Skip to content

GitLab

"example/42_groupnorm/groupnorm_blockwise.cpp" did not exist on "7f216620896909e254284e418d08f4d20f938a01"
Commit bc641634 authored by Jun Liu's avatar Jun Liu
Browse files

Merge branch 'develop-tmp' into amd-develop

parents f30e5975 a3d9a2cd
Hide whitespace changes
Inline Side-by-side
Showing with 288 additions and 112 deletions
script/redis-cli.conf 0 → 100644
View file @ bc641634
fips = no
setuid = root
setgid = root
pid = /var/run/stunnel.pid
debug = 7
options = NO_SSLv2
options = NO_SSLv3
[redis-cli]
client = yes
accept = 127.0.0.1:6379
script/sccache_wrapper.sh 0 → 100755
View file @ bc641634
#!/bin/bash
set -e
COMPILERS_HASH_DIR=${COMPILERS_HASH_DIR:-"/tmp/.sccache"}
SCCACHE_EXTRAFILES=${SCCACHE_EXTRAFILES:-"${COMPILERS_HASH_DIR}/rocm_compilers_hash_file"}
SCCACHE_BIN=${SCCACHE_BIN:-"${SCCACHE_INSTALL_LOCATION}/sccache"}
ENFORCE_REDIS="false"
while [ "$1" != "" ];
do
case $1 in
--enforce_redis )
shift; ENFORCE_REDIS="true" ;;
--no-hipcc )
shift ;;
*)
break ;;
esac
done
setup_rocm_compilers_hash_file() {
mkdir -p "$COMPILERS_HASH_DIR"
HIPCC_MD5="$(md5sum "${ROCM_PATH}/bin/hipcc")"
pushd "${ROCM_PATH}/amdgcn/bitcode"
DEVICELIBS_BITCODES_MD5="$(find . -type f -exec md5sum {} \; | sort | md5sum)"
popd
HIPCC_HASH_VALUE="${HIPCC_MD5%% *}"
DEVICELIBS_BITCODES_HASH_VALUE="${DEVICELIBS_BITCODES_MD5%% *}"
# MD5 checksums of clang and clang-offload-bundler cannot be used since they will keep changing
# if the ROCM_PATH changes, ie; for every mainline build.
# This is because ROCM_PATH gets encoded into the clang/clang-offload-bundler binaries as part
# of RPATH.
# The versions themselves contain the commit hash of the compiler repo at the time of building.
# Hence, this should be a viable alternative to using the binary checksum itself.
CLANG_VERSION="$("${ROCM_PATH}/llvm/bin/clang" --version | head -n 1)"
CLANG_OFFLOAD_BUNDLER_VERSION="$("${ROCM_PATH}/llvm/bin/clang-offload-bundler" --version | head -n 1)"
printf '%s: %s\n' 'clang version' "${CLANG_VERSION}" | tee -a "$SCCACHE_EXTRAFILES"
printf '%s: %s\n' 'clang-offload-bundler version' "${CLANG_OFFLOAD_BUNDLER_VERSION}" | tee -a "$SCCACHE_EXTRAFILES"
printf '%s: %s\n' 'hipcc md5sum' "${HIPCC_HASH_VALUE}" | tee -a "$SCCACHE_EXTRAFILES"
printf '%s: %s\n' 'devicelibs bitcode md5sum' "${DEVICELIBS_BITCODES_HASH_VALUE}" | tee -a "$SCCACHE_EXTRAFILES"
echo "sccache-wrapper: compilers hash file set up at ${SCCACHE_EXTRAFILES}"
cat "$SCCACHE_EXTRAFILES"
}
if [ "${ENFORCE_REDIS}" == "true" ]; then
if [ -z "${SCCACHE_REDIS}" ]; then
echo "SCCACHE_REDIS not set. Not wrapping compilers with sccache."
exit 10
else
response=$(redis-cli -u ${SCCACHE_REDIS} ping) || true
if [ "${response}" != "PONG" ]; then
echo "Redis server unreachable. Not wrapping compilers with sccache."
