xxq250
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

upgrade ascend 20220211

This commit is contained in:
shenwei41 2022-02-12 15:31:38 +08:00
parent f8ab74ef29
commit ff75314719
29 changed files with 114 additions and 173 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
graphengine

@ -1 +1 @@
Subproject commit 008fafbb15d6e1126b424dbb8689a4795578b5b6
Subproject commit 6b57c837b3f33a479817d9ddfeebfe4d19afe66b

2
mindspore/ccsrc/backend/common/optimizer/common_backend_optimization.cc
View File

@ -24,7 +24,6 @@
#include "backend/common/pass/convert_tuple_output_to_maketuple.h"
#include "backend/common/pass/convert_const_input_to_tensor_input.h"
#include "backend/common/pass/convert_tuple_input_to_dynamic_input.h"
#include "backend/common/pass/const_to_attr_strided_slice_grad.h"
#include "backend/common/pass/convert_const_scalar_to_tensor.h"
#include "backend/common/pass/convert_attr_to_unify_mindir.h"
#include "backend/common/pass/add_training_attr.h"
@ -61,7 +60,6 @@ void BackendCommonOptimization(const std::shared_ptr<session::KernelGraph> &kern
common_pm->AddPass(std::make_shared<CustomOpConstInputToAttr>());
common_pm->AddPass(std::make_shared<SparseProcess>());
common_pm->AddPass(std::make_shared<ConvertAttrToUnifyMindIR>());
common_pm->AddPass(std::make_shared<ConstToAttrStridedSliceGradPass>());
common_pm->AddPass(std::make_shared<ConvertConstInputToTensorInput>());
common_pm->AddPass(std::make_shared<ConvertTupleOutputToMaketuple>());
common_pm->AddPass(std::make_shared<ConvertConstScalarToTensor>());

11
mindspore/ccsrc/backend/common/pass/convert_const_input_to_attr.cc
View File

@ -67,6 +67,17 @@ const AnfNodePtr ConvertConstInputToAttr::Process(const FuncGraphPtr &, const An
}
}
}
if (device == kAscendDevice &&
NeedConvertToValueNodeSet.find(AnfAlgo::GetCNodeName(cnode)) != NeedConvertToValueNodeSet.end() &&
!AnfAlgo::HasNodeAttr(kAttrNeedConvertToValueNode, cnode)) {
auto input_attrs = reg.GetConstInputAttrInfo();
std::vector<size_t> need_convert_to_constant;
std::transform(input_attrs.begin(), input_attrs.end(), std::back_inserter(need_convert_to_constant),
[](size_t i) { return i + 1; });
AnfAlgo::SetNodeAttr(kAttrNeedConvertToValueNode, MakeValue(need_convert_to_constant), cnode);
return nullptr;
}
ConstInputToAttr(cnode, reg.GetConstInputAttrInfo());
return node;

4
mindspore/ccsrc/backend/common/session/anf_runtime_algorithm.cc
View File

@ -149,7 +149,9 @@ static std::map<std::string, std::pair<std::map<size_t, size_t>, std::map<size_t
{prim::kPrimMaximumGrad->name(), {{{0, 2}, {1, 0}, {2, 1}}, {{2, 0}, {0, 1}, {1, 2}}}},
{prim::kPrimApplyCenteredRMSProp->name(),
{{{0, 0}, {1, 1}, {2, 2}, {3, 3}, {4, 5}, {5, 6}, {6, 7}, {7, 8}, {8, 4}},
{{0, 0}, {1, 1}, {2, 2}, {3, 3}, {5, 4}, {6, 5}, {7, 6}, {8, 7}, {4, 8}}}}};
{{0, 0}, {1, 1}, {2, 2}, {3, 3}, {5, 4}, {6, 5}, {7, 6}, {8, 7}, {4, 8}}}},
{prim::kPrimStridedSliceGrad->name(),
{{{0, 1}, {1, 2}, {2, 3}, {3, 4}, {4, 0}}, {{1, 0}, {2, 1}, {3, 2}, {4, 3}, {0, 4}}}}};
std::vector<KernelWithIndex> GetAllOutputWithIndexInner(const AnfNodePtr &node) {
std::vector<KernelWithIndex> ret;

