PaddlePaddle · qili93 · May 22, 2024 · May 20, 2024 · Apr 19, 2024 · May 11, 2024
diff --git a/.gitmodules b/.gitmodules
@@ -110,3 +110,11 @@
 	path = third_party/cccl
 	url = https://github.com/NVIDIA/cccl.git
 	ignore = dirty
+[submodule "third_party/cryptopp"]
+	path = third_party/cryptopp
+	url = https://github.com/weidai11/cryptopp.git
+	ignore = dirty
+[submodule "third_party/cryptopp-cmake"]
+	path = third_party/cryptopp-cmake
+	url = https://github.com/noloader/cryptopp-cmake.git
+	ignore = dirty
diff --git a/README.md b/README.md
@@ -20,7 +20,7 @@ PaddlePaddle is originated from industrial practices with dedication and commitm
 
 ## Installation
 
-### Latest PaddlePaddle Release: [v2.5](https://github.com/PaddlePaddle/Paddle/tree/release/2.5)
+### Latest PaddlePaddle Release: [v2.6](https://github.com/PaddlePaddle/Paddle/tree/release/2.6)
 
 Our vision is to enable deep learning for everyone via PaddlePaddle.
 Please refer to our [release announcement](https://github.com/PaddlePaddle/Paddle/releases) to track the latest features of PaddlePaddle.

diff --git a/README_cn.md b/README_cn.md
@@ -18,9 +18,9 @@
 
 ## 安装
 
-### PaddlePaddle最新版本: [v2.5](https://github.com/PaddlePaddle/Paddle/tree/release/2.5)
+### PaddlePaddle 最新版本: [v2.6](https://github.com/PaddlePaddle/Paddle/tree/release/2.6)
 
-跟进PaddlePaddle最新特性请参考我们的[版本说明](https://github.com/PaddlePaddle/Paddle/releases)
+跟进 PaddlePaddle 最新特性请参考我们的[版本说明](https://github.com/PaddlePaddle/Paddle/releases)
 
 ### 安装最新稳定版本:
 ```

diff --git a/README_ja.md b/README_ja.md
@@ -20,7 +20,7 @@ PaddlePaddle は、工業化に対するコミットメントを持つ工業的
 
 ## インストール
 
-### PaddlePaddle の最新リリース: [v2.5](https://github.com/PaddlePaddle/Paddle/tree/release/2.5)
+### PaddlePaddle の最新リリース: [v2.6](https://github.com/PaddlePaddle/Paddle/tree/release/2.6)
 
 私たちのビジョンは、PaddlePaddle を通じて、誰もが深層学習を行えるようにすることです。
 PaddlePaddle の最新機能を追跡するために、私たちの[リリースのお知らせ](https://github.com/PaddlePaddle/Paddle/releases)を参照してください。

diff --git a/cmake/external/cryptopp.cmake b/cmake/external/cryptopp.cmake
@@ -14,12 +14,13 @@
 
 include(ExternalProject)
 
+set(CRYPTOPP_SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/cryptopp)
+set(CRYPTOPP_CMAKE_SOURCE_DIR ${PADDLE_SOURCE_DIR}/third_party/cryptopp-cmake)
 set(CRYPTOPP_PREFIX_DIR ${THIRD_PARTY_PATH}/cryptopp)
 set(CRYPTOPP_INSTALL_DIR ${THIRD_PARTY_PATH}/install/cryptopp)
 set(CRYPTOPP_INCLUDE_DIR
     "${CRYPTOPP_INSTALL_DIR}/include"
     CACHE PATH "cryptopp include directory." FORCE)
-set(CRYPTOPP_REPOSITORY ${GIT_URL}/weidai11/cryptopp.git)
 set(CRYPTOPP_TAG CRYPTOPP_8_2_0)
 
 if(WIN32)
@@ -63,17 +64,16 @@ include_directories(${CRYPTOPP_INCLUDE_DIR})
 ExternalProject_Add(
   extern_cryptopp
   ${EXTERNAL_PROJECT_LOG_ARGS} ${SHALLOW_CLONE}
-  GIT_REPOSITORY ${CRYPTOPP_REPOSITORY}
-  GIT_TAG ${CRYPTOPP_TAG}
   PREFIX ${CRYPTOPP_PREFIX_DIR}
+  SOURCE_DIR ${CRYPTOPP_SOURCE_DIR}
   UPDATE_COMMAND ""
   PATCH_COMMAND
-  COMMAND ${CMAKE_COMMAND} -E remove_directory "<SOURCE_DIR>/cmake/"
-  COMMAND git clone ${GIT_URL}/noloader/cryptopp-cmake "<SOURCE_DIR>/cmake"
-  COMMAND cd "<SOURCE_DIR>/cmake" && git checkout tags/${CRYPTOPP_TAG} -b
-          ${CRYPTOPP_TAG}
-  COMMAND ${CMAKE_COMMAND} -E copy_directory "<SOURCE_DIR>/cmake/"
-          "<SOURCE_DIR>/"
+  COMMAND ${CMAKE_COMMAND} -E copy "${CRYPTOPP_CMAKE_SOURCE_DIR}/CMakeLists.txt"
+          "<SOURCE_DIR>/CMakeLists.txt"
+  COMMAND
+    ${CMAKE_COMMAND} -E copy
+    "${CRYPTOPP_CMAKE_SOURCE_DIR}/cryptopp-config.cmake"
