dtype, dst_dtype, rtol, atol, params) + + # # Simplifying inference is essential right now, as batch norms (which get + # # removed) are broken with custom datatypes. + # #expr = relay.ir_pass.simplify_inference(expr) + +def run_conv2d(src_dtype, dst_dtype): + def run_test_conv2d(src_dtype, + dst_dtype, + scale, + dshape, + kshape, + padding=(1, 1), + fref=None, + groups=1, + dilation=(1, 1), + except_targets=None, + **attrs): + if except_targets is None: + except_targets = [] + + x = relay.var("x", shape=dshape, dtype=src_dtype) + w = relay.var("w", shape=kshape, dtype=src_dtype) + y = relay.nn.conv2d(x, + w, + padding=padding, + dilation=dilation, + groups=groups, + **attrs) + module = tvm.IRModule.from_expr(relay.Function([x, w], y)) + data = rs.uniform(-scale, scale, size=dshape).astype(src_dtype) + kernel = rs.uniform(-scale, scale, + size=kshape).astype(src_dtype) + dkernel = topi.testing.dilate_python(kernel, (1, 1) + dilation) + if fref is None: + ref_res = topi.testing.conv2d_nchw_python( + data.astype(src_dtype), + dkernel.astype(src_dtype), + 1, + padding, + groups=groups) + else: + ref_res = fref(data.astype(src_dtype), dkernel.astype(src_dtype)) + + for target, ctx in [("llvm", tvm.cpu(0))]: + if target in except_targets: + continue + intrp1 = relay.create_executor("graph", + ctx=ctx, + target=target, + mod=module) + module, _ = change_dtype(src_dtype, dst_dtype, module, []) + data_converted = convert_ndarray(dst_dtype, data) + kernel_converted = convert_ndarray(dst_dtype, kernel) + with tvm.transform.PassContext( + config={"tir.disable_vectorize": True}): + op_res1 = intrp1.evaluate()(data_converted, kernel_converted) + op_res1_converted = convert_ndarray(src_dtype, op_res1) + tvm.testing.assert_allclose(op_res1_converted.asnumpy(), ref_res) + + # depthwise conv2d + dshape = (1, 32, 18, 18) + kshape = (32, 1, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=32, + groups=32, + kernel_size=(3, 3), + fref=lambda x, w: topi.testing. + depthwise_conv2d_python_nchw(x, w, (1, 1), "SAME")) + + # CUDA is disabled for 'direct' schedule: + # https://github.com/dmlc/tvm/pull/3070#issuecomment-486597553 + # group conv2d + dshape = (1, 32, 18, 18) + kshape = (32, 4, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=32, + groups=8, + kernel_size=(3, 3), + except_targets=['cuda']) + # also group conv2d + dshape = (1, 32, 18, 18) + kshape = (64, 1, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=64, + groups=32, + kernel_size=(3, 3), + except_targets=['cuda']) + + # normal conv2d + dshape = (1, 3, 224, 224) + kshape = (10, 3, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=10, + kernel_size=(3, 3)) + + # dilated conv2d + dshape = (1, 3, 18, 18) + kshape = (10, 3, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=10, + kernel_size=(3, 3), + dilation=(3, 3)) + + +def test_ops(): + # TODO(gus) these tolerances are high, and still sometimes fail; + # this is expected, b/c we're comparing between 32bit float and 8 + # bit posit. + # Figure out a more logical way to test here. + run_ops('float32', 'custom[posites2]8', rtol=1, atol=1) + run_ops('float32', 'custom[posites2]16', rtol=0.01, atol=1) + run_ops('float32', 'custom[posites2]32') + + +def test_conv2d(): + # TODO(@gussmith23) slow and broken, needing refactor! + # run_conv2d('float32', 'custom[posit32]32') + pass + + +def test_models(): + # Expected posit8 might be faster, but it's not. + # run_model(get_mobilenet, (3, 224, 224), 'float32', 'custom[posit8]8') + # run_model(get_mobilenet, (3, 224, 224), 'float32', 'custom[posit32]32') + # run_model(get_inception, (3, 299, 299), 'float32', 'custom[posit32]32') + # run_model(get_resnet, (3, 224, 224), 'float32', 'custom[posit32]32') + + # Run cifar-10 sizes to be a little faster... + run_model(get_mobilenet, (3, 32, 32), + 'float32', + 'custom[posites2]32', + num_classes=10) + # run_model(get_inception, (3, 32, 32), + # 'float32', + # 'custom[posit32]32', + # num_classes=10) + # run_model(get_resnet, (3, 32, 32), + # 'float32', + # 'custom[posit32]32', + # num_classes=10) + + # Meanwhile, noptype is not slow. + run_model(get_mobilenet, (3, 