zoukankan      html  css  js  c++  java
  • TVM Pass IR如何使用

    TVM Pass IR如何使用

    随着Relay / tir中优化遍数的增加,执行并手动维护其依赖关系变得很棘手。引入了一个基础结构来管理优化过程,并应用于TVM堆栈中IR的不同层。

    Relay / tir程序的优化可以以各种粒度应用,即分别使用tvm.relay.transform.FunctionPasstvm.tir.transform.PrimFuncPass和的功能级别和模块级别tvm.transform.ModulePass。或者,用户可以依靠在tvm.transform.Sequential中继/ tir程序上应用一系列pass,其中pass之间的依赖性可以通过pass下文解决。有关这些pass的每种类型的更多详细信息,请参阅pass基础结构

    本文主要说明开发人员如何使用pass infra进行特定的优化,创建用于Relay程序的优化管道。同样的方法也可以用于tir。

    import numpy as np

    import tvm

    from tvm import te

    import tvm.relay as relay

    创建一个示例Relay中继程序

    首先,创建一个简单的Relay程序。该程序将用于示例的各种优化。类似地,用户可以编写一个tir基本函数并应用tirpass。

    def example():

        shape = (1, 64, 54, 54)

        c_data = np.empty(shape).astype("float32")

        c = relay.const(c_data)

        weight = relay.var("weight", shape=(64, 64, 3, 3))

        x = relay.var("x", relay.TensorType((1, 64, 56, 56), "float32"))

        conv = relay.nn.conv2d(x, weight)

        y = relay.add(c, c)

        y = relay.multiply(y, relay.const(2, "float32"))

        y = relay.add(conv, y)

        z = relay.add(y, c)

        z1 = relay.add(y, c)

        z2 = relay.add(z, z1)

        return relay.Function([x, weight], z2)

    让为conv2d op注册布局更改,以便可以在示例中应用布局更改通道。alter layout pass如何工作不在本文的讨论范围之内。

    @relay.op.register_alter_op_layout("nn.conv2d", level=101)

    def alter_conv2d(attrs, inputs, tinfos, out_type):

        data, weight = inputs

        new_attrs = dict(attrs)

        new_attrs["data_layout"] = "NCHW16c"

        return relay.nn.conv2d(data, weight, **new_attrs)

    优化程序

    现在要优化程序。Relay中继具有许多优化功能。将选择其中一些以应用于此示例程序。

    有多种方法可以优化中继程序。下面将为每个示例提供示例。

    手动应用优化pass

    # Let's first create a relay Module which contains one or multiple Relay

    # functions for optimization.

    f = example()

    mod = tvm.IRModule.from_expr(f)

     

    # Now we can apply constant folding on the module.

    # fold_const here is a callback that doesn't take any parameters.

    fold_const = relay.transform.FoldConstant()

    # Then, we can invoke the pass on the given module. Note that the constant

    # folding pass works at the function-level. That being said, each function in

    # the module will be applied with the optimization. Users don't need to iterate

    # through individual functions manually to apply this pass.

    mod = fold_const(mod)

    # We can see from the updated program that the constants are folded.

    print(mod)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    可以以类似方式应用更多优化。例如,可以消除zz1使用的通用表达式。

    mod = relay.transform.EliminateCommonSubexpr()(mod)

    print(mod)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    一些优化(例如融合)也是参数化的。例如,选择级别0不允许将算子融合在一起。用户可以传递 fuse_opt_level来启用此功能。

    mod = relay.transform.FuseOps(fuse_opt_level=0)(mod)

     

    # We can observe that the optimized module contains functions that only have

    # a signle primitive op.

    print(mod)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %1 = %0(%x, %weight) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = fn (%p01: Tensor[(1, 64, 54, 54), float32], %p11: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p01, %p11) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %3 = %2(%1, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %4 = fn (%p02: Tensor[(1, 64, 54, 54), float32], %p12: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p02, %p12) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %5 = %4(%3, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %6 = fn (%p03: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p03, %p03) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %6(%5) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    使用序列来应用pass序列

    如上所述,应用pass实际上是乏味的,并且可能需要用户更好地了解依赖性。例如,融合目前不适用于let绑定。如果relay.transform.ToANormalForm()在融合之前应用算子,将无法将融合在一起,因为此过程会为每个表达式生成let绑定,以规范化Relay程序。

    Relaytvm.transform.Sequential通过指定每个遍历,将打包为整体来缓解开发人员显式处理这些问题的麻烦。例如,可以使用以下序列样式应用相同遍历。tvm.transform.Sequentialtorch.nn.sequential 和mxnet.gluon.block类似。例如,torch.nn.sequential用于包含一系列PyTorch模块,这些模块将被添加,以构建网络,着重于网络层。取而代之的是tvm.transform.Sequential,下面的过程中的基础工作于优化过程。

    # Now let's execute some passes through :py:class:`tvm.transform.Sequential`

    f = example()

    mod = tvm.IRModule.from_expr(f)

    # Glob the interested passes.

    seq = tvm.transform.Sequential(

        [

            relay.transform.FoldConstant(),

            relay.transform.EliminateCommonSubexpr(),

            relay.transform.FuseOps(fuse_opt_level=2),

        ]

    )

    mod1 = seq(mod)

    print(mod1)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %4 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %3 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %4(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    从转换后的Relay程序中,可以看到仍然有两个相同的加法运算。这是因为EliminateCommonSubexpr 未实际执行。默认情况下,只有优化级别小于或等于2的过程才被执行 tvm.transform.Sequential。下面的pass提供了一个配置界面,供用户自定义要执行的优化级别。

    with tvm.transform.PassContext(opt_level=3):

        mod2 = seq(mod)

    print(mod2)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %3 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %3(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    可以看到仅保留了两个相同的加法之一。

    In addition, users can selectively disable some passes using the disabled_pass config, which is similar to the -fno-xxx option used the general purpose compilers, such as Clang and GCC. For example, we can disable EliminateCommonSubexpr as following. The printed module will again show two identical addition operations.

    with tvm.transform.PassContext(opt_level=3, disabled_pass=["EliminateCommonSubexpr"]):

        mod3 = seq(mod)

    print(mod3)

    Out:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %4 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %3 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %4(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    The passes applied so far are target independent. The pass infra also provides a means to make pass target-aware. For example, the layout alteration pass falls in such category.

    with tvm.transform.PassContext(opt_level=3):

        mod4 = seq(mod)

    print(mod4)

     

    seq1 = tvm.transform.Sequential([relay.transform.AlterOpLayout()])

    with tvm.transform.PassContext(opt_level=3):

        with tvm.target.Target("llvm"):

            mod5 = seq1(mod)

    print(mod5)

    Out:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %3 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %3(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = layout_transform(%x, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 56, 56, 16), float32] */;

      %1 = nn.conv2d(%0, %weight, padding=[0, 0, 0, 0], data_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %2 = add(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = multiply(%2, 2f /* ty=float32 */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %4 = layout_transform(%3, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %5 = add(%1, %4) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %6 = layout_transform(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %7 = add(%5, %6) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %8 = add(%5, %6) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      %9 = add(%7, %8) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

      layout_transform(%9, src_layout="NCHW16c", dst_layout="NCHW") /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    Implement a Pass Using Python Decorator

    下一个示例说明了如何使用Python装饰器,通过传递基础流程来编排定制的优化管道。此功能极大地简化了pass的实施。例如,用户可以简单地定义一个修饰的类,进行功能级别的优化,如以下示例所示。transform_function包装一个类,以用c的倍数替换所有常量。稍后,当调用自定义过程时,将访问给定模块中的每个函数,并且将替换函数中的每个常量。

    @relay.transform.function_pass(opt_level=1)

    class CustomPipeline:

        """Simple test function to replace one argument to another."""