exit 20
fi
fi
fi
setup_rocm_compilers_hash_file
$SCCACHE_BIN --version
$SCCACHE_BIN --start-server
test/CMakeLists.txt
View file @ bc641634
...@@ -139,7 +139,7 @@ add_subdirectory(grouped_convnd_fwd) ...@@ -139,7 +139,7 @@ add_subdirectory(grouped_convnd_fwd)
add_subdirectory(grouped_convnd_bwd_weight) add_subdirectory(grouped_convnd_bwd_weight)
add_subdirectory(block_to_ctile_map) add_subdirectory(block_to_ctile_map)
add_subdirectory(softmax) add_subdirectory(softmax)
add_subdirectory(normalization) add_subdirectory(normalization_fwd)
add_subdirectory(data_type) add_subdirectory(data_type)
add_subdirectory(elementwise_normalization) add_subdirectory(elementwise_normalization)
add_subdirectory(batchnorm) add_subdirectory(batchnorm)
......
test/contraction/test_contraction.cpp
View file @ bc641634
...@@ -10,9 +10,12 @@ ...@@ -10,9 +10,12 @@
#include <gtest/gtest.h> #include <gtest/gtest.h>
#include "profiler/profile_contraction_impl.hpp" #include "profiler/profile_contraction_impl.hpp"
#include "profiler/profile_contraction_utils.hpp"
using F32 = float; using F16 = ck::half_t;
using F64 = double; using BF16 = ck::bhalf_t;
using F32 = float;
using F64 = double;
using Row = ck::tensor_layout::gemm::RowMajor; using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor; using Col = ck::tensor_layout::gemm::ColumnMajor;
...@@ -20,49 +23,49 @@ using Col = ck::tensor_layout::gemm::ColumnMajor; ...@@ -20,49 +23,49 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
using Bilinear = ck::tensor_operation::element_wise::Bilinear; using Bilinear = ck::tensor_operation::element_wise::Bilinear;
using Scale = ck::tensor_operation::element_wise::Scale; using Scale = ck::tensor_operation::element_wise::Scale;
struct MemoryParams struct Dimensions
{ {
std::vector<ck::index_t> M; std::vector<ck::index_t> M;
std::vector<ck::index_t> N; std::vector<ck::index_t> N;
std::vector<ck::index_t> K; std::vector<ck::index_t> K;
std::vector<ck::index_t> StridesA;
std::vector<ck::index_t> StridesB;
std::vector<ck::index_t> StridesC;
std::vector<ck::index_t> StridesD;
}; };
template <typename Tuple> template <typename Tuple>
class TestContraction : public ::testing::Test class TestContraction : public ::testing::Test
{ {
protected: protected:
using ALayout = std::tuple_element_t<0, Tuple>; using ALayout = std::tuple_element_t<0, Tuple>;
using BLayout = std::tuple_element_t<1, Tuple>; using BLayout = std::tuple_element_t<1, Tuple>;
using CDLayout = std::tuple_element_t<2, Tuple>; using CDLayout = std::tuple_element_t<2, Tuple>;
using DataType = std::tuple_element_t<3, Tuple>; using DataType = std::tuple_element_t<3, Tuple>;
using DTupleDataType = std::tuple_element_t<4, Tuple>; using DTupleDataType = std::tuple_element_t<4, Tuple>;
using CDElementOp = std::tuple_element_t<5, Tuple>; using ComputeDataType = std::tuple_element_t<5, Tuple>;
using CDElementOp = std::tuple_element_t<6, Tuple>;
std::vector<MemoryParams> list_of_memory_params = {{{32, 32},
{32, 32}, std::vector<Dimensions> dimension_list = {{{32, 32}, {32, 32}, {32, 32}},
{32, 32}, {{16, 16}, {32, 32}, {16, 16}}};
{32768, 1024, 32, 1},
{32768, 1024, 32, 1}, std::vector<ck::index_t> init_methods = {1, 2};
{32768, 1024, 32, 1},
{32768, 1024, 32, 1}},
{{16, 16},
{32, 32},
{16, 16},
{4096, 256, 16, 1},
{16, 1, 8192, 256},
{16384, 1024, 32, 1},
{16384, 1024, 32, 1}}};
std::vector<ck::index_t> init_methods = {0, 1, 2};
std::unique_ptr<CDElementOp> p_cd_element_op; std::unique_ptr<CDElementOp> p_cd_element_op;
void Run() void Run()
{ {
for(auto& memory_params : list_of_memory_params) for(auto& dimension_params : dimension_list)
{ {
std::vector<ck::index_t> StridesA;
std::vector<ck::index_t> StridesB;
std::vector<ck::index_t> StridesC;
std::vector<ck::index_t> StridesD;
const auto& M = dimension_params.M;
const auto& N = dimension_params.N;
const auto& K = dimension_params.K;
