Merge branch 'main' into yc/0814-buffer-native-handle-null-device-1

.github/workflows/cmake.yml

@@ -20,7 +20,7 @@ jobs:
         compiler: [{c: gcc, cxx: g++}]
         libbacktrace: ['-DVAL_USE_LIBBACKTRACE_BACKTRACE=OFF']
         pool_tracking: ['-DUMF_ENABLE_POOL_TRACKING=ON', '-DUMF_ENABLE_POOL_TRACKING=OFF']
-        latency_tracking: ['-DUMF_ENABLE_LATENCY_TRACKING=OFF']
+        latency_tracking: ['-DUR_ENABLE_LATENCY_HISTOGRAM=OFF']
         include:
           - os: 'ubuntu-22.04'
             build_type: Release
@@ -40,7 +40,7 @@ jobs:
           - os: 'ubuntu-22.04'
             build_type: Release
             compiler: {c: clang, cxx: clang++}
-            latency_tracking: '-DUMF_ENABLE_LATENCY_TRACKING=ON'
+            latency_tracking: '-DUR_ENABLE_LATENCY_HISTOGRAM=ON'
     runs-on: ${{ (matrix.os == 'ubuntu-22.04' && github.repository_owner == 'oneapi-src') && 'intel-ubuntu-22.04' || matrix.os }}
 
     steps:

.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"

cmake/helpers.cmake

@@ -70,6 +70,7 @@ function(add_ur_target_compile_options name)
         )
         if (CMAKE_BUILD_TYPE STREQUAL "Release")
             target_compile_definitions(${name} PRIVATE -D_FORTIFY_SOURCE=2)
+            target_compile_options(${name} PRIVATE -fvisibility=hidden)
         endif()
         if(UR_DEVELOPER_MODE)
             target_compile_options(${name} PRIVATE

cmake/match.py

@@ -1,4 +1,4 @@
-#!/usr/bin/env python
+#!/usr/bin/env python3
 
 # Copyright (C) 2023 Intel Corporation
 # Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
@@ -12,6 +12,8 @@
 # List of available special tags:
 # {{OPT}}    - makes content in the same line as the tag optional
 # {{IGNORE}} - ignores all content until the next successfully matched line or the end of the input
+# {{NONDETERMINISTIC}} - order of match rules isn't important - each (non OPT) input line is paired with a match line
+# in any order
 # Special tags are mutually exclusive and are expected to be located at the start of a line.
 #
 
@@ -20,15 +22,25 @@
 import re
 from enum import Enum
 
+## @brief print a sequence of lines
+def print_lines(lines, hint = None):
+    counter = 1
+    for l in lines:
+        hint_char = " "
+        if hint == counter - 1:
+            hint_char = ">"
+        print("{}{:4d}| {}".format(hint_char, counter, l.strip()))
+        counter += 1
+
 
 ## @brief print the whole content of input and match files
-def print_content(input_lines, match_lines, ignored_lines):
-    print("--- Input Lines " + "-" * 64)
-    print("".join(input_lines).strip())
-    print("--- Match Lines " + "-" * 64)
-    print("".join(match_lines).strip())
-    print("--- Ignored Lines " + "-" * 62)
-    print("".join(ignored_lines).strip())
+def print_content(input_lines, match_lines, ignored_lines, hint_input = None, hint_match = None):
+    print("------ Input Lines " + "-" * 61)
+    print_lines(input_lines, hint_input)
+    print("------ Match Lines " + "-" * 61)
+    print_lines(match_lines, hint_match)
+    print("------ Ignored Lines " + "-" * 59)
+    print_lines(ignored_lines)
     print("-" * 80)
 
 
@@ -39,6 +51,24 @@ def print_incorrect_match(match_line, present, expected):
     print("expected: " + expected)
 