+    "<SOURCE_DIR>/cryptopp-config.cmake"
   COMMAND ${CRYPTOPP_PATCH_COMMAND}
   INSTALL_DIR ${CRYPTOPP_INSTALL_DIR}
   CMAKE_ARGS ${CRYPTOPP_CMAKE_ARGS}

diff --git a/cmake/generic.cmake b/cmake/generic.cmake
@@ -882,12 +882,6 @@ function(hip_library TARGET_NAME)
     cmake_parse_arguments(hip_library "${options}" "${oneValueArgs}"
                           "${multiValueArgs}" ${ARGN})
     if(hip_library_SRCS)
-      # FindHIP.cmake defined hip_add_library, HIP_SOURCE_PROPERTY_FORMAT is requried if no .cu files found
-      if(NOT (${CMAKE_CURRENT_SOURCE_DIR} MATCHES ".*/operators"
-              OR ${CMAKE_CURRENT_SOURCE_DIR} MATCHES ".*/phi/kernels"))
-        set_source_files_properties(${hip_library_SRCS}
-                                    PROPERTIES HIP_SOURCE_PROPERTY_FORMAT 1)
-      endif()
       if(hip_library_SHARED OR hip_library_shared) # build *.so
         hip_add_library(${TARGET_NAME} SHARED ${hip_library_SRCS})
       else()
@@ -901,6 +895,10 @@ function(hip_library TARGET_NAME)
       endif()
       # cpplint code style
       foreach(source_file ${hip_library_SRCS})
+        if(NOT ${source_file} MATCHES "\\.cu$")
+          set_source_files_properties(${source_file}
+                                      PROPERTIES HIP_SOURCE_PROPERTY_FORMAT 1)
+        endif()
         string(REGEX REPLACE "\\.[^.]*$" "" source ${source_file})
         if(EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/${source}.h)
           list(APPEND hip_library_HEADERS

diff --git a/cmake/inference_lib.cmake b/cmake/inference_lib.cmake
@@ -237,6 +237,16 @@ copy_part_of_thrid_party(inference_lib_dist ${PADDLE_INFERENCE_INSTALL_DIR})
 
 set(src_dir "${PADDLE_SOURCE_DIR}/paddle/fluid")
 
+if(WIN32)
+  set(paddle_common_lib ${PADDLE_BINARY_DIR}/paddle/common/common.*)
+else()
+  set(paddle_common_lib ${PADDLE_BINARY_DIR}/paddle/common/libcommon.*)
+endif()
+copy(
+  inference_lib_dist
+  SRCS ${paddle_common_lib}
+  DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib)
+
 if(WIN32)
   if(WITH_STATIC_LIB)
     set(paddle_inference_lib
@@ -268,11 +278,6 @@ else()
       SRCS ${paddle_phi_lib}
       DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib)
   endif()
-  set(paddle_common_lib ${PADDLE_BINARY_DIR}/paddle/common/libcommon.*)
-  copy(
-    inference_lib_dist
-    SRCS ${paddle_common_lib}
-    DSTS ${PADDLE_INFERENCE_INSTALL_DIR}/paddle/lib)
 endif()
 
 copy(

diff --git a/paddle/cinn/ir/ir_base.h b/paddle/cinn/ir/ir_base.h
@@ -110,16 +110,23 @@ class Dim;
   macro__(Product)                          \
   macro__(Sum)                              \
   macro__(PrimitiveNode)                    \
-  macro__(IntrinsicOp)                      \
   macro__(_BufferRange_)                    \
   macro__(ScheduleBlock)                    \
   macro__(ScheduleBlockRealize)             \
   macro__(_Dim_)                            \
 
+#define NODETY_CONTROL_OP_FOR_INTRINSIC(macro__) \
+  macro__(IntrinsicOp)                      \
 
 #define NODETY_FORALL(__m)              \
   NODETY_PRIMITIVE_TYPE_FOR_EACH(__m)   \
   NODETY_OP_FOR_EACH(__m)               \
+  NODETY_CONTROL_OP_FOR_INTRINSIC(__m)  \
+  NODETY_CONTROL_OP_FOR_EACH(__m)
+
+#define NODETY_FORALL_EXCEPT_INTRINSIC(__m)              \
+  NODETY_PRIMITIVE_TYPE_FOR_EACH(__m)                    \
+  NODETY_OP_FOR_EACH(__m)                                \
   NODETY_CONTROL_OP_FOR_EACH(__m)
 // clang-format on
 

diff --git a/paddle/cinn/ir/utils/ir_nodes_collector.cc b/paddle/cinn/ir/utils/ir_nodes_collector.cc