224, 224), Review comment: Let's remove these as well ########## File path: tests/python/unittest/test_custom_datatypes_change_dtype.py ########## @@ -81,163 +81,109 @@ def setup(): # You can pick a code for your datatype arbitrarily, as long as it is # greater than 128 and has not already been chosen. - register("posit32", 131) - - register_op(create_lower_func("FloatToPosit32es2"), "Cast", "llvm", - "posit32", "float") - register_op(create_lower_func("Posit32es2ToFloat"), "Cast", "llvm", - "float", "posit32") - register_op(create_lower_func("IntToPosit32es2"), "Cast", "llvm", - "posit32", "int") - register_op(create_lower_func("Posit32es2Add"), "Add", "llvm", "posit32") - register_op(create_lower_func("Posit32es2Sub"), "Sub", "llvm", "posit32") - register_op(create_lower_func("FloatToPosit32es2"), "FloatImm", "llvm", - "posit32") - register_op(create_lower_func("Posit32es2Mul"), "Mul", "llvm", "posit32") - register_op(create_lower_func("Posit32es2Div"), "Div", "llvm", "posit32") - register_op(create_lower_func("Posit32es2Max"), "Max", "llvm", "posit32") - register_op(create_lower_func("Posit32es2Sqrt"), - "Call", - "llvm", - "posit32", - intrinsic_name="sqrt") - # TODO(gus) not sure if this will work... - register_op(lower_ite, - "Call", - "llvm", - "posit32", - intrinsic_name="tvm_if_then_else") - register_op(create_lower_func("Posit32es2Exp"), - "Call", - "llvm", - "posit32", - intrinsic_name="exp") - register_op(create_lower_func("Posit32es2Log"), - "Call", - "llvm", - "posit32", - intrinsic_name="log") - register_op(create_lower_func("Posit32es2Sigmoid"), - "Call", - "llvm", - "posit32", - intrinsic_name="sigmoid") - register_op(create_lower_func("Posit32es2Tanh"), - "Call", - "llvm", - "posit32", - intrinsic_name="tanh") - register_min_func(lambda num_bits: -1.329227995784915872903807060280344576e36, "posit32") - - register("posit8", 132) - register_op(create_lower_func("FloatToPosit8es2"), "Cast", "llvm", - "posit8", "float") - register_op(create_lower_func("Posit8es2ToFloat"), "Cast", "llvm", "float", - "posit8") - register_op(create_lower_func("IntToPosit8es2"), "Cast", "llvm", "posit8", - "int") - register_op(create_lower_func("Posit8es2Add"), "Add", "llvm", "posit8") - register_op(create_lower_func("Posit8es2Sub"), "Sub", "llvm", "posit8") - register_op(create_lower_func("FloatToPosit8es2"), "FloatImm", "llvm", - "posit8") - register_op(create_lower_func("Posit8es2Mul"), "Mul", "llvm", "posit8") - register_op(create_lower_func("Posit8es2Div"), "Div", "llvm", "posit8") - register_op(create_lower_func("Posit8es2Max"), "Max", "llvm", "posit8") - register_op(create_lower_func("Posit8es2Sqrt"), - "Call", - "llvm", - "posit8", - intrinsic_name="sqrt") - # TODO(gus) not sure if this will work... - register_op(lower_ite, - "Call", - "llvm", - "posit8", - intrinsic_name="tvm_if_then_else") - register_op(create_lower_func("Posit8es2Exp"), - "Call", - "llvm", - "posit8", - intrinsic_name="exp") - register_op(create_lower_func("Posit8es2Log"), - "Call", - "llvm", - "posit8", - intrinsic_name="log") - register_op(create_lower_func("Posit8es2Sigmoid"), - "Call", - "llvm", - "posit8", - intrinsic_name="sigmoid") - register_op(create_lower_func("Posit8es2Tanh"), - "Call", - "llvm", - "posit8", - intrinsic_name="tanh") - register_min_func(lambda num_bits: -16777216, "posit8") - - register("posit16", 133) - register_op(create_lower_func("FloatToPosit16es2"), "Cast", "llvm", - "posit16", "float") - register_op(create_lower_func("Posit16es2ToFloat"), "Cast", "llvm", - "float", "posit16") - register_op(create_lower_func("IntToPosit16es2"), "Cast", "llvm", - "posit16", "int") - register_op(create_lower_func("Posit16es2Add"), "Add", "llvm", "posit16") - register_op(create_lower_func("Posit16es2Sub"), "Sub", "llvm", "posit16") - register_op(create_lower_func("FloatToPosit16es2"), "FloatImm", "llvm", - "posit16") - register_op(create_lower_func("Posit16es2Mul"), "Mul", "llvm", "posit16") - register_op(create_lower_func("Posit16es2Div"), "Div", "llvm", "posit16") - register_op(create_lower_func("Posit16es2Max"), "Max", "llvm", "posit16") - register_op(create_lower_func("Posit16es2Sqrt"), - "Call", - "llvm", - "posit16", - intrinsic_name="sqrt") - # TODO(gus) not