Merge remote-tracking branch 'upstream/develop' into cinn-pass-ir_sim…

…plify

CONTRIBUTING.md

@@ -39,7 +39,7 @@ PaddlePaddle uses this [Git branching model](http://nvie.com/posts/a-successful-
    pre-commit install
    ```
 
-   Our pre-commit configuration requires clang-format 3.8 for auto-formating C/C++ code and yapf for Python.
+   Our pre-commit configuration requires clang-format 3.8 for auto-formatting C/C++ code and yapf for Python.
 
    Once installed, `pre-commit` checks the style of code and documentation in every commit.  We will see something like the following when you run `git commit`:
 
@@ -52,7 +52,7 @@ PaddlePaddle uses this [Git branching model](http://nvie.com/posts/a-successful-
    Check for broken symlinks................................................Passed
    Detect Private Key...................................(no files to check)Skipped
    Fix End of Files.....................................(no files to check)Skipped
-   clang-formater.......................................(no files to check)Skipped
+   clang-format.........................................(no files to check)Skipped
    [my-cool-stuff c703c041] add test file
     1 file changed, 0 insertions(+), 0 deletions(-)
     create mode 100644 233

_typos.toml

@@ -7,11 +7,22 @@ extend-exclude = [
     # Skip `intermidiate` check in these files
     "test/cpp/eager/task_tests/CMakeLists.txt",
     "test/cpp/eager/task_tests/hook_test_intermidiate.cc",
+    # Skip `creater` check in these files
+    "paddle/fluid/inference/tensorrt/convert/CMakeLists.txt",
+    "paddle/fluid/inference/tensorrt/convert/generic_and_custom_plugin_creater.cc",
+    "paddle/fluid/inference/tensorrt/convert/test_custom_plugin_creater.cc",
 ]
 
 [default]
 # Ignore 1-3 letter words, refer to https://github.com/crate-ci/typos/issues/1079
 extend-ignore-words-re = ["^[a-zA-Z]{1,3}$"]
+# refer to https://github.com/crate-ci/typos/blob/master/docs/reference.md#example-configurations
+extend-ignore-re = [
+    # Ignore lines by `# typos: disable-line`
+    "(?Rm)^.*(#|//)\\s*typos:\\s*disable-line$",
+    # Ignore block by `# typos: off` and `# typos: on`
+    "(?s)(#|//)\\s*typos:\\s*off.*?\\n\\s*(#|//)\\s*typos:\\s*on"
+]
 
 [default.extend-words]
 # PaddlePaddle specific words
@@ -34,54 +45,11 @@ pash = 'pash'
 unpacket = "unpacket"
 
 # These words need to be fixed
-Creater = 'Creater'
-creater = 'creater'
-dito = 'dito'
-devide = 'devide'
-documention = 'documention'
-doens = 'doens'
-doen = 'doen'
-doubel = 'doubel'
-dobule = 'dobule'
-Dowloading = 'Dowloading'
-downsteram = 'downsteram'
 fetchs = 'fetchs'
-Flattend = 'Flattend'
-flattend = 'flattend'
-flattern = 'flattern'
-Flattern = 'Flattern'
-filpped = 'filpped'
-flaot = 'flaot'
-follwed = 'follwed'
-folowing = 'folowing'
-formater = 'formater'
-formating = 'formating'
-foramt = 'foramt'
-formate = 'formate'
-formt = 'formt'
-forwrad = 'forwrad'
-forwad = 'forwad'
-forword = 'forword'
-founf = 'founf'
-framwork = 'framwork'
-frequence = 'frequence'
-fron = 'fron'
-fullfill = 'fullfill'
 Indexs = 'Indexs'
 indexs = 'indexs'
-indiates = 'indiates'
-indeces = 'indeces'
-inferrence = 'inferrence'
 Infered = 'Infered'
 infered = 'infered'
-infering = 'infering'
-informations = 'informations'
-imformation = 'imformation'
-infomation = 'infomation'
-Infor = 'Infor'
-infor = 'infor'
-inheritted = 'inheritted'
-initilization = 'initilization'
 initilized = 'initilized'
 initalized = 'initalized'
 initalize = 'initalize'
@@ -190,25 +158,6 @@ protocal = 'protocal'
 PROTOCAL = 'PROTOCAL'
 pyrhon = 'pyrhon'
 pthon = 'pthon'
-qucik = 'qucik'
-quitted = 'quitted'
-runned = 'runned'
-readed = 'readed'
-recived = 'recived'
-recevied = 'recevied'
-reveived = 'reveived'
-recieved = 'recieved'
-recieves = 'recieves'
-recive = 'recive'
-receving = 'receving'
-recommand = 'recommand'
-recomplie = 'recomplie'
-recored = 'recored'
-Recusive = 'Recusive'
-recusive = 'recusive'
-recursivly = 'recursivly'
-Recursivly = 'Recursivly'
-reduntant = 'reduntant'
 Refered = 'Refered'
 refered = 'refered'
 registed = 'registed'
@@ -322,14 +271,6 @@ suppport = 'suppport'
 SWTICH = 'SWTICH'
 Swith = 'Swith'
 sysyem = 'sysyem'
-targt = 'targt'
-Taget = 'Taget'
-templat = 'templat'
-temporaily = 'temporaily'
-temporily = 'temporily'
-Temperarily = 'Temperarily'
-temporaly = 'temporaly'
-Temperary = 'Temperary'
 tenosr = 'tenosr'
 iterm = 'iterm'
 termiante = 'termiante'