@@ -15,6 +15,8 @@
 #include "paddle/cinn/ir/utils/ir_nodes_collector.h"
 #include <glog/logging.h>
 
+#include "paddle/cinn/ir/intrinsic_ops.h"
+#include "paddle/cinn/ir/ir.h"
 #include "paddle/cinn/ir/ir_mutator.h"
 #include "paddle/cinn/ir/ir_printer.h"
 
@@ -71,8 +73,71 @@ struct IrNodesCollector : public IRVisitorRequireReImpl<void> {
     }                                  \
   }
 
-  NODETY_FORALL(__m)
+  NODETY_FORALL_EXCEPT_INTRINSIC(__m)
 #undef __m
+
+  void Visit(const ir::IntrinsicOp* op) {
+    switch (op->getKind()) {
+#define __(x)                                     \
+  case ir::IntrinsicKind::k##x:                   \
+    Visit(llvm::dyn_cast<ir::intrinsics::x>(op)); \
+    break;
+
+      INTRINSIC_KIND_FOR_EACH(__)
+#undef __
+    }
+  }
+
+  void Visit(const ir::intrinsics::GetAddr* x) {
+    if (x->data.defined()) {
+      Visit(&(x->data));
+    }
+  }
+
+  void Visit(const ir::intrinsics::BufferGetDataHandle* x) {
+    if (x->buffer.defined()) {
+      Visit(&(x->buffer));
+    }
+  }
+
+  void Visit(const ir::intrinsics::BufferGetDataConstHandle* x) {
+    if (x->buffer.defined()) {
+      Visit(&(x->buffer));
+    }
+  }
+
+  void Visit(const ir::intrinsics::PodValueToX* x) {
+    if (x->pod_value_ptr.defined()) {
+      Visit(&(x->pod_value_ptr));
+    }
+  }
+
+  void Visit(const ir::intrinsics::BufferCreate* x) {
+    if (x->buffer.defined()) {
+      Visit(&(x->buffer));
+    }
+  }
+
+  void Visit(const ir::intrinsics::ArgsConstruct* x) {
+    if (x->var.defined()) {
+      Expr convert = Expr(x->var);
+      Visit(&convert);
+    }
+    for (int i = 0; i < x->args.size(); ++i) {
+      if (x->args[i].defined()) {
+        Visit(&(x->args[i]));
+      }
+    }
+  }
+
+  void Visit(const ir::intrinsics::BuiltinIntrin* x) {
+    for (int i = 0; i < x->args.size(); ++i) {
+      if (x->args[i].defined()) {
+        Visit(&(x->args[i]));
+      }
+    }
+  }
+
   std::set<void*> visited_;
 };
 

diff --git a/paddle/fluid/distributed/common/chunk_allocator.h b/paddle/fluid/distributed/common/chunk_allocator.h
@@ -14,6 +14,7 @@
 
 #pragma once
 #include <glog/logging.h>
+#include "paddle/fluid/platform/enforce.h"
 
 namespace paddle {
 namespace distributed {
@@ -77,9 +78,16 @@ class ChunkAllocator {
 
   void create_new_chunk() {
     Chunk* chunk;
-    posix_memalign(reinterpret_cast<void**>(&chunk),
-                   std::max<size_t>(sizeof(void*), alignof(Chunk)),
-                   sizeof(Chunk) + sizeof(Node) * _chunk_size);
+    size_t alloc_size = sizeof(Chunk) + sizeof(Node) * _chunk_size;
+    int error = posix_memalign(reinterpret_cast<void**>(&chunk),
+                               std::max<size_t>(sizeof(void*), alignof(Chunk)),
+                               alloc_size);
+    PADDLE_ENFORCE_EQ(error,
+                      0,
+                      paddle::platform::errors::ResourceExhausted(
+                          "Fail to alloc memory of %ld size, error code is %d.",
+                          alloc_size,
+                          error));
     chunk->next = _chunks;
     _chunks = chunk;
 

diff --git a/paddle/fluid/eager/api/manual/eager_manual/forwards/multiply_fwd_func.cc b/paddle/fluid/eager/api/manual/eager_manual/forwards/multiply_fwd_func.cc
@@ -61,8 +61,9 @@ paddle::Tensor multiply_ad_func(const paddle::Tensor& x,
   // Type promotion Logic
   if (phi::NeedTypePromotion(x.dtype(), y.dtype())) {
     VLOG(5) << "got different data type, run type protmotion automatically.";
