Method to get attention bitmask from threads vector

Signed-off-by: Mateusz Hoppe <[email protected]>

Author: HoppeMateusz

level_zero/core/source/debugger/debugger_l0.cpp

@@ -167,6 +167,27 @@ void DebuggerL0::getAttentionBitmaskForThread(uint32_t slice, uint32_t subslice,
     }
 }
 
+void DebuggerL0::getAttentionBitmaskForSingleThreads(std::vector<ze_device_thread_t> &threads, const NEO::HardwareInfo &hwInfo, std::unique_ptr<uint8_t[]> &bitmask, size_t &bitmaskSize) {
+    const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.MaxSubSlicesSupported / hwInfo.gtSystemInfo.MaxSlicesSupported;
+    const uint32_t numEuPerSubslice = hwInfo.gtSystemInfo.MaxEuPerSubSlice;
+    const uint32_t numThreadsPerEu = (hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.EUCount);
+    const uint32_t bytesPerEu = alignUp(numThreadsPerEu, 8) / 8;
+    const uint32_t threadsSizePerSlice = numSubslicesPerSlice * numEuPerSubslice * bytesPerEu;
+
+    bitmaskSize = hwInfo.gtSystemInfo.MaxSubSlicesSupported * hwInfo.gtSystemInfo.MaxEuPerSubSlice * bytesPerEu;
+    bitmask = std::make_unique<uint8_t[]>(bitmaskSize);
+
+    memset(bitmask.get(), 0, bitmaskSize);
+
+    for (auto &thread : threads) {
+        uint8_t *sliceData = ptrOffset(bitmask.get(), threadsSizePerSlice * thread.slice);
+        uint8_t *subsliceData = ptrOffset(sliceData, numEuPerSubslice * bytesPerEu * thread.subslice);
+        uint8_t *euData = ptrOffset(subsliceData, bytesPerEu * thread.eu);
+        UNRECOVERABLE_IF(thread.thread > 7);
+        *euData |= (1 << thread.thread);
+    }
+}
+
 std::vector<ze_device_thread_t> DebuggerL0::getThreadsFromAttentionBitmask(const NEO::HardwareInfo &hwInfo, const uint8_t *bitmask, const size_t bitmaskSize) {
     const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.MaxSubSlicesSupported / hwInfo.gtSystemInfo.MaxSlicesSupported;
     const uint32_t numEuPerSubslice = hwInfo.gtSystemInfo.MaxEuPerSubSlice;

level_zero/core/source/debugger/debugger_l0.h

@@ -91,6 +91,7 @@ class DebuggerL0 : public NEO::Debugger, NEO::NonCopyableOrMovableClass {
     virtual void programSbaTrackingCommands(NEO::LinearStream &cmdStream, const SbaAddresses &sba) = 0;
 
     static void getAttentionBitmaskForThread(uint32_t slice, uint32_t subslice, uint32_t eu, uint32_t thread, const NEO::HardwareInfo &hwInfo, std::unique_ptr<uint8_t[]> &bitmask, size_t &bitmaskSize);
+    static void getAttentionBitmaskForSingleThreads(std::vector<ze_device_thread_t> &threads, const NEO::HardwareInfo &hwInfo, std::unique_ptr<uint8_t[]> &bitmask, size_t &bitmaskSize);
     static std::vector<ze_device_thread_t> getThreadsFromAttentionBitmask(const NEO::HardwareInfo &hwInfo, const uint8_t *bitmask, const size_t bitmaskSize);
 
   protected:

level_zero/core/test/unit_tests/sources/debugger/test_l0_debugger.cpp

@@ -14,6 +14,7 @@
 #include "shared/source/os_interface/os_context.h"
 #include "shared/test/common/cmd_parse/gen_cmd_parse.h"
 #include "shared/test/common/helpers/variable_backup.h"
+#include "shared/test/common/test_macros/matchers.h"
 
 #include "test.h"
 
@@ -1526,6 +1527,24 @@ TEST(DebuggerL0, givenAllSliceSubsliceEuAndThreadIdsWhenGettingBitmaskThenCorrec
     EXPECT_EQ(0, memcmp(bitmask.get(), expectedBitmask.get(), size));
 }
 
+TEST(DebuggerL0, givenSingleThreadsWhenGettingBitmaskThenCorrectBitsAreSet) {
+    auto hwInfo = *NEO::defaultHwInfo.get();
+    std::unique_ptr<uint8_t[]> bitmask;
+    size_t size = 0;
+
+    std::vector<ze_device_thread_t> threads;
+    threads.push_back({0, 0, 0, 3});
+    threads.push_back({0, 0, 1, 0});
+
+    DebuggerL0::getAttentionBitmaskForSingleThreads(threads, hwInfo, bitmask, size);
+
+    auto data = bitmask.get();
+    EXPECT_EQ(1u << 3, data[0]);
+    EXPECT_EQ(1u, data[1]);
+
+    EXPECT_THAT(&data[2], MemoryZeroed(size - 2));
+}
+
 TEST(DebuggerL0, givenBitmaskWithAttentionBitsForSingleThreadWhenGettingThreadsThenSingleCorrectThreadReturned) {
     auto hwInfo = *NEO::defaultHwInfo.get();
     std::unique_ptr<uint8_t[]> bitmask;

opencl/test/unit_test/command_stream/command_stream_receiver_tests.cpp

@@ -24,14 +24,14 @@
 #include "shared/test/common/mocks/mock_execution_environment.h"
 #include "shared/test/common/mocks/mock_graphics_allocation.h"
 #include "shared/test/common/mocks/ult_device_factory.h"
+#include "shared/test/common/test_macros/matchers.h"
 #include "shared/test/common/test_macros/test_checks_shared.h"
 
