Merge from 'sycl' to 'sycl-web'

Author: iclsrc

.github/CODEOWNERS

@@ -28,9 +28,9 @@ sycl/doc/extensions/ @intel/dpcpp-specification-reviewers
 
 # Sub-groups
 sycl/include/CL/sycl/detail/spirv.hpp @Pennycook @AlexeySachkov
-sycl/include/CL/sycl/intel/group_algorithm.hpp @Pennycook @AlexeySachkov
-sycl/include/CL/sycl/intel/sub_group.hpp @Pennycook @AlexeySachkov
-sycl/include/CL/sycl/intel/sub_group_host.hpp @Pennycook @AlexeySachkov
+sycl/include/sycl/ext/intel/group_algorithm.hpp @Pennycook @AlexeySachkov
+sycl/include/sycl/ext/intel/sub_group.hpp @Pennycook @AlexeySachkov
+sycl/include/sycl/ext/intel/sub_group_host.hpp @Pennycook @AlexeySachkov
 
 # PI API
 sycl/include/CL/sycl/detail/pi.def @smaslov-intel
@@ -53,17 +53,17 @@ sycl/source/detail/stream_impl.cpp @againull
 sycl/source/stream.cpp @againull
 
 # FPGA extensions
-sycl/include/CL/sycl/intel/fpga_device_selector.hpp @MrSidims
-sycl/include/CL/sycl/intel/fpga_extensions.hpp @MrSidims
-sycl/include/CL/sycl/intel/fpga_reg.hpp @MrSidims
-sycl/include/CL/sycl/intel/pipes.hpp @MrSidims
+sycl/include/sycl/ext/intel/fpga_device_selector.hpp @MrSidims
+sycl/include/sycl/ext/intel/fpga_extensions.hpp @MrSidims
+sycl/include/sycl/ext/intel/fpga_reg.hpp @MrSidims
+sycl/include/sycl/ext/intel/pipes.hpp @MrSidims
 sycl/include/CL/sycl/pipes.hpp @MrSidims
 
 # Reduction extension
-sycl/include/CL/sycl/intel/reduction.hpp @v-klochkov
+sycl/include/sycl/ext/intel/reduction.hpp @v-klochkov
 
 # Function pointers
-sycl/include/CL/sycl/intel/function_pointer.hpp @AlexeySachkov
+sycl/include/sycl/ext/intel/function_pointer.hpp @AlexeySachkov
 sycl/source/function_pointer.cpp @AlexeySachkov
 
 # Half Type

clang/include/clang/Basic/Attr.td

@@ -1261,6 +1261,14 @@ def SYCLSimdAccessorPtr : InheritableAttr {
   let Documentation = [Undocumented];
 }
 
+// Used to mark readonly accessors. It is not to be used directly in the source.
+def SYCLAccessorReadonly : Attr {
+  // This attribute has no spellings as it is only ever created implicitly.
+  let Spellings = [];
+  let SemaHandler = 0;
+  let Documentation = [Undocumented];
+}
+
 // The attribute denotes that it is a function written in a scalar fashion, which
 // is used in ESIMD context and needs to be vectorized by a vector backend compiler.
 // For now, this attribute will be used only in internal implementation of

clang/include/clang/Basic/AttrDocs.td

@@ -415,6 +415,10 @@ def SYCLSimdDocs : Documentation {
     The compiler may decide to compile such functions using different optimization
     and code generation pipeline. Also, this attribute is used to distinguish
     ESIMD private globals from regular SYCL global variables.
+
+    In SYCL 1.2.1 mode, the ``intel::sycl_explicit_simd`` attribute is propagated
+    from the function it is applied to onto the kernel which calls the function.
+    In SYCL 2020 mode, the attribute is not propagated to the kernel.
   }];
 }
 
@@ -2443,8 +2447,9 @@ lambda capture, or function object member, of the callable to which the
 attribute was applied. This effect is equivalent to annotating restrict on
 **all** kernel pointer arguments in an OpenCL or SPIR-V kernel.
 
-If ``intel::kernel_args_restrict`` is applied to a function called from a device
-kernel, the attribute is not ignored and it is propagated to the kernel.
+In SYCL 1.2.1 mode, the ``intel::kernel_args_restrict`` attribute is propagated
+from the function it is applied to onto the kernel which calls the function.
+In SYCL 2020 mode, the attribute is not propagated to the kernel.
 
 The attribute forms an unchecked assertion, in that implementations
 do not need to check/confirm the pre-condition in any way. If a user applies
@@ -2482,8 +2487,10 @@ def SYCLIntelNumSimdWorkItemsAttrDocs : Documentation {
   let Content = [{
 Applies to a device function/lambda function. Indicates the number of work
 items that should be processed in parallel. Valid values are positive integers.
-If ``intel::num_simd_work_items`` is applied to a function called from a
-device kernel, the attribute is not ignored and it is propagated to the kernel.
+
+In SYCL 1.2.1 mode, the ``intel::num_simd_work_items`` attribute is propagated
+from the function it is applied to onto the kernel which calls the function.
+In SYCL 2020 mode, the attribute is not propagated to the kernel.
 
 .. code-block:: c++
 
@@ -2656,6 +2663,11 @@ allows the Y and Z arguments to be optional. If not provided by the user, the
 value of Y and Z defaults to 1. See section 5.8.1 Kernel Attributes for more
 details.
 
+In SYCL 1.2.1 mode, the ``intel::reqd_work_group_size``,
+``cl::reqd_work_group_size``, and ``sycl::reqd_work_group_size`` attributes are
+propagated from the function they are applied to onto the kernel which calls the
+function. In SYCL 2020 mode, the attributes are not propagated to the kernel.
+
 .. code-block:: c++
 
   [[sycl::reqd_work_group_size(4, 4, 4)]] void foo() {}
@@ -2800,8 +2812,10 @@ Applies to a device function/lambda function. Indicates the maximum dimensions
 of a work group. Values must be positive integers. This is similar to
 reqd_work_group_size, but allows work groups that are smaller or equal to the
 specified sizes.
-If ``intel::max_work_group_size`` is applied to a function called from a
-device kernel, the attribute is not ignored and it is propagated to the kernel.
+
+In SYCL 1.2.1 mode, the ``intel::max_work_group_size`` attribute is propagated
+from the function it is applied to onto the kernel which calls the function.
+In SYCL 2020 mode, the attribute is not propagated to the kernel.
 
 .. code-block:: c++
 
@@ -2832,8 +2846,10 @@ Applies to a device function/lambda function or function call operator (of a
 function object). Indicates the largest valid global work dimension that will be
 accepted when running the kernel on a device. Valid values are integers in a
 range of [0, 3].
-If ``intel::max_global_work_dim`` is applied to a function called from a
-device kernel, the attribute is not ignored and it is propagated to the kernel.
+
+In SYCL 1.2.1 mode, the ``intel::max_global_work_dim`` attribute is propagated
+from the function it is applied to onto the kernel which calls the function.
+In SYCL 2020 mode, the attribute is not propagated to the kernel.
 
 .. code-block:: c++
 
@@ -2890,6 +2906,10 @@ device operation, guiding the FPGA backend to insert the appropriate number of
 registers to break-up the combinational logic circuit, and thereby controlling
 the length of the longest combinational path.
 
+In SYCL 1.2.1 mode, the ``intel::scheduler_target_fmax_mhz`` attribute is
+propagated from the function it is applied to onto the kernel which calls the
+function. In SYCL 2020 mode, the attribute is not propagated to the kernel.
+
 .. code-block:: c++
 
   [[intel::scheduler_target_fmax_mhz(4)]] void foo() {}
@@ -2920,6 +2940,10 @@ function object). If 1, compiler doesn't use the global work offset values for
 the device function. Valid values are 0 and 1. If used without argument, value
 of 1 is set implicitly.
 
+In SYCL 1.2.1 mode, the ``intel::no_global_work_offset`` attribute is
+propagated from the function it is applied to onto the kernel which calls the
+function. In SYCL 2020 mode, the attribute is not propagated to the kernel.
+
 .. code-block:: c++
 
   [[intel::no_global_work_offset]]
@@ -4607,6 +4631,10 @@ the ``[[intel::named_sub_group_size(NAME)]]`` documentation for clarification.
 This attribute is mutually exclusive with ``[[intel::named_sub_group_size(NAME)]]``
 and ``[[intel::sycl_explicit_simd]]``.
 
+In SYCL 1.2.1 mode, the ``intel::reqd_sub_group_size`` attribute is propagated
+from the function it is applied to onto the kernel which calls the function.
+In SYCL 2020 mode, the attribute is not propagated to the kernel.
+
 In addition to device functions, the required sub-group size attribute may also
 be specified in the definition of a named functor object and lambda functions,
 as in the examples below:

clang/include/clang/Driver/Action.h

@@ -772,7 +772,14 @@ class FileTableTformJobAction : public JobAction {
 
 public:
   struct Tform {
-    enum Kind { EXTRACT, EXTRACT_DROP_TITLE, REPLACE, RENAME };
+    enum Kind {
+      EXTRACT,
+      EXTRACT_DROP_TITLE,
+      REPLACE,
+      REPLACE_CELL,
+      RENAME,
+      COPY_SINGLE_FILE
+    };
 
     Tform() = default;
     Tform(Kind K, std::initializer_list<StringRef> Args) : TheKind(K) {
@@ -794,10 +801,19 @@ class FileTableTformJobAction : public JobAction {
   // <To> from another file table passed as input to this action.
   void addReplaceColumnTform(StringRef From, StringRef To);
 
+  // Replaces a cell in this table with column title <ColumnName> and row <Row>
+  // with the file name passed as input to this action.
+  void addReplaceCellTform(StringRef ColumnName, int Row);
+
   // Renames a column with title <From> in this table with a column with title
   // <To> passed as input to this action.
   void addRenameColumnTform(StringRef From, StringRef To);
 
+  // Specifies that, instead of generating a new table, the transformation
+  // should copy the file at column <ColumnName> and row <Row> into the
+  // output file.
+  void addCopySingleFileTform(StringRef ColumnName, int Row);
+
   static bool classof(const Action *A) {
     return A->getKind() == FileTableTformJobClass;
   }
@@ -806,6 +822,9 @@ class FileTableTformJobAction : public JobAction {
 
 private:
   SmallVector<Tform, 2> Tforms; // transformation actions requested
+
+  // column to copy single file from if requested
+  std::string CopySingleFileColumnName;
 };
 
 class AppendFooterJobAction : public JobAction {

clang/lib/CodeGen/CGCall.cpp

@@ -2672,6 +2672,9 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
     unsigned FirstIRArg, NumIRArgs;
     std::tie(FirstIRArg, NumIRArgs) = IRFunctionArgs.getIRArgs(ArgNo);
 
+    if (Arg->hasAttr<SYCLAccessorReadonlyAttr>())
+      Fn->getArg(FirstIRArg)->addAttr(llvm::Attribute::ReadOnly);
+
     switch (ArgI.getKind()) {
     case ABIArgInfo::InAlloca: {
       assert(NumIRArgs == 0);

clang/lib/Driver/Action.cpp

@@ -507,11 +507,23 @@ void FileTableTformJobAction::addReplaceColumnTform(StringRef From,
   Tforms.emplace_back(Tform(Tform::REPLACE, {From, To}));
 }
 
+void FileTableTformJobAction::addReplaceCellTform(StringRef ColumnName,
+                                                  int Row) {
+  Tforms.emplace_back(
+      Tform(Tform::REPLACE_CELL, {ColumnName, std::to_string(Row)}));
+}
+
 void FileTableTformJobAction::addRenameColumnTform(StringRef From,
                                                    StringRef To) {
   Tforms.emplace_back(Tform(Tform::RENAME, {From, To}));
 }
 
+void FileTableTformJobAction::addCopySingleFileTform(StringRef ColumnName,
+                                                     int Row) {
+  Tforms.emplace_back(
+      Tform(Tform::COPY_SINGLE_FILE, {ColumnName, std::to_string(Row)}));
+}
+
 void AppendFooterJobAction::anchor() {}
 
 AppendFooterJobAction::AppendFooterJobAction(Action *Input, types::ID Type)

clang/lib/Driver/Driver.cpp

@@ -4378,33 +4378,33 @@ class OffloadingActionBuilder final {
         //         .--------------------------------------.
         //         |               PostLink               |
         //         .--------------------------------------.
-        //         [.n]                [+*]           [+*]
+        //         [+n]                [+*]            [+]
         //           |                   |              |
-        //           |          .-----------------.     |
-        //           |          | FileTableTform  |     |
-        //           |          | (extract "Code")|     |
-        //           |          .-----------------.     |
-        //           |                  [-]             |
+        //   .----------------. .-----------------.     |
+        //   | FileTableTform | | FileTableTform  |     |
+        //   | (copy "Code")  | | (extract "Code")|     |
+        //   .----------------. .-----------------.     |
+        //          [.]                 [-]             |
         //           |                   |              |
-        //           |                 [-*]             |
-        //    .-------------.   .-------------------.   |
-        //    |finalizeNVPTX|   |  SPIRVTranslator  |   |
-        //    .-------------.   .-------------------.   |
-        //           |              [-as]      [-!a]    |
+        //          [.]                [-*]             |
+        //   .---------------.  .-------------------.   |
+        //   | finalizeNVPTX |  |  SPIRVTranslator  |   |
+        //   .---------------.  .-------------------.   |
+        //          [.]             [-as]      [-!a]    |
         //           |                |          |      |
         //           |              [-s]         |      |
         //           |       .----------------.  |      |
         //           |       | BackendCompile |  |      |
         //           |       .----------------.  |      |
         //           |              [-s]         |      |
         //           |                |          |      |
-        //           |              [-a]      [-!a]    [+]
-        //           |              .--------------------.
-        //           |              |   FileTableTform   |
-        //           |              |  (replace "Code")  |
-        //           |              .--------------------.
-        //           |                          |
-        //         [.n]                       [+*]
+        //          [.]             [-a]      [-!a]    [+]
+        //          .------------------------------------.
+        //          |           FileTableTform           |
+        //          |          (replace "Code")          |
+        //          .------------------------------------.
+        //                            |
+        //                           [+]
         //         .--------------------------------------.
         //         |            OffloadWrapper            |
         //         .--------------------------------------.
@@ -4451,24 +4451,40 @@ class OffloadingActionBuilder final {
         ActionList WrapperInputs;
         // post link is not optional - even if not splitting, always need to
         // process specialization constants
-        types::ID PostLinkOutType =
-            isNVPTX || isAMDGCN ? types::TY_LLVM_BC : types::TY_Tempfiletable;
         auto *PostLinkAction = C.MakeAction<SYCLPostLinkJobAction>(
-            FullDeviceLinkAction, PostLinkOutType);
+            FullDeviceLinkAction, types::TY_Tempfiletable);
         PostLinkAction->setRTSetsSpecConstants(!isAOT);
 
-        if (isNVPTX) {
-          Action *FinAction =
-              finalizeNVPTXDependences(PostLinkAction, (*TC)->getTriple());
-          WrapperInputs.push_back(FinAction);
-        } else if (isAMDGCN) {
-          Action *FinAction =
-              finalizeAMDGCNDependences(PostLinkAction, (*TC)->getTriple());
-          WrapperInputs.push_back(FinAction);
+        constexpr char COL_CODE[] = "Code";
+
+        if (isNVPTX || isAMDGCN) {
+          // Make extraction copy the only remaining code file instead of
+          // creating a new table with a single entry.
+          // TODO: Process all PTX code files in file table to enable code
+          //       splitting for PTX target.
+          auto *ExtractIRFilesAction = C.MakeAction<FileTableTformJobAction>(
+              PostLinkAction, types::TY_LLVM_BC);
+          ExtractIRFilesAction->addCopySingleFileTform(COL_CODE, 0);
+
+          Action *FinAction;
+          if (isNVPTX) {
+            FinAction = finalizeNVPTXDependences(ExtractIRFilesAction,
+                                                 (*TC)->getTriple());
+          } else /* isAMDGCN */ {
+            FinAction = finalizeAMDGCNDependences(ExtractIRFilesAction,
+                                                  (*TC)->getTriple());
+          }
+          ActionList TformInputs{PostLinkAction, FinAction};
+
+          // Replace the only code entry in the table, as confirmed by the
+          // previous transformation.
+          auto *ReplaceFilesAction = C.MakeAction<FileTableTformJobAction>(
+              TformInputs, types::TY_Tempfiletable);
+          ReplaceFilesAction->addReplaceCellTform(COL_CODE, 0);
+          WrapperInputs.push_back(ReplaceFilesAction);
         } else {
           // For SPIRV-based targets - translate to SPIRV then optionally
           // compile ahead-of-time to native architecture
-          constexpr char COL_CODE[] = "Code";
           auto *ExtractIRFilesAction = C.MakeAction<FileTableTformJobAction>(
               PostLinkAction, types::TY_Tempfilelist);
           // single column w/o title fits TY_Tempfilelist format
@@ -4513,6 +4529,7 @@ class OffloadingActionBuilder final {
           ReplaceFilesAction->addReplaceColumnTform(COL_CODE, COL_CODE);
           WrapperInputs.push_back(ReplaceFilesAction);
         }
+
         // After the Link, wrap the files before the final host link
         auto *DeviceWrappingAction = C.MakeAction<OffloadWrapperJobAction>(
             WrapperInputs, types::TY_Object);
@@ -5619,11 +5636,9 @@ Action *Driver::ConstructPhaseAction(
         !Args.hasArg(options::OPT_fno_sycl_use_footer) &&
         TargetDeviceOffloadKind == Action::OFK_None) {
       // Performing a host compilation with -fsycl.  Append the integration
-      // footer to the preprocessed source file.  We then add another
-      // preprocessed step to complete the action chain.
-      auto *Preprocess = C.MakeAction<PreprocessJobAction>(Input, HostPPType);
+      // footer to the source file.
       auto *AppendFooter =
-          C.MakeAction<AppendFooterJobAction>(Preprocess, types::TY_CXX);
+          C.MakeAction<AppendFooterJobAction>(Input, types::TY_CXX);
       // FIXME: There are 2 issues with dependency generation in regards to
       // the integration footer that need to be addressed.
       // 1) Input file referenced on the RHS of a dependency is based on the