Skip to content

T
tic
Unverified Commit f08d5d78 by Tianqi Chen Committed by GitHub
Options

[TIR] Refactor MakePackedAPI to target dependent stage. (#5326) 

Previously MakePackedAPI was in the target independent stage,
but never the less requires the device_type information that will be
binded at a later target dependent stage.

The previous implementation was due to the limitation of LoweredFunc
which can not carry buffer_map info(so they have to be lowered right away).
This is no longer the case after the unified IR refactor.

This PR migrates MakePackedAPI to a target dependent stage
and removes the un-necessary BindDevice pass.
parent 4720cf85
Showing with 114 additions and 187 deletions
include/tvm/ir/transform.h
......@@ -352,12 +352,19 @@ class Sequential : public Pass {
*
* \return The created module pass.
*/
Pass CreateModulePass(
TVM_DLL Pass CreateModulePass(
const runtime::TypedPackedFunc<IRModule(IRModule, PassContext)>& pass_func,
int opt_level,
const std::string& name,
const Array<runtime::String>& required);
/*!
* \brief A special trace pass that prints the header and IR to LOG(INFO).
* \return The pass.
*/
TVM_DLL Pass PrintIR(std::string header);
} // namespace transform
} // namespace tvm
......
include/tvm/runtime/device_api.h
......@@ -193,6 +193,15 @@ class TVM_DLL DeviceAPI {
* \return The corresponding device API.
*/
static DeviceAPI* Get(TVMContext ctx, bool allow_missing = false);
/*!
* \brief Whether a certian device type requires set device context
* before launching the kernel function.
* \param device_type The device type.
*/
static bool NeedSetDeviceContext(int device_type) {
return device_type != kDLCPU && device_type != kDLMicroDev;
}
};
/*! \brief The device type bigger than this is RPC device */
......
include/tvm/tir/transform.h
......@@ -112,15 +112,6 @@ TVM_DLL Pass RemapThreadAxis(Map<runtime::String, IterVar> axis_map);
*/
TVM_DLL Pass LowerCustomDatatypes();
/*!
* \brief Bind the device type ofthe function to be
* the device_type specified in the target attribute.
*
* \return The pass.
*/
TVM_DLL Pass BindDeviceType();
/*!
* \brief Split the function into a host function and device functions.
*
......
python/tvm/driver/build_module.py
......@@ -200,7 +200,7 @@ def lower(sch,
if cfg.restricted_func:
f = f.with_attr("tir.noalias", True)
mod = tvm.IRModule({name: f})
return tvm.tir.transform.MakePackedAPI()(mod)
return mod
def _build_for_device(input_mod, target, target_host):
......@@ -243,13 +243,13 @@ def _build_for_device(input_mod, target, target_host):
tvm.tir.transform.ThreadSync("warp"),
tvm.tir.transform.InferFragment(),
tvm.tir.transform.LowerThreadAllreduce(),
tvm.tir.transform.BindDeviceType(),
tvm.tir.transform.MakePackedAPI(),
tvm.tir.transform.SplitHostDevice()]
mod_mixed = tvm.ir.transform.Sequential(opt_mixed)(mod_mixed)
mod_mixed = tvm.transform.Sequential(opt_mixed)(mod_mixed)
# device optimizations
opt_device = tvm.ir.transform.Sequential(
opt_device = tvm.transform.Sequential(
[tvm.tir.transform.Filter(
lambda f: "calling_conv" in f.attrs and
f.attrs["calling_conv"].value == CallingConv.DEVICE_KERNEL_LAUNCH),
......@@ -259,7 +259,7 @@ def _build_for_device(input_mod, target, target_host):
mod_dev = opt_device(mod_mixed)
# host optimizations
opt_host = tvm.ir.transform.Sequential(
opt_host = tvm.transform.Sequential(
[tvm.tir.transform.Filter(
lambda f: "calling_conv" not in f.attrs or
f.attrs["calling_conv"].value != CallingConv.DEVICE_KERNEL_LAUNCH),
......
python/tvm/ir/transform.py
......@@ -22,13 +22,13 @@ import functools
import tvm._ffi
from tvm._ffi.runtime_ctypes import TVMContext
from tvm.runtime import Object, ndarray as _nd
import tvm.runtime
from tvm.runtime import ndarray as _nd
from . import _ffi_transform_api
@tvm._ffi.register_object("transform.PassInfo")
class PassInfo(Object):
class PassInfo(tvm.runtime.Object):
"""The class contains the meta data required by a pass. It is the
container of information needed by running an optimization or analysis.
