Merge branch 'develop' into cinn/margin

Author: crazyxiaoxi

.github/CODEOWNERS

@@ -1,7 +1,7 @@
 # This file is migrated from CI script, it's an effort of modenizing our dev infra.
 # Code owners are expected to take responsibility for review patches to respective file.
 
-/CMakeLists.txt @wanghuancoder @Aurelius84 @XiaoguangHu01 @qili93
+/CMakeLists.txt @wanghuancoder @Aurelius84 @XiaoguangHu01
 paddle/fluid/distributed/collective @sneaxiy @ForFishes
 paddle/fluid/eager/autograd_meta.cc @JiabinYang @phlrain
 paddle/fluid/eager/autograd_meta.h @JiabinYang @phlrain
@@ -12,20 +12,20 @@ paddle/fluid/eager/grad_node_info.h @JiabinYang @phlrain
 paddle/fluid/eager/grad_tensor_holder.cc @JiabinYang @phlrain
 paddle/fluid/eager/grad_tensor_holder.h @JiabinYang @phlrain
 paddle/fluid/eager/tensor_wrapper.h @JiabinYang @phlrain
-paddle/fluid/framework/block_desc.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/details/op_registry.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/framework.proto @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/grad_op_desc_maker.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/ir/graph.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/ir/node.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/lod_tensor.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/op_desc.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/operator.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/scope.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/selected_rows.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
-paddle/fluid/framework/tensor.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
+paddle/fluid/framework/block_desc.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/details/op_registry.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/framework.proto @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/grad_op_desc_maker.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/ir/graph.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/ir/node.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/lod_tensor.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/op_desc.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/operator.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/scope.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/selected_rows.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
+paddle/fluid/framework/tensor.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
 paddle/fluid/framework/unused_var_check.cc @zhiqiu @phlrain
-paddle/fluid/framework/var_desc.h @XiaoguangHu01 @zhiqiu @Xreki @qili93 @Aurelius84
+paddle/fluid/framework/var_desc.h @XiaoguangHu01 @zhiqiu @Xreki @Aurelius84
 paddle/fluid/operators/distributed/send_recv.proto.in @gongweibao @seiriosPlus
 paddle/fluid/prim/api/api.yaml @cxxly @xiaoguoguo626807 @Charles-hit @cyber-pioneer @JiabinYang
 paddle/fluid/prim/api/composite_backward/composite_backward_api.h @cxxly @xiaoguoguo626807 @Charles-hit @cyber-pioneer @JiabinYang

.github/PULL_REQUEST_TEMPLATE.md

@@ -2,7 +2,7 @@
 <!-- Demo: https://github.com/PaddlePaddle/Paddle/pull/24810 -->
 
 ### PR Category
-<!-- One of [ User Experience | Execute Infrastructure | Operator Mechanism | CINN | Custom Device | Performance Optimization | Distributed Strategy | Parameter Server | Communication Library | Auto Parallel | Inference | Environment Adaptation | Others ] -->
+<!-- One of [ User Experience | Execute Infrastructure | Operator Mechanism | CINN | Custom Device | Performance Optimization | Distributed Strategy | Parameter Server | Communication Library | Auto Parallel | Inference | Environment Adaptation ] -->
 
 
 ### PR Types

cmake/external/xpu.cmake

@@ -30,7 +30,7 @@ if(NOT DEFINED XPU_XRE_BASE_VERSION)
   set(XPU_XRE_BASE_VERSION "4.32.0.1")
 endif()
 if(NOT DEFINED XPU_XHPC_BASE_DATE)
-  set(XPU_XHPC_BASE_DATE "20240712")
+  set(XPU_XHPC_BASE_DATE "20240804")
 endif()
 set(XPU_XCCL_BASE_VERSION "1.2.5")
 if(NOT DEFINED XPU_XFT_BASE_VERSION)

cmake/generic.cmake

@@ -1312,7 +1312,7 @@ function(math_library TARGET)
   set(cc_srcs)
   set(cu_srcs)
   set(hip_srcs)
-  set(math_common_deps device_context framework_proto enforce)
+  set(math_common_deps device_context framework_proto phi common)
   if(WITH_GPU)
     if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0)
       list(APPEND math_common_deps cub)

paddle/cinn/adt/adt.h

@@ -283,7 +283,7 @@ struct Ok final {
   bool operator!=(const Ok&) const { return false; }
 };
 
