C-host 代码生成简介

import set_env

import tvm
import tvm.testing

from tvm import te
from tvm.contrib import utils
from tvm.script import tir as T, ir as I

import numpy as np

add c-host 代码生成

nn = 1024
n = tvm.runtime.convert(nn)
A = te.placeholder((n,), name="A")
B = te.placeholder((n,), name="B")
C = te.compute(A.shape, lambda *i: A(*i) + B(*i), name="C")
s = te.create_schedule(C.op)

mhost = tvm.build(s, [A, B, C], "c", name="test_fadd")
print(mhost.get_source())

// tvm target: c -keys=cpu 
#define TVM_EXPORTS
#include "tvm/runtime/c_runtime_api.h"
#include "tvm/runtime/c_backend_api.h"
#include <math.h>
#include <stdbool.h>
#ifdef __cplusplus
extern "C"
#endif
TVM_DLL int32_t test_fadd(void* args, int32_t* arg_type_ids, int32_t num_args, void* out_ret_value, int32_t* out_ret_tcode, void* resource_handle);
#ifdef __cplusplus
extern "C"
#endif
TVM_DLL int32_t test_fadd(void* args, int32_t* arg_type_ids, int32_t num_args, void* out_ret_value, int32_t* out_ret_tcode, void* resource_handle) {
  int32_t A_code = arg_type_ids[0];
  int32_t B_code = arg_type_ids[1];
  int32_t C_code = arg_type_ids[2];
  void* A = (((TVMValue*)args)[0].v_handle);
  void* B = (((TVMValue*)args)[1].v_handle);
  void* C = (((TVMValue*)args)[2].v_handle);
  void* test_fadd_A_shape = (((DLTensor*)A)[0].shape);
  void* test_fadd_A_strides = (((DLTensor*)A)[0].strides);
  int32_t dev_id = (((DLTensor*)A)[0].device.device_id);
  void* A_1 = (((DLTensor*)A)[0].data);
  void* test_fadd_B_shape = (((DLTensor*)B)[0].shape);
  void* test_fadd_B_strides = (((DLTensor*)B)[0].strides);
  void* B_1 = (((DLTensor*)B)[0].data);
  void* test_fadd_C_shape = (((DLTensor*)C)[0].shape);
  void* test_fadd_C_strides = (((DLTensor*)C)[0].strides);
  void* C_1 = (((DLTensor*)C)[0].data);
  if (!(test_fadd_A_strides == NULL)) {
  }
  if (!(test_fadd_B_strides == NULL)) {
  }
  if (!(test_fadd_C_strides == NULL)) {
  }
  for (int32_t i0 = 0; i0 < 1024; ++i0) {
    ((float*)C_1)[i0] = (((float*)A_1)[i0] + ((float*)B_1)[i0]);
  }
  return 0;
}

// CodegenC: NOTE: Auto-generated entry function
#ifdef __cplusplus
extern "C"
#endif
TVM_DLL int32_t __tvm_main__(void* args, int* arg_type_ids, int num_args, void* out_ret_value, int* out_ret_tcode, void* resource_handle) {
  return test_fadd(args, arg_type_ids, num_args, out_ret_value, out_ret_tcode, resource_handle);
}

temp = utils.tempdir()
path_dso = temp.relpath("temp.so")
mhost.export_library(path_dso)
m = tvm.runtime.load_module(path_dso)
fadd = m["test_fadd"]
dev = tvm.cpu(0)
# launch the kernel.
n = nn
a = tvm.nd.array(np.random.uniform(size=n).astype(A.dtype), dev)
b = tvm.nd.array(np.random.uniform(size=n).astype(B.dtype), dev)
c = tvm.nd.array(np.zeros(n, dtype=C.dtype), dev)
fadd(a, b, c)
tvm.testing.assert_allclose(c.numpy(), a.numpy() + b.numpy())