 
+## @brief print missing match line
+def print_input_not_found(input_line, input):
+    print("Input line " + str(input_line) + " has no match line")
+    print("is:       " + input)
+
+
+## @brief print missing input line
+def print_match_not_found(match_line, input):
+    print("Match line " + str(match_line) + " has no input line")
+    print("is:       " + input)
+
+
+## @brief print general syntax error
+def print_error(text, match_line):
+    print("Line " + str(match_line) + " encountered an error")
+    print(text)
+
+
 ## @brief pattern matching script status values
 class Status(Enum):
     INPUT_END = 1
@@ -63,6 +93,7 @@ def check_status(input_lines, match_lines):
 class Tag(Enum):
     OPT = "{{OPT}}"         # makes the line optional
     IGNORE = "{{IGNORE}}"   # ignores all input until next match or end of input file
+    NONDETERMINISTIC = "{{NONDETERMINISTIC}}" # switches on "deterministic mode"
     COMMENT = "#"           # comment - line ignored
 
 
@@ -88,32 +119,53 @@ def main():
         )
 
     ignored_lines = []
+    matched_lines = set()
 
     input_idx = 0
     match_idx = 0
     tags_in_effect = []
+    deterministic_mode = False
     while True:
         # check file status
         status = check_status(input_lines[input_idx:], match_lines[match_idx:])
-        if (status == Status.INPUT_AND_MATCH_END) or (status == Status.MATCH_END and Tag.IGNORE in tags_in_effect):
-            # all lines matched or the last line in match file is an ignore tag
-            sys.exit(0)
-        elif status == Status.MATCH_END:
-            print_incorrect_match(match_idx + 1, input_lines[input_idx].strip(), "");
-            print_content(input_lines, match_lines, ignored_lines)
-            sys.exit(1)
-        elif status == Status.INPUT_END:
-        # If we get to the end of the input, but still have pending matches,
-        # then that's a failure unless all pending matches are optional -
-        # otherwise we're done
-            while match_idx < len(match_lines):
-                if not (match_lines[match_idx].startswith(Tag.OPT.value) or
-                        match_lines[match_idx].startswith(Tag.IGNORE.value)):
-                    print_incorrect_match(match_idx + 1, "", match_lines[match_idx]);
-                    print_content(input_lines, match_lines, ignored_lines)
+        if deterministic_mode:
+            if status == Status.INPUT_END:
+                # Convert the list of seen matches to the list of unseen matches
+                remaining_matches = set(range(len(match_lines))) - matched_lines
+                for m in remaining_matches:
+                    line = match_lines[m]
+                    if line.startswith(Tag.OPT.value) or line.startswith(Tag.NONDETERMINISTIC.value):
+                        continue
+                    print_match_not_found(m + 1, match_lines[m])
+                    print_content(input_lines, match_lines, ignored_lines, hint_match=m)
                     sys.exit(1)
-                match_idx += 1
-            sys.exit(0)
+
+                sys.exit(0)
+            elif status == Status.MATCH_END:
+                print_input_not_found(input_idx + 1, input_lines[input_idx])
+                print_content(input_lines, match_lines, ignored_lines, hint_input=input_idx)
+                sys.exit(1)
+        else:
+            if (status == Status.INPUT_AND_MATCH_END) or (status == Status.MATCH_END and Tag.IGNORE in tags_in_effect):
+                # all lines matched or the last line in match file is an ignore tag
+                sys.exit(0)
+            elif status == Status.MATCH_END:
+                print_incorrect_match(input_idx + 1, input_lines[input_idx].strip(), "")
+                print_content(input_lines, match_lines, ignored_lines, hint_input=input_idx)
+                sys.exit(1)
+            elif status == Status.INPUT_END:
+            # If we get to the end of the input, but still have pending matches,
+            # then that's a failure unless all pending matches are optional -
+            # otherwise we're done
+                while match_idx < len(match_lines):
+                    if not (match_lines[match_idx].startswith(Tag.OPT.value) or
+                            match_lines[match_idx].startswith(Tag.IGNORE.value) or
+                            match_lines[match_idx].startswith(Tag.NONDETERMINISTIC.value)):
+                        print_incorrect_match(match_idx + 1, "", match_lines[match_idx])
+                        print_content(input_lines, match_lines, ignored_lines, hint_match=match_idx)
+                        sys.exit(1)
+                    match_idx += 1
+                sys.exit(0)
 
         input_line = input_lines[input_idx].strip() if input_idx < len(input_lines) else ""
         match_line = match_lines[match_idx]
@@ -122,7 +174,15 @@ def main():
         if match_line.startswith(Tag.OPT.value):
             tags_in_effect.append(Tag.OPT)
             match_line = match_line[len(Tag.OPT.value):]
+        elif match_line.startswith(Tag.NONDETERMINISTIC.value) and not deterministic_mode:
+            deterministic_mode = True
+            match_idx = 0
+            input_idx = 0
+            continue
         elif match_line.startswith(Tag.IGNORE.value):
+            if deterministic_mode:
+                print_error(r"Can't use \{{IGNORE\}} in deterministic mode")
+                sys.exit(2)
             tags_in_effect.append(Tag.IGNORE)
             match_idx += 1
             continue # line with ignore tag should be skipped
@@ -137,20 +197,29 @@ def main():
                 pattern += part
 
         # match or process tags
-        if re.fullmatch(pattern, input_line):
-            input_idx += 1
-            match_idx += 1
-            tags_in_effect = []
-        elif Tag.OPT in tags_in_effect:
-            match_idx += 1
-            tags_in_effect.remove(Tag.OPT)
-        elif Tag.IGNORE in tags_in_effect:
-            ignored_lines.append(input_line + os.linesep)
-            input_idx += 1
+        if deterministic_mode:
+            if re.fullmatch(pattern, input_line) and match_idx not in matched_lines:
+                input_idx += 1
+                matched_lines.add(match_idx)
+                match_idx = 0
+                tags_in_effect = []
+            else:
+                match_idx += 1
         else:
-            print_incorrect_match(match_idx + 1, input_line, match_line.strip())
-            print_content(input_lines, match_lines, ignored_lines)
-            sys.exit(1)
+            if re.fullmatch(pattern, input_line):
+                input_idx += 1
+                match_idx += 1
+                tags_in_effect = []
+            elif Tag.OPT in tags_in_effect:
+                match_idx += 1
+                tags_in_effect.remove(Tag.OPT)
+            elif Tag.IGNORE in tags_in_effect:
+                ignored_lines.append(input_line + os.linesep)
+                input_idx += 1
+            else:
+                print_incorrect_match(match_idx + 1, input_line, match_line.strip())
+                print_content(input_lines, match_lines, ignored_lines, hint_match=match_idx, hint_input=input_idx)
+                sys.exit(1)
 
 
 if __name__ == "__main__":

examples/collector/CMakeLists.txt

@@ -17,6 +17,6 @@ target_link_libraries(${TARGET_NAME} PRIVATE ${TARGET_XPTI})
 target_include_directories(${TARGET_NAME} PRIVATE ${xpti_SOURCE_DIR}/include)
 
 if(MSVC)
-    target_compile_definitions(${TARGET_NAME} PRIVATE
-        XPTI_STATIC_LIBRARY XPTI_CALLBACK_API_EXPORTS)
+    target_compile_definitions(${TARGET_NAME} PRIVATE XPTI_STATIC_LIBRARY)
 endif()
+target_compile_definitions(${TARGET_NAME} PRIVATE XPTI_CALLBACK_API_EXPORTS)

include/ur_api.h

@@ -332,9 +332,17 @@ typedef enum ur_structure_type_t {
 #if defined(_WIN32)
 /// @brief Microsoft-specific dllexport storage-class attribute
 #define UR_APIEXPORT __declspec(dllexport)
+#endif // defined(_WIN32)
+#endif // UR_APIEXPORT
+
+///////////////////////////////////////////////////////////////////////////////
+#ifndef UR_APIEXPORT
+#if __GNUC__ >= 4
+/// @brief GCC-specific dllexport storage-class attribute
+#define UR_APIEXPORT __attribute__((visibility("default")))
 #else
 #define UR_APIEXPORT
-#endif // defined(_WIN32)
+#endif // __GNUC__ >= 4
 #endif // UR_APIEXPORT
 
 ///////////////////////////////////////////////////////////////////////////////

include/ur_print.hpp

@@ -17403,6 +17403,11 @@ inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct
     return os;
 }
 
+inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const ur_bool_t value) {
+    os << (value ? "true" : "false");
+    return os;
+}
+
 namespace ur::details {
 ///////////////////////////////////////////////////////////////////////////////
 // @brief Print pointer value

scripts/core/common.yml

@@ -39,6 +39,12 @@ desc: "Microsoft-specific dllexport storage-class attribute"
 condition: "defined(_WIN32)"
 name: $X_APIEXPORT
 value: __declspec(dllexport)
+--- #--------------------------------------------------------------------------
+type: macro
+desc: "GCC-specific dllexport storage-class attribute"
+condition: "__GNUC__ >= 4"
+name: $X_APIEXPORT
+value: __attribute__ ((visibility ("default")))
 altvalue: ""
 --- #--------------------------------------------------------------------------
 type: macro

scripts/templates/print.hpp.mako

@@ -411,6 +411,11 @@ inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct
 %endfor
 %endfor
 
+inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const ur_bool_t value) {
+    os << (value ? "true" : "false");
+    return os;
+}
+
 namespace ${x}::details {
 ///////////////////////////////////////////////////////////////////////////////
 // @brief Print pointer value

scripts/templates/trcddi.cpp.mako

@@ -45,15 +45,19 @@ namespace ur_tracing_layer
         ${th.make_pfncb_param_type(n, tags, obj)} params = { &${",&".join(th.make_param_lines(n, tags, obj, format=["name"]))} };
         uint64_t instance = getContext()->notify_begin(${th.make_func_etor(n, tags, obj)}, "${th.make_func_name(n, tags, obj)}", &params);
 
-        getContext()->logger.info("---> ${th.make_func_name(n, tags, obj)}");
+        auto &logger = getContext()->logger;
+
+        logger.info("---> ${th.make_func_name(n, tags, obj)}");
 
         ${x}_result_t result = ${th.make_pfn_name(n, tags, obj)}( ${", ".join(th.make_param_lines(n, tags, obj, format=["name"]))} );
 
         getContext()->notify_end(${th.make_func_etor(n, tags, obj)}, "${th.make_func_name(n, tags, obj)}", &params, &result, instance);
 
-        std::ostringstream args_str;
-        ur::extras::printFunctionParams(args_str, ${th.make_func_etor(n, tags, obj)}, &params);
-        getContext()->logger.info("({}) -> {};\n", args_str.str(), result);
+        if (logger.getLevel() <= logger::Level::INFO) {
+            std::ostringstream args_str;
+            ur::extras::printFunctionParams(args_str, ${th.make_func_etor(n, tags, obj)}, &params);
+            logger.info("({}) -> {};\n", args_str.str(), result);
+        }
 
         return result;
     }

source/adapters/cuda/device.cpp

@@ -57,12 +57,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
     return ReturnValue(4318u);
   }
   case UR_DEVICE_INFO_MAX_COMPUTE_UNITS: {
-    int ComputeUnits = 0;
-    UR_CHECK_ERROR(cuDeviceGetAttribute(
-        &ComputeUnits, CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT,
-        hDevice->get()));
-    detail::ur::assertion(ComputeUnits >= 0);
-    return ReturnValue(static_cast<uint32_t>(ComputeUnits));
+    return ReturnValue(hDevice->getNumComputeUnits());
   }
   case UR_DEVICE_INFO_MAX_WORK_ITEM_DIMENSIONS: {
     return ReturnValue(MaxWorkItemDimensions);