cmake/external/openvino.cmake

@@ -103,7 +103,7 @@ file(TO_NATIVE_PATH ${PADDLE_SOURCE_DIR}/patches/openvino/convert.patch
      native_convert)
 
 set(OPENVINO_PATCH_COMMAND
-    git checkout -- . && git fetch --depth=1 origin <OPENVINO_COMMIT> && git
+    git checkout -- . && git fetch --depth=1 origin ${OPENVINO_COMMIT} && git
     checkout ${OPENVINO_COMMIT} && patch -Np1 -d ${SOURCE_DIR} <
     ${native_convert} || true)
 

cmake/simd.cmake

@@ -13,6 +13,11 @@ if(CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
   set(AVX512F_FLAG "-mavx512f")
   set(Wno_Maybe_Uninitialized "-Wno-maybe-uninitialized")
   set(FMA_FLAG "-mfma")
+  if(${CMAKE_CXX_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0)
+    set(NO_INLINE "-fno-inline")
+  else()
+    set(NO_INLINE "")
+  endif()
 elseif(MSVC)
   set(MMX_FLAG "/arch:MMX")
   set(SSE2_FLAG "/arch:SSE2")
@@ -22,6 +27,11 @@ elseif(MSVC)
   set(AVX512F_FLAG "/arch:AVX512")
   set(Wno_Maybe_Uninitialized "/wd4701")
   set(FMA_FLAG "/arch:AVX2")
+  if(${CMAKE_CXX_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0)
+    set(NO_INLINE "/Ob0")
+  else()
+    set(NO_INLINE "")
+  endif()
 endif()
 
 set(CMAKE_REQUIRED_FLAGS_RETAINED ${CMAKE_REQUIRED_FLAGS})

paddle/cinn/backends/codegen_gpu_dev.cc

@@ -115,7 +115,7 @@ std::vector<Expr> FilterDeallocTempBuffers(const std::vector<Expr> &frees) {
     bool has_symbolic_constant = false;
     const ir::_Buffer_ *buffer = op->destination.As<ir::_Buffer_>();
     for (Expr shape : buffer->shape) {
-      shape = common::AutoSimplify(shape);
+      shape = optim::ArithSimplify(shape);
       ir::ir_utils::CollectIRNodes(shape, [&](const Expr *x) {
         if (x->as_var()) {
           PADDLE_ENFORCE_EQ(
@@ -539,7 +539,7 @@ ir::Expr CalculateSharedMemory(const ir::LoweredFunc &func) {
       shm_size = shm_size + CalculateSharedMemory(buffer);
     }
   }
-  return common::AutoSimplify(shm_size);
+  return optim::ArithSimplify(shm_size);
 }
 
 }  // namespace backends

paddle/cinn/backends/compiler.cc

@@ -430,7 +430,6 @@ void Compiler::CompileCudaModule(const Module& module,
     device_fn_name_.emplace_back(kernel_fn_name);
   }
   engine_->Link<CodeGenGpuHost>(host_module);
-
 #else
   CINN_NOT_IMPLEMENTED
 #endif

paddle/cinn/backends/llvm/codegen_llvm.cc

@@ -928,7 +928,7 @@ llvm::Value *CodeGenLLVM::Visit(const ir::Store *op) {
       // fit the total_lanes in native_lanes(split into multiple native steps)
       for (int offset = 0; offset < total_lanes; offset += total_lanes) {
         int lanes = total_lanes;
-        Expr base = cinn::common::AutoSimplify(ramp->base + offset);
+        Expr base = optim::ArithSimplify(ramp->base + offset);
         optim::VarModSimplify(&base);
         auto *ptr =
             CreateBufferPtr(op->type().ElementOf(), buffer, Visit(&base));
@@ -1242,10 +1242,8 @@ llvm::Value *CodeGenLLVM::DenseVectorLoad(const ir::Load *op) {
 
   for (int i = 0; i < load_lanes; i += load_lanes) {
     int slice_lanes = load_lanes;
-    auto slice_base = cinn::common::AutoSimplify(ramp->base + i);
+    auto slice_base = optim::ArithSimplify(ramp->base + i);
     optim::VarModSimplify(&slice_base);
-    auto slide_stride = Expr(1);
-    auto slide_index = slice_base;
 
 #if LLVM_VERSION_MAJOR >= 11
     const llvm::ElementCount elem_count(slice_lanes, /*scalable*/ false);

paddle/cinn/backends/llvm/execution_engine.cc

@@ -171,8 +171,10 @@ std::unique_ptr<llvm::MemoryBuffer> NaiveObjectCache::getObject(
 
 template <typename CodeGenT>
 void ExecutionEngine::Link(const ir::Module &module) {
+  if (module.functions().size() == 0) {
+    return;
+  }
   utils::RecordEvent("ExecutionEngine Link", utils::EventType::kOrdinary);
-
   auto ir_emitter = std::make_unique<CodeGenT>(m.get(), b.get());
   VLOG(3) << "ir_emitter->Compile(module) Begin";
   ir_emitter->Compile(module);
@@ -211,6 +213,16 @@ void ExecutionEngine::Link(const ir::Module &module) {
   }
 }
 
+template <>
+void ExecutionEngine::Link<CodeGenGpuHost>(const ir::Module &module) {
+  if (module.functions().size() == 0) {
+    return;
+  }
+  utils::RecordEvent("ExecutionEngine Link", utils::EventType::kOrdinary);
+  auto ir_emitter = std::make_unique<CodeGenGpuHost>(m.get(), b.get());
+  ir_emitter->Compile(module);
+}
+
 bool ExecutionEngine::AddModule(std::unique_ptr<llvm::Module> module,
                                 std::unique_ptr<llvm::LLVMContext> context) {
   utils::RecordEvent("ExecutionEngine AddModule", utils::EventType::kOrdinary);

paddle/cinn/common/ir_util.cc

@@ -101,8 +101,8 @@ Expr RampRelatedAdd(ir::Ramp *ramp, ir::Ramp *other) {
                           ::common::errors::InvalidArgument(
                               "Other ramp pointer should not be null."));
   if (ramp->lanes == other->lanes) {
-    Expr base_add = cinn::common::AutoSimplify(ramp->base + other->base);
-    Expr stride_add = cinn::common::AutoSimplify(ramp->stride + other->stride);
+    Expr base_add = optim::ArithSimplify(ramp->base + other->base);
+    Expr stride_add = optim::ArithSimplify(ramp->stride + other->stride);
     VLOG(2) << base_add;
     VLOG(2) << stride_add;
     return ir::Ramp::Make(base_add, stride_add, ramp->lanes);
@@ -641,8 +641,7 @@ ir::IndexExpr SimplifySymbolicDivide(const ir::IndexExpr &lhs,
 
 bool ProveDivisible(const ir::IndexExpr &lhs, const ir::IndexExpr &rhs) {
   if (IsZero(lhs % rhs)) return true;
-  // remove AutoSimplify later.
-  if (IsZero(AutoSimplify(lhs % rhs))) return true;
+  if (IsZero(optim::ArithSimplify(lhs % rhs))) return true;
   return false;
 }
 

paddle/cinn/common/ir_util.h

@@ -191,7 +191,7 @@ inline void UnpackReduction(const ir::IndexExpr &expr, FLeaf fleaf) {
 }
 
 /*!
- * \brief Flattern the expression into a vector of expressions splited by `Add`
+ * \brief Flatten the expression into a vector of expressions splited by `Add`
  * or `Mul`.
  *
  * For example (Add):

paddle/cinn/hlir/dialect/operator/ir/generate_shape_util.cc

@@ -666,7 +666,7 @@ bool MakeGenerateShapeOpAttribute(
       }
     }
     if (!has_symbol_binding) {
-      LOG(WARNING) << "no symbol binding found for dim expr: " << symbol_name;
+      VLOG(2) << "no symbol binding found for dim expr: " << symbol_name;
       return false;
     }
   }

paddle/cinn/hlir/dialect/operator/ir/ops.yaml

@@ -73,7 +73,7 @@
   output : Tensor(out)
   infer_meta :
     func : ReduceSumInferMeta
-    param : [x, axis, keepdim]
+    param : [x, axis, keepdim, dtype]
   kernel :
     func : frobenius_norm
     param : [x, axis, keepdim]

paddle/cinn/hlir/dialect/operator/transforms/check_infer_symbolic_util.cc

@@ -54,7 +54,7 @@ std::ostream& operator<<(std::ostream& stream,
 }
 
 DimExprs4ValueT MakeDimExprs4Value(
-    pir::Program* program, const PassManagerCreater& CreatePassManager) {
+    pir::Program* program, const PassManagerCreator& CreatePassManager) {
   std::shared_ptr<pir::PassManager> pass_manager = CreatePassManager();
   pass_manager->AddPass(pir::CreateShapeOptimizationPass());
   pass_manager->Run(program);
@@ -404,13 +404,13 @@ struct ShapeSignatureGenerator {
                        const DoEachT& DoEach) {
     if (set_size <= 0) return DoEach(is_subset_flags);
 
-    const auto& RecusiveVisit = [&](bool is_subset) {
+    const auto& RecursiveVisit = [&](bool is_subset) {
       std::vector<IsSubset> current_is_subset_flags(is_subset_flags);
       current_is_subset_flags.push_back(static_cast<int>(is_subset));
       VisitEachSubSet(set_size - 1, current_is_subset_flags, DoEach);
     };
-    RecusiveVisit(true);
-    RecusiveVisit(false);
+    RecursiveVisit(true);
+    RecursiveVisit(false);
   }
 
   std::optional<ConstrainedSymbolNamesList> GetConstrainedSymbolNamesList(
@@ -623,7 +623,7 @@ void CheckProgramDimExprConstraints(
 }  // namespace
 
 void CheckInferSymbolicIfNeed(pir::Program* program,
-                              const PassManagerCreater& CreatePassManager) {
+                              const PassManagerCreator& CreatePassManager) {
   if (!FLAGS_prim_all || !FLAGS_check_infer_symbolic) return;
   const auto& GraphDimExprs4Value =
       MakeDimExprs4Value(program, CreatePassManager);

paddle/cinn/hlir/dialect/operator/transforms/check_infer_symbolic_util.h

@@ -24,10 +24,10 @@ namespace cinn {
 namespace dialect {
 namespace ir {
 
-using PassManagerCreater = std::function<std::shared_ptr<pir::PassManager>()>;
+using PassManagerCreator = std::function<std::shared_ptr<pir::PassManager>()>;
 
 void CheckInferSymbolicIfNeed(pir::Program* program,
-                              const PassManagerCreater& CreatePassManager);
+                              const PassManagerCreator& CreatePassManager);
 
 }  // namespace ir
 }  // namespace dialect

paddle/cinn/hlir/dialect/operator/transforms/fuse_shape_ops_into_generate_shape_op_pass.cc

@@ -76,7 +76,7 @@ std::vector<pir::Value> FindSourceDenseTensorOfDimTensor(
       [](const symbol::NullShapeOrDataDimExpr& null_shape_or_data) {
         return false;
       }};
-  // For TensorListShapeOrDataDimExprs case, we should recursivly visit its
+  // For TensorListShapeOrDataDimExprs case, we should recursively visit its
   // each dim_expr, which is automatically in next step.
   const auto& NeedTrackUpstream = [&](pir::Value value) -> bool {
     const auto& sym_shape = ShapeOrDataDimExprs4Value(value);

paddle/cinn/hlir/dialect/operator/transforms/lowering_pass/utils.cc

@@ -117,7 +117,7 @@ OpLoweringGroupPtr BuildOpLoweringGroup(pir::Operation* fusion_op_ptr) {
   if (FLAGS_cinn_enable_map_expr) {
     cinn::adt::TryGenerateMapExprFromGroup(group);
   }
-  // Rebuild other informations
+  // Rebuild other information
   // TODO(zhangyuqin1998): Do we need group.master_ops?
   return group;
 }