Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into ApiEnhance6

Author: Manfredss

.github/workflows/rerun.yml

@@ -296,3 +296,33 @@ jobs:
           OWNER: ${{ github.repository_owner }}
           REPO: ${{ github.event.repository.name }}
           JOB_NAME: 'Slice / Check bypass / Check bypass'
+
+      - name: Rerun Windows-GPU
+        if: ${{ contains(github.event.comment.body, 'win') && contains(github.event.comment.body, 'gpu') }}
+        uses: ./.github/actions/rerun-workflow
+        with:
+          PR_ID: ${{ github.event.issue.number }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          OWNER: ${{ github.repository_owner }}
+          REPO: ${{ github.event.repository.name }}
+          JOB_NAME: 'Windows-GPU / Check bypass / Check bypass'
+
+      - name: Rerun Windows-Inference
+        if: ${{ contains(github.event.comment.body, 'win') && contains(github.event.comment.body, 'infer')}}
+        uses: ./.github/actions/rerun-workflow
+        with:
+          PR_ID: ${{ github.event.issue.number }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          OWNER: ${{ github.repository_owner }}
+          REPO: ${{ github.event.repository.name }}
+          JOB_NAME: 'Windows-Inference / Check bypass / Check bypass'
+
+      - name: Rerun Windows-OPENBLAS
+        if: ${{ contains(github.event.comment.body, 'win') && contains(github.event.comment.body, 'openblas')}}
+        uses: ./.github/actions/rerun-workflow
+        with:
+          PR_ID: ${{ github.event.issue.number }}
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          OWNER: ${{ github.repository_owner }}
+          REPO: ${{ github.event.repository.name }}
+          JOB_NAME: 'Windows-OPENBLAS / Check bypass / Check bypass'

_typos.toml

@@ -49,13 +49,3 @@ UNEXPECT = 'UNEXPECT'
 tood = 'tood'
 unpacket = "unpacket"
 vaccum = 'vaccum'
-
-# need to fix
-Detecter = 'Detecter'
-dismatch = 'dismatch'
-distributer = 'distributer'
-duoble = 'duoble'
-Grap = 'Grap'
-Simplied = 'Simplied'
-precess = 'precess'
-sittin = 'sittin'

ci/h-test.sh

@@ -161,7 +161,10 @@ concurrency_list="^test_fp8_deep_gemm$|\
 ^test_dist_fuse_gemm_epilogue_pass$|\
 ^test_fuse_allreduce_split_to_reducescatter_pass$|\
 ^test_ps_server_pass$|\
-^test_white_lists$"
+^test_white_lists$|\
+^test_scaled_dot_product_attention$|\
+^test_compat_scaled_dot_product_attention$|\
+^test_flash_attention$"
 
 cd ${work_dir}/build
 tmp_dir=`mktemp -d`

ci/rule-tests/no-deeply-nested-loops-test.yml

@@ -169,7 +169,7 @@ valid:
                         for e in range(10):
                             consume(a, b, c, d, e)
 
-  # 5 levels breaked by extra code - OK
+  # 5 levels broken by extra code - OK
   - |
     for i in range(10):
         for j in range(10):

paddle/cinn/optim/eliminate_common_factor_of_local_index.cc

@@ -166,9 +166,9 @@ CollectLocalVarToIndexes(ir::stmt::BlockRef func_body) {
 }
 
 int ExtractMulNumberFromExpr(const ir::Expr& expr) {
-  ir::Expr simplied_expr = optim::ArithSimplify(expr);
-  if (simplied_expr.is_constant()) {
-    return static_cast<int>(simplied_expr.get_constant());
+  ir::Expr simplified_expr = optim::ArithSimplify(expr);
+  if (simplified_expr.is_constant()) {
+    return static_cast<int>(simplified_expr.get_constant());
   } else if (expr.As<ir::Mul>()) {
     auto mul = expr.As<ir::Mul>();
     return ExtractMulNumberFromExpr(mul->a()) *
@@ -181,9 +181,9 @@ int ExtractMulNumberFromExpr(const ir::Expr& expr) {
 }
 
 int ExtractAddNumberFromExpr(const ir::Expr& expr) {
-  ir::Expr simplied_expr = optim::ArithSimplify(expr);
-  if (simplied_expr.is_constant()) {
-    return static_cast<int>(simplied_expr.get_constant());
+  ir::Expr simplified_expr = optim::ArithSimplify(expr);
+  if (simplified_expr.is_constant()) {
+    return static_cast<int>(simplified_expr.get_constant());
   } else if (expr.As<ir::Add>()) {
     auto add = expr.As<ir::Add>();
     return ExtractAddNumberFromExpr(add->a()) +
@@ -203,8 +203,8 @@ int gcd(int a, int b) {
 }
 
