Merge remote-tracking branch 'origin/intel' into instr-device-code

Author: Artem Gindinson

.github/CODEOWNERS

@@ -29,7 +29,7 @@ SYCL/DeviceLib/ITTAnnotations @vzakhari @MrSidims @AGindinson
 SYCL/DotProduct @rdeodhar
 
 # Explicit SIMD
-SYCL/ESIMD @kbobrovs @DenisBakhvalov
+SYCL/ESIMD @kbobrovs @v-klochkov
 
 # Functor
 SYCL/Functor @AlexeySachkov

SYCL/AOT/accelerator.cpp

@@ -8,5 +8,5 @@
 
 // REQUIRES: aoc, accelerator
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_fpga-unknown-unknown-sycldevice %S/Inputs/aot.cpp -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_fpga %S/Inputs/aot.cpp -o %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out

SYCL/AOT/cpu.cpp

@@ -8,8 +8,8 @@
 
 // REQUIRES: opencl-aot, cpu
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %S/Inputs/aot.cpp -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64 %S/Inputs/aot.cpp -o %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 
 // Test that opencl-aot can handle multiple build options.
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice  %S/Inputs/aot.cpp -Xsycl-target-backend "--bo=-g" -Xsycl-target-backend "--bo=-cl-opt-disable" -o %t2.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64  %S/Inputs/aot.cpp -Xsycl-target-backend "--bo=-g" -Xsycl-target-backend "--bo=-cl-opt-disable" -o %t2.out

SYCL/AOT/gpu.cpp

@@ -10,5 +10,5 @@
 // UNSUPPORTED: cuda
 // CUDA is not compatible with SPIR.
 //
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen-unknown-unknown-sycldevice -Xsycl-target-backend=spir64_gen-unknown-unknown-sycldevice "-device *" %S/Inputs/aot.cpp -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_gen -Xsycl-target-backend=spir64_gen "-device *" %S/Inputs/aot.cpp -o %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out

SYCL/AOT/multiple-devices.cpp

@@ -12,7 +12,7 @@
 
 // 1-command compilation case
 // Targeting CPU, GPU, FPGA
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice,spir64_gen-unknown-unknown-sycldevice,spir64_fpga-unknown-unknown-sycldevice -Xsycl-target-backend=spir64_gen-unknown-unknown-sycldevice "-device *" %S/Inputs/aot.cpp -o %t_all.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64,spir64_gen,spir64_fpga -Xsycl-target-backend=spir64_gen "-device *" %S/Inputs/aot.cpp -o %t_all.out
 // RUN: %HOST_RUN_PLACEHOLDER %t_all.out
 // RUN: %CPU_RUN_PLACEHOLDER %t_all.out
 // RUN: %GPU_RUN_PLACEHOLDER %t_all.out
@@ -26,7 +26,7 @@
 // Produce object file, spirv, device images to combine these differently
 // at link-time, thus testing various AOT-compiled images configurations
 // RUN: %clangxx -fsycl %S/Inputs/aot.cpp -c -o %t.o
-// RUN: %clangxx -fsycl -fsycl-link-targets=spir64-unknown-unknown-sycldevice %t.o -o %t.spv
+// RUN: %clangxx -fsycl -fsycl-link-targets=spir64 %t.o -o %t.spv
 // AOT-compile device binary images
 // Neither of AOT tools can compile several files, hence, here is this
 // workaround

SYCL/AOT/spec_const_aot.cpp

@@ -1,6 +1,6 @@
 // REQUIRES: opencl-aot, cpu
 //
-// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64-unknown-unknown-sycldevice %s -o %t.out
+// RUN: %clangxx -fsycl -fsycl-targets=spir64_x86_64 %s -o %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 //
 // The test checks that the specialization constant feature works with ahead

SYCL/AOT/with-llvm-bc.cpp

@@ -8,8 +8,8 @@
 
 // REQUIRES: cpu, dump_ir
 
-// RUN: %clangxx -fsycl -fsycl-targets=spir64-unknown-unknown-sycldevice -c %S/Inputs/aot.cpp -o %t.o
-// RUN: %clangxx -fsycl -fsycl-link-targets=spir64-unknown-unknown-sycldevice %t.o -o %t.spv
+// RUN: %clangxx -fsycl -fsycl-targets=spir64 -c %S/Inputs/aot.cpp -o %t.o
+// RUN: %clangxx -fsycl -fsycl-link-targets=spir64 %t.o -o %t.spv
 // RUN: llvm-spirv -r %t.spv -o %t.bc
 // RUN: %clangxx -fsycl -fsycl-add-targets=spir64:%t.bc %t.o -o %t.out
 //

