intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-08 22:12:59 +08:00
Code Issues Packages Projects Releases Wiki Activity

test: use realistic values in topology query tests

Browse Source 
Related-To: NEO-9489

Signed-off-by: Wenbin Lu <wenbin.lu@intel.com>
This commit is contained in:
Wenbin Lu
2024-07-26 23:52:13 +00:00
committed by Compute-Runtime-Automation
parent 6796f41c77
commit e2f1735cc5
8 changed files with 295 additions and 110 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
level_zero/tools/test/unit_tests/sources/debug/debug_session_common.cpp
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2022-2023 Intel Corporation
* Copyright (C) 2022-2024 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -63,11 +63,22 @@ void initStateSaveArea(std::vector<char> &stateSaveArea, SIP::version version, L
fillRegsetForThread(&pStateSaveAreaHeader->regHeader.grf, 0, 0, 4, 0, 'a');
if (version.major < 2) {
// grfs for 0/3/7/3 - somewhere in the middle
fillRegsetForThread(&pStateSaveAreaHeader->regHeader.grf, 0, 3, 7, 3, 'a');
auto &hwInfo = *NEO::defaultHwInfo.get();
const uint32_t numSlices = hwInfo.gtSystemInfo.SliceCount > 2 ? 2 : hwInfo.gtSystemInfo.SliceCount;
const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.SubSliceCount / hwInfo.gtSystemInfo.SliceCount;
const uint32_t numEusPerSubslice = hwInfo.gtSystemInfo.EUCount / hwInfo.gtSystemInfo.SubSliceCount;
const uint32_t numThreadsPerEu = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.EUCount;
// grfs for 0/5/15/6 - very last eu thread
fillRegsetForThread(&pStateSaveAreaHeader->regHeader.grf, 0, 5, 15, 6, 'a');
const uint32_t midSlice = (numSlices > 1) ? (numSlices / 2) : 0;
const uint32_t midSubslice = (numSubslicesPerSlice > 1) ? (numSubslicesPerSlice / 2) : 0;
const uint32_t midEu = (numEusPerSubslice > 1) ? (numEusPerSubslice / 2) : 0;
const uint32_t midThread = (numThreadsPerEu > 1) ? (numThreadsPerEu / 2) : 0;
// grfs for an eu thread that is somewhere in the middle
fillRegsetForThread(&pStateSaveAreaHeader->regHeader.grf, midSlice, midSubslice, midEu, midThread, 'a');
// grfs for the very last eu thread
fillRegsetForThread(&pStateSaveAreaHeader->regHeader.grf, numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1, 'a');
}
}

98
level_zero/tools/test/unit_tests/sources/debug/linux/prelim/test_debug_api_linux.cpp
View File

@@ -132,16 +132,22 @@ TEST(DebugSessionLinuxi915Test, GivenDebugSessionWhenExtractingCpuVaFromUuidThen
TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithSingleSliceThenSubsliceIsCorrectlyRemapped) {
auto hwInfo = *NEO::defaultHwInfo.get();
const auto maxSubslicesPerSlice = hwInfo.gtSystemInfo.MaxSubSlicesSupported / hwInfo.gtSystemInfo.MaxSlicesSupported;
hwInfo.gtSystemInfo.SliceCount = 1;
hwInfo.gtSystemInfo.SubSliceCount = 8;
hwInfo.gtSystemInfo.SubSliceCount = hwInfo.gtSystemInfo.SliceCount * maxSubslicesPerSlice;
hwInfo.gtSystemInfo.DualSubSliceCount = hwInfo.gtSystemInfo.SliceCount * maxSubslicesPerSlice;
hwInfo.gtSystemInfo.EUCount = hwInfo.gtSystemInfo.SubSliceCount * hwInfo.gtSystemInfo.MaxEuPerSubSlice;
hwInfo.gtSystemInfo.MaxSlicesSupported = hwInfo.gtSystemInfo.SliceCount * 2;
hwInfo.gtSystemInfo.MaxSubSlicesSupported = hwInfo.gtSystemInfo.SubSliceCount * 2;
hwInfo.gtSystemInfo.MaxDualSubSlicesSupported = hwInfo.gtSystemInfo.DualSubSliceCount * 2;
NEO::MockDevice *neoDevice(NEO::MockDevice::createWithNewExecutionEnvironment<NEO::MockDevice>(&hwInfo, 0));
auto mockDrm = new DrmQueryMock(*neoDevice->executionEnvironment->rootDeviceEnvironments[0]);