sure if this will work... + register("posites2", 131) + + register_op(create_lower_func( + { + (32, 32): "FloatToPosit32es2", + (32, 16): "FloatToPosit16es2", + (32, 8): 'FloatToPosit8es2', + }), + "Cast", "llvm", "posites2", "float") Review comment: Make sure to change it everywhere (i.e. also in the python notebook that we want to merge in to this PR) ########## File path: python/tvm/target/datatype.py ########## @@ -134,10 +152,33 @@ def lower(op): dtype = "uint" + str(t.bits) if t.lanes > 1: dtype += "x" + str(t.lanes) - if isinstance(op, (_Cast, _FloatImm)): - return _Call(dtype, extern_func_name, convert([op.value]), + if isinstance(op, _Cast): + src_bits = bit_length(op.value.dtype) + return _Call(dtype, extern_func_map[(src_bits, t.bits)], convert([op.value]), + _Call.Extern) + elif isinstance(op, _FloatImm): + return _Call(dtype, extern_func_map[t.bits], convert([op.value]), _Call.Extern) - return _Call(dtype, extern_func_name, convert([op.a, op.b]), + elif isinstance(op, _Call) and (op.call_type == _Call.Intrinsic or + op.call_type == _Call.PureIntrinsic): + return _Call(dtype, extern_func_map[t.bits], convert(op.args), + _Call.Extern) + return _Call(dtype, extern_func_map[t.bits], convert([op.a, op.b]), Review comment: Let's figure out -- what happens in the unary case? How does this work for the unary case? Are all of the unary ops intrinsics? Does `convert([op.a, op.b])` handle the case where `op.b` is none? ########## File path: Makefile ########## @@ -90,6 +90,10 @@ scalalint: lint: cpplint pylint jnilint +# Test scripts +pyunittest: + ./tests/scripts/task_python_unittest.sh Review comment: I might have added this. We can probably remove it. ########## File path: tests/python/unittest/test_custom_datatypes_change_dtype.py ########## @@ -0,0 +1,481 @@ +# 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. +"""Utilities for changing datatypes of models.""" +import tvm +import topi.testing +import numpy as np +from numpy.random import MT19937, RandomState, SeedSequence +from tvm import relay +from tvm.relay.testing.inception_v3 import get_workload as get_inception +from tvm.relay.testing.resnet import get_workload as get_resnet +from tvm.relay.testing.mobilenet import get_workload as get_mobilenet +from tvm.target.datatype import register, register_min_func, register_op, create_lower_func, lower_ite +from nose.tools import nottest + +tgt = "llvm" +# we use a random seed to generate input_data +# to guarantee stable tests +rs = RandomState(MT19937(SeedSequence(123456789))) + +def convert_ndarray(dst_dtype, *arrays): + """Converts NDArray(s) into the specified datatype""" + def convert(array): + x = relay.var('x', shape=array.shape, dtype=str(array.dtype)) + cast = relay.Function([x], x.astype(dst_dtype)) + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + return relay.create_executor('graph').evaluate(cast)(array) + + return tuple([convert(x) for x in arrays]) + + +def change_dtype(src, dst, module, params): + module = relay.frontend.ChangeDatatype(src, dst)(module) + module = relay.transform.InferType()(module) + params = dict((p, convert_ndarray(dst, params[p])) for p in params) + return module, params + +def compare(module, input, src_dtype, dst_dtype, rtol, atol, params = {}): + ex = relay.create_executor("graph", mod=module) + + correct = ex.evaluate()(*input, **params) + + module, _ = change_dtype(src_dtype, dst_dtype, module, []) + ex = relay.create_executor("graph", mod=module) + # converts all inputs to dst_dtype + x_converted = convert_ndarray(dst_dtype, *input) + + # Vectorization is not implemented with custom datatypes + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + maybe_correct = ex.evaluate()(*x_converted, **params) + # TODO(andrew) this only works on single output + maybe_correct_converted = convert_ndarray(src_dtype, maybe_correct)[0] + np.testing.assert_allclose(maybe_correct_converted.asnumpy(), + correct.asnumpy(), + rtol=rtol, + atol=atol) + +def setup(): + """Set up tests + + Currently, this