-#define ADT_TODO() PADDLE_THROW(phi::errors::Fatal("TODO"))
+#define ADT_TODO() PADDLE_THROW(::common::errors::Fatal("TODO"))
 
 inline std::size_t hash_combine(std::size_t lhs, std::size_t rhs) {
   return lhs ^= rhs + 0x9e3779b9 + (lhs << 6) + (lhs >> 2);

paddle/cinn/adt/anchor_sd_equation_context.cc

@@ -31,7 +31,7 @@ void GenerateScheduleMeshEquationsImpl(const List<ScheduleDim>& sched_dims,
   PADDLE_ENFORCE_EQ(
       input_iterators->size() == output_iterators->size(),
       true,
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of input iterators and output iterators should be equal, "
           "but got input iterators size = %d, output iterators size = %d.",
           input_iterators->size(),
@@ -53,7 +53,7 @@ void GenerateScheduleMeshEquationsImpl(
   PADDLE_ENFORCE_EQ(
       shape.value()->size() == output_iterators->size(),
       true,
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of shape and output iterators should be equal, but got "
           "shape size = %d, output iterators size = %d.",
           shape.value()->size(),
@@ -75,7 +75,7 @@ void GenerateScheduleMeshEquationsImpl(
   const auto& [sched_mesh, perm] = sched_transpose.tuple();
   PADDLE_ENFORCE_EQ(GetOutputRank(sched_mesh) == output_iterators->size(),
                     true,
-                    phi::errors::InvalidArgument(
+                    ::common::errors::InvalidArgument(
                         "The size of output iterators should be equal to the "
                         "rank of the schedule mesh, but got output iterators "
                         "size = %d, rank of the schedule mesh = %d.",
@@ -99,7 +99,7 @@ void GenerateScheduleMeshEquationsImpl(
   const auto& [sched_mesh, _] = sched_padding.tuple();
   PADDLE_ENFORCE_EQ(GetOutputRank(sched_mesh) == output_iterators->size(),
                     true,
-                    phi::errors::InvalidArgument(
+                    ::common::errors::InvalidArgument(
                         "The size of output iterators should be equal to the "
                         "rank of the schedule mesh, but got output iterators "
                         "size = %d, rank of the schedule mesh = %d.",

paddle/cinn/adt/equation_solver.cc

@@ -186,7 +186,7 @@ std::unordered_map<Variable, Value> InferValuesImpl(
   PADDLE_ENFORCE_EQ(
       out_msg_in_indexes.value()->size() == in_msg_in_indexes.value()->size(),
       true,
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of out_msg_in_indexes should be equal to the size of "
           "in_msg_in_indexes, but got out_msg_in_indexes size = %d, "
           "in_msg_in_indexes size = %d.",
@@ -195,7 +195,7 @@ std::unordered_map<Variable, Value> InferValuesImpl(
   PADDLE_ENFORCE_EQ(
       out_msg_out_indexes.value()->size() == in_msg_out_indexes.value()->size(),
       true,
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of out_msg_out_indexes should be equal to the size of "
           "in_msg_out_indexes, but got out_msg_out_indexes size = %d, "
           "in_msg_out_indexes size = %d.",
@@ -288,8 +288,8 @@ void CheckEquationsSolvable(
         [&](const auto& opt_old_value, const auto& simplified_value) {
           LOG(ERROR) << "old_value: " << ToTxtString(opt_old_value);
           LOG(ERROR) << "simplified_value: " << ToTxtString(simplified_value);
-          PADDLE_THROW(
-              phi::errors::InvalidArgument("CheckEquationsSolvable Failed"));
+          PADDLE_THROW(::common::errors::InvalidArgument(
+              "CheckEquationsSolvable Failed"));
           return tValueInferSuccess<bool>{false};
         });
   };

paddle/cinn/adt/generate_map_expr.cc

@@ -105,7 +105,7 @@ bool HasDynamicShape(const ::pir::Value& tensor) {
       PADDLE_ENFORCE_EQ(
           dim,
           -1UL,
-          phi::errors::InvalidArgument(
+          ::common::errors::InvalidArgument(
               "The dynamic shape dim should be -1, but got %d.", dim));
       return true;
     }
@@ -170,7 +170,11 @@ hlir::framework::OpPatternKind GetOpPatternKind(const ::pir::Operation* node) {
 bool CollectRewrittenReductionOpStmts(const OpStmt& op_stmt,
                                       List<OpStmt>* ret) {
   const auto& [op, inputs, outputs] = op_stmt.tuple();
-  CHECK(op.Has<const ::pir::Operation*>());
+  PADDLE_ENFORCE_EQ(
+      op.Has<const ::pir::Operation*>(),
+      true,
+      phi::errors::InvalidArgument(
+          "The op should have a value of type ::pir::Operation*"));
   if (GetOpPatternKind(op.Get<const ::pir::Operation*>()) ==
       hlir::framework::OpPatternKind::kReduction) {
     tReduceInit<const ::pir::Operation*> init_op{
@@ -234,7 +238,10 @@ std::vector<std::shared_ptr<IGroup>> GenerateIGroups(
   std::vector<std::shared_ptr<IGroup>> ret{};
 