 ir::Expr ExtractSymbolicFromExpr(const ir::Expr& expr) {
-  ir::Expr simplied_expr = optim::ArithSimplify(expr);
-  if (simplied_expr.is_constant()) {
+  ir::Expr simplified_expr = optim::ArithSimplify(expr);
+  if (simplified_expr.is_constant()) {
     return ir::Expr(0);
   } else if (expr.As<ir::_Var_>()) {
     auto var = expr.As<ir::_Var_>();

paddle/fluid/distributed/collective/reducer.cc

@@ -781,7 +781,7 @@ void EagerReducer::AddDistHook(size_t var_index) {
   }
 
   VLOG(3) << "Tensor[" << var_index << "] [" << tensors_[var_index].name()
-          << "@Grad] arrived and triggered disthook";
+          << "@GRAD] arrived and triggered DistHook";
 
   local_used_vars_[var_index] = 1;
 

paddle/fluid/eager/auto_code_generator/generator/eager_gen.py

@@ -93,6 +93,7 @@
     "acos_double_grad",
     "put_along_axis_double_grad",
     "masked_fill_double_grad",
+    "index_elementwise_put_with_tensor_double_grad",
 ]
 
 # white ops list whose kernel can automatically do type promotion.

paddle/fluid/framework/ir/coalesce_grad_tensor_pass.cc

@@ -157,25 +157,25 @@ class CoalesceGradTensorPass : public ir::Pass {
       PADDLE_ENFORCE_EQ(iter != vars_info.end(),
                         true,
                         common::errors::NotFound(
-                            "Parameter@Grad %s is not found.", p_g.second));
+                            "Parameter@GRAD %s is not found.", p_g.second));
       PADDLE_ENFORCE_EQ(
           !iter->second.empty(),
           true,
           common::errors::InvalidArgument(
-              "Parameter@Grad %s's var node is empty.", p_g.second));
+              "Parameter@GRAD %s's var node is empty.", p_g.second));
       for (auto it : iter->second) {
         PADDLE_ENFORCE_NOT_NULL(
             it->Var(),
             common::errors::InvalidArgument(
-                "A node of Parameter@Grad %s does not hold variable.",
+                "A node of Parameter@GRAD %s does not hold variable.",
                 p_g.second));
         pinned_var_set->insert(it->Var()->Name());
       }
       PADDLE_ENFORCE_EQ(
           IsDenseTensorType(GetTypeOfVar(vars_info, p_g.second)),
           true,
           common::errors::InvalidArgument(
-              "Parameter@Grad %s is not phi::DenseTensor.", p_g.second));
+              "Parameter@GRAD %s is not phi::DenseTensor.", p_g.second));
     }
   }
 
@@ -541,7 +541,7 @@ class CoalesceGradTensorPass : public ir::Pass {
       PADDLE_ENFORCE_EQ(next_dtype,
                         dtype,
                         common::errors::InvalidArgument(
-                            "All Parameter@Grad should have same dtype, but "
+                            "All Parameter@GRAD should have same dtype, but "
                             "there are two different type: %s, %s.",
                             DataTypeToString(next_dtype),
                             DataTypeToString(dtype)));

paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc

@@ -513,7 +513,7 @@ void FuseOptimizerOpPass::SortParametersAndAuxVars(
         iter != grad_vec.end(),
         true,
         common::errors::NotFound(
-            "Parameter@Grad(%s) is not found in gradient vector.", p_g.second));
+            "Parameter@GRAD(%s) is not found in gradient vector.", p_g.second));
     auto idx = std::distance(grad_vec.begin(), iter);
     grad_sort_idx.emplace_back(idx);
   }

paddle/fluid/inference/api/analysis_predictor.cc

@@ -3567,7 +3567,7 @@ USE_TRT_CONVERTER(merge_layernorm)
 USE_TRT_CONVERTER(trans_layernorm)
 USE_TRT_CONVERTER(skip_merge_layernorm)
 USE_TRT_CONVERTER(generic_plugin_creator)
-USE_TRT_CONVERTER(custom_plugin_creater)  // typos: disable-line
+USE_TRT_CONVERTER(custom_plugin_creator)
 USE_TRT_CONVERTER(custom_generic_plugin_creator)
 USE_TRT_CONVERTER(fuse_eleadd_transpose)
 USE_TRT_CONVERTER(tanh_shrink)