     

        def __init__(self, multiplier):

            self.multiplier = multiplier

     

        # This function can define a pass.

        def transform_function(self, func, mod, ctx):

            obj = self

     

            class ReplaceConstant(tvm.relay.ExprMutator):

                def visit_constant(self, c):

                    return relay.multiply(obj.multiplier, c)

     

            return ReplaceConstant().visit(func)

     

     

    f = example()

    mod = tvm.IRModule.from_expr(f)

    custom_pass = CustomPipeline(multiplier=relay.const(3, "float32"))

    assert custom_pass.info.name == "CustomPipeline"

    mod3 = custom_pass(mod)

    print(mod3)

    输出:

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = multiply(3f /* ty=float32 */, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, %1) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = multiply(3f /* ty=float32 */, 2f /* ty=float32 */) /* ty=float32 */;

      %4 = multiply(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %5 = add(%0, %4) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %6 = add(%5, %1) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %7 = add(%5, %1) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%6, %7) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    调试pass

    TVM为用户提供了一种即插即用式的调试通道,该通道在通过特殊通道(PrintIR)来转储整个模块的IR之后,将IR打印出来。序列传递示例的略微修改版本,可能类似于以下内容,以启用IR转储以进行FoldConstant优化。

    f = example()

    mod = tvm.IRModule.from_expr(f)

    seq = tvm.transform.Sequential(

        [

            relay.transform.FoldConstant(),

            tvm.transform.PrintIR(),

            relay.transform.EliminateCommonSubexpr(),

            relay.transform.FuseOps(),

            relay.transform.AlterOpLayout(),

        ]

    )

     

    # By inserting the ``PrintIR`` pass after ``FoldConstant``, the pass infra will

    # dump out the module IR when ``FoldConstant`` is done. Users can plug in this

    # pass after any pass they want to debug for viewing the optimization effect.

    #

    # There is a more flexible debugging mechanism also exposed by the build configuration

    # object. One can pass a tracing function which can be used to execute arbitrary code

    # before and/or after each pass. A tracing function will receive a :py::class:`tvm.IRModule`,

    # a :py:class:`tvm.transform.PassInfo` object,

    # and a boolean indicating whether you are executing before, or after a pass.

    # An example is below.

     

     

    def print_ir(mod, info, is_before):

        """Print the name of the pass, the IR, only before passes execute."""

        if is_before:

            print("Running pass: {}", info)

            print(mod)

     

     

    with tvm.transform.PassContext(opt_level=3, trace=print_ir):

        with tvm.target.Target("llvm"):

            # Perform the optimizations.

            mod = seq(mod)

    print(mod)

     

    print("done")

    输出:

    Running pass: {} The meta data of the pass: pass name: FoldConstantopt_level: 2required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]);

      %1 = add(meta[relay.Constant][0], meta[relay.Constant][0]);

      %2 = multiply(%1, 2f);

      %3 = add(%0, %2);

      %4 = add(%3, meta[relay.Constant][0]);

      %5 = add(%3, meta[relay.Constant][0]);

      add(%4, %5)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() {

      add(meta[relay.Constant][0], meta[relay.Constant][0])

    }

     

     

    Running pass: {} The meta data of the pass: pass name: FuseOpsopt_level: 1required passes: [

    InferType, ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      add(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      %0 = fn (%p0: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p0, %p0)

      };

      %0(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: ToANormalFormopt_level: 1required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      %0 = fn (%p0: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p0, %p0) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %0(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      let %x = meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */;

      let %x1 = fn (%p0: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        add(%p0, %p0) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      let %x2 = %x1(%x);

      %x2

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() {

      multiply(meta[relay.Constant][0], 2f)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: FuseOpsopt_level: 1required passes: [

    InferType, ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      multiply(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, 2f /* ty=float32 */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      %0 = fn (%p0: Tensor[(1, 64, 54, 54), float32], %p1: float32, Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        multiply(%p0, %p1)

      };

      %0(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, 2f /* ty=float32 */)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: ToANormalFormopt_level: 1required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      %0 = fn (%p0: Tensor[(1, 64, 54, 54), float32], %p1: float32, Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        multiply(%p0, %p1) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %0(meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, 2f /* ty=float32 */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main() -> Tensor[(1, 64, 54, 54), float32] {

      let %x = meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */;

      let %x1 = 2f /* ty=float32 */;

      let %x2 = fn (%p0: Tensor[(1, 64, 54, 54), float32], %p1: float32, Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        multiply(%p0, %p1) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      let %x3 = %x2(%x, %x1);

      %x3

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]);

      %1 = add(%0, meta[relay.Constant][0]);