   List<OpStmt> op_stmts = MakeOpStmts(group);
-  CHECK(!op_stmts->empty());
+  PADDLE_ENFORCE_EQ(
+      !op_stmts->empty(),
+      true,
+      phi::errors::InvalidArgument("The op_stmts should not be empty"));
 
   PartitionIGroupOpStmts(op_stmts, [&](const auto& igroup_spec) {
     ret.push_back(MakeIGroup(igroup_spec));
@@ -249,7 +256,7 @@ std::shared_ptr<KGroup> GenerateKGroups(
   PADDLE_ENFORCE_EQ(
       igroups.size(),
       1UL,
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of igroups should be 1, but got %d.", igroups.size()));
   return std::make_shared<KGroup>(group, igroups);
 }
@@ -271,9 +278,12 @@ std::unordered_map<Variable, const Value> MakeSdIterator2Iterator(
   std::unordered_map<Variable, const Value> ret{};
 
   for (std::size_t i = 0; i < igroup.loop_iterators()->size(); ++i) {
-    CHECK(ret.emplace(igroup.loop_iterators()->at(i),
-                      igroup.loop_iterators()->at(i))
-              .second);
+    PADDLE_ENFORCE_EQ(
+        ret.emplace(igroup.loop_iterators()->at(i),
+                    igroup.loop_iterators()->at(i))
+            .second,
+        true,
+        phi::errors::InvalidArgument("The loop iterator should be unique"));
   }
 
   return ret;
@@ -326,15 +336,18 @@ LoopDescriptor4IterVarT MakeGetterLoopDescriptor4IterVar(
   PADDLE_ENFORCE_EQ(
       loop_iters->size(),
       sd->size(),
-      phi::errors::InvalidArgument(
+      ::common::errors::InvalidArgument(
           "The size of loop iterators and loop descriptors should be equal, "
           "but got loop iterators size = %d, loop descriptors size = %d.",
           loop_iters->size(),
           sd->size()));
   using Cache = std::unordered_map<Iterator, LoopDescriptor>;
   const auto& sd_iter2sd = std::make_shared<Cache>();
   for (std::size_t i = 0; i < loop_iters->size(); ++i) {
-    CHECK(sd_iter2sd->emplace(loop_iters->at(i), sd->at(i)).second);
+    PADDLE_ENFORCE_EQ(
+        sd_iter2sd->emplace(loop_iters->at(i), sd->at(i)).second,
+        true,
+        phi::errors::InvalidArgument("The loop iterator should be unique"));
   }
   return [sd_iter2sd](const auto& sd_iter) { return sd_iter2sd->at(sd_iter); };
 }
@@ -343,7 +356,10 @@ TreeMerger<Stmt> MakeTreeMerger(const MapIr& map_ir) {
   using Cache = std::unordered_map<OpStmt, LoopIterators>;
   auto cache = std::make_shared<Cache>();
   for (const auto& op_stmt : *(map_ir.op_stmts())) {
-    CHECK(cache->emplace(op_stmt, map_ir.loop_iterators()).second);
+    PADDLE_ENFORCE_EQ(
+        cache->emplace(op_stmt, map_ir.loop_iterators()).second,
+        true,
+        phi::errors::InvalidArgument("The op_stmt should be unique"));
   }
 
