mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-06-30 11:47:48 +00:00
295 lines
14 KiB
C++
295 lines
14 KiB
C++
// This test is designed to verify that the ConvBuilder can instantiate the same
|
|
// kernel classes that are used in production code. Production code may have
|
|
// hundreds or thousands of kernel instances, so this test uses a GTest typed
|
|
// test suite to efficiently test a representative set of these kernel examples.
|
|
// Each test case defines a specific convolution algorithm configuration and the
|
|
// expected kernel type string that the builder should generate.
|
|
#include <gtest/gtest.h>
|
|
|
|
#include <ck_tile/builder/conv_builder.hpp>
|
|
|
|
namespace {
|
|
|
|
namespace ckb = ck_tile::builder;
|
|
using P = ckb::BlockGemmPipelineVersion;
|
|
|
|
// Defines the signature of the convolution operation to be tested.
|
|
// This includes dimensionality, direction, data layout, and data type.
|
|
struct ConvSignature
|
|
{
|
|
int spatial_dim = 2;
|
|
ckb::ConvDirection direction = ckb::ConvDirection::FORWARD;
|
|
ckb::GroupConvLayout layout = ckb::GroupConvLayout::CHANNELS_LAST;
|
|
ckb::DataType data_type = ckb::DataType::FP16;
|
|
};
|
|
static_assert(ckb::ConvSignatureDescriptor<ConvSignature>);
|
|
|
|
// Defines the tunable algorithmic parameters for the convolution kernel.
|
|
// This includes thread block configuration, tuning parameters, data transfer
|
|
// settings, and the GEMM pipeline version.
|
|
struct FwdConvAlgorithm
|
|
{
|
|
ckb::ThreadBlock thread_block;
|
|
ckb::ConvTuningParams tuning_params;
|
|
struct BlockTransfer
|
|
{
|
|
ckb::BlockATransferLengths thread_cluster_dims_a;
|
|
ckb::BlockBTransferLengths thread_cluster_dims_b;
|
|
ckb::BlockCTransferLengths thread_cluster_dims_c;
|
|
} block_transfer;
|
|
ckb::BlockGemmPipelineVersion pipeline_version;
|
|
};
|
|
static_assert(ckb::ConvAlgorithmDescriptor<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesThreadBlock<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesConvTuning<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesBlockATransfer<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesBlockBTransfer<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesBlockCTransfer<FwdConvAlgorithm>);
|
|
static_assert(ckb::SpecifiesGemmPipelineVersion<FwdConvAlgorithm>);
|
|
|
|
// A container for a single test case, bundling a descriptive name, the
|
|
// algorithm configuration, and the expected generated kernel type string.
|
|
struct TestCase
|
|
{
|
|
std::string_view name;
|
|
FwdConvAlgorithm algorithm;
|
|
std::string_view expected_type;
|
|
};
|
|
|
|
// Helper function to set the sub_matrix size.
|
|
constexpr ckb::ThreadBlock set_submatrix(int m, int n, int k)
|
|
{
|
|
return {.block_size = 256, .submatrix = {.m = m, .n = n, .k = k}};
|
|
}
|
|
|
|
// Helper function to set the thread cluster dimensions.
|
|
constexpr FwdConvAlgorithm::BlockTransfer set_thread_cluster_dims(int k0, int m, int k1)
|
|
{
|
|
return {.thread_cluster_dims_a = {.k0 = k0, .m = m, .k1 = k1},
|
|
.thread_cluster_dims_b = {.k0 = k0, .n = m, .k1 = k1},
|
|
.thread_cluster_dims_c = {
|
|
.m_block = 1, .m_wave_per_xdl = 32, .n_block = 1, .n_wave_per_xdl = 8}};
|
|
}
|
|
|
|
// An array of test cases that drive the typed test suite. Each entry
|
|
// represents a unique kernel instance to be verified.
|
|
constexpr std::array TEST_CASES = {
|
|
TestCase{
|
|
// double rate mfma instances on gfx950
|
|
.name = "ConvFwdXdlBf16CompInstances2x_0",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(256, 128, 64),
|
|
.tuning_params = {.ak1 = 16, .bk1 = 16, .m_xdl_per_wave = 2, .n_xdl_per_wave = 2},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V4},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 256, 128, 64, Default, 32, 32, "
|
|
"2, 2, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v4>",
|
|
},
|
|
TestCase{
|
|
// Compute-friendly.