This class can be extended by adding new members when more meta data is
......@@ -52,7 +52,7 @@ class PassInfo(Object):
@tvm._ffi.register_object("transform.PassContext")
class PassContext(Object):
class PassContext(tvm.runtime.Object):
"""The basis where a Relay optimization/analysis runs on.
Each pass context contains a number of auxiliary information that is used
to help an optimization pass. Such information includes the error reporter
......@@ -79,7 +79,7 @@ class PassContext(Object):
trace=None):
if isinstance(fallback_device, str):
fallback_device = _nd.context(fallback_device).device_type
elif isinstance(fallback_device, TVMContext):
elif isinstance(fallback_device, tvm.runtime.TVMContext):
fallback_device = fallback_device.device_type
if not isinstance(fallback_device, int):
raise TypeError("fallback_device is expected to be the type of " +
......@@ -113,7 +113,7 @@ class PassContext(Object):
@tvm._ffi.register_object("transform.Pass")
class Pass(Object):
class Pass(tvm.runtime.Object):
"""The base class of all passes. All methods here are just simple wrappers
that are implemented in the backend. They are defined for users to
conveniently interact with the base class.
......@@ -327,3 +327,18 @@ def module_pass(pass_func=None, opt_level=None, name=None, required=None):
if pass_func:
return create_module_pass(pass_func)
return create_module_pass
def PrintIR(header):
"""A special trace pass that prints the header and IR.
Parameters
----------
header : str
The header to be displayed along with the dump.
Returns
--------
The pass
"""
return _ffi_transform_api.PrintIR(header)
python/tvm/testing.py
......@@ -195,13 +195,14 @@ def MakeAPILegacy(stmt, name, args, num_unpacked_args, noalias):
mod : IRModule
The created IRModule.
"""
assert num_unpacked_args == 0
f = tvm.tir.PrimFunc(args, stmt).with_attr(
"global_symbol", tvm.runtime.String(name))
f = f.with_attr("tir.is_entry_func", True)
if noalias:
f = f.with_attr("tir.noalias", True)
mod = tvm.IRModule({name: f})
return tvm.tir.transform.MakePackedAPI(num_unpacked_args)(mod)
return mod
tvm._ffi._init_api("testing", __name__)
python/tvm/tir/transform/transform.py
......@@ -32,7 +32,7 @@ def Apply(ftransform):
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
# pylint: disable=unused-argument
......@@ -51,7 +51,7 @@ def Filter(fcond):
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
# pylint: disable=unused-argument
......@@ -67,7 +67,7 @@ def LowerCustomDatatypes():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.LowerCustomDatatypes()
......@@ -84,30 +84,18 @@ def MakePackedAPI(num_unpacked_params=0):
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.MakePackedAPI(num_unpacked_params)
def BindDeviceType():
"""Bind the device type of the function to be
the device_type specified in the target attribute.
Returns
-------
fpass : tvm.ir.transform.Pass
The result pass
"""
return _ffi_api.BindDeviceType()
def SplitHostDevice():
"""Split the function into a host function and device functions.
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.SplitHostDevice()
......@@ -118,7 +106,7 @@ def SkipAssert():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.SkipAssert()
......@@ -134,7 +122,7 @@ def ThreadSync(storage_scope):
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.ThreadSync(storage_scope)
......@@ -145,7 +133,7 @@ def LowerThreadAllreduce():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.LowerThreadAllreduce()
......@@ -156,7 +144,7 @@ def InferFragment():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.InferFragment()
......@@ -167,7 +155,7 @@ def LowerWarpMemory():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.LowerWarpMemory()
......@@ -178,7 +166,7 @@ def LowerTVMBuiltin():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.LowerTVMBuiltin()
......@@ -189,7 +177,7 @@ def LowerIntrin():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.LowerIntrin()
......@@ -200,7 +188,7 @@ def LowerDeviceStorageAccessInfo():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
Note
......@@ -215,7 +203,7 @@ def CombineContextCall():
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
"""
return _ffi_api.CombineContextCall()
......@@ -231,7 +219,7 @@ def NarrowDataType(target_bits):
Returns
-------
fpass : tvm.ir.transform.Pass
fpass : tvm.transform.Pass
The result pass
Note
......