   TreeMerger<Stmt> tree_merger{};
@@ -363,9 +379,12 @@ MapStmt<Stmt> MakeMapStmt(const MapIrList& map_irs) {
   PADDLE_ENFORCE_EQ(
       stmts->size(),
       1UL,
-      phi::errors::InvalidArgument("The size of stmts should be 1, but got %d.",
-                                   stmts->size()));
-  CHECK(stmts->at(0).Has<MapStmt<Stmt>>());
+      ::common::errors::InvalidArgument(
+          "The size of stmts should be 1, but got %d.", stmts->size()));
+  PADDLE_ENFORCE_EQ(stmts->at(0).Has<MapStmt<Stmt>>(),
+                    true,
+                    phi::errors::InvalidArgument(
+                        "The stmts should have a value of type MapStmt<Stmt>"));
   return stmts->at(0).Get<MapStmt<Stmt>>();
 }
 

paddle/cinn/adt/get_sub_reshape_dim_ranges.cc

@@ -37,8 +37,13 @@ GetSubReshapeDimRanges(const List<DimExpr>& lhs_dims,
   if (GetNumel(lhs_dims) != GetNumel(rhs_dims)) {
     return std::nullopt;
   }
-  CHECK(!lhs_dims->empty());
-  CHECK(!rhs_dims->empty());
+  PADDLE_ENFORCE_EQ(
+      !lhs_dims->empty(),
+      true,
+      phi::errors::InvalidArgument("Sorry,but lhs_dims is empty"));
+  PADDLE_ENFORCE_EQ(!rhs_dims->empty(),
+                    true,
+                    phi::errors::InvalidArgument("Sory,but rhs_dims is empty"));
   std::vector<std::pair<int, int>> lhs_ranges{};
   std::vector<std::pair<int, int>> rhs_ranges{};
   int lhs_start = 0;
@@ -51,7 +56,10 @@ GetSubReshapeDimRanges(const List<DimExpr>& lhs_dims,
     end = (end > dims->size() ? dims->size() : end);
     std::int64_t ret = 1;
     for (std::size_t i = 0; i < end; ++i) {
-      CHECK(dims->at(i).Has<std::int64_t>());
+      PADDLE_ENFORCE_EQ(
+          dims->at(i).Has<std::int64_t>(),
+          true,
+          phi::errors::InvalidArgument("dims->at(i) is not int64_t"));
       ret *= dims->at(i).Get<std::int64_t>();
     }
     return ret;
@@ -82,10 +90,14 @@ GetSubReshapeDimRanges(const List<DimExpr>& lhs_dims,
     } else if (LhsAcc() > RhsAcc()) {
       rhs_end++;
     } else {
-      PADDLE_THROW(phi::errors::Fatal("Dead code"));
+      PADDLE_THROW(::common::errors::Fatal("Dead code"));
     }
   }
-  CHECK(lhs_end == lhs_dims->size() && rhs_end == rhs_dims->size());
+  PADDLE_ENFORCE_EQ(lhs_end == lhs_dims->size() && rhs_end == rhs_dims->size(),
+                    true,
+                    phi::errors::InvalidArgument(
+                        "lhs_end is not equal to lhs_dims->size() and rhs_end "
+                        "is not equal to rhs_dims->size()"));
   if (lhs_start < lhs_end && rhs_start < rhs_end) {
     lhs_ranges.emplace_back(std::make_pair(lhs_start, lhs_end));
     rhs_ranges.emplace_back(std::make_pair(rhs_start, rhs_end));

paddle/cinn/adt/igroup.cc

@@ -27,9 +27,14 @@ std::shared_ptr<IndexExprInferContext> MakeIndexExprInferContext(
   const auto& anchor_iterators = igroup.GetAnchorIterators();
 
   for (std::size_t i = 0; i < anchor_iterators->size(); ++i) {
-    CHECK(anchor_iterator2value
-              .emplace(anchor_iterators->at(i), anchor_iterators->at(i))
-              .second);
+    PADDLE_ENFORCE_EQ(
+        anchor_iterator2value
+            .emplace(anchor_iterators->at(i), anchor_iterators->at(i))
+            .second,
+        true,
+        phi::errors::InvalidArgument(
+            "The element in anchor iterators failed to insert in anchor "
+            "iterator2value! Please check."));
   }
 
   return std::make_shared<IndexExprInferContext>(anchor_iterator2value);
@@ -102,10 +107,10 @@ List<Iterator> IGroup::GetIndexIterators(const Index& index) const {
     } else if (arg_pos.Has<Undefined>()) {
       // do nothing
     } else {
-      PADDLE_THROW(phi::errors::Fatal("Dead code"));
+      PADDLE_THROW(::common::errors::Fatal("Dead code"));
     }
   }
-  PADDLE_THROW(phi::errors::Fatal("Can not find anchor iterators"));
+  PADDLE_THROW(::common::errors::Fatal("Can not find anchor iterators"));
 }
 
 }  // namespace cinn::adt