mockDrm->storedSVal = 2; // slice 0 disabled in topology
mockDrm->storedSSVal = hwInfo.gtSystemInfo.SubSliceCount;
mockDrm->storedEUVal = hwInfo.gtSystemInfo.EUCount;
mockDrm->storedSSVal = mockDrm->storedSVal * maxSubslicesPerSlice;
mockDrm->storedEUVal = mockDrm->storedSSVal * hwInfo.gtSystemInfo.MaxEuPerSubSlice;
mockDrm->disableSomeTopology = true;
NEO::DrmQueryTopologyData topologyData = {};
@@ -157,13 +163,15 @@ TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithSingleSliceT
auto sessionMock = std::make_unique<MockDebugSessionLinuxi915>(zet_debug_config_t{0x1234}, &deviceImp, 10);
ASSERT_NE(nullptr, sessionMock);
ze_device_thread_t thread = {UINT32_MAX, 1, 0, 0};
// Only one SS will be active if maxSS == 2
EXPECT_GE(maxSubslicesPerSlice, 2u);
ze_device_thread_t thread = {UINT32_MAX, maxSubslicesPerSlice > 2u ? 1u : 0u, 0u, 0u};
uint32_t deviceIndex = 0;
auto physicalThread = sessionMock->convertToPhysicalWithinDevice(thread, deviceIndex);
EXPECT_EQ(1u, physicalThread.slice);
EXPECT_EQ(3u, physicalThread.subslice);
EXPECT_EQ(maxSubslicesPerSlice > 2u ? 2u : 1u, physicalThread.subslice);
EXPECT_EQ(0u, physicalThread.eu);
EXPECT_EQ(0u, physicalThread.thread);
@@ -171,7 +179,7 @@ TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithSingleSliceT
physicalThread = sessionMock->convertToPhysicalWithinDevice(thread, deviceIndex);
EXPECT_EQ(1u, physicalThread.slice);
EXPECT_EQ(3u, physicalThread.subslice);
EXPECT_EQ(maxSubslicesPerSlice > 2u ? 2u : 1u, physicalThread.subslice);
EXPECT_EQ(0u, physicalThread.eu);
EXPECT_EQ(0u, physicalThread.thread);
}
@@ -179,8 +187,13 @@ TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithSingleSliceT
TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithMultipleSlicesThenSubsliceIsNotRemapped) {
auto hwInfo = *NEO::defaultHwInfo.get();
hwInfo.gtSystemInfo.SliceCount = 8;
hwInfo.gtSystemInfo.SubSliceCount = 8;
hwInfo.gtSystemInfo.SliceCount = std::min(static_cast<uint32_t>(GT_MAX_SLICE), hwInfo.gtSystemInfo.SliceCount * 8);
hwInfo.gtSystemInfo.SubSliceCount = hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.DualSubSliceCount = hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.EUCount = hwInfo.gtSystemInfo.SubSliceCount * hwInfo.gtSystemInfo.MaxEuPerSubSlice;
hwInfo.gtSystemInfo.MaxSlicesSupported = hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.MaxSubSlicesSupported = hwInfo.gtSystemInfo.SubSliceCount;
hwInfo.gtSystemInfo.MaxDualSubSlicesSupported = hwInfo.gtSystemInfo.DualSubSliceCount;
NEO::MockDevice *neoDevice(NEO::MockDevice::createWithNewExecutionEnvironment<NEO::MockDevice>(&hwInfo, 0));
@@ -226,7 +239,14 @@ TEST(DebugSessionLinuxi915Test, WhenConvertingThreadIDsForDeviceWithMultipleSlic
TEST(DebugSessionLinuxi915Test, GivenDeviceWithSingleSliceWhenCallingAreRequestedThreadsStoppedForSliceAllThenCorrectValuesAreReturned) {
auto hwInfo = *NEO::defaultHwInfo.get();
hwInfo.gtSystemInfo.EUCount /= hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.ThreadCount /= hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.SubSliceCount /= hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.DualSubSliceCount /= hwInfo.gtSystemInfo.SliceCount;
hwInfo.gtSystemInfo.SliceCount = 1;
hwInfo.gtSystemInfo.MaxSubSlicesSupported /= hwInfo.gtSystemInfo.MaxSlicesSupported;
hwInfo.gtSystemInfo.MaxDualSubSlicesSupported /= hwInfo.gtSystemInfo.MaxSlicesSupported;
hwInfo.gtSystemInfo.MaxSlicesSupported = 1;
NEO::MockDevice *neoDevice(NEO::MockDevice::createWithNewExecutionEnvironment<NEO::MockDevice>(&hwInfo, 0));