assign_default_strides(ALayout{}, StridesA, {M[0], M[1], K[0], K[1]});
assign_default_strides(BLayout{}, StridesB, {N[0], N[1], K[0], K[1]});
assign_default_strides(CDLayout{}, StridesC, {M[0], M[1], N[0], N[1]});
assign_default_strides(CDLayout{}, StridesD, {M[0], M[1], N[0], N[1]});
for(const ck::index_t init_method : init_methods) for(const ck::index_t init_method : init_methods)
{ {
bool pass = bool pass =
...@@ -70,19 +73,20 @@ class TestContraction : public ::testing::Test ...@@ -70,19 +73,20 @@ class TestContraction : public ::testing::Test
BLayout, BLayout,
CDLayout, CDLayout,
DataType, DataType,
ComputeDataType,
DTupleDataType, DTupleDataType,
CDElementOp>(true /*do_verification*/, CDElementOp>(true /*do_verification*/,
init_method, init_method,
false /*do_logs*/, false /*do_logs*/,
false /*time_kernel*/, false /*time_kernel*/,
*p_cd_element_op, *p_cd_element_op,
memory_params.M, dimension_params.M,
memory_params.N, dimension_params.N,
memory_params.K, dimension_params.K,
memory_params.StridesA, StridesA,
memory_params.StridesB, StridesB,
memory_params.StridesC, StridesC,
memory_params.StridesD); StridesD);
EXPECT_TRUE(pass); EXPECT_TRUE(pass);
} }
} }
...@@ -99,24 +103,18 @@ class TestContractionBilinear : public TestContraction<Tuple> ...@@ -99,24 +103,18 @@ class TestContractionBilinear : public TestContraction<Tuple>
{ {
}; };
#define ALL_LAYOUT_COMBINATIONS(dt, tuple_dt, compute_dt, op) \
std::tuple<Row, Row, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Row, Col, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Col, Row, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Col, Col, Row, dt, tuple_dt, compute_dt, op>
using BilinearKernelTypes = using BilinearKernelTypes =
::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<F32>, Bilinear>, ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F32, Bilinear),
std::tuple<Row, Col, Row, F32, ck::Tuple<F32>, Bilinear>, ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F64, Bilinear)>;
std::tuple<Col, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
std::tuple<Col, Col, Row, F32, ck::Tuple<F32>, Bilinear>, using ScaleKernelTypes = ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F32, Scale),
std::tuple<Row, Row, Row, F64, ck::Tuple<F32>, Bilinear>, ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F64, Scale)>;
std::tuple<Row, Col, Row, F64, ck::Tuple<F32>, Bilinear>,
std::tuple<Col, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
std::tuple<Col, Col, Row, F64, ck::Tuple<F32>, Bilinear>>;
using ScaleKernelTypes = ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<>, Scale>,
std::tuple<Row, Col, Row, F32, ck::Tuple<>, Scale>,
std::tuple<Col, Row, Row, F32, ck::Tuple<>, Scale>,
std::tuple<Col, Col, Row, F32, ck::Tuple<>, Scale>,
std::tuple<Row, Row, Row, F64, ck::Tuple<>, Scale>,
std::tuple<Row, Col, Row, F64, ck::Tuple<>, Scale>,
std::tuple<Col, Row, Row, F64, ck::Tuple<>, Scale>,
std::tuple<Col, Col, Row, F64, ck::Tuple<>, Scale>>;
TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes); TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes); TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
...@@ -136,3 +134,46 @@ TYPED_TEST(TestContractionScale, scale) ...@@ -136,3 +134,46 @@ TYPED_TEST(TestContractionScale, scale)
this->p_cd_element_op = std::make_unique<Scale>(0.5f); this->p_cd_element_op = std::make_unique<Scale>(0.5f);
this->Run(); this->Run();
} }
template <typename Tuple>
class TestContractionScaleMixedPrecision : public TestContraction<Tuple>
{
};
template <typename Tuple>
class TestContractionBilinearMixedPrecision : public TestContraction<Tuple>
{
};
using BilinearKernelTypesMixedPrecision =
::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F16, Bilinear),
ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, BF16, Bilinear),
ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F32, Bilinear),
ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<F16>, F32, Bilinear),
ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<BF16>, F32, Bilinear)>;
using ScaleKernelTypesMixedPrecision =
::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F16, Scale),
ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, BF16, Scale),
ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F32, Scale),
ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<>, F32, Scale),
ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<>, F32, Scale)>;
TYPED_TEST_SUITE(TestContractionBilinearMixedPrecision, BilinearKernelTypesMixedPrecision);
TYPED_TEST_SUITE(TestContractionScaleMixedPrecision, ScaleKernelTypesMixedPrecision);
TYPED_TEST(TestContractionBilinearMixedPrecision, bilinear)
{
this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
this->Run();
this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
this->Run();
}
TYPED_TEST(TestContractionScaleMixedPrecision, scale)
{
this->p_cd_element_op = std::make_unique<Scale>(1.f);
this->Run();
this->p_cd_element_op = std::make_unique<Scale>(0.5f);
this->Run();
}
test/contraction/test_contraction_interface.cpp
View file @ bc641634
...@@ -34,11 +34,11 @@ class ContractionInstanceWrapper ...@@ -34,11 +34,11 @@ class ContractionInstanceWrapper
static constexpr ck::index_t NumDim = 2; static constexpr ck::index_t NumDim = 2;
// clang-format off // clang-format off
using ContractionDeviceInstance = ck::tensor_operation::device:: using ContractionDeviceInstance = ck::tensor_operation::device::
//#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle| DsData| EData| A| B| CDE| GEMM| NumGemmK| Block| MPer| NPer| KPer| AK1| BK1| MPer| NPer| MXdl| NXdl| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds| BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffle| CShuffle| CBlockTransferClusterLengths| CBlockTransfer| Compute|
//#####################################| | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| //#####################################| | | | Type| Type| Type| DataType| Type| Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch| Size| Block| Block| Block| | | XDL| XDL| Per| Per| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraM| ThreadCluster| ThreadCluster| SrcAccessOrder| SrcVectorDim| SrcScalar| DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MWaveMPerXdl| ScalarPerVector| Data|
//#####################################| | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| //#####################################| | | | | | | | | | Operation| Operation| Operation| | Stage| | | | | | | | | Wave| Wave| Lengths_K0_M_K1| ArrangeOrder| | | PerVector| PerVector_K1| | Lengths_K0_N_K1| ArrangeOrder| | | PerVector| PerVector_K1| | PerShuffle| PerShuffle| _NBlock_NWaveNPerXdl| _NWaveNPerXdl| Type|
//#####################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | //#####################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
DeviceContractionMultipleD_Xdl_CShuffle< NumDim, NumDim, NumDim, F32, F32, F32, F32, ck::Tuple<F32>, F32, Pass, Pass, Bilinear, GemmSpec, 1, 256, 256, 128, 16, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, ABlockTransferSrcVectorDim, 4, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, BBlockTransferSrcVectorDim, 4, 4, 1, 1, 1, S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>; DeviceContractionMultipleD_Xdl_CShuffle< NumDim, NumDim, NumDim, F32, F32, F32, F32, ck::Tuple<F32>, F32, Pass, Pass, Bilinear, GemmSpec, 1, 256, 256, 128, 16, 4, 4, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, ABlockTransferSrcVectorDim, 4, 4, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, BBlockTransferSrcVectorDim, 4, 4, 1, 1, 1, S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector, F32>;
// clang-format on // clang-format on
bool isSupported(std::vector<ck::index_t>& ADims, bool isSupported(std::vector<ck::index_t>& ADims,
......
test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
View file @ bc641634
...@@ -45,14 +45,20 @@ class TestConvTensorRearrange : public ::testing::Test ...@@ -45,14 +45,20 @@ class TestConvTensorRearrange : public ::testing::Test
using namespace ck::tensor_layout::convolution; using namespace ck::tensor_layout::convolution;
using namespace ck::conv_tensor_rearrange_op; using namespace ck::conv_tensor_rearrange_op;
using KernelTypes1d = using KernelTypes1d = ::testing::Types<std::tuple<GNWC, ImageToColumn>,
::testing::Types<std::tuple<GNWC, ImageToColumn>, std::tuple<GNWC, ColumnToImage>>; std::tuple<GNWC, ColumnToImage>,
std::tuple<NWGC, ImageToColumn>,
std::tuple<NWGC, ColumnToImage>>;
using KernelTypes2d = using KernelTypes2d = ::testing::Types<std::tuple<GNHWC, ImageToColumn>,
::testing::Types<std::tuple<GNHWC, ImageToColumn>, std::tuple<GNHWC, ColumnToImage>>; std::tuple<GNHWC, ColumnToImage>,
std::tuple<NHWGC, ImageToColumn>,
std::tuple<NHWGC, ColumnToImage>>;
using KernelTypes3d = using KernelTypes3d = ::testing::Types<std::tuple<GNDHWC, ImageToColumn>,
::testing::Types<std::tuple<GNDHWC, ImageToColumn>, std::tuple<GNDHWC, ColumnToImage>>; std::tuple<GNDHWC, ColumnToImage>,
std::tuple<NDHWGC, ImageToColumn>,
std::tuple<NDHWGC, ColumnToImage>>;
template <typename Tuple> template <typename Tuple>
class TestConvTensorRearrange1d : public TestConvTensorRearrange<Tuple> class TestConvTensorRearrange1d : public TestConvTensorRearrange<Tuple>
...@@ -77,16 +83,16 @@ TYPED_TEST(TestConvTensorRearrange1d, Test1D) ...@@ -77,16 +83,16 @@ TYPED_TEST(TestConvTensorRearrange1d, Test1D)
{ {
this->conv_params.clear(); this->conv_params.clear();
this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}}); this->conv_params.push_back({1, 2, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}}); this->conv_params.push_back({1, 2, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {3}, {1}, {0}, {0}}); this->conv_params.push_back({1, 2, 64, 1, 64, {1}, {7}, {3}, {1}, {0}, {0}});
this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}}); this->conv_params.push_back({1, 2, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
// ScalarPerVector should be 1 // ScalarPerVector should be 1
this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}}); this->conv_params.push_back({1, 2, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
// stride != 1 // stride != 1
this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}}); this->conv_params.push_back({1, 2, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
// dilation != 1 // dilation != 1
this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}}); this->conv_params.push_back({1, 2, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
#ifdef CK_ENABLE_FP32 #ifdef CK_ENABLE_FP32
this->template Run<1, float, float>(); this->template Run<1, float, float>();
#endif #endif
...@@ -106,13 +112,13 @@ TYPED_TEST(TestConvTensorRearrange2d, Test2D) ...@@ -106,13 +112,13 @@ TYPED_TEST(TestConvTensorRearrange2d, Test2D)
this->conv_params.clear(); this->conv_params.clear();
this->conv_params.push_back( this->conv_params.push_back(
{2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}}); {2, 2, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back( this->conv_params.push_back(
{2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}}); {2, 2, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {3, 3}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {3, 3}, {1, 1}, {0, 0}, {0, 0}});
this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}}); this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
this->conv_params.push_back( this->conv_params.push_back(
{2, 1, 64, 1, 64, {3, 3}, {28, 28}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}); {2, 2, 64, 1, 64, {3, 3}, {28, 28}, {2, 2}, {2, 2}, {1, 1}, {1, 1}});
#ifdef CK_ENABLE_FP32 #ifdef CK_ENABLE_FP32
this->template Run<2, float, float>(); this->template Run<2, float, float>();
#endif #endif
...@@ -131,13 +137,13 @@ TYPED_TEST(TestConvTensorRearrange3d, Test3D) ...@@ -131,13 +137,13 @@ TYPED_TEST(TestConvTensorRearrange3d, Test3D)
{ {
this->conv_params.clear(); this->conv_params.clear();
this->conv_params.push_back( this->conv_params.push_back(
{3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {3, 3, 3}, {0, 0, 0}, {0, 0, 0}}); {3, 2, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {3, 3, 3}, {0, 0, 0}, {0, 0, 0}});
this->conv_params.push_back( this->conv_params.push_back(
{3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}); {3, 2, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->conv_params.push_back( this->conv_params.push_back(
{3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}}); {3, 2, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
this->conv_params.push_back( this->conv_params.push_back(
{3, 1, 64, 1, 64, {3, 3, 3}, {14, 14, 14}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}); {3, 2, 64, 1, 64, {3, 3, 3}, {14, 14, 14}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}});
#ifdef CK_ENABLE_FP32 #ifdef CK_ENABLE_FP32
this->template Run<3, float, float>(); this->template Run<3, float, float>();
#endif #endif
......
test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
View file @ bc641634
...@@ -53,7 +53,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test ...@@ -53,7 +53,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
template <typename ConvTensorRearrangeOp> template <typename ConvTensorRearrangeOp>
bool Run() bool Run()
{ {
const auto G = conv_param.G_;
const auto N = conv_param.N_; const auto N = conv_param.N_;
const auto C = conv_param.C_; const auto C = conv_param.C_;
const auto FakeC = const auto FakeC =
...@@ -71,13 +71,13 @@ class TestConvTensorRearrangeInterface : public ::testing::Test ...@@ -71,13 +71,13 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
const auto image_desc = const auto image_desc =
ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>( ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(
conv_param); conv_param);
const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX}); const auto gemm_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
std::array<ck::index_t, NDimSpatial> input_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
std::array<ck::index_t, NDimSpatial> output_spatial_lengths{}; std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{}; std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
std::array<ck::index_t, 2> output_m_k_strides{}; std::array<ck::index_t, 3> output_g_m_k_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_strides{}; std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
std::array<ck::index_t, NDimSpatial> conv_filter_dilations{}; std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
std::array<ck::index_t, NDimSpatial> input_left_pads{}; std::array<ck::index_t, NDimSpatial> input_left_pads{};
...@@ -89,7 +89,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test ...@@ -89,7 +89,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths); copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
copy(conv_param.output_spatial_lengths_, output_spatial_lengths); copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
copy(image_desc.GetStrides(), input_g_n_c_wis_strides); copy(image_desc.GetStrides(), input_g_n_c_wis_strides);
copy(gemm_desc.GetStrides(), output_m_k_strides); copy(gemm_desc.GetStrides(), output_g_m_k_strides);
copy(conv_param.conv_filter_strides_, conv_filter_strides); copy(conv_param.conv_filter_strides_, conv_filter_strides);
copy(conv_param.conv_filter_dilations_, conv_filter_dilations); copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
copy(conv_param.input_left_pads_, input_left_pads); copy(conv_param.input_left_pads_, input_left_pads);
...@@ -100,13 +100,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test ...@@ -100,13 +100,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
auto img2col = DeviceImgToColInstance{}; auto img2col = DeviceImgToColInstance{};
auto argument = img2col.MakeArgument(nullptr, auto argument = img2col.MakeArgument(nullptr,
nullptr, nullptr,
G,
N, N,
IsCPacked ? C : FakeC, IsCPacked ? C : FakeC,
input_spatial_lengths, input_spatial_lengths,
filter_spatial_lengths, filter_spatial_lengths,
output_spatial_lengths, output_spatial_lengths,
input_g_n_c_wis_strides, input_g_n_c_wis_strides,
output_m_k_strides, output_g_m_k_strides,
conv_filter_strides, conv_filter_strides,
conv_filter_dilations, conv_filter_dilations,
input_left_pads, input_left_pads,
...@@ -119,13 +120,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test ...@@ -119,13 +120,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
auto col2img = DeviceColToimgInstance{}; auto col2img = DeviceColToimgInstance{};
auto argument = col2img.MakeArgument(nullptr, auto argument = col2img.MakeArgument(nullptr,
nullptr, nullptr,
G,
N, N,
IsCPacked ? C : FakeC, IsCPacked ? C : FakeC,
input_spatial_lengths, input_spatial_lengths,
filter_spatial_lengths, filter_spatial_lengths,
output_spatial_lengths, output_spatial_lengths,
input_g_n_c_wis_strides, input_g_n_c_wis_strides,
output_m_k_strides, output_g_m_k_strides,
conv_filter_strides, conv_filter_strides,
conv_filter_dilations, conv_filter_dilations,
input_left_pads, input_left_pads,
......
test/grouped_gemm/test_grouped_gemm_interface.cpp
View file @ bc641634
...@@ -108,6 +108,10 @@ TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops) ...@@ -108,6 +108,10 @@ TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops)
// kloops % 2 // kloops % 2
Ks = std::vector<int>{256, 512, 320, 768}; Ks = std::vector<int>{256, 512, 320, 768};
EXPECT_FALSE(
DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
Ks = std::vector<int>{256, 512, 384, 768};
EXPECT_TRUE( EXPECT_TRUE(
DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch)); DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
......
