Drop ICC support in code (#3279)

* Drop ICC from code

Fixes: #3278

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>

Author: bernhardmgruber

CMakePresets.json

@@ -22,7 +22,6 @@
         "CCCL_ENABLE_TESTING": false,
         "CCCL_ENABLE_EXAMPLES": false,
         "CCCL_ENABLE_C": false,
-        "CCCL_SUPPRESS_ICC_DEPRECATION_WARNING": true,
         "CCCL_SUPPRESS_MSVC2017_DEPRECATION_WARNING": true,
         "libcudacxx_ENABLE_INSTALL_RULES": true,
         "CUB_ENABLE_INSTALL_RULES": true,

cmake/CCCLBuildCompilerTargets.cmake

@@ -23,7 +23,6 @@ set(CMAKE_MSVC_DEBUG_INFORMATION_FORMAT Embedded)
 option(CCCL_ENABLE_EXCEPTIONS "Enable exceptions within CCCL libraries." ON)
 option(CCCL_ENABLE_RTTI "Enable RTTI within CCCL libraries." ON)
 option(CCCL_ENABLE_WERROR "Treat warnings as errors for CCCL targets." ON)
-option(CCCL_SUPPRESS_ICC_DEPRECATION_WARNING "Suppress Intel Compiler deprecation warnings" OFF)
 option(CCCL_SUPPRESS_MSVC2017_DEPRECATION_WARNING "Suppress Visual Studio 2017 deprecation warnings" OFF)
 
 function(cccl_build_compiler_interface interface_target cuda_compile_options cxx_compile_options compile_defs)
@@ -69,10 +68,6 @@ function(cccl_build_compiler_targets)
     list(APPEND cxx_compile_definitions "CCCL_DISABLE_RTTI")
   endif()
 
-  if (CCCL_SUPPRESS_ICC_DEPRECATION_WARNING)
-    list(APPEND cxx_compile_definitions "CCCL_SUPPRESS_ICC_DEPRECATION_WARNING")
-  endif()
-
   if (CCCL_SUPPRESS_MSVC2017_DEPRECATION_WARNING)
     list(APPEND cxx_compile_definitions "CCCL_SUPPRESS_MSVC2017_DEPRECATION_WARNING")
   endif()
@@ -160,16 +155,6 @@ function(cccl_build_compiler_targets)
     endif()
   endif()
 
-  if ("Intel" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
-    # Do not flush denormal floats to zero
-    append_option_if_available("-no-ftz" cxx_compile_options)
-    # Disable warning that inlining is inhibited by compiler thresholds.
-    append_option_if_available("-diag-disable=11074" cxx_compile_options)
-    append_option_if_available("-diag-disable=11076" cxx_compile_options)
-    # Disable warning about deprecated classic compiler
-    append_option_if_available("-diag-disable=10441" cxx_compile_options)
-  endif()
-
   cccl_build_compiler_interface(cccl.compiler_interface
     "${cuda_compile_options}"
     "${cxx_compile_options}"

cub/cub/detail/fast_modulo_division.cuh

@@ -109,8 +109,6 @@ multiply_extract_higher_bits(T value, R multiplier)
 {
   static_assert(supported_integral<T>::value, "unsupported type");
   static_assert(supported_integral<R>::value, "unsupported type");
-  _CCCL_DIAG_PUSH
-  _CCCL_DIAG_SUPPRESS_ICC(186) // pointless comparison of unsigned integer with zero
   _CCCL_IF_CONSTEXPR (_CCCL_TRAIT(::cuda::std::is_signed, T))
   {
     _CCCL_ASSERT(value >= 0, "value must be non-negative");
@@ -119,7 +117,6 @@ multiply_extract_higher_bits(T value, R multiplier)
   {
     _CCCL_ASSERT(multiplier >= 0, "multiplier must be non-negative");
   }
-  _CCCL_DIAG_POP
   static constexpr int NumBits = sizeof(DivisorType) * CHAR_BIT;
   using unsigned_t             = unsigned_implicit_prom_t<DivisorType>;
   using larger_t               = larger_unsigned_type_t<DivisorType>;

cub/cub/util_macro.cuh

@@ -112,12 +112,6 @@ _CCCL_DIAG_SUPPRESS_CLANG("-Wattributes")
 #  if !_CCCL_CUDA_COMPILER(NVHPC)
 _CCCL_DIAG_SUPPRESS_NVHPC(attribute_requires_external_linkage)
 #  endif // !_CCCL_CUDA_COMPILER(NVHPC)
-#  if _CCCL_COMPILER(ICC)
-#    pragma nv_diag_suppress 1407 // the "__visibility__" attribute can only appear on functions and
-                                  // variables with external linkage'
-#    pragma warning(disable : 1890) // the "__visibility__" attribute can only appear on functions and
-                                    // variables with external linkage'
-#  endif // _CCCL_COMPILER(ICC)
 #endif // !CUB_DISABLE_KERNEL_VISIBILITY_WARNING_SUPPRESSION
 
 #ifndef CUB_DEFINE_KERNEL_GETTER

cub/cub/util_type.cuh

@@ -82,9 +82,9 @@ CUB_NAMESPACE_BEGIN
 #    endif // !defined(__CUDACC_RTC_INT128__)
 #  else // !defined(__CUDACC_RTC__)
 #    if _CCCL_CUDACC_AT_LEAST(11, 5)
-#      if _CCCL_COMPILER(GCC) || _CCCL_COMPILER(CLANG) || _CCCL_COMPILER(ICC) || _CCCL_COMPILER(NVHPC)
+#      if _CCCL_COMPILER(GCC) || _CCCL_COMPILER(CLANG) || _CCCL_COMPILER(NVHPC)
 #        define CUB_IS_INT128_ENABLED 1
-#      endif // GCC || CLANG || ICC || NVHPC
+#      endif // GCC || CLANG || NVHPC
 #    endif // _CCCL_CUDACC_AT_LEAST(11, 5)
 #  endif // !defined(__CUDACC_RTC__)
 #endif // !defined(CUB_IS_INT128_ENABLED)

cub/test/catch2_test_device_for_each_in_extents.cu

@@ -55,7 +55,7 @@ static auto fill_linear_impl(c2h::host_vector<T>& vector, const ExtentType&, siz
   _CCCL_TRAILING_REQUIRES(void)((Rank == ExtentType::rank()))
 {
   vector[pos++] = {indices...};
-  return void(); // Intel and nvc++ require a return statement
+  return void(); // nvc++ requires a return statement
 }
 
 template <int Rank = 0, typename T, typename ExtentType, typename... IndicesType>
@@ -67,7 +67,7 @@ static auto fill_linear_impl(c2h::host_vector<T>& vector, const ExtentType& ext,
   {
     fill_linear_impl<Rank + 1>(vector, ext, pos, indices..., i);
   }
-  return void(); // Intel and nvc++ require a return statement
+  return void(); // nvc++ requires a return statement
 }
 
 template <typename T, typename IndexType, size_t... Extents>

cudax/include/cuda/experimental/__detail/config.cuh

@@ -35,15 +35,9 @@
 // two attributes:
 // - `_CUDAX_API` declares the function host/device and hides the symbol from the ABI
 // - `_CUDAX_TRIVIAL_API` does the same while also forcing inlining and hiding the function from debuggers
-#if _CCCL_COMPILER(ICC) // ICC has issues with visibility attributes on symbols with internal linkage
-#  define _CUDAX_API        _CCCL_HOST_DEVICE
-#  define _CUDAX_HOST_API   _CCCL_HOST
-#  define _CUDAX_DEVICE_API _CCCL_DEVICE
-#else // ^^^ _CCCL_COMPILER(ICC) ^^^ / vvv !_CCCL_COMPILER(ICC) vvv
-#  define _CUDAX_API        _CCCL_HOST_DEVICE _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
-#  define _CUDAX_HOST_API   _CCCL_HOST _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
-#  define _CUDAX_DEVICE_API _CCCL_DEVICE _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
-#endif // !_CCCL_COMPILER(ICC)
+#define _CUDAX_API        _CCCL_HOST_DEVICE _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
+#define _CUDAX_HOST_API   _CCCL_HOST _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
+#define _CUDAX_DEVICE_API _CCCL_DEVICE _CCCL_VISIBILITY_HIDDEN _CCCL_EXCLUDE_FROM_EXPLICIT_INSTANTIATION
 
 // _CUDAX_TRIVIAL_API force-inlines a function, marks its visibility as hidden, and causes debuggers to skip it.
 // This is useful for trivial internal functions that do dispatching or other plumbing work. It is particularly

docs/repo.toml

@@ -157,7 +157,6 @@ doxygen_predefined = [
     "_CCCL_DIAG_POP=",
     "_CCCL_DIAG_SUPPRESS_CLANG(x)=",
     "_CCCL_DIAG_SUPPRESS_GCC(x)=",
-    "_CCCL_DIAG_SUPPRESS_ICC(x)=",
     "_CCCL_DIAG_SUPPRESS_MSVC(x)=",
     "_CCCL_DIAG_SUPPRESS_NVHPC(x)=",
     "_CCCL_DOXYGEN_INVOKED",
@@ -270,7 +269,6 @@ doxygen_predefined = [
   "_CCCL_DIAG_POP=",
   "_CCCL_DIAG_SUPPRESS_CLANG(x)=",
   "_CCCL_DIAG_SUPPRESS_GCC(x)=",
-  "_CCCL_DIAG_SUPPRESS_ICC(x)=",
   "_CCCL_DIAG_SUPPRESS_MSVC(x)=",
   "_CCCL_DIAG_SUPPRESS_NVHPC(x)=",
   "CUDASTF_HOST=",
@@ -435,7 +433,6 @@ doxygen_predefined = [
   "_CCCL_DIAG_POP=",
   "_CCCL_DIAG_SUPPRESS_CLANG(x)=",
   "_CCCL_DIAG_SUPPRESS_GCC(x)=",
-  "_CCCL_DIAG_SUPPRESS_ICC(x)=",
   "_CCCL_DIAG_SUPPRESS_MSVC(x)=",
   "_CCCL_DIAG_SUPPRESS_NVHPC(x)=",
   "_CCCL_REQUIRES(x)= ::cuda::std::enable_if_t<x, int> = 0>",

libcudacxx/cmake/DetermineGCCCompatible.cmake

@@ -7,7 +7,5 @@ if(NOT DEFINED LLVM_COMPILER_IS_GCC_COMPATIBLE)
     set(LLVM_COMPILER_IS_GCC_COMPATIBLE OFF)
   elseif( "${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang" )
     set(LLVM_COMPILER_IS_GCC_COMPATIBLE ON)
-  elseif( "${CMAKE_CXX_COMPILER_ID}" MATCHES "Intel" )
-    set(LLVM_COMPILER_IS_GCC_COMPATIBLE ON)
   endif()
 endif()

libcudacxx/include/cuda/std/__cccl/assert.h

@@ -79,15 +79,9 @@ _CCCL_HOST_DEVICE
   __attribute__((__noreturn__));
 }
 #  endif // NDEBUG
-// ICC cannot deal with `__builtin_expect` in the constexpr evaluator, so just drop it
-#  if _CCCL_COMPILER(ICC)
-#    define _CCCL_ASSERT_IMPL_HOST(expression, message) \
-      static_cast<bool>(expression) ? (void) 0 : __assert_fail(message, __FILE__, __LINE__, __func__);
-#  else // ^^^ _CCCL_COMPILER(ICC) ^^^ / vvv !_CCCL_COMPILER(ICC) vvv
-#    define _CCCL_ASSERT_IMPL_HOST(expression, message)      \
-      _CCCL_BUILTIN_EXPECT(static_cast<bool>(expression), 1) \
-      ? (void) 0 : __assert_fail(message, __FILE__, __LINE__, __func__)
-#  endif // !_CCCL_COMPILER(ICC)
+#  define _CCCL_ASSERT_IMPL_HOST(expression, message)      \
+    _CCCL_BUILTIN_EXPECT(static_cast<bool>(expression), 1) \
+    ? (void) 0 : __assert_fail(message, __FILE__, __LINE__, __func__)
 #endif // !MSVC STL
 
 //! Use custom implementations with nvcc on device and the host ones with clang-cuda and nvhpc