Fix typos (PaddlePaddle#1455)

cinn/common/cas.cc

@@ -129,15 +129,15 @@ Expr ProductGetNonConstantPart(Expr u) {
 
 namespace detail {
 
-// Is a Divisiable to b.
+// Is a Divisible to b.
 // @{
 bool IsDivisible(int64_t a, int64_t b) {
   CHECK_NE(b, 0);
   return a % b == 0;
 }
 bool IsDivisible(const Sum* a, int b);
 
-// If int a Divisiable to any operands of product b
+// If int a Divisible to any operands of product b
 bool IsDivisible(int a, const Product* b) {
   if (a < 0) return false;
   for (auto& item : b->operands()) {
@@ -694,7 +694,7 @@ std::vector<Expr> CasSimplifyMutator::SimplifyBinarySum(Expr left, Expr right) {
     if (bi && bi->value == 0) return {a};
     if (bf && bf->value == 0.f) return {a};
 
-    // customied case for Mod
+    // customized case for Mod
     {
       auto* am = a.As<Mod>();
       auto* bm = b.As<Mod>();
@@ -1879,11 +1879,11 @@ Expr CasSimplifyMutator::SimplifyFracOp(Expr expr) {
   }
 
   // case 2
-  // sum/x or product/x is divisiable
+  // sum/x or product/x is divisible
   if (bi) {
     auto* a_sum     = a.As<Sum>();
     auto* a_product = a.As<Product>();
-    // disiviable
+    // divisible
     if (a_sum && IsDivisible(a_sum, bi->value)) return Divide(a_sum, bi->value);
     if (a_product) {
       if (IsDivisible(a_product, bi->value) || IsDivisible(bi->value, a_product)) {
@@ -1915,7 +1915,7 @@ Expr CasSimplifyMutator::SimplifyFracOp(Expr expr) {
       }
     }
 
-    // not divisiable
+    // not divisible
     /*
     if (a_sum) {
       auto expr = DividePartially(a_sum, bi->value);

cinn/common/cas.h

@@ -37,7 +37,7 @@ Expr ReplaceMaxToConstant(Expr expr);
 struct CasInterval {
   template <typename T>
   CasInterval(T l, T r) : l(l), r(r) {
-    CHECK_LE(l, r) << "left shoud not be larger than right";
+    CHECK_LE(l, r) << "left should not be larger than right";
   }
 
   /**

cinn/frontend/decomposer/batch_norm_test.cc

@@ -128,7 +128,7 @@ void ComputeBatchNormTrainRef(const std::vector<T>& x,
   };
   Loop(func_y, n, c, h, w);
 
-  // new moving runnning and variance
+  // new moving running and variance
   float factor_0 = momentum;
   float factor_1 = static_cast<float>(1.0f - momentum);
   for (int ic = 0; ic < c; ++ic) {

cinn/frontend/op_mappers/paddle/batchnorm.cc

@@ -145,7 +145,7 @@ void BatchNormGradOpMapper(const paddle::cpp::OpDesc& op_desc, const OpMapperCon
 
   // batch norm grad, output(grad_x, grad_scale, grad_bias)
   auto outs = ctx.Builder()->BatchNormGrad(dy, x, scale, saved_mean, saved_variance, epsilon, data_layout);
-  CHECK_EQ(outs.size(), 3ul) << "batch_norm_grad API's should return 3 Variable!";
+  CHECK_EQ(outs.size(), 3ul) << "batch_norm_grad APIs should return 3 Variable!";
 
   for (int i = 0; i < outs.size(); i++) {
     if (output_names[i].empty()) {

cinn/frontend/pass/cast_collapsing.cc

@@ -171,17 +171,17 @@ class CastCollapsingPass : public ProgramPass {
 
     if (CheckCastBorder(cast, in2instr)) {
       if (can_remove) {
-        VLOG(4) << "The op " << cast_info << " is a output op of graph, connot fuse, remove.";
+        VLOG(4) << "The op " << cast_info << " is a output op of graph, cannot fuse, remove.";
         // this cast not used by any other op, remove
         remove_instrs->insert(cast);
       } else {
         if (input_dtype == output_dtype) {
           VLOG(4) << "The cast op " << cast_info << " is fetched but useless, replace with identity.";
-          // cannot remove, however, the transpsoe is useless, we can replace the cast with indentiy for more
+          // cannot remove, however, the transpose is useless, we can replace the cast with identity for more
           // fusion opportunity
           ReplaceWithIdentity(cast);
         }
-        // else the transpsoe is fetched and helpful, ignore
+        // else the transpose is fetched and helpful, ignore
       }
       return;
     }

cinn/hlir/op/elementwise.cc

@@ -295,7 +295,7 @@ std::shared_ptr<OpStrategy> StrategyForFillConstant(const framework::NodeAttr &a
 
     if (force_cpu && target != common::DefaultHostTarget()) {
       LOG(WARNING) << "The attribute \"force_cpu\" of \"fill_constant\" not supported in CINN! The \"fill_constant\"'s "
-                      "output tensot will placed on "
+                      "output tensor will placed on "
                    << target;
     }
 