-    LOG(WARNING) << "got different data type, run type protmotion "
-                    "automatically, this may cause data type been changed.";
+    LOG_FIRST_N(WARNING, 1)
+        << "got different data type, run type protmotion "
+           "automatically, this may cause data type been changed.";
     auto op_name = phi::TransToFluidOpName("multiply");
     auto promotion_type = phi::GetPromoteDtype(op_name, x.dtype(), y.dtype());
 
@@ -407,8 +408,9 @@ paddle::Tensor multiply_ad_func(const paddle::Tensor& x,
   // Type promotion Logic
   if (phi::NeedTypePromotion(x.dtype(), y.dtype())) {
     VLOG(5) << "got different data type, run type protmotion automatically.";
-    LOG(WARNING) << "got different data type, run type protmotion "
-                    "automatically, this may cause data type been changed.";
+    LOG_FIRST_N(WARNING, 1)
+        << "got different data type, run type protmotion "
+           "automatically, this may cause data type been changed.";
     auto op_name = phi::TransToFluidOpName("multiply");
     auto promotion_type = phi::GetPromoteDtype(op_name, x.dtype(), y.dtype());
 

diff --git a/paddle/fluid/eager/auto_code_generator/generator/eager_gen.py b/paddle/fluid/eager/auto_code_generator/generator/eager_gen.py
@@ -528,7 +528,7 @@ class {} : public egr::GradNodeBase {{
 
 TYPE_PROMOTION_LOGIC_TEMPLATE = """   if (phi::NeedTypePromotion({x}.dtype(), {y}.dtype())) {{
     VLOG(5) << "got different data type, run type protmotion automatically.";
-    LOG(WARNING) << "got different data type, run type protmotion automatically, this may cause data type been changed.";
+    LOG_FIRST_N(WARNING, 1) << "got different data type, run type protmotion automatically, this may cause data type been changed.";
     {op_name}
     auto promotion_type = phi::GetPromoteDtype(op_name, {x}.dtype(), {y}.dtype());
 

diff --git a/paddle/fluid/inference/CMakeLists.txt b/paddle/fluid/inference/CMakeLists.txt
@@ -35,7 +35,7 @@ get_property(fluid_modules GLOBAL PROPERTY FLUID_MODULES)
 get_property(phi_modules GLOBAL PROPERTY PHI_MODULES)
 get_property(ir_targets GLOBAL PROPERTY IR_TARGETS)
 get_property(not_infer_modules GLOBAL PROPERTY NOT_INFER_MODULES)
-set(utils_modules pretty_log string_helper benchmark utf8proc)
+set(utils_modules pretty_log string_helper utf8proc)
 
 if(NOT WITH_GFLAGS)
   set(utils_modules ${utils_modules} paddle_flags)

diff --git a/paddle/fluid/inference/api/analysis_config.cc b/paddle/fluid/inference/api/analysis_config.cc
@@ -180,6 +180,11 @@ void AnalysisConfig::EnableXpu(int l3_size,
                                bool transformer_encoder_adaptive_seqlen,
                                bool enable_multi_stream) {
 #if defined(PADDLE_WITH_XPU) || defined(LITE_SUBGRAPH_WITH_XPU)
+  LOG_FIRST_N(WARNING, 1)
+      << "Parameters in EnableXpu/enable_xpu is deprecated since version "
+         "2.6.1, and will be removed in version 3.0! Please use "
+         "EnableXpu/enable_xpu without parameters, and use "
+         "SetXpuConfig/set_xpu_config to set options.";
   use_xpu_ = true;
   xpu_config_.l3_size = l3_size;
   xpu_config_.conv_autotune_level = conv_autotune;

diff --git a/paddle/fluid/inference/api/paddle_analysis_config.h b/paddle/fluid/inference/api/paddle_analysis_config.h
@@ -111,6 +111,7 @@ struct PD_INFER_DECL XpuConfig {
   bool conv_autotune_file_writeback{false};
 
   // Fc autotune level. The Optional values are 0-9. Default 0 means no
+  // autotune.
   int fc_autotune_level{0};
   // Base fc autotune info is read from fc_autotune_file.
   std::string fc_autotune_file;
@@ -367,7 +368,7 @@ struct PD_INFER_DECL AnalysisConfig {
   ///
   void EnableXpu(int l3_size = 0xfffc00,
                  bool l3_locked = false,
-                 bool conv_autotune = true,
+                 bool conv_autotune = false,
                  const std::string& conv_autotune_file = "",
                  const std::string& transformer_encoder_precision = "int16",
                  bool transformer_encoder_adaptive_seqlen = false,

diff --git a/paddle/fluid/inference/tensorrt/op_teller.cc b/paddle/fluid/inference/tensorrt/op_teller.cc
@@ -47,6 +47,7 @@ struct SimpleOpTypeSetTeller : public Teller {
 #endif
 #if IS_TRT_VERSION_GE(7000)
     teller_set.insert("tile");
+    int8_teller_set.insert("tile");
     teller_set.insert("flatten_contiguous_range");
     int8_teller_set.insert("flatten_contiguous_range");
     teller_set.insert("rnn");
@@ -2302,15 +2303,20 @@ struct SimpleOpTypeSetTeller : public Teller {
       if (!with_dynamic_shape) {
         if (tile_inputs.find("repeat_times_tensor") != tile_inputs.end()) {
           if (!desc.Input("repeat_times_tensor").empty()) {
+            VLOG(3) << "Tile op: repeat_times_tensor is not empty.";
             return false;
           }
         }
         if (tile_inputs.find("RepeatTimes") != tile_inputs.end()) {
           if (!desc.Input("RepeatTimes").empty()) {
+            VLOG(3) << "Tile op: RepeatTimes is not empty.";
             return false;
           }
         }
-        if (!desc.HasAttr("repeat_times")) return false;
+        if (!desc.HasAttr("repeat_times")) {
+          VLOG(3) << "Tile op:`repeat_times` is not set.";
+          return false;
+        }
       }
     }
 #endif

diff --git a/paddle/fluid/inference/utils/CMakeLists.txt b/paddle/fluid/inference/utils/CMakeLists.txt
@@ -1,8 +1,3 @@
-cc_library(
-  benchmark
-  SRCS benchmark.cc
-  DEPS enforce common)
-paddle_test(test_benchmark SRCS benchmark_tester.cc DEPS benchmark)
 cc_library(
   infer_io_utils
   SRCS io_utils.cc
@@ -13,13 +8,5 @@ cc_library(
   DEPS proto_desc enforce common)
 
 cc_library(table_printer SRCS table_printer.cc)
-paddle_test(test_table_printer SRCS table_printer_tester.cc)
 
 proto_library(shape_range_info_proto SRCS shape_range_info.proto)
-
-if(WITH_ONNXRUNTIME AND WIN32)
-  # Copy onnxruntime for some c++ test in Windows, since the test will
-  # be build only in CI, so suppose the generator in Windows is Ninja.
-  copy_onnx(test_benchmark)
-  copy_onnx(test_table_printer)
-endif()