 #include "opencl/source/command_stream/definitions/command_stream_receiver_simulated_hw.h"
 #include "opencl/source/mem_obj/buffer.h"
 #include "opencl/source/platform/platform.h"
 #include "opencl/test/unit_test/fixtures/cl_device_fixture.h"
 #include "opencl/test/unit_test/fixtures/multi_root_device_fixture.h"
-#include "opencl/test/unit_test/gen_common/matchers.h"
 #include "opencl/test/unit_test/helpers/raii_hw_helper.h"
 #include "opencl/test/unit_test/mocks/mock_allocation_properties.h"
 #include "opencl/test/unit_test/mocks/mock_buffer.h"

opencl/test/unit_test/device_queue/device_queue_tests.cpp

@@ -6,10 +6,10 @@
  */
 
 #include "shared/source/device/device_info.h"
+#include "shared/test/common/test_macros/matchers.h"
 
 #include "opencl/source/helpers/dispatch_info.h"
 #include "opencl/test/unit_test/fixtures/device_host_queue_fixture.h"
-#include "opencl/test/unit_test/gen_common/matchers.h"
 #include "opencl/test/unit_test/mocks/mock_context.h"
 #include "opencl/test/unit_test/mocks/mock_kernel.h"
 #include "opencl/test/unit_test/mocks/mock_program.h"

opencl/test/unit_test/gen_common/CMakeLists.txt

@@ -9,7 +9,6 @@ set(IGDRCL_SRCS_tests_gen_common
     ${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_declare.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_exclude.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/gen_commands_common_validation.h
-    ${CMAKE_CURRENT_SOURCE_DIR}/matchers.h
     ${NEO_SHARED_TEST_DIRECTORY}/common/test_macros/header${BRANCH_DIR_SUFFIX}/test.h
 )
 target_sources(igdrcl_tests PRIVATE ${IGDRCL_SRCS_tests_gen_common})

opencl/test/unit_test/kernel/kernel_reflection_surface_tests.cpp

@@ -6,6 +6,7 @@
  */
 
 #include "shared/test/common/helpers/debug_manager_state_restore.h"
+#include "shared/test/common/test_macros/matchers.h"
 
 #include "opencl/source/execution_model/device_enqueue.h"
 #include "opencl/source/kernel/kernel.h"
@@ -15,7 +16,6 @@
 #include "opencl/test/unit_test/fixtures/execution_model_kernel_fixture.h"
 #include "opencl/test/unit_test/fixtures/image_fixture.h"
 #include "opencl/test/unit_test/fixtures/multi_root_device_fixture.h"
-#include "opencl/test/unit_test/gen_common/matchers.h"
 #include "opencl/test/unit_test/helpers/gtest_helpers.h"
 #include "opencl/test/unit_test/mocks/mock_allocation_properties.h"
 #include "opencl/test/unit_test/mocks/mock_context.h"

opencl/test/unit_test/mem_obj/buffer_tests.cpp

@@ -15,13 +15,13 @@
 #include "shared/test/common/mocks/mock_device.h"
 #include "shared/test/common/mocks/mock_execution_environment.h"
 #include "shared/test/common/mocks/ult_device_factory.h"
+#include "shared/test/common/test_macros/matchers.h"
 
 #include "opencl/extensions/public/cl_ext_private.h"
 #include "opencl/source/command_queue/command_queue_hw.h"
 #include "opencl/test/unit_test/fixtures/cl_device_fixture.h"
 #include "opencl/test/unit_test/fixtures/memory_management_fixture.h"
 #include "opencl/test/unit_test/fixtures/multi_root_device_fixture.h"
-#include "opencl/test/unit_test/gen_common/matchers.h"
 #include "opencl/test/unit_test/mocks/mock_allocation_properties.h"
 #include "opencl/test/unit_test/mocks/mock_buffer.h"
 #include "opencl/test/unit_test/mocks/mock_command_queue.h"

shared/test/common/test_macros/CMakeLists.txt

@@ -7,6 +7,7 @@
 set(NEO_CORE_test_macros
     ${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
     ${CMAKE_CURRENT_SOURCE_DIR}/header${BRANCH_DIR_SUFFIX}/test.h
+    ${CMAKE_CURRENT_SOURCE_DIR}/matchers.h
     ${CMAKE_CURRENT_SOURCE_DIR}/test_checks_shared.cpp
     ${CMAKE_CURRENT_SOURCE_DIR}/test_checks_shared.h
 )