|
|
.name = "GroupedConvFwdXdlBf16CompInstance0",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(256, 256, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 4, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V4},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 256, 256, 32, Default, 32, 32, "
|
|
"4, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v4>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance1",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(128, 128, 64),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 2},
|
|
.block_transfer = set_thread_cluster_dims(8, 32, 1),
|
|
.pipeline_version = P::V4},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 128, 128, 64, Default, 32, 32, "
|
|
"2, 2, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v4>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance2",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(128, 128, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 2},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V4},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 128, 128, 32, Default, 32, 32, "
|
|
"2, 2, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v4>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance3",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(256, 256, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 4, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V3},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 256, 256, 32, Default, 32, 32, "
|
|
"4, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v3>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance4",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(256, 256, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 4, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V5},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 256, 256, 32, Default, 32, 32, "
|
|
"4, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v5>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance5",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(256, 128, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V1},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 256, 128, 32, Default, 32, 32, "
|
|
"2, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v1>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance7",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(128, 256, 32),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V1},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 128, 256, 32, Default, 32, 32, "
|
|
"2, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v1>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance8",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(128, 128, 64),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V1},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 128, 128, 64, Default, 32, 32, "
|
|
"2, 4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v1>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance9",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(128, 64, 64),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V3},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 128, 64, 64, Default, 32, 32, 2, "
|
|
"4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v3>",
|
|
},
|
|
TestCase{
|
|
.name = "GroupedConvFwdXdlBf16CompInstance9",
|
|
.algorithm =
|
|
{.thread_block = set_submatrix(64, 128, 64),
|
|
.tuning_params = {.ak1 = 8, .bk1 = 8, .m_xdl_per_wave = 2, .n_xdl_per_wave = 4},
|
|
.block_transfer = set_thread_cluster_dims(4, 64, 1),
|
|
.pipeline_version = P::V3},
|
|
.expected_type =
|
|
"DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle_V3<256, 64, 128, 64, Default, 32, 32, 2, "
|
|
"4, 8, 8, 8, 1, 1, BlkGemmPipelineScheduler: Intrawave, BlkGemmPipelineVersion: v3>",
|
|
},
|
|
};
|
|
|
|
static constexpr int NUM_TEST_CASES = std::tuple_size_v<decltype(TEST_CASES)>;
|
|
|
|
// Helper to generate testing::Types<TestIndex<0>, TestIndex<1>, ..., TestIndex<N-1>>.
|
|
template <int N>
|
|
struct TestingIndices
|
|
{
|
|
template <int INDEX>
|
|
struct TestIndex
|
|
{
|
|
static constexpr int index = INDEX;
|
|
};
|
|
template <typename T, T... Indices>
|
|
static auto GenerateTypes(std::integer_sequence<T, Indices...>)
|
|
{
|
|
return ::testing::Types<TestIndex<Indices>...>{};
|
|
}
|
|
// testing::Types sequence of TestIndex types.
|
|
using Types = decltype(GenerateTypes(std::make_integer_sequence<int, N>{}));
|
|
};
|
|
|
|
// A typed test suite that will be instantiated for each type in TestingIndices::Types.
|
|
// This creates a separate test for each entry in the TEST_CASES array, allowing
|
|
// GTest to run and report on them individually.
|
|
template <typename T>
|
|
class ConvBuilderFwdInstancesTest : public ::testing::Test
|
|
{
|
|
protected:
|
|
static constexpr int N = T::index;
|
|
static constexpr const std::string_view& NAME = TEST_CASES[N].name;
|
|
static constexpr auto& ALGORITHM = TEST_CASES[N].algorithm;
|
|
static constexpr const std::string_view& EXPECTED_TYPE = TEST_CASES[N].expected_type;
|
|
};
|
|
|
|
// Custom test name generator to provide more descriptive names for each
|
|
// typed test instance, incorporating the index and the name from the TestCase.
|
|
struct TestNameGenerator
|
|
{
|
|
template <typename T>
|
|
static std::string GetName(int index)
|
|
{
|
|
return std::to_string(index) + "." + std::string(TEST_CASES[index].name);
|
|
}
|
|
};
|
|
|
|
TYPED_TEST_SUITE(ConvBuilderFwdInstancesTest,
|
|
TestingIndices<NUM_TEST_CASES>::Types,
|
|
TestNameGenerator);
|
|
|
|
// This is the body of the typed test. It will be executed for each TestCase.
|
|
// It verifies that the ConvBuilder, when configured with a specific algorithm,
|
|
// generates the correct kernel type string and correctly configures the
|
|
// underlying factory parameters.
|
|
TYPED_TEST(ConvBuilderFwdInstancesTest, KernelParamsConfigured)
|
|
{
|
|
static constexpr const FwdConvAlgorithm& ALGORITHM =
|
|
ConvBuilderFwdInstancesTest<TypeParam>::ALGORITHM;
|
|
static constexpr const ConvSignature SIGNATURE;
|
|
using Builder = ckb::ConvBuilder<SIGNATURE, ALGORITHM>;
|
|
EXPECT_EQ(Builder::Instance::TypeString(), ConvBuilderFwdInstancesTest<TypeParam>::EXPECTED_TYPE);
|
|
const auto& tp = ALGORITHM.tuning_params;
|
|
EXPECT_EQ(Builder::Factory::TUNING.ak1, tp.ak1);
|
|
EXPECT_EQ(Builder::Factory::TUNING.bk1, tp.bk1);
|
|
const auto& tcda = ALGORITHM.block_transfer.thread_cluster_dims_a;
|
|
EXPECT_EQ(Builder::Factory::A_BLOCK_TRANSFER.thread_cluster_dims[0], tcda.k0);
|
|
EXPECT_EQ(Builder::Factory::A_BLOCK_TRANSFER.thread_cluster_dims[1], tcda.m);
|
|
EXPECT_EQ(Builder::Factory::A_BLOCK_TRANSFER.thread_cluster_dims[2], tcda.k1);
|
|
const auto& tcdb = ALGORITHM.block_transfer.thread_cluster_dims_b;
|
|
EXPECT_EQ(Builder::Factory::B_BLOCK_TRANSFER.thread_cluster_dims[0], tcdb.k0);
|
|
EXPECT_EQ(Builder::Factory::B_BLOCK_TRANSFER.thread_cluster_dims[1], tcdb.n);
|
|
EXPECT_EQ(Builder::Factory::B_BLOCK_TRANSFER.thread_cluster_dims[2], tcdb.k1);
|
|
const auto& tcdc = ALGORITHM.block_transfer.thread_cluster_dims_c;
|
|
EXPECT_EQ(Builder::Factory::C_BLOCK_TRANSFER.thread_cluster_dims[0], tcdc.m_block);
|
|
EXPECT_EQ(Builder::Factory::C_BLOCK_TRANSFER.thread_cluster_dims[1], tcdc.m_wave_per_xdl);
|
|
EXPECT_EQ(Builder::Factory::C_BLOCK_TRANSFER.thread_cluster_dims[2], tcdc.n_block);
|
|
EXPECT_EQ(Builder::Factory::C_BLOCK_TRANSFER.thread_cluster_dims[3], tcdc.n_wave_per_xdl);
|
|
}
|
|
|
|
// A standard GTest to ensure that all `expected_type` strings in the
|
|
// TEST_CASES array are unique. This helps prevent copy-paste errors and
|
|
// ensures that each test case is meaningful.
|
|
TEST(ConvBuilderFwdInstancesTest, TypeStringsAreUnique)
|
|
{
|
|
std::set<std::string> strings;
|
|
for(int i = 0; i < NUM_TEST_CASES; ++i)
|
|
{
|
|
const auto& [iter, inserted] = strings.insert(std::string(TEST_CASES[i].expected_type));
|
|
EXPECT_TRUE(inserted) << "Duplicate expected_string " << *iter;
|
|
}
|
|
EXPECT_EQ(strings.size(), NUM_TEST_CASES)
|
|
<< "Found fewer unique expected_strings than test cases";
|
|
}
|
|
|
|
} // namespace
|