[SYCL][CUDA] Explicitly adding extension headers in tests.

SYCL/BFloat16/bfloat16_builtins.cpp

@@ -6,14 +6,14 @@
 //
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out -Xsycl-target-backend --cuda-gpu-arch=sm_80
 // RUN: %t.out
-
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
 #include <sycl/sycl.hpp>
 
 #include <cmath>
 #include <vector>
 
-using namespace cl::sycl;
-using sycl::ext::oneapi::experimental::bfloat16;
+using namespace sycl;
+using namespace sycl::ext::oneapi::experimental;
 
 constexpr int N = 60; // divisible by all tested array sizes
 constexpr float bf16_eps = 0.00390625;

SYCL/DeviceLib/built-ins/ext_native_math.cpp

@@ -7,8 +7,9 @@
 // OpenCL CPU driver does not support cl_khr_fp16 extension for this reason this
 // test is compiled with the -fsycl-device-code-split flag
 
-#include <CL/sycl.hpp>
 #include <cassert>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 template <typename T> void assert_out_of_bound(T val, T lower, T upper) {
   assert(sycl::all(lower < val && val < upper));

SYCL/Matrix/element_wise_all_ops_cuda.cpp

@@ -10,7 +10,7 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX=3 %s -o %t.out
 // RUN: %t.out
 
-#include <CL/sycl.hpp>
+#include <sycl/sycl.hpp>
 
 using namespace sycl;
 using namespace sycl::ext::oneapi::experimental::matrix;

SYCL/Matrix/element_wise_wi_marray.cpp

@@ -10,11 +10,12 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -Xsycl-target-backend --cuda-gpu-arch=sm_80 -DSYCL_EXT_ONEAPI_MATRIX=3 %s -o %t.out
 // RUN: %t.out
 
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
 #include <sycl/sycl.hpp>
 
 using namespace sycl;
+using namespace sycl::ext::oneapi::experimental;
 using namespace sycl::ext::oneapi::experimental::matrix;
-using sycl::ext::oneapi::experimental::bfloat16;
 
 #define SG_SZ 32
 

SYCL/Matrix/joint_matrix_tensorcore.cpp

@@ -7,11 +7,11 @@
 // supported for the Nvidia matrix extension, although some JIT optimizations
 // are performed at the level of the PTX assembly code.
 
-#include <CL/sycl.hpp>
+#include <sycl/sycl.hpp>
 
 using namespace sycl;
+using namespace sycl::ext::oneapi::experimental;
 using namespace sycl::ext::oneapi::experimental::matrix;
-using sycl::ext::oneapi::experimental::bfloat16;
 constexpr float bf16_eps = 0.00390625;
 
 // Example usage of Nvidia matrix multiply.

SYCL/Printf/char.cpp

@@ -25,7 +25,8 @@
 // CHECK: literal strings: s=Hello World!
 // CHECK_DISABLED: non-literal strings: s=Hello, World! ls=
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include <cstring>
 

SYCL/Printf/double.cpp

@@ -28,7 +28,8 @@
 
 #include <iostream>
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include "helper.hpp"
 

SYCL/Printf/float.cpp

@@ -31,7 +31,8 @@
 
 #include <iostream>
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include "helper.hpp"
 

SYCL/Printf/int.cpp

@@ -20,7 +20,8 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.constant.out %GPU_CHECK_PLACEHOLDER
 // RUN: %ACC_RUN_PLACEHOLDER %t.constant.out %ACC_CHECK_PLACEHOLDER
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include "helper.hpp"
 

SYCL/Printf/long.cpp

@@ -20,7 +20,8 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.constant.out %GPU_CHECK_PLACEHOLDER
 // RUN: %ACC_RUN_PLACEHOLDER %t.constant.out %ACC_CHECK_PLACEHOLDER
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include "helper.hpp"
 

SYCL/Printf/mixed-address-space.cpp

@@ -12,7 +12,8 @@
 // CHECK: Constant addrspace literal
 // CHECK: Generic addrspace literal
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include "helper.hpp"
 

SYCL/Printf/percent-symbol.cpp

@@ -23,7 +23,8 @@
 // CHECK: %c %s %d %i %o %x %X %u
 // CHECK-NEXT: %f %F %e %E %a %A %g %G %n %p
 
-#include <CL/sycl.hpp>
+#include <sycl/ext/oneapi/experimental/builtins.hpp>
+#include <sycl/sycl.hpp>
 
 #include <cstring>