registers some custom datatypes using the Bring Your + Own Datatypes framework. + """ + + # To use datatype operations in an external library, you should first load + # the library containing the datatype implementation: + # CDLL("libposit.so", RTLD_GLOBAL) + # In this case, the datatype library we are using is built right into TVM, + # so we do not need to explicitly load any library. + + # You can pick a code for your datatype arbitrarily, as long as it is + # greater than 128 and has not already been chosen. + + register("posites2", 131) + + register_op(create_lower_func( + { + (32, 32): "FloatToPosit32es2", + (32, 16): "FloatToPosit16es2", + (32, 8): 'FloatToPosit8es2', + }), + "Cast", "llvm", "posites2", "float") + register_op(create_lower_func( + { + (32, 32): "Posit32es2ToFloat", + (16, 32): 'Posit16es2ToFloat', + (8, 32): 'Posit8es2ToFloat', + }), + "Cast", "llvm", "float", "posites2") + register_op(create_lower_func( + { + (4, 32): 'IntToPosit32es2', + (4, 16): 'IntToPosit16es2', + (4, 8): 'IntToPosit8es2' + }), + "Cast", "llvm", "posites2", "int") + register_op(create_lower_func({ + 32: 'Posit32es2Add', + 16: 'Posit16es2Add', + 8: 'Posit8es2Add' + }), "Add", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sub', + 16: 'Posit16es2Sub', + 8: 'Posit8es2Sub' + }), "Sub", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'FloatToPosit32es2', + 16: 'FloatToPosit16es2', + 8: 'FloatToPosit8es2' + }), "FloatImm", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Mul', + 16: 'Posit16es2Mul', + 8: 'Posit8es2Mul' + }), "Mul", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Div', + 16: 'Posit16es2Div', + 8: 'Posit8es2Div' + }), "Div", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Max', + 16: 'Posit16es2Max', + 8: 'Posit8es2Max' + }), "Max", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sqrt', + 16: 'Posit16es2Sqrt', + 8: 'Posit8es2Sqrt' + }), "Call", "llvm", "posites2", intrinsic_name="sqrt") + register_op(lower_ite, + "Call", + "llvm", + "posites2", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({ + 32: 'Posit32es2Exp', + 16: 'Posit16es2Exp', + 8: 'Posit8es2Exp' + }), "Call", "llvm", "posites2", intrinsic_name="exp") + register_op(create_lower_func({ + 32: 'Posit32es2Log', + 16: 'Posit16es2Log', + 8: 'Posit8es2Log' + }), "Call", "llvm", "posites2", intrinsic_name="log") + register_op(create_lower_func({ + 32: 'Posit32es2Sigmoid', + 16: 'Posit16es2Sigmoid', + 8: 'Posit8es2Sigmoid' + }), "Call", "llvm", "posites2", intrinsic_name="sigmoid") + register_op(create_lower_func({ + 32: 'Posit32es2Tanh', + 16: 'Posit16es2Tanh', + 8: 'Posit8es2Tanh' + }), "Call", "llvm", "posites2", intrinsic_name="tanh") + register_min_func(lambda num_bits: - (2 ** 2 ** 2) ** (num_bits - 2), "posites2") + + register("noptype", 132) + register_op(create_lower_func({ + (32, 32): "FloatToNop32" + }), "Cast", "llvm", "noptype", "float") + register_op(create_lower_func({ + (32, 32): 'Nop32ToFloat' + }), "Cast", "llvm", "float", "noptype") + register_op(create_lower_func({ + (4, 32): "IntToNop32" + }), "Cast", "llvm", "noptype", "int") + register_op(create_lower_func({32: 'IntToNop32'}), "Add", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sub'}), "Sub", "llvm", "noptype") + register_op(create_lower_func({32: 'FloatToNop32'}), "FloatImm", "llvm", + "noptype") + register_op(create_lower_func({32: 'Nop32Mul'}), "Mul", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Div'}), "Div", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Max'}), "Max", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sqrt'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sqrt") + # TODO(gus) not sure if this will work... + register_op(lower_ite, + "Call", + "llvm", + "noptype", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({32: 'Nop32Exp'}), + "Call", + "llvm", + "noptype", + intrinsic_name="exp") + register_op(create_lower_func({32: 'Nop32Log'}), + "Call", + "llvm", + "noptype", + intrinsic_name="log") + register_op(create_lower_func({32: 'Nop32Sigmoid'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sigmoid") + register_op(create_lower_func({32: 