Merge remote-tracking branch 'origin/develop' into stream-k-initial-impl

test/grouped_gemm/grouped_gemm_fp16.cpp

-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <random>
-
-#include "profiler/profile_grouped_gemm_impl.hpp"
-
-namespace {
-
-using ADataType   = ck::half_t;
-using BDataType   = ck::half_t;
-using CDataType   = ck::half_t;
-using AccDataType = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <typename ALayout, typename BLayout, typename CLayout>
-bool TestGroupedGemm()
-{
-
-    std::mt19937 gen(19391);
-    std::uniform_int_distribution<> distrib(1, 10);
-    int group_count = distrib(gen);
-
-    // GEMM shape
-    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
-    std::vector<const void*> p_a, p_b;
-    std::vector<void*> p_c;
-
-    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideCs;
-
-    for(int i = 0; i < group_count; i++)
-    {
-        Ms.push_back(256 + 256 * distrib(gen));
-        Ns.push_back(256 + 256 * distrib(gen));
-        Ks.push_back(128 + 128 * distrib(gen));
-
-        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
-        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
-        StrideCs.push_back(std::is_same<Row, CLayout>::value ? Ns[i] : Ms[i]);
-    }
-
-    return ck::profiler::profile_grouped_gemm_impl<ADataType,
-                                                   BDataType,
-                                                   CDataType,
-                                                   AccDataType,
-                                                   ALayout,
-                                                   BLayout,
-                                                   CLayout>(
-        true, 1, false, 1, Ms, Ns, Ks, StrideAs, StrideBs, StrideCs);
-}
-
-} // anonymous namespace
-
-int main()
-{
-    bool res = true;
-
-    res = res && TestGroupedGemm<Row, Row, Row>();
-    res = res && TestGroupedGemm<Row, Col, Row>();
-    res = res && TestGroupedGemm<Col, Row, Row>();
-    res = res && TestGroupedGemm<Col, Col, Row>();
-
-    std::cout << "TestGroupedGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-
-    return res ? 0 : 1;
-}

test/grouped_gemm/test_grouped_gemm_splitk.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "gtest/gtest.h"
+#include "test_grouped_gemm_util.hpp"
+
+using F16 = ck::half_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using RRR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+using RRR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+const std::vector<int> KBATCH{1, 2, 3, 5, 8};
+
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_KN, RRR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_NK, RCR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_KN,
+                         RRR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_NK,
+                         RCR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+
+#include "test_grouped_gemm_ut_cases.inc"