18
mindspore/ccsrc/backend/common/session/session_basic.cc
View File

@ -1454,7 +1454,9 @@ void SessionBasic::GetForwardOpOutputRefCount(const KernelGraph *graph, const st
MS_EXCEPTION_IF_NULL(real_input);
if (real_input->isa<ValueNode>()) {
const auto &tensor = GetValueNodeOutputTensor(real_input, kernel_with_index.second);
MS_EXCEPTION_IF_NULL(tensor);
if (tensor == nullptr) {
continue;
}
if (forward_op_output_id.find(tensor->id()) != forward_op_output_id.end()) {
(*forward_op_output_tensor_id)[tensor->id()] += 1;
}
@ -1630,6 +1632,11 @@ void SessionBasic::GetOpInputTensors(const CNodePtr &cnode,
InputTensorInfo *input_tensor_info) {
MS_EXCEPTION_IF_NULL(cnode);
MS_EXCEPTION_IF_NULL(input_tensor_info);
auto has_const_input_to_attr = AnfAlgo::HasNodeAttr(kAttrNeedConvertToValueNode, cnode);
std::vector<size_t> const_input_attr_index = {};
if (has_const_input_to_attr) {
const_input_attr_index = AnfAlgo::GetNodeAttr<std::vector<size_t>>(cnode, kAttrNeedConvertToValueNode);
}
const auto input_tensor_num = AnfAlgo::GetInputTensorNum(cnode);
for (size_t i = 1; i <= input_tensor_num; i += 1) {
const auto &input = cnode->input(i);
@ -1641,8 +1648,13 @@ void SessionBasic::GetOpInputTensors(const CNodePtr &cnode,
tensor = GetValueNodeOutputTensor(real_input, kernel_with_index.second);
const auto &value_ptr = GetValueNode(real_input);
MS_EXCEPTION_IF_NULL(value_ptr);
input_tensor_info->input_tensors_mask.emplace_back(value_ptr->isa<StringImm>() ? kValueNodeTensorMask
: kParameterDataTensorMask);
auto is_value_node = value_ptr->isa<StringImm>();
if (has_const_input_to_attr) {
is_value_node =
std::find(const_input_attr_index.begin(), const_input_attr_index.end(), i) != const_input_attr_index.end();
}
input_tensor_info->input_tensors_mask.emplace_back(is_value_node ? kValueNodeTensorMask
: kParameterDataTensorMask);
} else if (real_input->isa<Parameter>()) {
tensor = GetParameterOutputTensor(real_input, parameter_index, graph_inputs);
input_tensor_info->input_tensors_mask.emplace_back(tensor->is_parameter() ? kParameterWeightTensorMask

1
mindspore/ccsrc/backend/graph_compiler/backend.cc
View File

@ -867,7 +867,6 @@ void MindRTBackend::RunGraphBySingleOp(const std::vector<KernelGraphPtr> &graphs
for (const auto &kernel : graph->execution_order()) {
InputTensorInfo input_tensor_info;
VectorRef op_outputs;
if (!AnfAlgo::IsControlOpExecInBackend(kernel)) {
OpRunInfo op_run_info;
GraphInfo graph_info;

1
mindspore/ccsrc/kernel/kash/kernel_pack.cc
View File

@ -203,6 +203,7 @@ void KernelPack::ParseKernelJson(const nlohmann::json &js) {
}
}
kernel_json_info_.sha256 = js["sha256"];
kernel_json_info_.has_kernel_list = js.find("kernelList") != js.end();
}
bool KernelPack::LoadKernelMeta(const std::string &json_f) {

1
mindspore/ccsrc/kernel/kernel.h
View File

@ -128,6 +128,7 @@ struct KernelJsonInfo {
std::vector<size_t> parameters;
std::string sha256;
std::vector<size_t> workspaces;
bool has_kernel_list = false;
uint32_t op_para_size;
KernelJsonInfo() : block_dim(0), op_para_size(0) {}
};

3
mindspore/ccsrc/plugin/device/ascend/hal/device/executor/ai_core_dynamic_kernel.cc
View File

@ -94,8 +94,7 @@ void AiCoreDynamicKernel::Execute() {
rtL2Ctrl_t *l2ctrl = nullptr;
auto args_size = static_cast<uint32_t>(UlongToUint(sizeof(void *)) * runtime_args_.size());
if (handle_ != nullptr) {
const auto dev_func =
origin_key_.find("kernel0") != origin_key_.npos ? origin_key_ : origin_key_ + "_" + std::to_string(tiling_key_);
const auto dev_func = std::to_string(tiling_key_);
const auto kernel_info = node_info + "/" + std::to_string(tiling_key_);
if (RT_ERROR_NONE != rtKernelLaunchWithHandle(handle_, dev_func.c_str(), block_dim_, runtime_args_.data(),
args_size, l2ctrl, stream_, kernel_info.c_str())) {

2
mindspore/ccsrc/plugin/device/ascend/hal/device/ge_runtime/runtime_model.cc
View File

@ -122,7 +122,7 @@ void RuntimeModel::InitLabel(const std::shared_ptr<DavinciModel> &davinci_model)
}
rtLabel_t rt_label = nullptr;
rtError_t rt_ret = rtLabelCreateEx(&rt_label, stream_list_[label_set_task_info->stream_id()]);
rtError_t rt_ret = rtLabelCreateExV2(&rt_label, rt_model_handle_, stream_list_[label_set_task_info->stream_id()]);
if (rt_ret != RT_ERROR_NONE) {
MS_LOG(EXCEPTION) << "Call rt api rtLabelCreate failed, ret: " << rt_ret;
}

8
mindspore/ccsrc/plugin/device/ascend/kernel/aicpu/aicpu_util.h
View File

@ -77,10 +77,10 @@ constexpr auto kEnvironGet = "EnvironGet";
constexpr auto kEnvironDestroyAll = "EnvironDestroyAll";
const std::set<std::string> kCpuKernelOps{kIdentity, kMaskedSelect, kMaskedSelectGrad, kDynamicStitch,
kSearchSorted, kResizeBilinear, kResizeBilinearGrad, kScatterElements};
const std::set<std::string> kCacheKernelOps{kUpdateCache, kCacheSwapTable, kSubAndFilter, kPadAndShift,
kDropout3D, kDropout2D, kNonMaxSuppressionV3};
const std::set<std::string> kCpuKernelBaseOps{kGetNext, kInitData, kRandomChoiceWithMask, kEnvironCreate,
kEnvironSet, kEnvironGet, kEnvironDestroyAll};
const std::set<std::string> kCacheKernelOps{kUpdateCache, kCacheSwapTable, kSubAndFilter, kPadAndShift, kDropout3D,
kDropout2D, kNonMaxSuppressionV3, kGetNext, kInitData, kPrint};
const std::set<std::string> kCpuKernelBaseOps{kRandomChoiceWithMask, kEnvironCreate, kEnvironSet, kEnvironGet,
kEnvironDestroyAll};
const std::set<std::string> kDynamicInputOps{
kPrint, kPack, kMeshgrid, kStackInitOpName, kStackDestroyOpName, kStackPushOpName, kStackPopOpName, kDynamicStitch};
struct AicpuParamHead {

42
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/dynamic_tbe_kernel_mod.cc
View File

@ -102,10 +102,17 @@ void DynamicTbeKernelMod::InitOp() {
}
// gen FuncStub
if (handle_ == nullptr) {
if (func_stub_ == nullptr || handle_ == nullptr) {
auto func_stub = KernelManager::GenFuncStub(*kernel_pack_, false, &block_dim_, true, &handle_, &origin_key_);
if (func_stub != 1) {
MS_LOG(EXCEPTION) << "GenFuncStub failed.";
if (kernel_pack_->kernel_json_info().has_kernel_list) {
if (func_stub != 1) {
MS_LOG(EXCEPTION) << "GenFuncStub failed.";
}
} else {
if (func_stub == 0) {
MS_LOG(EXCEPTION) << "GenFuncStub failed.";
}
func_stub_ = reinterpret_cast<void *>(func_stub);
}
}
@ -286,16 +293,25 @@ bool DynamicTbeKernelMod::Launch(const std::vector<AddressPtr> &inputs, const st
rtL2Ctrl_t *l2ctrl = nullptr;
auto args_size = static_cast<uint32_t>(UlongToUint(sizeof(void *)) * runtimeargs.size());
auto node_info = cnode->fullname_with_scope();
const auto dev_func =
origin_key_.find("kernel0") != origin_key_.npos ? origin_key_ : origin_key_ + "_" + std::to_string(tiling_key_);
const auto kernel_info = node_info + "/" + std::to_string(tiling_key_);
// cppcheck-suppress unreadVariable
auto lock = device::KernelRuntime::LockRuntime();
auto ret = rtKernelLaunchWithHandle(handle_, dev_func.c_str(), block_dim_, runtimeargs.data(), args_size, l2ctrl,
stream_ptr, kernel_info.c_str());
if (ret != RT_ERROR_NONE) {
MS_LOG(ERROR) << "Call runtime rtKernelLaunchWithHandle error. Node info: " << node_info;
return false;
if (kernel_pack_->kernel_json_info().has_kernel_list) {
const auto dev_func = std::to_string(tiling_key_);
const auto kernel_info = node_info + "/" + std::to_string(tiling_key_);
// cppcheck-suppress unreadVariable
auto lock = device::KernelRuntime::LockRuntime();
auto ret = rtKernelLaunchWithHandle(handle_, dev_func.c_str(), block_dim_, runtimeargs.data(), args_size, l2ctrl,
stream_ptr, kernel_info.c_str());
if (ret != RT_ERROR_NONE) {
MS_LOG(ERROR) << "Call runtime rtKernelLaunchWithHandle error. Node info: " << node_info;
return false;
}
} else {
// cppcheck-suppress unreadVariable
auto lock = device::KernelRuntime::LockRuntime();
auto ret = rtKernelLaunch(func_stub_, block_dim_, runtimeargs.data(), args_size, l2ctrl, stream_ptr);
if (ret != RT_ERROR_NONE) {
MS_LOG(ERROR) << "Call runtime rtKernelLaunch error. Node info: " << node_info;
return false;
}
}
return true;

1
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/dynamic_tbe_kernel_mod.h
View File

@ -57,6 +57,7 @@ class DynamicTbeKernelMod : public TbeKernelMod {
void *tiling_data_ptr_ = nullptr;
uint64_t tiling_key_{0};
void *handle_ = nullptr;
void *func_stub_ = nullptr;
std::string origin_key_{""};
std::string op_compile_info_{};
bool need_skip_execute_ = false;

1
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_adapter.cc
View File

@ -73,6 +73,7 @@ std::unordered_set<std::string> TbeAdapter::input_order_adjusted_ops_ = {kConv2D
kLayerNormBetaGammaBackpropOpName,
kMinimumGradOpName,
kMaximumGradOpName,
kStridedSliceGradOpName,
kApplyCenteredRMSPropOpName};
bool TbeAdapter::IsSpecialFusionComputeNode(const std::vector<mindspore::AnfNodePtr> &compute_nodes) {

5
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_adapter.h
View File

@ -102,6 +102,11 @@ class TbeAdapter {
inputs_json->push_back(inputs_list[kIndex7]);
inputs_json->push_back(inputs_list[kIndex8]);
inputs_json->push_back(inputs_list[kIndex4]);
} else if (op_name == kStridedSliceGradOpName) {
for (size_t i = 1; i < inputs_list.size(); ++i) {
inputs_json->push_back(inputs_list[i]);
}
inputs_json->push_back(inputs_list[kIndex0]);
} else {
if (inputs_list.size() < kIndex2) {
MS_LOG(EXCEPTION) << "Op " << op_name << " should have at least " << kIndex2 << " inputs, but got "

5
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_json/tbe_json_creator.cc
View File

@ -522,6 +522,11 @@ void TbeJsonCreator::GenInputConstValue(const AnfNodePtr &anf_node, size_t real_
auto value = value_node->value();
MS_EXCEPTION_IF_NULL(value);
ParseConstValue(value, input_desc);
} else if (input_node->isa<Parameter>()) {
auto param = input_node->cast<ParameterPtr>();
auto value = param->default_param();
MS_EXCEPTION_IF_NULL(value);
ParseConstValue(value, input_desc);
} else {
MS_LOG(ERROR) << "The operator " << anf_node->fullname_with_scope() << "'s input" << real_input_index
<< "'s value depend is " << value_depend << ", but its input node is a " << input_node->type_name()

2
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_kernel_mod.cc
View File

@ -165,7 +165,7 @@ device::DynamicKernelPtr TbeKernelMod::GenDynamicKernel(const CNodePtr &cnode_pt
std::string origin_key;
void *handle = nullptr;
auto func_stub = KernelManager::GenFuncStub(*kernel_pack_, false, &block_dim, dynamic_flag, &handle, &origin_key);
if (dynamic_flag) {
if (kernel_json_info.has_kernel_list) {
if (func_stub != 1) {
MS_LOG(EXCEPTION) << "GenFuncStub failed.";
}

10
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_utils.cc
View File

@ -236,7 +236,7 @@ KernelPackPtr TbeUtils::InsertCache(const std::string &kernel_name, const std::s
}
int KernelManager::BinaryRegister(const mindspore::kernel::FlexArray &kernel_buffer, void **module, const string &magic,
const bool dynamic_flag) {
bool has_kernel_list) {
static std::map<string, uint32_t> magic_maps = {{"RT_DEV_BINARY_MAGIC_PLAIN", RT_DEV_BINARY_MAGIC_PLAIN},
{"RT_DEV_BINARY_MAGIC_PLAIN_AICPU", RT_DEV_BINARY_MAGIC_PLAIN_AICPU},
{"RT_DEV_BINARY_MAGIC_PLAIN_AIVEC", RT_DEV_BINARY_MAGIC_PLAIN_AIVEC},
@ -255,7 +255,7 @@ int KernelManager::BinaryRegister(const mindspore::kernel::FlexArray &kernel_buf
dev_bin.magic = iter->second;
dev_bin.length = kernel_buffer.len;
dev_bin.version = 0;
auto ret = dynamic_flag ? rtRegisterAllKernel(&dev_bin, module) : rtDevBinaryRegister(&dev_bin, module);
auto ret = has_kernel_list ? rtRegisterAllKernel(&dev_bin, module) : rtDevBinaryRegister(&dev_bin, module);
if (RT_ERROR_NONE != ret) {
MS_LOG(INFO) << "Call runtime rtDevBinaryRegister error.";
return -1;
@ -288,20 +288,20 @@ uintptr_t KernelManager::GenFuncStub(const mindspore::kernel::KernelPack &kernel
if (iter != info_table_.end()) {
auto kernelmeta = iter->second;
*block_dim = kernelmeta->block_dim_;
if (!dynamic_flag) {
if (!kernel_json_info.has_kernel_list) {
return kernelmeta->func_stub_;
}
}
}
void *module = nullptr;
if (BinaryRegister((*kernel_pack.GetKernel()), &module, magic, dynamic_flag) != 0) {
if (BinaryRegister((*kernel_pack.GetKernel()), &module, magic, kernel_json_info.has_kernel_list) != 0) {
MS_LOG(INFO) << "Call runtime BinaryRegister error.";
if (module != nullptr) {
(void)rtDevBinaryUnRegister(module);
}
return 0;
}
if (dynamic_flag) {
if (kernel_json_info.has_kernel_list) {
*handle = module;
*origin_key = func_name;
return 1;

3
mindspore/ccsrc/plugin/device/ascend/kernel/tbe/tbe_utils.h
View File

@ -104,8 +104,7 @@ class KernelManager {
private:
KernelManager() = default;
~KernelManager() = default;
static int BinaryRegister(const FlexArray &kernel_buffer, void **module, const string &magic,
const bool dynamic_flag);
static int BinaryRegister(const FlexArray &kernel_buffer, void **module, const string &magic, bool has_kernel_list);
static std::unordered_map<string, KernelMetaPtr> info_table_;
static std::atomic<uintptr_t> kernel_stub_gen_;
static std::mutex info_table_mutex_;

3
mindspore/ccsrc/plugin/device/ascend/optimizer/ascend_backend_optimization.cc
View File

@ -580,9 +580,6 @@ void AscendUnifyMindIR(const std::shared_ptr<session::KernelGraph> &graph) {
unify_mindir_pm->AddPass(std::make_shared<opt::MaxPool2MaxPoolWithArgmax>());
unify_mindir_pm->AddPass(std::make_shared<opt::MaxPoolWithArgmaxUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::MaxPoolGradWithArgmaxUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::Conv2DUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::Conv2DBackpropInputUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::Conv2DBackpropFilterUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::SliceGradUnifyMindIR>());
unify_mindir_pm->AddPass(std::make_shared<opt::StridedSliceGradUpdateInputNames>());
unify_mindir_pm->AddPass(std::make_shared<opt::AvgPoolGradUnifyMindIR>());

7
mindspore/ccsrc/transform/graph_ir/convert.cc
View File

@ -597,7 +597,12 @@ DfGraphConvertor &DfGraphConvertor::ConvertAllNode() {
MS_LOG(INFO) << "Dataset param is " << param.ToString() << ".";
// GetNext
auto iter_getnext_op = make_shared<ge::op::GetNext>("get_next_tmp");
(void)iter_getnext_op->set_attr_output_types(param.ge_types());
std::vector<enum ge::DataType> getnext_types;
const auto &origin_ge_types = param.ge_types();
std::transform(
origin_ge_types.begin(), origin_ge_types.end(), std::back_inserter(getnext_types),
[](int64_t t_num) -> enum ge::DataType { return static_cast<enum ge::DataType>(t_num); });
(void)iter_getnext_op->set_attr_output_types(getnext_types);
(void)iter_getnext_op->set_attr_output_shapes(param.shapes());
(void)iter_getnext_op->set_attr_channel_name(param.queue_name());

3
mindspore/ccsrc/utils/utils.h
View File

@ -527,6 +527,7 @@ constexpr auto kAttrPlaceHolderIndex = "placeholder_index";
constexpr auto kAttrMicro = "micro";
constexpr auto kAttrJsonFileName = "json_file_name";
constexpr auto kAttrNeedDropInput = "need_drop_input";
constexpr auto kAttrNeedConvertToValueNode = "need_convert_to_value_node";
constexpr auto kAttrSendSrcNodeName = "send_src_node_name";
constexpr auto kAttrSendDstNodeName = "send_dst_node_name";
constexpr auto kAttrSendDstRanks = "send_dst_ranks";
@ -773,6 +774,8 @@ const std::set<std::string> DynamicShapeConstInputToAttr = {kCastOpName, kE
kReduceMinOpName, kReduceMeanOpName, kReduceMaxOpName,
kReduceAllOpName, kReduceAnyOpName, kConcatOpName};
const std::set<std::string> NeedConvertToValueNodeSet = {kStridedSliceGradOpName};
const std::set<std::string> DynamicShapeConstInputToAttrCPU = {
kCastOpName, kExpandDimsOpName, kEmbeddingLookupOpName, kReduceMinOpName, kReduceMeanOpName,
kReduceMaxOpName, kReduceAllOpName, kReduceAnyOpName, kConcatOpName, kReduceSumOpName};

6
mindspore/python/mindspore/_extends/parallel_compile/tbe_compiler/tbe_adapter.py
View File

@ -403,7 +403,6 @@ def _pre_build_compute_op_info(compute_op, job):
_normalize_module_name(op_module_name, py_module_path)
unknown_shape = compute_op["unknown_shape"]
int64_mode = compute_op["int64mode"]
dynamic_compile_static = compute_op["dynamic_compile_static"]
res = check_op_impl_mode(op_module_name, op_func_name)
op_impl_mode = job.content["SocInfo"]["op_impl_mode"]
op_impl_mode_list = job.content["SocInfo"]["op_impl_mode_list"]
@ -416,7 +415,7 @@ def _pre_build_compute_op_info(compute_op, job):
options = get_options_info(job.content)
dispatch_prebuild_task(job.source_id, job.id, l1_size, op_module_name, op_full_name,
op_type, op_func_name, unknown_shape,
(inputs, outputs, attrs, options), int64_mode, dynamic_compile_static, unknown_shape,
(inputs, outputs, attrs, options), int64_mode, unknown_shape,
None, job.pass_list)
@ -477,13 +476,12 @@ def build_single_pre_op(job: TbeJob):
_normalize_module_name(op_module_name, py_module_path)
unknown_shape = compute_op_info["unknown_shape"]
int64_mode = compute_op_info["int64mode"]
dynamic_compile_static = compute_op_info["dynamic_compile_static"]
op_pattern = compute_op_info["pattern"]
options = get_options_info(job.content)
fuzz_build_info = get_fuzz_build_info(job.content)
dispatch_single_op_compile_task(job.source_id, job.id, l1_size, op_module_name, op_name, op_type, op_func_name,
op_kernel_name, unknown_shape, (inputs, outputs, attrs, options), int64_mode,
None, None, dynamic_compile_static, unknown_shape, op_pattern,
None, None, unknown_shape, op_pattern,
json.dumps(fuzz_build_info), None, job.pass_list)
return True

24
mindspore/python/mindspore/ops/_op_impl/tbe/strided_slice_grad_d.py
View File

@ -19,26 +19,28 @@ from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
strided_slice_grad_d_op_info = TBERegOp("StridedSliceGrad") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("strided_slice_grad_d.so") \
.binfile_name("strided_slice_grad.so") \
.compute_cost(10) \
.kernel_name("strided_slice_grad_d") \
.kernel_name("strided_slice_grad") \
.partial_flag(True) \
.attr("shapex", "required", "listInt", "all") \
.attr("begin", "required", "listInt", "all") \
.attr("end", "required", "listInt", "all") \
.attr("strides", "required", "listInt", "all") \
.dynamic_compile_static(True) \
.attr("begin_mask", "optional", "int", "all", "0") \
.attr("end_mask", "optional", "int", "all", "0") \
.attr("ellipsis_mask", "optional", "int", "all", "0") \
.attr("new_axis_mask", "optional", "int", "all", "0") \
.attr("shrink_axis_mask", "optional", "int", "all", "0") \
.input(0, "dy", False, "required", "all") \
.input(1, "shapex", False, "required", "all", "optional") \
.input(2, "begin", False, "required", "all", "optional") \
.input(3, "end", False, "required", "all", "optional") \
.input(4, "strides", False, "required", "all", "optional") \
.output(0, "output", False, "required", "all") \
.dtype_format(DataType.I8_Default, DataType.I8_Default) \
.dtype_format(DataType.U8_Default, DataType.U8_Default) \
.dtype_format(DataType.I32_Default, DataType.I32_Default) \
.dtype_format(DataType.F16_Default, DataType.F16_Default) \
.dtype_format(DataType.F32_Default, DataType.F32_Default) \
.dtype_format(DataType.I32_Default, DataType.I32_Default, DataType.I32_Default, DataType.I32_Default,
DataType.I32_Default, DataType.I32_Default) \
.dtype_format(DataType.F16_Default, DataType.I32_Default, DataType.I32_Default, DataType.I32_Default,
DataType.I32_Default, DataType.F16_Default) \
.dtype_format(DataType.F32_Default, DataType.I32_Default, DataType.I32_Default, DataType.I32_Default,
DataType.I32_Default, DataType.F32_Default) \
.get_op_info()

2
mindspore/python/mindspore/run_check/_check_version.py
View File

@ -206,7 +206,7 @@ class AscendEnvChecker(EnvChecker):
"""ascend environment check"""
def __init__(self):
self.version = ["1.80"]
self.version = ["1.81"]
atlas_nnae_version = "/usr/local/Ascend/nnae/latest/fwkacllib/version.info"
atlas_toolkit_version = "/usr/local/Ascend/ascend-toolkit/latest/fwkacllib/version.info"
hisi_fwk_version = "/usr/local/Ascend/fwkacllib/version.info"

77
tests/ut/cpp/pre_activate/pass/const_to_attr_strided_slice_grad_test.cc
View File

@ -1,77 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed 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.
*/
#include "common/backend_common_test.h"
#include "frontend/operator/ops.h"
#include "debug/anf_ir_dump.h"
#include "common/py_func_graph_fetcher.h"
#include "backend/common/session/anf_runtime_algorithm.h"
#include "backend/common/optimizer/optimizer.h"
#include "backend/common/optimizer/pass_manager.h"
#include "backend/common/pass/const_to_attr_strided_slice_grad.h"
#include "utils/utils.h"
#include "utils/ms_utils.h"
namespace mindspore {
namespace opt {
class TestHWConstToAttrStridedSliceGrad : public BackendCommon {
public:
TestHWConstToAttrStridedSliceGrad() : getPyFun_("gtest_input.pre_activate.const_to_attr_strided_slice_grad", true) {}
~TestHWConstToAttrStridedSliceGrad() override = default;
public:
UT::PyFuncGraphFetcher getPyFun_;
};
TEST_F(TestHWConstToAttrStridedSliceGrad, test_strided_slice_grad) {
FuncGraphPtr g = getPyFun_.CallAndParseRet("test_const_to_attr_strided_slice_grad", "before");
ASSERT_TRUE(g != nullptr);
FuncGraphPtr g_after = getPyFun_.CallAndParseRet("test_const_to_attr_strided_slice_grad", "after");
ASSERT_TRUE(g_after != nullptr);
auto ret = g->get_return();
ASSERT_TRUE(ret != nullptr);
EXPECT_NE(ret->input(1), nullptr);
EXPECT_NE(ret->input(1)->cast<CNodePtr>(), nullptr);
auto cnode = ret->input(1)->cast<CNodePtr>();
EXPECT_FALSE(AnfAlgo::HasNodeAttr("shapex", cnode));
EXPECT_FALSE(AnfAlgo::HasNodeAttr("begin", cnode));
EXPECT_FALSE(AnfAlgo::HasNodeAttr("end", cnode));
EXPECT_FALSE(AnfAlgo::HasNodeAttr("strides", cnode));
EXPECT_FALSE(CheckEqualGraph(g, g_after));
std::vector<int64_t> shp_x{16, 1, 1024};
auto x_abstract = std::make_shared<abstract::AbstractTensor>(kFloat32, shp_x);
AbstractBasePtrList args_spec_list{x_abstract};
auto kg = GetKernelGraph(g, args_spec_list);
ASSERT_TRUE(kg != nullptr);
ret = kg->get_return();
ASSERT_TRUE(ret != nullptr);
EXPECT_NE(ret->input(1), nullptr);
EXPECT_NE(ret->input(1)->cast<CNodePtr>(), nullptr);
auto make_tuple = ret->input(1)->cast<CNodePtr>();
ASSERT_TRUE(make_tuple != nullptr);
EXPECT_NE(make_tuple->input(1), nullptr);
EXPECT_NE(make_tuple->input(1)->cast<CNodePtr>(), nullptr);
cnode = make_tuple->input(1)->cast<CNodePtr>();
EXPECT_TRUE(AnfAlgo::HasNodeAttr("shapex", cnode));
EXPECT_TRUE(AnfAlgo::HasNodeAttr("begin", cnode));
EXPECT_TRUE(AnfAlgo::HasNodeAttr("end", cnode));
EXPECT_TRUE(AnfAlgo::HasNodeAttr("strides", cnode));
EXPECT_TRUE(CheckEqualGraph(kg, g_after));
}
} // namespace opt
} // namespace mindspore

2
tests/ut/cpp/stub/runtime/runtime_stub.cc
View File

@ -185,6 +185,8 @@ RTS_API rtError_t rtModelGetId(rtModel_t model, uint32_t *modelId) { return RT_E
RTS_API rtError_t rtLabelCreateEx(rtLabel_t *label, rtStream_t stream) { return RT_ERROR_NONE; }
RTS_API rtError_t rtLabelCreateExV2(rtLabel_t *lbl, rtModel_t mdl, rtStream_t stm) { return RT_ERROR_NONE; }
RTS_API rtError_t rtCpuKernelLaunchWithFlag(const void *soName, const void *kernelName, uint32_t blockDim,
const void *args, uint32_t argsSize, rtSmDesc_t *smDesc, rtStream_t stream,
uint32_t flags) {

39
tests/ut/cpp/stub/tdt/tdt_mock.cc
View File

@ -1,39 +0,0 @@
/**
* Copyright 2020 Huawei Technologies Co., Ltd
*
* Licensed 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.
*/
#ifndef TDT_MOCK_H
#define TDT_MOCK_H
#include "tdt/tsd_client.h"
namespace tdt {
StatusFactory* StatusFactory::GetInstance() {
static StatusFactory instance;
return &instance;
}
void StatusFactory::RegisterErrorNo(const uint32_t err, const std::string& desc) { return; }
std::string StatusFactory::GetErrDesc(const uint32_t err) { return "Error"; }
std::string StatusFactory::GetErrCodeDesc(uint32_t errCode) { return "Error"; }
StatusFactory::StatusFactory() {}
std::mutex& StatusFactory::GetMutex() { return GetInstance()->rwMutex_; }
} // namespace tdt
#endif // TDT_MOCK_H

2
version.txt
View File

@ -1 +1 @@
1.6.0
1.7.0