add_pipeline c host 代码生成

nn = 1024
n = tvm.runtime.convert(nn)
A = te.placeholder((n,), name="A")
B = te.placeholder((n,), name="B")
AA = te.compute((n,), lambda *i: A(*i), name="A")
BB = te.compute((n,), lambda *i: B(*i), name="B")
T = te.compute(A.shape, lambda *i: AA(*i) + BB(*i), name="T")
C = te.compute(A.shape, lambda *i: T(*i), name="C")
s = te.create_schedule(C.op)
xo, xi = s[C].split(C.op.axis[0], factor=4)
xo1, xo2 = s[C].split(xo, factor=13)
s[C].parallel(xo2)
s[C].pragma(xo1, "parallel_launch_point")
s[C].pragma(xo2, "parallel_stride_pattern")
s[C].pragma(xo2, "parallel_barrier_when_finish")
# FIXME(tvm-team): vector operators are not supported for codegen to C yet
# s[C].vectorize(xi)

def check_c():
    # Specifically allow offset to test codepath when offset is available
    Ab = tvm.tir.decl_buffer(
        A.shape, A.dtype, elem_offset=te.size_var("Aoffset"), offset_factor=8, name="A"
    )
    binds = {A: Ab}
    # BUILD and invoke the kernel.
    f1 = tvm.lower(s, [A, B, C], name="test_fadd_pipeline")
    mhost = tvm.build(f1, target="c")

    temp = utils.tempdir()
    path_dso = temp.relpath("temp.so")
    mhost.export_library(path_dso)
    m = tvm.runtime.load_module(path_dso)
    fadd = m["test_fadd_pipeline"]
    dev = tvm.cpu(0)
    # launch the kernel.
    n = nn
    a = tvm.nd.array(np.random.uniform(size=n).astype(A.dtype), dev)
    b = tvm.nd.array(np.random.uniform(size=n).astype(B.dtype), dev)
    c = tvm.nd.array(np.zeros(n, dtype=C.dtype), dev)
    fadd(a, b, c)
    tvm.testing.assert_allclose(c.numpy(), a.numpy() + b.numpy())

check_c()

tir.reinterpret c host

nn = 1024
n = tvm.runtime.convert(nn)
A = te.placeholder((n,), name="A", dtype="int32")
B = te.compute(
    A.shape, lambda *i: tvm.tir.call_intrin("float32", "tir.reinterpret", 2 + A(*i)), name="B"
)
s = te.create_schedule(B.op)

def check_c():
    mhost = tvm.build(s, [A, B], "c", name="test_reinterpret")
    temp = utils.tempdir()
    path_dso = temp.relpath("temp.so")
    mhost.export_library(path_dso)
    m = tvm.runtime.load_module(path_dso)
    fadd = m["test_reinterpret"]
    dev = tvm.cpu(0)
    n = nn
    a = tvm.nd.array(np.random.randint(-(2**30), 2**30, size=n).astype(A.dtype), dev)
    b = tvm.nd.array(np.zeros(n, dtype=B.dtype), dev)
    fadd(a, b)
    tvm.testing.assert_allclose(b.numpy(), (2 + a.numpy()).view("float32"))

check_c()

tir.ceil c host

def test_ceil():
    nn = 1024
    n = tvm.runtime.convert(nn)
    A = te.placeholder((n,), name="A", dtype="float32")
    B = te.compute(A.shape, lambda *i: tvm.tir.call_intrin("float32", "tir.ceil", A(*i)), name="B")
    s = te.create_schedule(B.op)

    def check_c():
        mhost = tvm.build(s, [A, B], "c", name="test_ceil")
        temp = utils.tempdir()
        path_dso = temp.relpath("temp.so")
        mhost.export_library(path_dso)
        m = tvm.runtime.load_module(path_dso)
        fceil = m["test_ceil"]
        dev = tvm.cpu(0)
        n = nn
        a = tvm.nd.array(np.random.rand(n).astype(A.dtype), dev)
        b = tvm.nd.array(np.zeros(n, dtype=B.dtype), dev)
        fceil(a, b)
        tvm.testing.assert_allclose(b.numpy(), (np.ceil(a.numpy()).view("float32")))

    check_c()

tir.floor c host

def test_floor():
    nn = 1024
    n = tvm.runtime.convert(nn)
    A = te.placeholder((n,), name="A", dtype="float32")
    B = te.compute(A.shape, lambda *i: tvm.tir.call_intrin("float32", "tir.floor", A(*i)), name="B")
    s = te.create_schedule(B.op)