      %2 = add(%1, meta[relay.Constant][1]);

      %3 = add(%1, meta[relay.Constant][1]);

      add(%2, %3)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: PrintIRopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: EliminateCommonSubexpropt_level: 3required passes: [

    InferType, ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %3 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %3) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %2)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: FuseOpsopt_level: 1required passes: [

    InferType, ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %0 = nn.conv2d(%x, %weight, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %1 = add(%0, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      %2 = add(%1, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */;

      add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %3 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]);

        %1 = add(%0, %p2);

        %2 = add(%1, %p3);

        add(%2, %2)

      };

      %3(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */)

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %3 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %3(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: AlterOpLayoutopt_level: 3required passes: [

    InferType, ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %3 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = nn.conv2d(%p0, %p1, padding=[0, 0, 0, 0]) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %1 = add(%0, %p2) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        %2 = add(%1, %p3) /* ty=Tensor[(1, 64, 54, 54), float32] */;

        add(%2, %2) /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %3(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

     

     

    Running pass: {} The meta data of the pass: pass name: InferTypeopt_level: 0required passes: [

    ]

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %7 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = layout_transform(%p0, src_layout="NCHW", dst_layout="NCHW16c");

        %1 = nn.conv2d(%0, %p1, padding=[0, 0, 0, 0], data_layout="NCHW16c");

        %2 = layout_transform(%p2, src_layout="NCHW", dst_layout="NCHW16c");

        %3 = add(%1, %2);

        %4 = layout_transform(%p3, src_layout="NCHW", dst_layout="NCHW16c");

        %5 = add(%3, %4);

        %6 = add(%5, %5);

        layout_transform(%6, src_layout="NCHW16c", dst_layout="NCHW")

      };

      %7(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */)

    }

     

     

    def @main(%x: Tensor[(1, 64, 56, 56), float32], %weight: Tensor[(64, 64, 3, 3), float32]) -> Tensor[(1, 64, 54, 54), float32] {

      %7 = fn (%p0: Tensor[(1, 64, 56, 56), float32], %p1: Tensor[(64, 64, 3, 3), float32], %p2: Tensor[(1, 64, 54, 54), float32], %p3: Tensor[(1, 64, 54, 54), float32], Primitive=1) -> Tensor[(1, 64, 54, 54), float32] {

        %0 = layout_transform(%p0, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 56, 56, 16), float32] */;

        %1 = nn.conv2d(%0, %p1, padding=[0, 0, 0, 0], data_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        %2 = layout_transform(%p2, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        %3 = add(%1, %2) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        %4 = layout_transform(%p3, src_layout="NCHW", dst_layout="NCHW16c") /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        %5 = add(%3, %4) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        %6 = add(%5, %5) /* ty=Tensor[(1, 4, 54, 54, 16), float32] */;

        layout_transform(%6, src_layout="NCHW16c", dst_layout="NCHW") /* ty=Tensor[(1, 64, 54, 54), float32] */

      };

      %7(%x, %weight, meta[relay.Constant][0] /* ty=Tensor[(1, 64, 54, 54), float32] */, meta[relay.Constant][1] /* ty=Tensor[(1, 64, 54, 54), float32] */) /* ty=Tensor[(1, 64, 54, 54), float32] */

    }

    done

    概括

    本文介绍了如何使用pass基础,更加方便地在TVM中编写和调用pass。还讨论了调用pass的不同方法。使用tvm.transform.Sequential,可以极大地帮助用户简化处理多个优化过程及其依赖项的工作。另外,提供了一个示例来说明如何使用PrintIR和跟踪调试过程。

    人工智能芯片与自动驾驶
  • 相关阅读:
    区分/不区分大小写的比较,查找字符串在另一字符串中的位置,字符串开头是否包括另一字符串 hasPrefix
    获取文件名以及后缀
    监听Documents文件夹内文件发生改变
    根据路径获取文件大小
    获取视频第一帧的方法
    判断图片格式
    iTunes文件共享
    iOS 10 隐私权限设置
    uCos 学习:0-有关概念
    ALSA 有关文档
  • 原文地址:https://www.cnblogs.com/wujianming-110117/p/14580172.html
Copyright © 2011-2022 走看看