Added a define to enable subgroup shuffling if supported by the device

src/utilities/compile.cpp

@@ -57,6 +57,11 @@ Program CompileFromSource(const std::string &source_string, const Precision prec
     header_string += "#define GLOBAL_MEM_FENCE 1\n";
   }
 
+  // For Intel GPUs with subgroup support, use subgroup shuffling.
+  if (device.IsGPU() && device.HasExtension(kKhronosIntelSubgroups)) {
+    header_string += "#define USE_SUBGROUP_SHUFFLING 1\n";
+  }
+
   // Optionally adds a translation header from OpenCL kernels to CUDA kernels
   #ifdef CUDA_API
     header_string +=

src/utilities/utilities.hpp

@@ -47,6 +47,7 @@ using double2 = std::complex<double>;
 // Khronos OpenCL extensions
 const std::string kKhronosAttributesAMD = "cl_amd_device_attribute_query";
 const std::string kKhronosAttributesNVIDIA = "cl_nv_device_attribute_query";
+const std::string kKhronosIntelSubgroups = "cl_intel_subgroups";
 
 // Catched an unknown error
 constexpr auto kUnknownError = -999;