@@ -7186,7 +7206,6 @@ TEST_F(DebugApiLinuxAsyncThreadTest, GivenInterruptedThreadsWhenNoAttentionEvent
struct DebugApiRegistersAccessFixture : public DebugApiLinuxPrelimFixture {
void setUp() {
hwInfo = *NEO::defaultHwInfo.get();
hwInfo.gtSystemInfo.SubSliceCount = 6 * hwInfo.gtSystemInfo.SliceCount;
DebugApiLinuxPrelimFixture::setUp(&hwInfo);
@@ -7304,6 +7323,16 @@ TEST_F(DebugApiRegistersAccessTest, givenReadRegistersCalledCorrectValuesRead) {
char grf[32] = {0};
char grfRef[32] = {0};
const uint32_t numSlices = hwInfo.gtSystemInfo.SliceCount > 2 ? 2 : hwInfo.gtSystemInfo.SliceCount;
const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.SubSliceCount / hwInfo.gtSystemInfo.SliceCount;
const uint32_t numEusPerSubslice = hwInfo.gtSystemInfo.EUCount / hwInfo.gtSystemInfo.SubSliceCount;
const uint32_t numThreadsPerEu = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.EUCount;
const uint32_t midSlice = (numSlices > 1) ? (numSlices / 2) : 0;
const uint32_t midSubslice = (numSubslicesPerSlice > 1) ? (numSubslicesPerSlice / 2) : 0;
const uint32_t midEu = (numEusPerSubslice > 1) ? (numEusPerSubslice / 2) : 0;
const uint32_t midThread = (numThreadsPerEu > 1) ? (numThreadsPerEu / 2) : 0;
session->ensureThreadStopped({0, 0, 0, 0});
for (uint32_t reg = 0; reg < 20; ++reg) {
memset(grfRef, 'a' + reg, 32);
@@ -7311,17 +7340,17 @@ TEST_F(DebugApiRegistersAccessTest, givenReadRegistersCalledCorrectValuesRead) {
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
}
session->ensureThreadStopped({0, 3, 7, 3});
session->ensureThreadStopped({midSlice, midSubslice, midEu, midThread});
for (uint32_t reg = 0; reg < 20; ++reg) {
memset(grfRef, 'a' + reg, 32);
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 3, 7, 3}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, reg, 1, grf));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {midSlice, midSubslice, midEu, midThread}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, reg, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
}
session->ensureThreadStopped({0, 5, 15, 6});
session->ensureThreadStopped({numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1});
for (uint32_t reg = 0; reg < 20; ++reg) {
memset(grfRef, 'a' + reg, 32);
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 5, 15, 6}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, reg, 1, grf));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, reg, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
}
}
@@ -7406,6 +7435,16 @@ TEST_F(DebugApiRegistersAccessTest, givenWriteRegistersCalledCorrectValuesRead)
char grf[32] = {0};
char grfRef[32] = {0};
const uint32_t numSlices = hwInfo.gtSystemInfo.SliceCount > 2 ? 2 : hwInfo.gtSystemInfo.SliceCount;
const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.SubSliceCount / hwInfo.gtSystemInfo.SliceCount;
const uint32_t numEusPerSubslice = hwInfo.gtSystemInfo.EUCount / hwInfo.gtSystemInfo.SubSliceCount;
const uint32_t numThreadsPerEu = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.EUCount;
const uint32_t midSlice = (numSlices > 1) ? (numSlices / 2) : 0;
const uint32_t midSubslice = (numSubslicesPerSlice > 1) ? (numSubslicesPerSlice / 2) : 0;
const uint32_t midEu = (numEusPerSubslice > 1) ? (numEusPerSubslice / 2) : 0;
const uint32_t midThread = (numThreadsPerEu > 1) ? (numThreadsPerEu / 2) : 0;
// grfs for 0/0/0/0 - very first eu thread
memset(grfRef, 'k', 32); // 'a' + 10, r10
session->ensureThreadStopped(stoppedThread);
@@ -7416,24 +7455,24 @@ TEST_F(DebugApiRegistersAccessTest, givenWriteRegistersCalledCorrectValuesRead)
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), stoppedThread, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
// grfs for 0/3/7/3 - somewhere in the middle
// grfs for an eu thread that is somewhere in the middle
memset(grfRef, 'k', 32);
session->ensureThreadStopped({0, 3, 7, 3});
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 3, 7, 3}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
session->ensureThreadStopped({midSlice, midSubslice, midEu, midThread});
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {midSlice, midSubslice, midEu, midThread}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
memset(grfRef, 'y', 32);
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugWriteRegisters(session->toHandle(), {0, 3, 7, 3}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grfRef));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 3, 7, 3}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugWriteRegisters(session->toHandle(), {midSlice, midSubslice, midEu, midThread}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grfRef));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {midSlice, midSubslice, midEu, midThread}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
// grfs for 0/5/15/6 - very last eu thread
// grfs for the very last eu thread
memset(grfRef, 'k', 32);
session->ensureThreadStopped({0, 5, 15, 6});
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 5, 15, 6}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
session->ensureThreadStopped({numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1});
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
memset(grfRef, 'z', 32);
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugWriteRegisters(session->toHandle(), {0, 5, 15, 6}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grfRef));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {0, 5, 15, 6}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugWriteRegisters(session->toHandle(), {numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grfRef));
EXPECT_EQ(ZE_RESULT_SUCCESS, zetDebugReadRegisters(session->toHandle(), {numSlices - 1, numSubslicesPerSlice - 1, numEusPerSubslice - 1, numThreadsPerEu - 1}, ZET_DEBUG_REGSET_TYPE_GRF_INTEL_GPU, 10, 1, grf));
EXPECT_EQ(0, memcmp(grf, grfRef, 32));
}
@@ -7771,7 +7810,18 @@ TEST_F(DebugApiRegistersAccessTest, GivenScratchPointerAndZeroAddressInSurfaceSt
session->ioctlHandler.reset(ioctlHandler);
session->vmHandle = 7;
session->clientHandleToConnection[MockDebugSessionLinuxi915::mockClientHandle]->vmToStateBaseAreaBindInfo[vmHandle] = {stateSaveAreaGpuVa + maxDbgSurfaceSize, sizeof(SbaTrackedAddresses)};
ze_device_thread_t thread{0, 3, 7, 3};
const uint32_t numSlices = hwInfo.gtSystemInfo.SliceCount > 2 ? 2 : hwInfo.gtSystemInfo.SliceCount;
const uint32_t numSubslicesPerSlice = hwInfo.gtSystemInfo.SubSliceCount / hwInfo.gtSystemInfo.SliceCount;
const uint32_t numEusPerSubslice = hwInfo.gtSystemInfo.EUCount / hwInfo.gtSystemInfo.SubSliceCount;
const uint32_t numThreadsPerEu = hwInfo.gtSystemInfo.ThreadCount / hwInfo.gtSystemInfo.EUCount;
const uint32_t midSlice = (numSlices > 1) ? (numSlices / 2) : 0;
const uint32_t midSubslice = (numSubslicesPerSlice > 1) ? (numSubslicesPerSlice / 2) : 0;
const uint32_t midEu = (numEusPerSubslice > 1) ? (numEusPerSubslice / 2) : 0;
const uint32_t midThread = (numThreadsPerEu > 1) ? (numThreadsPerEu / 2) : 0;
ze_device_thread_t thread{midSlice, midSubslice, midEu, midThread};
session->ensureThreadStopped(thread);
SbaTrackedAddresses sbaExpected{};