@@ -339,10 +339,10 @@ std::vector<Type> InferDtypeForFillConstant(const std::vector<Type> &inputs_type
     out_type       = common::Str2Type(dtype_str);
     VLOG(3) << "FillConstant output dtype (from [dtype]): " << dtype_str;
   } else {
-    // attribute [dtype] no given, infered by value's type
+    // attribute [dtype] no given, inferred by value's type
     auto scalar = GetScalarExpr(attrs.at("value"));
     out_type    = scalar->type();
-    VLOG(3) << "FillConstant scalar type (from [vaule]): " << common::Type2Str(out_type);
+    VLOG(3) << "FillConstant scalar type (from [value]): " << common::Type2Str(out_type);
   }
   return {out_type};
 }

cinn/optim/buffer_assign.cc

@@ -95,7 +95,7 @@ std::map<std::string, ir::Tensor> InitialAssignBuffer(Expr* expr,
     }
   }
 
-  // determine which tensor to have the initial buffer, and will share accross the cluser, we take a topological order
+  // determine which tensor to have the initial buffer, and will share across the cluster, we take a topological order
   // of the computational graph, and find out which tensor comes first in a cluster.
 
   auto _topo_order_topo_edges_ = comp_graph->topological_order();

cinn/runtime/cinn_runtime.h

@@ -62,7 +62,7 @@ typedef enum cinn_type_code_t {
 #endif
 
 /**
- * A tuntime tag for type in CINN system.
+ * A runtime tag for type in CINN system.
  */
 typedef struct cinn_type_t {
 #if __cplusplus >= 201103L

cinn/utils/event.cc

@@ -66,7 +66,7 @@ std::string Summary::Format(const std::vector<HostEvent> &events) {
       unique_items[e.annotation_]                    = &items.back();
       unique_items.at(e.annotation_)->info.duration_ = 0.0;
     }
-    // Sum cost for categorey
+    // Sum cost for category
     category_cost[e.type_] += e.duration_;
     total_cost += e.duration_;
     max_annot_size = std::max(max_annot_size, e.annotation_.size());

cmake/core.cmake

@@ -254,7 +254,7 @@ function(merge_static_libs TARGET_NAME)
       COMMAND ${CMAKE_COMMAND} -E touch ${target_SRCS}
       DEPENDS ${libs} ${target_OBJS})
 
-    # Generate dummy staic lib
+    # Generate dummy static lib
     file(WRITE ${target_SRCS} "const char *dummy_${TARGET_NAME} = \"${target_SRCS}\";")
     add_library(${TARGET_NAME} STATIC ${target_SRCS})
     target_link_libraries(${TARGET_NAME} ${libs_deps})
@@ -275,7 +275,7 @@ function(merge_static_libs TARGET_NAME)
       COMMAND ${CMAKE_COMMAND} -E touch ${target_SRCS}
       DEPENDS ${libs})
 
-    # Generate dummy staic lib
+    # Generate dummy static lib
     file(WRITE ${target_SRCS} "const char *dummy_${TARGET_NAME} = \"${target_SRCS}\";")
     add_library(${TARGET_NAME} STATIC ${target_SRCS})
     target_link_libraries(${TARGET_NAME} ${libs_deps})
@@ -284,7 +284,7 @@ function(merge_static_libs TARGET_NAME)
       # Get the file names of the libraries to be merged
       set(libfiles ${libfiles} $<TARGET_FILE:${lib}>)
     endforeach()
-    # msvc will put libarary in directory of "/Release/xxxlib" by default
+    # msvc will put library in directory of "/Release/xxxlib" by default
     #       COMMAND cmake -E remove "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_BUILD_TYPE}/${TARGET_NAME}.lib"
     add_custom_command(TARGET ${TARGET_NAME} POST_BUILD
       COMMAND cmake -E make_directory "${CMAKE_CURRENT_BINARY_DIR}/${CMAKE_BUILD_TYPE}"

docs/design.org

@@ -315,7 +315,7 @@ It by default can be resized to proper shape by binding to multiple tensors.
 
 All the buffers will be maintained in global scope, and alloc or dealloc in local scopes.
 
-The benefit is buffer is easy to shared accross multiple statements.
+The benefit is buffer is easy to shared across multiple statements.
 ** Module
 Module is the container of LoweredFuncs and Buffers.
 There might be more than one module in an generated execution.
@@ -334,7 +334,7 @@ then replace call with the statement.
 
 There are several steps to vectorize a forloop:
 
-- the foorloop will first been split with a small factor(better to be times of the SIMD width
+- the forloop will first been split with a small factor(better to be times of the SIMD width
 - convert the PolyFor to For
 - Substitute the iterator variable of the forloop to a Ramp node
 - Transform all the operations related to the Ramp

infrt/common/axis.h

@@ -7,7 +7,7 @@
 namespace infrt {
 namespace common {
 
-//! Get the predifined axis name.
+//! Get the predefined axis name.
 const std::string& axis_name(int level);
 bool IsAxisNameReserved(const std::string& x);