test/normalization/CMakeLists.txt deleted 100644 → 0
View file @ f30e5975
add_custom_target(test_normalization)
add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
if(result EQUAL 0)
target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_layernorm2d_fp32)
endif()
add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_groupnorm_fp32)
endif()
add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_layernorm2d_fp16)
endif()
add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
add_dependencies(test_normalization test_groupnorm_fp16)
endif()
test/normalization_fwd/CMakeLists.txt 0 → 100644
View file @ bc641634
add_custom_target(test_normalization_fwd)
add_gtest_executable(test_layernorm2d_fwd_fp32 test_layernorm2d_fwd_fp32.cpp)
if(result EQUAL 0)
target_link_libraries(test_layernorm2d_fwd_fp32 PRIVATE utility device_normalization_fwd_instance)
add_dependencies(test_normalization_fwd test_layernorm2d_fwd_fp32)
endif()
add_gtest_executable(test_groupnorm_fwd_fp32 test_groupnorm_fwd_fp32.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fwd_fp32 PRIVATE utility device_normalization_fwd_instance)
add_dependencies(test_normalization_fwd test_groupnorm_fwd_fp32)
endif()
add_gtest_executable(test_layernorm2d_fwd_fp16 test_layernorm2d_fwd_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_layernorm2d_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
add_dependencies(test_normalization_fwd test_layernorm2d_fwd_fp16)
endif()
add_gtest_executable(test_layernorm4d_fwd_fp16 test_layernorm4d_fwd_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_layernorm4d_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
add_dependencies(test_normalization_fwd test_layernorm4d_fwd_fp16)
endif()
add_gtest_executable(test_groupnorm_fwd_fp16 test_groupnorm_fwd_fp16.cpp)
if(result EQUAL 0)
target_link_libraries(test_groupnorm_fwd_fp16 PRIVATE utility device_normalization_fwd_instance)
add_dependencies(test_normalization_fwd test_groupnorm_fwd_fp16)
endif()
test/normalization/test_groupnorm_fp16.cpp → test/normalization_fwd/test_groupnorm_fwd_fp16.cpp
View file @ bc641634
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "profiler/profile_groupnorm_impl.hpp" #include "profiler/profile_groupnorm_fwd_impl.hpp"
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float; using F32 = float;
......
test/normalization/test_groupnorm_fp32.cpp → test/normalization_fwd/test_groupnorm_fwd_fp32.cpp
View file @ bc641634
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "profiler/profile_groupnorm_impl.hpp" #include "profiler/profile_groupnorm_fwd_impl.hpp"
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float; using F32 = float;
......
test/normalization/test_layernorm2d_fp16.cpp → test/normalization_fwd/test_layernorm2d_fwd_fp16.cpp
View file @ bc641634
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "profiler/profile_layernorm_impl.hpp" #include "profiler/profile_layernorm_fwd_impl.hpp"
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float; using F32 = float;
......
test/normalization/test_layernorm2d_fp32.cpp → test/normalization_fwd/test_layernorm2d_fwd_fp32.cpp
View file @ bc641634
...@@ -2,7 +2,7 @@ ...@@ -2,7 +2,7 @@
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved. // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h" #include "gtest/gtest.h"
#include "profiler/profile_layernorm_impl.hpp" #include "profiler/profile_layernorm_fwd_impl.hpp"
using F16 = ck::half_t; using F16 = ck::half_t;
using F32 = float; using F32 = float;
......
test/normalization_fwd/test_layernorm4d_fwd_fp16.cpp 0 → 100644
View file @ bc641634
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include "gtest/gtest.h"
#include "profiler/profile_layernorm_fwd_impl.hpp"
using F16 = ck::half_t;
using F32 = float;
using ck::index_t;
template <typename Tuple>
class TestLayernorm4d : public ::testing::Test
{
protected:
using XDataType = std::tuple_element_t<0, Tuple>;
using GammaDataType = std::tuple_element_t<1, Tuple>;
using BetaDataType = std::tuple_element_t<2, Tuple>;
using ComputeDataType = std::tuple_element_t<3, Tuple>;
using YDataType = std::tuple_element_t<4, Tuple>;
using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
void Run()
{
// [N, D], reduce D
std::vector<std::vector<ck::index_t>> lengths = {
{1, 1, 1, 1}, {7, 7, 7, 7}, {256, 16, 16, 8}};
for(auto length : lengths)
{
bool success = ck::profiler::profile_layernorm_impl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
SaveMeanInvStdDataType,
true,
4>(true, 2, false, false, length);
EXPECT_TRUE(success);
}
}
};
using KernelTypes = ::testing::Types<
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
std::tuple<F16, F16, F16, F32, F16, F32>>;
TYPED_TEST_SUITE(TestLayernorm4d, KernelTypes);
TYPED_TEST(TestLayernorm4d, Test_FP16) { this->Run(); }
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment