Merge branch 'main' into compilation_fixes_for_sles

.github/workflows/e2e_core.yml

@@ -169,10 +169,6 @@ jobs:
         -DCMAKE_CXX_COMPILER="$(which clang++)"
         -DLLVM_LIT="${{github.workspace}}/sycl-repo/llvm/utils/lit/lit.py"
 
-    - name: Set LIT_XFAIL_NOT
-      if: inputs.xfail_not != ''
-      run: echo "LIT_XFAIL_NOT=${{inputs.xfail_not}}" >> $GITHUB_ENV
-
     - name: Set LIT_XFAIL
       if: inputs.xfail != ''
       run: echo "LIT_XFAIL=${{inputs.xfail}}" >> $GITHUB_ENV
@@ -181,6 +177,10 @@ jobs:
       if: inputs.filter_out != ''
       run: echo "LIT_FILTER_OUT=${{inputs.filter_out}}" >> $GITHUB_ENV
 
+    - name: Set LIT_XFAIL_NOT
+      if: inputs.xfail_not != ''
+      run: echo "LIT_XFAIL_NOT=${{inputs.xfail_not}}" >> $GITHUB_ENV
+
     # TODO: remove once intel/llvm lit tests can properly recognize the GPU
     - name: Configure hardware platform feature for L0
       if: matrix.adapter.name == 'L0'

.github/workflows/e2e_cuda.yml

@@ -21,3 +21,4 @@ jobs:
       config: "--cuda"
       unit: "gpu"
       extra_lit_flags: "-sv --max-time=3600"
+      xfail: "Regression/device_num.cpp"

.github/workflows/e2e_level_zero.yml

@@ -21,9 +21,11 @@ jobs:
       config: ""
       unit: "gpu"
       # Failing tests
-      xfail: "DeviceCodeSplit/grf.cpp;ESIMD/grf.cpp;ESIMD/mask_expand_load.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;  SYCL :: ESIMD/mask_expand_load.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_out_bounds.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_prefetch.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_rowmajorA_rowmajorB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/joint_matrix_bf16_fill_k_cache_prefetch.cpp;Matrix/joint_matrix_out_bounds.cpp;Matrix/joint_matrix_unaligned_k.cpp"
+      xfail: "DeviceCodeSplit/grf.cpp;ESIMD/mask_expand_load.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;ESIMD/mask_expand_load.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_out_bounds.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_prefetch.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_rowmajorA_rowmajorB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/joint_matrix_bf16_fill_k_cache_prefetch.cpp;Matrix/joint_matrix_out_bounds.cpp;Matrix/joint_matrix_unaligned_k.cpp;Matrix/SPVCooperativeMatrix/SG32/get_coord_int8_matB.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_1d.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_1d_cont.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_scalar.cpp;Matrix/SPVCooperativeMatrix/element_wise_ops.cpp;Matrix/SPVCooperativeMatrix/get_coord_int8_matB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_SLM.cpp;Matrix/joint_matrix_bf16_fill_k_cache_SLM.cpp"
+      # Unexpectedly Passed Tests
+      xfail_not: ""
       # Flaky tests
-      filter_out: "ESIMD/named_barriers/loop_extended.cpp;ESIMD/local_accessor_copy_to_from.cpp;"
+      filter_out: "Basic/accessor/accessor.cpp|DeviceArchitecture/device_architecture_comparison_on_device_aot.cpp|Graph/Explicit/interop-level-zero-launch-kernel.cpp|Graph/RecordReplay/interop-level-zero-launch-kernel.cpp|syclcompat/launch/launch_policy_lmem.cpp"
       # These runners by default spawn upwards of 260 workers.
       # We also add a time out just in case some test hangs
       extra_lit_flags: "--param gpu-intel-pvc=True --param gpu-intel-pvc-1T=True -sv -j 100 --max-time=3600"

.github/workflows/e2e_opencl.yml

@@ -20,5 +20,5 @@ jobs:
       prefix: ""
       config: ""
       unit: "cpu"
-      xfail: "AOT/double.cpp;AOT/half.cpp;AOT/reqd-sg-size.cpp;Basic/built-ins/marray_geometric.cpp;KernelCompiler/kernel_compiler_spirv.cpp;KernelCompiler/opencl_queries.cpp"
+      xfail: "AOT/double.cpp;AOT/half.cpp;AOT/reqd-sg-size.cpp;Basic/built-ins/marray_geometric.cpp;KernelCompiler/kernel_compiler_spirv.cpp;KernelCompiler/opencl_queries.cpp;NonUniformGroups/ballot_group.cpp;NonUniformGroups/ballot_group_algorithms.cpp;NonUniformGroups/fixed_size_group_algorithms.cpp;NonUniformGroups/opportunistic_group.cpp;NonUniformGroups/opportunistic_group_algorithms.cpp;NonUniformGroups/tangle_group.cpp;NonUniformGroups/tangle_group_algorithms.cpp"
       extra_lit_flags: "-sv --max-time=3600"

source/adapters/native_cpu/device.cpp

@@ -364,11 +364,23 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   case UR_DEVICE_INFO_MEMORY_CLOCK_RATE:
   case UR_DEVICE_INFO_MEMORY_BUS_WIDTH:
     return UR_RESULT_ERROR_INVALID_VALUE;
+  case UR_DEVICE_INFO_ATOMIC_FENCE_ORDER_CAPABILITIES: {
+    // Currently for Native CPU fences are implemented using OCK
+    // builtins, so we have different capabilities than atomic operations
+    ur_memory_order_capability_flags_t Capabilities =
+        UR_MEMORY_ORDER_CAPABILITY_FLAG_RELAXED |
+        UR_MEMORY_ORDER_CAPABILITY_FLAG_ACQUIRE |
+        UR_MEMORY_ORDER_CAPABILITY_FLAG_RELEASE |
+        UR_MEMORY_ORDER_CAPABILITY_FLAG_ACQ_REL |
+        UR_MEMORY_ORDER_CAPABILITY_FLAG_SEQ_CST;
+    return ReturnValue(Capabilities);
+  }
   case UR_DEVICE_INFO_ATOMIC_MEMORY_ORDER_CAPABILITIES: {
     ur_memory_order_capability_flags_t Capabilities =
         UR_MEMORY_ORDER_CAPABILITY_FLAG_RELAXED;
     return ReturnValue(Capabilities);
   }
+  case UR_DEVICE_INFO_ATOMIC_FENCE_SCOPE_CAPABILITIES:
   case UR_DEVICE_INFO_ATOMIC_MEMORY_SCOPE_CAPABILITIES: {
     uint64_t Capabilities = UR_MEMORY_SCOPE_CAPABILITY_FLAG_WORK_ITEM |
                             UR_MEMORY_SCOPE_CAPABILITY_FLAG_SUB_GROUP |

source/adapters/opencl/device.cpp

@@ -450,15 +450,16 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
       URValue[i].type = static_cast<ur_device_partition_t>(CLValue[0]);
       switch (URValue[i].type) {
       case UR_DEVICE_PARTITION_EQUALLY: {
-        URValue[i].value.equally = CLValue[i + 1];
+        URValue[i].value.equally = static_cast<uint32_t>(CLValue[i + 1]);
         break;
       }
       case UR_DEVICE_PARTITION_BY_COUNTS: {
-        URValue[i].value.count = CLValue[i + 1];
+        URValue[i].value.count = static_cast<uint32_t>(CLValue[i + 1]);
         break;
       }
       case UR_DEVICE_PARTITION_BY_AFFINITY_DOMAIN: {
-        URValue[i].value.affinity_domain = CLValue[i + 1];
+        URValue[i].value.affinity_domain =
+            static_cast<uint32_t>(CLValue[i + 1]);
         break;
       }
       default: {

source/adapters/opencl/kernel.cpp

@@ -302,7 +302,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urKernelSetExecInfo(
 
   switch (propName) {
   case UR_KERNEL_EXEC_INFO_USM_INDIRECT_ACCESS: {
-    if (*(static_cast<const ur_bool_t *>(pPropValue)) == true) {
+    if (*(static_cast<const ur_bool_t *>(pPropValue))) {
       UR_RETURN_ON_FAILURE(usmSetIndirectAccess(hKernel));
     }
     return UR_RESULT_SUCCESS;

test/conformance/enqueue/urEnqueueKernelLaunch.cpp

@@ -3,6 +3,7 @@
 // See LICENSE.TXT
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+#include <array>
 #include <uur/fixtures.h>
 
 struct urEnqueueKernelLaunchTest : uur::urKernelExecutionTest {

test/conformance/exp_command_buffer/buffer_saxpy_kernel_update.cpp

@@ -4,6 +4,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 #include "fixtures.h"
+#include <array>
 
 // Test that updating a command-buffer with a single kernel command
 // taking buffer & scalar arguments works correctly.

test/conformance/exp_command_buffer/commands.cpp

@@ -4,6 +4,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 #include "fixtures.h"
+#include <array>
 
 struct urCommandBufferCommandsTest
     : uur::command_buffer::urCommandBufferExpTest {

test/conformance/exp_command_buffer/ndrange_update.cpp

@@ -4,6 +4,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 #include "fixtures.h"
+#include <array>
 #include <cstring>
 
 // Test that updating a command-buffer with a single kernel command

test/conformance/exp_command_buffer/usm_fill_kernel_update.cpp

@@ -4,6 +4,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 #include "fixtures.h"
+#include <array>
 #include <cstring>
 
 // Test that updating a command-buffer with a single kernel command

test/conformance/exp_command_buffer/usm_saxpy_kernel_update.cpp

@@ -4,6 +4,7 @@
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 #include "fixtures.h"
+#include <array>
 #include <cstring>
 
 // Test that updating a command-buffer with a single kernel command

test/conformance/kernel/urKernelGetGroupInfo.cpp

@@ -3,6 +3,7 @@
 // See LICENSE.TXT
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
+#include <array>
 #include <uur/fixtures.h>
 
 using urKernelGetGroupInfoTest =