    def check_c():
        mhost = tvm.build(s, [A, B], "c", name="test_floor")
        temp = utils.tempdir()
        path_dso = temp.relpath("temp.so")
        mhost.export_library(path_dso)
        m = tvm.runtime.load_module(path_dso)
        ffloor = m["test_floor"]
        dev = tvm.cpu(0)
        n = nn
        a = tvm.nd.array(np.random.rand(n).astype(A.dtype), dev)
        b = tvm.nd.array(np.zeros(n, dtype=B.dtype), dev)
        ffloor(a, b)
        tvm.testing.assert_allclose(b.numpy(), (np.floor(a.numpy()).view("float32")))

    check_c()

tir.round c host

def test_round():
    nn = 1024
    n = tvm.runtime.convert(nn)
    A = te.placeholder((n,), name="A", dtype="float32")
    B = te.compute(A.shape, lambda *i: tvm.tir.call_intrin("float32", "tir.round", A(*i)), name="B")
    s = te.create_schedule(B.op)

    def check_c():
        mhost = tvm.build(s, [A, B], "c", name="test_round")
        temp = utils.tempdir()
        path_dso = temp.relpath("temp.so")
        mhost.export_library(path_dso)
        m = tvm.runtime.load_module(path_dso)
        fround = m["test_round"]
        dev = tvm.cpu(0)
        n = nn
        a = tvm.nd.array(np.random.rand(n).astype(A.dtype), dev)
        b = tvm.nd.array(np.zeros(n, dtype=B.dtype), dev)
        fround(a, b)
        tvm.testing.assert_allclose(b.numpy(), (np.round(a.numpy()).view("float32")))

    check_c()

subroutine c host

def test_subroutine_call():
    @I.ir_module
    class mod:
        @T.prim_func
        def main(A: T.Buffer(1, dtype="float32")):
            mod.subroutine(A.data)

        @T.prim_func(private=True)
        def subroutine(A_data: T.handle("float32")):
            A = T.decl_buffer(1, dtype="float32", data=A_data)
            A[0] = 42.0

    built = tvm.build(mod, target="c")

    func_names = list(built["get_func_names"]())
    assert (
        "main" in func_names
    ), "Externally exposed functions should be listed in available functions."
    assert (
        "subroutine" not in func_names
    ), "Internal function should not be listed in available functions."

    source = built.get_source()
    assert (
        source.count("main(void*") == 2
    ), "Expected two occurrences, for forward-declaration and definition"
    assert (
        source.count("subroutine(float*") == 2
    ), "Expected two occurrences, for forward-declaration and definition"
    assert (
        source.count("subroutine(") == 3
    ), "Expected three occurrences, for forward-declaration, definition, and call from main."

tvm.tir.call_intrin() c host

def test_call_packed():
    def fake_func(fname="fake.func"):
        ib = tvm.tir.ir_builder.create()
        A = ib.pointer("float32", name="A")
        fake_func1 = tvm.tir.call_packed(fname, A[0])

        ib.emit(fake_func1)
        body = ib.get()
        return A, body

    def check_global_packed_func():
        fname = "fake.func"
        A, body = fake_func(fname)
        func1 = tvm.tir.PrimFunc([A], body).with_attr("global_symbol", "func1")
        B, body = fake_func()
        func2 = tvm.tir.PrimFunc([B], body).with_attr("global_symbol", "func2")
        mod = tvm.IRModule({"fake_func1": func1, "fake_func2": func2})
        fcode = tvm.build(mod, None, "c")
        src = fcode.get_source()

        # there are two locations calling the packed func
        assert src.count(fname) == 2

        suffix = "_packed"
        packed_func_name = fname + suffix
        # func name will be standardized by GetUniqueName and not exists anymore
        assert src.find(packed_func_name) == -1

        packed_func_real_name = "_".join(fname.split(".")) + suffix
        func_declaration = "static void* %s = NULL;" % packed_func_real_name
        # src only has 1 valid declaration
        assert src.count(func_declaration) == 1

    check_global_packed_func()