src/driver/driver_api.cc
......@@ -216,8 +216,7 @@ IRModule lower(te::Schedule sch,
if (config->restricted_func) {
f = WithAttr(std::move(f), "tir.noalias", Integer(1));
}
auto mod = IRModule(Map<GlobalVar, BaseFunc>({{GlobalVar(name), f}}));
return tir::transform::MakePackedAPI(0)(mod);
return IRModule(Map<GlobalVar, BaseFunc>({{GlobalVar(name), f}}));
}
......@@ -237,7 +236,7 @@ split_dev_host_funcs(IRModule mod_mixed,
mixed_pass_list.push_back(tir::transform::ThreadSync("warp"));
mixed_pass_list.push_back(tir::transform::InferFragment());
mixed_pass_list.push_back(tir::transform::LowerThreadAllreduce());
mixed_pass_list.push_back(tir::transform::BindDeviceType());
mixed_pass_list.push_back(tir::transform::MakePackedAPI(0));
mixed_pass_list.push_back(tir::transform::SplitHostDevice());
auto opt_mixed = transform::Sequential(mixed_pass_list);
mod_mixed = opt_mixed(std::move(mod_mixed));
......
src/ir/transform.cc
......@@ -473,5 +473,18 @@ TVM_REGISTER_GLOBAL("transform.EnterPassContext")
TVM_REGISTER_GLOBAL("transform.ExitPassContext")
.set_body_typed(PassContext::Internal::ExitScope);
Pass PrintIR(std::string header) {
auto pass_func =[header](IRModule mod, const PassContext& ctx) {
LOG(INFO) << "PrintIR(" << header << "):\n"
<< mod;
return mod;
};
return CreateModulePass(pass_func, 0, "PrintIR", {});
}
TVM_REGISTER_GLOBAL("transform.PrintIR")
.set_body_typed(PrintIR);
} // namespace transform
} // namespace tvm
src/target/stackvm/codegen_stackvm.cc
......@@ -58,6 +58,8 @@ StackVM::StructFieldKind MapFieldKind(int64_t kind) {
}
StackVM CodeGenStackVM::Compile(const PrimFunc& f) {
CHECK_EQ(f->buffer_map.size(), 0U)
<< "Cannot codegen function with buffer_map, please lower them first";
for (size_t i = 0; i < f->params.size(); ++i) {
Var v = f->params[i];
int vid = AllocVarID(v.get());
......
src/tir/analysis/verify_memory.cc
......@@ -114,9 +114,11 @@ class MemoryAccessVerifier final : protected StmtExprVisitor {
/// Check if the value of a Variable comes from function argument.
bool IsFromFunctionArgs(const VarNode *var) const {
const VarNode *V = var;
while (true) {
CHECK(V) << "Invalid Variable\n";
for (auto kv : func_->buffer_map) {
if (V == kv.second->data.get()) return true;
}
while (true) {
// Variable is from function args. Return true.
if (V == func_->params[0].get()) return true;
......
src/tir/transforms/bind_device_type.cc deleted 100644 → 0
/*
* 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.
*/
/*!
* \file bind_device_type.cc
* \brief Bind the device type according to the target field.
*/
#include <tvm/ir/transform.h>
#include <tvm/tir/expr.h>
#include <tvm/tir/op.h>
#include <tvm/tir/transform.h>
#include <tvm/tir/stmt_functor.h>
#include <tvm/tir/analysis.h>
#include <tvm/target/target.h>
#include <tvm/runtime/registry.h>
namespace tvm {
namespace tir {
class DeviceTypeBinder: public StmtExprMutator {
public:
explicit DeviceTypeBinder(int device_type)
: device_type_(device_type) {}
Stmt VisitStmt_(const AttrStmtNode* op) final {
if (op->attr_key == attr::device_context_type) {
if (const VarNode* var = op->value.as<VarNode>()) {
var_ = var;
PrimExpr value = make_const(op->value.dtype(), device_type_);
Stmt body = StmtExprMutator::VisitStmt_(op);
var_ = nullptr;
std::ostringstream os;
os << "device_type need to be " << device_type_;
return AssertStmtNode::make(op->value == value, tvm::tir::StringImmNode::make(os.str()),
body);
}
}
return StmtExprMutator::VisitStmt_(op);
}
Stmt VisitStmt_(const IfThenElseNode* op) final {
// eager simplify if guard.