'Nop32Tanh'}), + "Call", + "llvm", + "noptype", + intrinsic_name="tanh") + # This can be anything, considering the type isn't functionally correct. + register_min_func(lambda num_bits: 0, "noptype") + + +def run_ops(src_dtype, dst_dtype, rtol=1e-7, atol=1e-7): + """Run the same op, but with two different datatypes""" + def check_unary_op(op, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + x = relay.var("x", t1) + z = op(x) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + + module = tvm.IRModule.from_expr(relay.Function([x], z)) + + compare(module, (x_data, ), src_dtype, dst_dtype, rtol, atol) + # print(maybe_correct_converted) + # print(correct) + + for op in [ + relay.nn.softmax, + tvm.relay.log, + tvm.relay.exp, + tvm.relay.sqrt, + tvm.relay.rsqrt, + tvm.relay.sigmoid, + tvm.relay.tanh, + relay.nn.relu, + ]: + check_unary_op(op, src_dtype, dst_dtype) + + def check_binary_op(opfunc, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + t2 = relay.TensorType((5, ), src_dtype) + x = relay.var("x", t1) + y = relay.var("y", t2) + z = opfunc(x, y) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + y_data = rs.rand(5).astype(t2.dtype) + module = tvm.IRModule.from_expr(relay.Function([x, y], z)) + + compare(module, (x_data, y_data), src_dtype, dst_dtype, rtol, atol) + + for op in [ + relay.add, + relay.subtract, + relay.divide, + relay.multiply, + ]: + check_binary_op(op, src_dtype, dst_dtype) + + # we would like to test tvm_if_then_else + # but Relay.IfNode is not lowered to this intrinsic, + # so to keep our tests consistent with relay, we decide to not unit test + # Note: tvm_if_then_else is tested as part of the mobile_net model + + +def run_model(get_workload, + input_shape, + src_dtype, + dst_dtype, + num_classes, + rtol=0.0001, + atol=0.0001): + module, params = get_workload(image_shape=input_shape, + num_classes=num_classes) + + # Convert the input into the correct format. + input = tvm.nd.array(rs.rand(*input_shape).astype(src_dtype)) + + compare(module, (input, ), src_dtype, dst_dtype, rtol, atol, params) + + # # Simplifying inference is essential right now, as batch norms (which get + # # removed) are broken with custom datatypes. + # #expr = relay.ir_pass.simplify_inference(expr) + +def run_conv2d(src_dtype, dst_dtype): + def run_test_conv2d(src_dtype, + dst_dtype, + scale, + dshape, + kshape, + padding=(1, 1), + fref=None, + groups=1, + dilation=(1, 1), + except_targets=None, + **attrs): + if except_targets is None: + except_targets = [] + + x = relay.var("x", shape=dshape, dtype=src_dtype) + w = relay.var("w", shape=kshape, dtype=src_dtype) + y = relay.nn.conv2d(x, + w, + padding=padding, + dilation=dilation, + groups=groups, + **attrs) + module = tvm.IRModule.from_expr(relay.Function([x, w], y)) + data = rs.uniform(-scale, scale, size=dshape).astype(src_dtype) + kernel = rs.uniform(-scale, scale, + size=kshape).astype(src_dtype) + dkernel = topi.testing.dilate_python(kernel, (1, 1) + dilation) + if fref is None: + ref_res = topi.testing.conv2d_nchw_python( + data.astype(src_dtype), + dkernel.astype(src_dtype), + 1, + padding, + groups=groups) + else: + ref_res = fref(data.astype(src_dtype), dkernel.astype(src_dtype)) + + for target, ctx in [("llvm", tvm.cpu(0))]: + if target in except_targets: + continue + intrp1 = relay.create_executor("graph", + ctx=ctx, + target=target, + mod=module) + module, _ = change_dtype(src_dtype, dst_dtype, module, []) + data_converted = convert_ndarray(dst_dtype, data) + kernel_converted = convert_ndarray(dst_dtype, kernel) + with tvm.transform.PassContext( + config={"tir.disable_vectorize": True}): + op_res1 = intrp1.evaluate()(data_converted, kernel_converted) + op_res1_converted = convert_ndarray(src_dtype, op_res1) + tvm.testing.assert_allclose(op_res1_converted.asnumpy(), ref_res) + + # depthwise conv2d + dshape = (1, 32, 18, 18) + kshape = (32, 1, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=32, + groups=32, + kernel_size=(3, 3), + fref=lambda x, w: topi.testing. + depthwise_conv2d_python_nchw(x, w, (1, 1), "SAME")) + + # CUDA is disabled for 'direct' schedule: + # https://github.com/dmlc/tvm/pull/3070#issuecomment-486597553 + # group conv2d + dshape = (1, 32, 18, 18) + kshape = (32, 4, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=32, + groups=8, + kernel_size=(3, 3), + except_targets=['cuda']) + # also group conv2d + dshape = (1, 32, 18, 18) + kshape = (64, 1, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=64, + groups=32, + kernel_size=(3, 3), + except_targets=['cuda']) + + # normal conv2d + dshape = (1, 3, 224, 224) + kshape = (10, 3, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=10, + kernel_size=(3, 3)) + + # dilated conv2d + dshape = (1, 3, 18, 18) + kshape = (10, 3, 3, 3) + run_test_conv2d(src_dtype, + dst_dtype, + 1, + dshape, + kshape, + padding=(1, 1), + channels=10, + kernel_size=(3, 3), + dilation=(3, 3)) + + +def test_ops(): + # TODO(gus) these tolerances are high, and still sometimes fail; + # this is expected, b/c we're comparing between 32bit float and 8 + # bit posit. + # Figure out a more logical way to test here. + run_ops('float32', 'custom[posites2]8', rtol=1, atol=1) + run_ops('float32', 'custom[posites2]16', rtol=0.01, atol=1) + run_ops('float32', 'custom[posites2]32') + + +def test_conv2d(): + # TODO(@gussmith23) slow and broken, needing refactor! + # run_conv2d('float32', 'custom[posit32]32') + pass + + +def test_models(): + # Expected posit8 might be faster, but it's not. + # run_model(get_mobilenet, (3, 224, 224), 'float32', 'custom[posit8]8') + # run_model(get_mobilenet, (3, 224, 224), 'float32', 'custom[posit32]32') + # run_model(get_inception, (3, 299, 299), 'float32', 'custom[posit32]32') + # run_model(get_resnet, (3, 224, 224), 'float32', 'custom[posit32]32') + + # Run cifar-10 sizes to be a little faster... + run_model(get_mobilenet, (3, 32, 32), + 'float32', + 'custom[posites2]32', + num_classes=10) + # run_model(get_inception, (3, 32, 32), + # 'float32', + # 'custom[posit32]32', + # num_classes=10) + # run_model(get_resnet, (3, 32, 32), + # 'float32', + # 'custom[posit32]32', + # num_classes=10) Review comment: Experiment with these -- are they slow? If they're slow -- we need a contingency plan. If we can't run whole models, then we should be testing layer by layer --> partly achieved by testing unary/binary ops --> but then we should also test the larger operators i.e. convolution (in which case we should beef up/update conv2d tests) ########## File path: 3rdparty/nop-type/nop-type.cc ########## @@ -0,0 +1,30 @@ +#include <tvm/runtime/c_runtime_api.h> Review comment: Yes, let's remove it ########## File path: python/tvm/relay/testing/mobilenet.py ########## @@ -129,6 +129,7 @@ def mobile_net(num_classes=1000, data_shape=(1, 3, 224, 224), bias = relay.var('fc_bias') fc = relay.nn.dense(data=flatten, weight=weight, units=num_classes) fc = relay.nn.bias_add(fc, bias) + # TODO(gus) i think softmax is broken Review comment: Yes, remove this ########## File path: python/tvm/target/datatype.py ########## @@ -73,7 +73,12 @@ def get_type_registered(type_code): return tvm.runtime._ffi_api._datatype_get_type_registered(type_code) -def register_op(lower_func, op_name, target, type_name, src_type_name=None): +def register_op(lower_func, + op_name, + target, + type_name, + src_type_name=None, + intrinsic_name=None): """Register an external function which computes the given op. Currently, this will only work with Casts and binary expressions Review comment: Does it make sense to mention `a` and `b`? Is it just confusing to the reader? ########## File path: tests/python/unittest/test_custom_datatypes_change_dtype.py ########## @@ -0,0 +1,481 @@ +# 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. +"""Utilities for changing datatypes of models.""" +import tvm +import topi.testing +import numpy as np +from numpy.random import MT19937, RandomState, SeedSequence +from tvm import relay +from tvm.relay.testing.inception_v3 import get_workload as get_inception +from tvm.relay.testing.resnet import get_workload as get_resnet +from tvm.relay.testing.mobilenet import get_workload as get_mobilenet +from tvm.target.datatype import register, register_min_func, register_op, create_lower_func, lower_ite +from nose.tools import nottest + +tgt = "llvm" +# we use a random seed to generate input_data +# to guarantee stable tests +rs = RandomState(MT19937(SeedSequence(123456789))) + +def convert_ndarray(dst_dtype, *arrays): + """Converts NDArray(s) into the specified datatype""" + def convert(array): + x = relay.var('x', shape=array.shape, dtype=str(array.dtype)) + cast = relay.Function([x], x.astype(dst_dtype)) + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + return relay.create_executor('graph').evaluate(cast)(array) + + return tuple([convert(x) for x in arrays]) + + +def change_dtype(src, dst, module, params): + module = relay.frontend.ChangeDatatype(src, dst)(module) + module = relay.transform.InferType()(module) + params = dict((p, convert_ndarray(dst, params[p])) for p in params) + return module, params + +def compare(module, input, src_dtype, dst_dtype, rtol, atol, params = {}): + ex = relay.create_executor("graph", mod=module) + + correct = ex.evaluate()(*input, **params) + + module, _ = change_dtype(src_dtype, dst_dtype, module, []) + ex = relay.create_executor("graph", mod=module) + # converts all inputs to dst_dtype + x_converted = convert_ndarray(dst_dtype, *input) + + # Vectorization is not implemented with custom datatypes + with tvm.transform.PassContext(config={"tir.disable_vectorize": True}): + maybe_correct = ex.evaluate()(*x_converted, **params) + # TODO(andrew) this only works on single output + maybe_correct_converted = convert_ndarray(src_dtype, maybe_correct)[0] + np.testing.assert_allclose(maybe_correct_converted.asnumpy(), + correct.asnumpy(), + rtol=rtol, + atol=atol) + +def setup(): + """Set up tests + + Currently, this registers some custom datatypes using the Bring Your + Own Datatypes framework. + """ + + # To use datatype operations in an external library, you should first load + # the library containing the datatype implementation: + # CDLL("libposit.so", RTLD_GLOBAL) + # In this case, the datatype library we are using is built right into TVM, + # so we do not need to explicitly load any library. + + # You can pick a code for your datatype arbitrarily, as long as it is + # greater than 128 and has not already been chosen. + + register("posites2", 131) + + register_op(create_lower_func( + { + (32, 32): "FloatToPosit32es2", + (32, 16): "FloatToPosit16es2", + (32, 8): 'FloatToPosit8es2', + }), + "Cast", "llvm", "posites2", "float") + register_op(create_lower_func( + { + (32, 32): "Posit32es2ToFloat", + (16, 32): 'Posit16es2ToFloat', + (8, 32): 'Posit8es2ToFloat', + }), + "Cast", "llvm", "float", "posites2") + register_op(create_lower_func( + { + (4, 32): 'IntToPosit32es2', + (4, 16): 'IntToPosit16es2', + (4, 8): 'IntToPosit8es2' + }), + "Cast", "llvm", "posites2", "int") + register_op(create_lower_func({ + 32: 'Posit32es2Add', + 16: 'Posit16es2Add', + 8: 'Posit8es2Add' + }), "Add", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sub', + 16: 'Posit16es2Sub', + 8: 'Posit8es2Sub' + }), "Sub", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'FloatToPosit32es2', + 16: 'FloatToPosit16es2', + 8: 'FloatToPosit8es2' + }), "FloatImm", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Mul', + 16: 'Posit16es2Mul', + 8: 'Posit8es2Mul' + }), "Mul", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Div', + 16: 'Posit16es2Div', + 8: 'Posit8es2Div' + }), "Div", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Max', + 16: 'Posit16es2Max', + 8: 'Posit8es2Max' + }), "Max", "llvm", "posites2") + register_op(create_lower_func({ + 32: 'Posit32es2Sqrt', + 16: 'Posit16es2Sqrt', + 8: 'Posit8es2Sqrt' + }), "Call", "llvm", "posites2", intrinsic_name="sqrt") + register_op(lower_ite, + "Call", + "llvm", + "posites2", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({ + 32: 'Posit32es2Exp', + 16: 'Posit16es2Exp', + 8: 'Posit8es2Exp' + }), "Call", "llvm", "posites2", intrinsic_name="exp") + register_op(create_lower_func({ + 32: 'Posit32es2Log', + 16: 'Posit16es2Log', + 8: 'Posit8es2Log' + }), "Call", "llvm", "posites2", intrinsic_name="log") + register_op(create_lower_func({ + 32: 'Posit32es2Sigmoid', + 16: 'Posit16es2Sigmoid', + 8: 'Posit8es2Sigmoid' + }), "Call", "llvm", "posites2", intrinsic_name="sigmoid") + register_op(create_lower_func({ + 32: 'Posit32es2Tanh', + 16: 'Posit16es2Tanh', + 8: 'Posit8es2Tanh' + }), "Call", "llvm", "posites2", intrinsic_name="tanh") + register_min_func(lambda num_bits: - (2 ** 2 ** 2) ** (num_bits - 2), "posites2") + + register("noptype", 132) + register_op(create_lower_func({ + (32, 32): "FloatToNop32" + }), "Cast", "llvm", "noptype", "float") + register_op(create_lower_func({ + (32, 32): 'Nop32ToFloat' + }), "Cast", "llvm", "float", "noptype") + register_op(create_lower_func({ + (4, 32): "IntToNop32" + }), "Cast", "llvm", "noptype", "int") + register_op(create_lower_func({32: 'IntToNop32'}), "Add", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sub'}), "Sub", "llvm", "noptype") + register_op(create_lower_func({32: 'FloatToNop32'}), "FloatImm", "llvm", + "noptype") + register_op(create_lower_func({32: 'Nop32Mul'}), "Mul", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Div'}), "Div", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Max'}), "Max", "llvm", "noptype") + register_op(create_lower_func({32: 'Nop32Sqrt'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sqrt") + # TODO(gus) not sure if this will work... + register_op(lower_ite, + "Call", + "llvm", + "noptype", + intrinsic_name="tvm_if_then_else") + register_op(create_lower_func({32: 'Nop32Exp'}), + "Call", + "llvm", + "noptype", + intrinsic_name="exp") + register_op(create_lower_func({32: 'Nop32Log'}), + "Call", + "llvm", + "noptype", + intrinsic_name="log") + register_op(create_lower_func({32: 'Nop32Sigmoid'}), + "Call", + "llvm", + "noptype", + intrinsic_name="sigmoid") + register_op(create_lower_func({32: 'Nop32Tanh'}), + "Call", + "llvm", + "noptype", + intrinsic_name="tanh") + # This can be anything, considering the type isn't functionally correct. + register_min_func(lambda num_bits: 0, "noptype") + + +def run_ops(src_dtype, dst_dtype, rtol=1e-7, atol=1e-7): + """Run the same op, but with two different datatypes""" + def check_unary_op(op, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + x = relay.var("x", t1) + z = op(x) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + + module = tvm.IRModule.from_expr(relay.Function([x], z)) + + compare(module, (x_data, ), src_dtype, dst_dtype, rtol, atol) + # print(maybe_correct_converted) + # print(correct) + + for op in [ + relay.nn.softmax, + tvm.relay.log, + tvm.relay.exp, + tvm.relay.sqrt, + tvm.relay.rsqrt, + tvm.relay.sigmoid, + tvm.relay.tanh, + relay.nn.relu, + ]: + check_unary_op(op, src_dtype, dst_dtype) + + def check_binary_op(opfunc, src_dtype, dst_dtype): + t1 = relay.TensorType((5, 10, 5), src_dtype) + t2 = relay.TensorType((5, ), src_dtype) + x = relay.var("x", t1) + y = relay.var("y", t2) + z = opfunc(x, y) + x_data = rs.rand(5, 10, 5).astype(t1.dtype) + y_data = rs.rand(5).astype(t2.dtype) + module = tvm.IRModule.from_expr(relay.Function([x, y], z)) + + compare(module, (x_data, y_data), src_dtype, dst_dtype, rtol, atol) + + for op in [ + relay.add, + relay.subtract, + relay.divide, + relay.multiply, + ]: + check_binary_op(op, src_dtype, dst_dtype) + + # we would like to test tvm_if_then_else + # but Relay.IfNode is not lowered to this intrinsic, + # so to keep our tests consistent with relay, we decide to not unit test + # Note: tvm_if_then_else is tested as part of the mobile_net model + + +def run_model(get_workload, + input_shape, + src_dtype, + dst_dtype, + num_classes, + rtol=0.0001, + atol=0.0001): + module, params = get_workload(image_shape=input_shape, + num_classes=num_classes) + + # Convert the input into the correct format. + input = tvm.nd.array(rs.rand(*input_shape).astype(src_dtype)) + + compare(module, (input, ), src_dtype, dst_dtype, rtol, atol, params) + + # # Simplifying inference is essential right now, as batch norms (which get + # # removed) are broken with custom datatypes. + # #expr = relay.ir_pass.simplify_inference(expr) Review comment: Let's delete this comment, but we should add a note somewhere about `SimplifyInference`. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