SYCL/Assert/assert_in_simultaneously_multiple_tus.cpp

@@ -1,9 +1,5 @@
 // FIXME unsuppoerted on CUDA until fallback libdevice becomes available
 // UNSUPPORTED: cuda
-// FIXME disable the test on Linux on OpenCL on FPGA as there's some funny
-// failure taking place in this configuration on CI. The test should be enabled
-// with this configuration once it's debugged.
-// UNSUPPORTED: linux && opencl && fpga
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -I %S/Inputs %s %S/Inputs/kernels_in_file2.cpp -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt

SYCL/Assert/assert_in_simultaneously_multiple_tus_one_ndebug.cpp

@@ -1,9 +1,5 @@
 // FIXME unsuppoerted on CUDA until fallback libdevice becomes available
 // UNSUPPORTED: cuda
-// FIXME disable the test on Linux on OpenCL on FPGA as there's some funny
-// failure taking place in this configuration on CI. The test should be enabled
-// with this configuration once it's debugged.
-// UNSUPPORTED: linux && opencl && fpga
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple -DDEFINE_NDEBUG_INFILE2 -I %S/Inputs %S/assert_in_simultaneously_multiple_tus.cpp %S/Inputs/kernels_in_file2.cpp -o %t.out %threads_lib
 // RUN: %CPU_RUN_PLACEHOLDER %t.out &> %t.txt || true
 // RUN: %CPU_RUN_PLACEHOLDER FileCheck %s --input-file %t.txt

SYCL/AtomicRef/add.cpp

@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-unnamed-lambda -fsycl-targets=%sycl_triple %s -o %t.out \
-// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda-sycldevice --cuda-gpu-arch=sm_60
+// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out

SYCL/AtomicRef/add_atomic64.cpp

@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-unnamed-lambda -fsycl-targets=%sycl_triple %s -o %t.out \
-// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda-sycldevice --cuda-gpu-arch=sm_60
+// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out

SYCL/AtomicRef/sub.cpp

@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-unnamed-lambda -fsycl-targets=%sycl_triple %s -o %t.out \
-// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda-sycldevice --cuda-gpu-arch=sm_60
+// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out

SYCL/AtomicRef/sub_atomic64.cpp

@@ -1,5 +1,5 @@
 // RUN: %clangxx -fsycl -fsycl-unnamed-lambda -fsycl-targets=%sycl_triple %s -o %t.out \
-// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda-sycldevice --cuda-gpu-arch=sm_60
+// RUN: -Xsycl-target-backend=nvptx64-nvidia-cuda --cuda-gpu-arch=sm_60
 // RUN: %HOST_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %CPU_RUN_PLACEHOLDER %t.out

SYCL/BFloat16/bfloat16_type.cpp

@@ -1,4 +1,4 @@
-// UNSUPPORTED: cuda
+// UNSUPPORTED: cuda || hip
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // TODO currently the feature isn't supported on most of the devices
 //      need to enable the test when the aspect and device_if feature are

SYCL/Basic/barrier_order.cpp

@@ -3,7 +3,7 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 
-// UNSUPPORTED: cuda
+// UNSUPPORTED: cuda || hip
 
 #include <CL/sycl.hpp>
 #include <stdlib.h>

SYCL/Basic/buffer/buffer_full_copy.cpp

@@ -5,6 +5,9 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t2.out
 // RUN: %GPU_RUN_PLACEHOLDER %t2.out
 // RUN: %ACC_RUN_PLACEHOLDER %t2.out
+//
+// Incorrect results with hip
+// XFAIL: hip
 
 //==------------- buffer_full_copy.cpp - SYCL buffer basic test ------------==//
 //

SYCL/Basic/buffer/subbuffer.cpp

@@ -4,6 +4,8 @@
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
 //
+// Test failing with hip on AMD
+// XFAIL: hip_amd
 //==---------- subbuffer.cpp --- sub-buffer basic test ---------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

SYCL/Basic/device_code_dae.cpp

@@ -1,7 +1,7 @@
 // NOTE A temporary test before this compilation flow is enabled by default in
 // driver
-// UNSUPPORTED: cuda
-// CUDA does not support SPIR-V.
+// UNSUPPORTED: cuda || hip
+// CUDA and HIP don't support SPIR-V.
 // RUN: %clangxx -fsycl-device-only -Xclang -fenable-sycl-dae -Xclang -fsycl-int-header=int_header.h %s -c -o device_code.bc -Wno-sycl-strict
 // RUN: %clangxx %cxx_std_optionc++17 %include_option int_header.h %debug_option -c %s -o host_code.o %sycl_options -Wno-sycl-strict
 // FIXME Added explicit offline linking step here until online-linking support

SYCL/Basic/device_event.cpp

@@ -4,6 +4,9 @@
 // RUN: %ACC_RUN_PLACEHOLDER %t.run
 // TODO: nd_item::barrier() is not implemented on HOST
 // RUNx: %HOST_RUN_PLACEHOLDER %t.run
+//
+// Crashes on AMD
+// XFAIL: hip_amd
 
 //==--------device_event.cpp - SYCL class device_event test ----------------==//
 //

SYCL/Basic/enqueue_barrier.cpp

@@ -5,7 +5,7 @@
 
 // The test is failing sporadically on Windows OpenCL RTs
 // Disabling on windows until fixed
-// UNSUPPORTED: cuda || windows
+// UNSUPPORTED: cuda || windows || hip
 
 #include <CL/sycl.hpp>
 #include <sycl/ext/intel/fpga_device_selector.hpp>

SYCL/Basic/free_function_queries/free_function_queries_sub_group.cpp

@@ -1,5 +1,5 @@
-// UNSUPPORTED: cuda
-// CUDA compilation and runtime do not yet support sub-groups.
+// UNSUPPORTED: cuda || hip
+// CUDA and HIP compilation and runtime do not yet support sub-groups.
 //
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: %HOST_RUN_PLACEHOLDER %t.out

SYCL/Basic/get_backend.cpp

@@ -1,6 +1,9 @@
 // RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
 // RUN: env SYCL_DEVICE_FILTER=%sycl_be %t.out
 //
+// Failing on HIP AMD
+// XFAIL: hip_amd
+//
 //==----------------- get_backend.cpp ------------------------==//
 // This is a test of get_backend().
 // Do not set SYCL_DEVICE_FILTER. We do not want the preferred

SYCL/Basic/group_async_copy.cpp

@@ -3,6 +3,9 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.run
 // RUN: %ACC_RUN_PLACEHOLDER %t.run
 // RUN: env SYCL_DEVICE_FILTER=host %t.run
+//
+// Crashes on AMD
+// XFAIL: hip_amd
 
 #include <CL/sycl.hpp>
 #include <iostream>

SYCL/Basic/handler/handler_set_args.cpp

@@ -2,6 +2,9 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
+//
+// Memory access fault on AMD
+// XFAIL: hip_amd
 //==--------------- handler_set_args.cpp -------------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.

SYCL/Basic/host_platform_avail.cpp

@@ -2,7 +2,7 @@
 // RUN: %HOST_RUN_PLACEHOLDER %t1.out
 // RUN: %CPU_RUN_PLACEHOLDER %t1.out
 // RUN: %GPU_RUN_PLACEHOLDER %t1.out
-// RUN: %ACC_RUN_PLACEHOLDER %t1.out
+// RUN: env SYCL_DEVICE_FILTER=acc,host %t1.out
 
 //==------ host_platform_avail.cpp - Host Platform Availability test -------==//
 //

SYCL/Basic/image/image_accessor_range.cpp

@@ -0,0 +1,121 @@
+// UNSUPPORTED: cuda || hip
+// CUDA does not support SYCL 1.2.1 images.
+//
+// RUN: %clangxx -fsycl -fsycl-targets=%sycl_triple %s -o %t.out
+// RUN: %HOST_RUN_PLACEHOLDER %t.out
+// RUN: %CPU_RUN_PLACEHOLDER %t.out
+// RUN: %GPU_RUN_PLACEHOLDER %t.out
+
+// XFAIL: level_zero
+
+#include <CL/sycl.hpp>
+#include <CL/sycl/accessor.hpp>
+#include <iostream>
+using namespace sycl;
+#define N 4   // dimensin
+#define M 128 // dimension
+#define C 4   // 4 channel
+#define L 2   // 2 images
+
+void try_1D(queue &Q) {
+  int X = -55;
+  buffer<int, 1> BX{&X, 1};
+  int *host_array = new int[M * C];
+  image im1(host_array, image_channel_order::rgba,
+            image_channel_type::unsigned_int8, range{M});
+
+  Q.submit([&](handler &h) {
+    accessor<int4, 1, access::mode::read, access::target::image> acs1(im1, h);
+    accessor ABX{BX, h};
+    auto R = acs1.get_range();
+    std::cout << "Host acs1.get_range()=" << R[0] << "\n";
+    assert(R[0] == M);
+    h.parallel_for(nd_range{range{M}, range{N}}, [=](nd_item<1> it) {
+      int idx = it.get_global_linear_id();
+      if (idx == 0) {
+        auto R = acs1.get_range();
+        ABX[0] = R[0];
+      }
+    });
+  });
+  {
+    host_accessor HABX{BX, read_only};
+    std::cout << "From Device acs1.get_range()=" << X << "\n";
+    assert(X == M);
+  }
+}
+
+void try_2D(queue &Q) {
+  range<2> X = {55, 66};
+  buffer<range<2>, 1> BX{&X, 1};
+  int *host_array = new int[M * N * C];
+  image im2(host_array, image_channel_order::rgba,
+            image_channel_type::unsigned_int8, range{M, N});
+
+  Q.submit([&](handler &h) {
+    accessor<int4, 2, access::mode::read, access::target::image> acs2(im2, h);
+    accessor ABX{BX, h};
+    auto R = acs2.get_range();
+    std::cout << "Host acs2.get_range()=" << R[0] << "," << R[1] << "\n";
+    assert(R[0] == M);
+    assert(R[1] == N);
+    h.parallel_for(nd_range{range{M, N}, range{N, N}}, [=](nd_item<2> it) {
+      int idx = it.get_global_linear_id();
+      if (idx == 0) {
+        ABX[0] = acs2.get_range();
+      }
+    });
+  });
+  {
+    host_accessor HABX{BX, read_only};
+    std::cout << "From Device acs2.get_range()=" << HABX[0][0] << ","
+              << HABX[0][1] << "\n";
+    assert(HABX[0][0] == M);
+    assert(HABX[0][1] == N);
+  }
+}
+
+void try_3D(queue &Q) {
+  range<3> X{55, 66, 77};
+  buffer<range<3>, 1> BX{&X, 1};
+  int *host_array3_2 = malloc_host<int>(N * M * C * L, Q);
+  image im3(host_array3_2, image_channel_order::rgba,
+            image_channel_type::unsigned_int8, range{M, N, L});
+
+  Q.submit([&](handler &h) {
+    accessor<int4, 2, access::mode::read, access::target::image_array> acs3(im3,
+                                                                            h);
+    accessor ABX{BX, h};
+    auto R = acs3.get_range();
+    std::cout << "Host acs3.get_range()=" << R[0] << "," << R[1] << "," << R[2]
+              << "\n";
+    assert(R[0] == M);
+    assert(R[1] == N);
+    assert(R[2] == L);
+    h.parallel_for(nd_range{range{M, N, L}, range{N, N, L}},
+                   [=](nd_item<3> it) {
+                     int idx = it.get_global_linear_id();
+                     if (idx == 0) {
+                       ABX[0] = acs3.get_range();
+                     }
+                   });
+  });
+  {
+    host_accessor HABX{BX, read_only};
+    std::cout << "From Device acs3.get_range()=" << HABX[0][0] << ","
+              << HABX[0][1] << "," << HABX[0][2] << "\n";
+    assert(HABX[0][0] == M);
+    assert(HABX[0][1] == N);
+    assert(HABX[0][2] == L);
+  }
+}
+
+int main() {
+  queue Q;
+
+  try_1D(Q);
+  try_2D(Q);
+  try_3D(Q);
+
+  return 0;
+}

SYCL/Basic/image/image_read.h

@@ -6,8 +6,6 @@
  * the OpenCL 1.2 specification, Section 8. Image Addressing and Filtering
  *
  * https://www.khronos.org/registry/OpenCL/specs/opencl-1.2.pdf#page=329
- *
- * Confirmed to pass with -fsycl-targets=spir64-unknown-linux-sycldevice
  */
 
 #include <CL/sycl.hpp>

SYCL/Basic/image/image_write.h

@@ -6,8 +6,6 @@
  * the OpenCL 1.2 specification, Section 8. Image Addressing and Filtering
  *
  * https://www.khronos.org/registry/OpenCL/specs/opencl-1.2.pdf#page=329
- *
- * Confirmed to pass with -fsycl-targets=spir64-unknown-linux-sycldevice
  */
 
 #include <CL/sycl.hpp>

SYCL/Basic/info.cpp

@@ -3,6 +3,9 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
+//
+// Hits a terminate on HIP AMD
+// XFAIL: hip_amd
 
 //==----------------info.cpp - SYCL objects get_info() test ----------------==//
 //

SYCL/Basic/kernel_info.cpp

@@ -5,6 +5,9 @@
 //
 // Fail is flaky for level_zero, enable when fixed.
 // UNSUPPORTED: level_zero
+//
+// Failing on HIP AMD
+// XFAIL: hip_amd
 
 //==--- kernel_info.cpp - SYCL kernel info test ----------------------------==//
 //

SYCL/Basic/linear-sub_group.cpp

@@ -3,6 +3,9 @@
 // RUN: %CPU_RUN_PLACEHOLDER %t.out
 // RUN: %GPU_RUN_PLACEHOLDER %t.out
 // RUN: %ACC_RUN_PLACEHOLDER %t.out
+//
+// Missing built-ins on AMD
+// XFAIL: hip_amd
 //==--------------- linear-sub_group.cpp - SYCL linear id test -------------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.