Stmt res = StmtExprMutator::VisitStmt_(op);
op = res.as<IfThenElseNode>();
if (is_zero(op->condition)) {
if (op->else_case.defined()) return op->else_case;
return EvaluateNode::make(0);
}
if (is_one(op->condition)) {
return op->then_case;
}
return res;
}
PrimExpr VisitExpr_(const NENode* op) final {
// eager check NE for device check
PrimExpr res = StmtExprMutator::VisitExpr_(op);
op = res.as<NENode>();
if (tir::ExprDeepEqual()(op->a, op->b)) {
return make_const(op->dtype, false);
}
return res;
}
PrimExpr VisitExpr_(const VarNode* op) final {
if (op == var_) {
return make_const(op->dtype, device_type_);
} else {
return GetRef<PrimExpr>(op);
}
}
public:
const VarNode* var_{nullptr};
int device_type_;
};
namespace transform {
Pass BindDeviceType() {
auto pass_func = [](PrimFunc f, IRModule m, PassContext ctx) {
auto* n = f.CopyOnWrite();
auto target = f->GetAttr<Target>(tvm::attr::kTarget);
CHECK(target.defined())
<< "BindDeviceType: Require the target attribute";
n->body = DeviceTypeBinder(target.value()->device_type)(std::move(n->body));
return f;
};
return CreatePrimFuncPass(pass_func, 0, "tir.BindDeviceType", {});
}
TVM_REGISTER_GLOBAL("tir.transform.BindDeviceType")
.set_body_typed(BindDeviceType);
} // namespace transform
} // namespace tir
} // namespace tvm
src/tir/transforms/make_packed_api.cc
......@@ -26,6 +26,7 @@
#include <tvm/tir/transform.h>
#include <tvm/tir/stmt_functor.h>
#include <tvm/tir/buffer.h>
#include <tvm/target/target.h>
#include <tvm/runtime/device_api.h>
#include <tvm/runtime/registry.h>
#include <tvm/runtime/container.h>
......@@ -50,6 +51,12 @@ PrimFunc MakePackedAPI(PrimFunc&& func,
auto global_symbol = func->GetAttr<String>(tvm::attr::kGlobalSymbol);
CHECK(global_symbol)
<< "MakePackedAPI: Expect PrimFunc to have the global_symbol attribute";
auto target = func->GetAttr<Target>(tvm::attr::kTarget);
CHECK(target.defined())
<< "MakePackedAPI: Require the target attribute";
int target_device_type = target.value()->device_type;
std::string name_hint = global_symbol.value();
auto* func_ptr = func.CopyOnWrite();
......@@ -68,7 +75,8 @@ PrimFunc MakePackedAPI(PrimFunc&& func,
// The arguments of the function.
Array<Var> args;
// The device context
Var device_type("dev_type"), device_id("dev_id");
Var device_id("dev_id");
Integer device_type(target_device_type);
// seq_init gives sequence of initialization
// seq_check gives sequence of later checks after init
std::vector<Stmt> seq_init, seq_check;
......@@ -195,17 +203,18 @@ PrimFunc MakePackedAPI(PrimFunc&& func,
// Set device context
if (vmap.count(device_id.get())) {
PrimExpr node = StringImmNode::make("default");
CHECK(vmap.count(device_type.get()));
seq_check.push_back(AttrStmtNode::make(
node, attr::device_context_id, device_id, nop));
seq_check.push_back(AttrStmtNode::make(
node, attr::device_context_type, device_type, nop));
Stmt set_device = IfThenElseNode::make(
device_type != kDLCPU, EvaluateNode::make(CallNode::make(
DataType::Int(32), intrinsic::tvm_call_packed,
{StringImmNode::make(runtime::symbol::tvm_set_device),
device_type, device_id}, CallNode::Intrinsic)));
body = SeqStmt({set_device, body});
if (runtime::DeviceAPI::NeedSetDeviceContext(target_device_type)) {
Stmt set_device = EvaluateNode::make(CallNode::make(
DataType::Int(32), intrinsic::tvm_call_packed,
{StringImmNode::make(runtime::symbol::tvm_set_device),
device_type, device_id}, CallNode::Intrinsic));
body = SeqStmt({set_device, body});
}
}
func_ptr->body = MergeNest(
{seq_init, binder.init_nest(), seq_check, binder.asserts()}, body);
......
tests/python/unittest/test_runtime_extension.py
......@@ -39,10 +39,8 @@ def test_dltensor_compatible():
A[i + 1] = A[i] + 1
stmt = ib.get()
mod = tvm.testing.MakeAPILegacy(stmt, "arange", [Ab], 0, True)
mod = tvm.tir.transform.LowerTVMBuiltin()(mod)
f = tvm.target.codegen.build_module(mod, "stackvm")
f = tvm.build(mod, target="stackvm")
a = tvm.nd.array(np.zeros(10, dtype=dtype))
aview = MyTensorView(a)
f(aview)
......
tests/python/unittest/test_target_codegen_llvm.py
......@@ -111,7 +111,7 @@ def test_llvm_lookup_intrin():
x = tvm.tir.call_llvm_intrin("uint8x8", "llvm.ctpop.v8i8", tvm.tir.const(1, 'uint32'), A[z])
ib.emit(x)
body = ib.get()
func = tvm.testing.MakeAPILegacy(body, "ctpop", [A], 1, True)
func = tvm.testing.MakeAPILegacy(body, "ctpop", [A], 0, True)
fcode = tvm.build(func, None, "llvm")
......
tests/python/unittest/test_tir_analysis_verify_memory.py
......@@ -44,7 +44,7 @@ def lower(sch, args):
f = tvm.tir.PrimFunc(arg_list, stmt).with_attr(
"global_symbol", tvm.runtime.String("test"))
mod = tvm.IRModule({"test": f})
return tvm.tir.transform.MakePackedAPI()(mod)
return mod
# All computations are bound.
......
tests/python/unittest/test_tir_pass_inject_double_buffer.py
......@@ -40,9 +40,10 @@ def test_double_buffer():
stmt = tvm.tir.ir_pass.Simplify(stmt)
assert isinstance(stmt.body.body, tvm.tir.Allocate)
assert stmt.body.body.extents[0].value == 2
mod = tvm.testing.MakeAPILegacy(stmt, "db", [A.asobject(), C.asobject()], 2, True)
mod = tvm.IRModule({
"db" : tvm.tir.PrimFunc([A.asobject(), C.asobject()], stmt)
})
f = tvm.tir.transform.ThreadSync("shared")(mod)["db"]
count = [0]
def count_sync(op):
if isinstance(op, tvm.tir.Call) and op.name == "tvm_storage_sync":
......
tests/python/unittest/test_tir_pass_storage_flatten.py
......@@ -92,7 +92,7 @@ def test_flatten_double_buffer():
stmt = tvm.tir.ir_pass.Simplify(stmt)
assert isinstance(stmt.body.body, tvm.tir.Allocate)
assert stmt.body.body.extents[0].value == 2
mod = tvm.testing.MakeAPILegacy(stmt, "db", [A.asobject(), C.asobject()], 2, True)
mod = tvm.testing.MakeAPILegacy(stmt, "db", [A.asobject(), C.asobject()], 0, True)
f = tvm.tir.transform.ThreadSync("shared")(mod)["db"]
count = [0]
......
tests/python/unittest/test_tir_transform_combine_context_call.py
......@@ -36,7 +36,10 @@ def test_for():
ib.emit(tvm.tir.call_extern
("int32", "fadd", device_context(0), A))
body = ib.get()
mod = tvm.testing.MakeAPILegacy(body, "func", [dev_type, n], 2, True)
mod = tvm.IRModule({
"func" : tvm.tir.PrimFunc([dev_type, n], body)
})
mod = tvm.tir.transform.CombineContextCall()(mod)
assert mod["func"].body.value.dtype == "handle"
......
tests/python/unittest/test_tir_transform_make_packed_api.py
......@@ -35,8 +35,10 @@ def test_makeapi():
stmt = tvm.tir.ir_pass.StorageFlatten(stmt, {A: Ab, B:Bb, C:Cb}, 64)
num_unpacked_args = 2
f = tvm.tir.PrimFunc([n, Ab, Bb, Cb], stmt).with_attr(
"tir.noalias", True).with_attr("global_symbol", tvm.runtime.String("myadd"))
f = tvm.tir.PrimFunc([n, Ab, Bb, Cb], stmt)
f = f.with_attr("global_symbol", "myadd")
f = f.with_attr("target", tvm.target.create("llvm"))
mod = tvm.IRModule.from_expr(f)
f = tvm.tir.transform.MakePackedAPI(num_unpacked_args)(mod)["main"]
assert(len(f.params) == 7)
......
tests/python/unittest/test_tir_transform_prim_func_pass.py
......@@ -60,7 +60,7 @@ def test_cow_pass():
del func
# copy on write
mod_hash = mod.__hash__()
mod = tvm.ir.transform.Sequential(
mod = tvm.transform.Sequential(
[pidentity, tvm.tir.transform.NarrowDataType(32)])(mod._move())
assert mod_hash == mod.__hash__()
assert func_hash == mod["main"].__hash__()
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment