Merge branch 'sycl' into llvmspirv_pulldown51

Author: Fznamznon

CONTRIBUTING.md

@@ -81,6 +81,15 @@ approved by an Intel representative.
   - A new approval is needed if the PR was updated (e.g. during code review).
 - Once the PR is approved and all checks have passed, the pull request is
 ready for merge.
+- Sometimes unrelated fails can be observed in the PR. It's author
+responsibility to find/guess the reason of these fails and post a comment in
+the PR with:
+  - possible reason of fails, ideally with a link to the PR, which caused fails
+  - link to other PR(s), which expected to fix fails
+  - person who is currently looking into fails
+  - link to existing open issue
+  - if author cannot identify any of these, the minimal action expected is to
+    open a new [issue](/../../issues)
 
 ### Merge
 
@@ -89,7 +98,8 @@ and merge] and using PR description as the commit message, replacing all
 individual comments made per commit. Authors of the change must ensure PR
 description is up to date at the merge stage, as sometimes comments addressed
 during code reviews can invalidate original PR description. Feel free to ping
-@intel/llvm-gatekeepers if your PR is green and can be merged.
+@intel/llvm-gatekeepers if your PR is green and can be merged. Note that
+gatekeepers will require explanation for any failures observed.
 
 Pulldown from LLVM upstream is done through merge commits to preserve hashes of
 the original commits pulled from the LLVM community repository.

README.md

@@ -6,6 +6,8 @@ Intel LLVM-based projects:
 - [oneAPI DPC++ compiler](#oneapi-dpc-compiler)
 - [Late-outline OpenMP and OpenMP Offload](#late-outline-openmp-and-openmp-offload)
 
+For general contribution process see [CONTRIBUTING.md](./CONTRIBUTING.md)
+
 ## oneAPI DPC++ compiler
 
 [![](https://spec.oneapi.io/oneapi-logo-white-scaled.jpg)](https://www.oneapi.io/)

libclc/cmake/modules/AddLibclc.cmake

@@ -105,7 +105,7 @@ macro(add_libclc_builtin_set arch_suffix)
   install(
     FILES ${LIBCLC_LIBRARY_OUTPUT_INTDIR}/${obj_suffix}
     DESTINATION ${CMAKE_INSTALL_DATADIR}/clc )
-  
+
   # Generate remangled variants if requested
   if( LIBCLC_GENERATE_REMANGLED_VARIANTS )
     set(dummy_in "${CMAKE_BINARY_DIR}/lib/clc/libclc_dummy_in.cc")
@@ -149,6 +149,29 @@ macro(add_libclc_builtin_set arch_suffix)
           DESTINATION ${CMAKE_INSTALL_DATADIR}/clc )
       endforeach()
     endforeach()
+
+    # For remangler tests we do not care about long_width, or signedness, as it
+    # performs no substitutions.
+    # Collect all remangler tests in libclc-remangler-tests to later add
+    # dependency against check-libclc.
+    set(libclc-remangler-tests)
+    set(libclc-remangler-test-no 0)
+    set(libclc-remangler-target-ir
+         "$<TARGET_PROPERTY:opt.${obj_suffix},TARGET_FILE>"
+         "${LIBCLC_LIBRARY_OUTPUT_INTDIR}/builtins.link.${obj_suffix}"
+         "$<TARGET_PROPERTY:prepare-${obj_suffix},TARGET_FILE>")
+    foreach(target-ir ${libclc-remangler-target-ir})
+      math(EXPR libclc-remangler-test-no "${libclc-remangler-test-no}+1")
+      set(current-test "libclc-remangler-test-${obj_suffix}-${libclc-remangler-test-no}")
+      add_custom_target(${current-test}
+        COMMAND libclc-remangler
+        --long-width=l32
+        --char-signedness=signed
+        --input-ir=${target-ir}
+        ${dummy_in} -t -o -
+        DEPENDS "${builtins_obj_path}" "prepare-${obj_suffix}" "${dummy_in}" libclc-remangler)
+      list(APPEND libclc-remangler-tests ${current-test})
+    endforeach()
   endif()
 
   # nvptx-- targets don't include workitem builtins

libclc/test/CMakeLists.txt

@@ -45,25 +45,9 @@ foreach( t ${LIBCLC_TARGET_TO_TEST} )
 endforeach( t )
 
 if(LIBCLC_GENERATE_REMANGLED_VARIANTS)
-  # Run remangler in test mode if generating remangled variants and make sure
-  # it depends on check-libclc target.
-  # Both `long_widths` and `char_signedness` are set in AddLibclc.cmake and can
-  # be used here.
-  foreach(long_width ${long_widths})
-    foreach(signedness ${char_signedness})
-      # In `-t` (TestRun) the remangler does not perform any substitutions, it
-      # needs to make sure that the remangled name matches the original mangled
-      # one.
-      set (test_target_name "test-remangled-${long_width}-${signedness}_char")
-      add_custom_target(${test_target_name}
-        COMMAND libclc-remangler
-        --long-width=${long_width}
-        --char-signedness=${signedness}
-        --input-ir="$<TARGET_PROPERTY:prepare-${obj_suffix},TARGET_FILE>"
-        ${dummy_in} -t -o -
-        DEPENDS "${builtins_obj_path}" "prepare-${obj_suffix}" "${dummy_in}" libclc-remangler)
-
-      add_dependencies(check-libclc ${test_target_name})
-    endforeach()
+  #Now that check-libclc is defined make sure that all remangler targets depend
+  # on it.
+  foreach(remangler-test ${libclc-remangler-tests})
+    add_dependencies(check-libclc ${remangler-test})
   endforeach()
 endif()

sycl/doc/developer/ContributeToDPCPP.md

@@ -1,5 +1,9 @@
 # Contributing to DPC++
 
+## General guidelines
+
+Read [CONTRIBUTING.md](/CONTRIBUTING.md) first.
+
 ## Maintaining stable ABI/API
 
 All changes made to the DPC++ compiler and runtime library should generally

sycl/include/CL/__spirv/spirv_types.hpp

@@ -108,13 +108,22 @@ enum class GroupOperation : uint32_t {
   ExclusiveScan = 2
 };
 
+#if (SYCL_EXT_ONEAPI_MATRIX_VERSION > 1)
+enum class MatrixLayout : uint32_t {
+  RowMajor = 0,
+  ColumnMajor = 1,
+  Packed = 2,
+  Dynamic = 3
+};
+#else
 enum class MatrixLayout : uint32_t {
   RowMajor = 0,
   ColumnMajor = 1,
   PackedA = 2,
   PackedB = 3,
   Unused = 4
 };
+#endif
 
 enum class MatrixUse : uint32_t { MatrixA = 0, MatrixB = 1, Accumulator = 2 };
 

sycl/include/sycl/detail/defines.hpp

@@ -21,7 +21,7 @@
 #endif
 #endif
 
-#if __has_attribute(sycl_special_class)
+#if __has_attribute(sycl_special_class) && (defined __SYCL_DEVICE_ONLY__)
 #define __SYCL_SPECIAL_CLASS __attribute__((sycl_special_class))
 #else
 #define __SYCL_SPECIAL_CLASS

sycl/include/sycl/ext/oneapi/device_global/device_global.hpp

@@ -15,6 +15,7 @@
 #include <sycl/exception.hpp>
 #include <sycl/ext/oneapi/device_global/properties.hpp>
 #include <sycl/ext/oneapi/properties/properties.hpp>
+#include <sycl/pointers.hpp>
 
 #ifdef __SYCL_DEVICE_ONLY__
 #define __SYCL_HOST_NOT_SUPPORTED(Op)
@@ -42,9 +43,27 @@ struct HasArrowOperator<
 template <typename T, typename PropertyListT, typename = void>
 class device_global_base {
 protected:
-  T *usmptr;
-  T *get_ptr() noexcept { return usmptr; }
-  const T *get_ptr() const noexcept { return usmptr; }
+  using pointer_t = typename decorated_global_ptr<T>::pointer;
+  pointer_t usmptr;
+  pointer_t get_ptr() noexcept { return usmptr; }
+  const pointer_t get_ptr() const noexcept { return usmptr; }
+
+public:
+  template <access::decorated IsDecorated>
+  multi_ptr<T, access::address_space::global_space, IsDecorated>
+  get_multi_ptr() noexcept {
+    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
+    return multi_ptr<T, access::address_space::global_space, IsDecorated>{
+        get_ptr()};
+  }
+
+  template <access::decorated IsDecorated>
+  multi_ptr<const T, access::address_space::global_space, IsDecorated>
+  get_multi_ptr() const noexcept {
+    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
+    return multi_ptr<const T, access::address_space::global_space, IsDecorated>{
+        get_ptr()};
+  }
 };
 
 // Specialization of device_global base class for when device_image_scope is in
@@ -58,6 +77,23 @@ class device_global_base<
   T val{};
   T *get_ptr() noexcept { return &val; }
   const T *get_ptr() const noexcept { return &val; }
+
+public:
+  template <access::decorated IsDecorated>
+  multi_ptr<T, access::address_space::global_space, IsDecorated>
+  get_multi_ptr() noexcept {
+    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
+    return address_space_cast<access::address_space::global_space, IsDecorated,
+                              T>(this->get_ptr());
+  }
+
+  template <access::decorated IsDecorated>
+  multi_ptr<const T, access::address_space::global_space, IsDecorated>
+  get_multi_ptr() const noexcept {
+    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
+    return address_space_cast<access::address_space::global_space, IsDecorated,
+                              const T>(this->get_ptr());
+  }
 };
 } // namespace detail
 
@@ -113,22 +149,6 @@ class
   device_global &operator=(const device_global &) = delete;
   device_global &operator=(const device_global &&) = delete;
 
-  template <access::decorated IsDecorated>
-  multi_ptr<T, access::address_space::global_space, IsDecorated>
-  get_multi_ptr() noexcept {
-    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
-    return address_space_cast<access::address_space::global_space, IsDecorated>(
-        this->get_ptr());
-  }
-
-  template <access::decorated IsDecorated>
-  multi_ptr<const T, access::address_space::global_space, IsDecorated>
-  get_multi_ptr() const noexcept {
-    __SYCL_HOST_NOT_SUPPORTED("get_multi_ptr()")
-    return address_space_cast<access::address_space::global_space, IsDecorated,
-                              const T>(this->get_ptr());
-  }
-
   T &get() noexcept {
     __SYCL_HOST_NOT_SUPPORTED("get()")
     return *this->get_ptr();