Skip to content

GitLab

Commit 9d2280d6 authored by rocking's avatar rocking
Browse files

1. Rename AccDatatype in normalization to computeData 

2. Rename AccElementwiseOperation to YElementwiseOperation in normalization
parent 1a38e362
Show whitespace changes
Inline Side-by-side
Showing with 109 additions and 107 deletions
client_example/05_layernorm/layernorm2d.cpp
View file @ 9d2280d6
......@@ -16,7 +16,7 @@ using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using AccDataType = float;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
......@@ -54,7 +54,7 @@ int main(int argc, char* argv[])
using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
......
example/27_layernorm/layernorm_blockwise.cpp
View file @ 9d2280d6
......@@ -24,7 +24,7 @@ using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using AccDataType = float;
using ConputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
......@@ -34,7 +34,7 @@ using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ConputeDataType,
YDataType,
PassThrough,
Rank,
......@@ -121,7 +121,7 @@ int main()
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
ConputeDataType,
PassThrough,
Rank,
NumReduceDim>;
......
example/42_groupnorm/groupnorm_sigmoid_fp16.cpp
View file @ 9d2280d6
......@@ -27,7 +27,7 @@ using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using AccDataType = float;
using ConputeDataType = float;
struct YElementOp
{
......@@ -50,7 +50,7 @@ using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ConputeDataType,
YDataType,
YElementOp,
Rank,
......@@ -157,7 +157,7 @@ int main(int argc, char* argv[])
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
ConputeDataType,
YElementOp>;
ReferenceInstance ref;
......
include/ck/tensor_operation/gpu/device/device_normalization.hpp
View file @ 9d2280d6
......@@ -14,9 +14,9 @@ namespace device {
template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename AccDataType,
typename ConputeDataType,
typename YDataType,
typename AccElementwiseOperation,
typename YElementwiseOperation,
index_t Rank,
index_t NumReduceDim>
struct DeviceNormalization : public BaseOperator
......@@ -35,7 +35,7 @@ struct DeviceNormalization : public BaseOperator
void* p_y,
void* p_savedMean,
void* p_savedInvVar,
AccElementwiseOperation acc_elementwise_op) = 0;
YElementwiseOperation y_elementwise_op) = 0;
virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
};
......@@ -43,17 +43,17 @@ struct DeviceNormalization : public BaseOperator
template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename AccDataType,
typename ConputeDataType,
typename YDataType,
typename AccElementwiseOperation,
typename YElementwiseOperation,
index_t Rank,
index_t NumReduceDim>
using DeviceNormalizationPtr = std::unique_ptr<DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ConputeDataType,
YDataType,
AccElementwiseOperation,
YElementwiseOperation,
Rank,
NumReduceDim>>;
......
include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
View file @ 9d2280d6
......@@ -21,20 +21,20 @@ template <typename GridwiseReduction,
typename GammaDataType,
typename BetaDataType,
typename YDataType,
typename AccDataType,
typename AccElementwiseOperation,
typename ConputeDataType,
typename YElementwiseOperation,
typename GridDesc_M_K>
__global__ void kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
const GridDesc_M_K gamma_grid_desc_m_k,
const GridDesc_M_K beta_grid_desc_m_k,
const GridDesc_M_K y_grid_desc_m_k,
index_t num_k_block_tile_iteration,
AccDataType epsilon,
ConputeDataType epsilon,
const XDataType* const __restrict__ p_x_global,
const GammaDataType* const __restrict__ p_gamma_global,
const BetaDataType* const __restrict__ p_beta_global,
YDataType* const __restrict__ p_y_global,
const AccElementwiseOperation acc_elementwise_op)
const YElementwiseOperation y_elementwise_op)
{
GridwiseReduction::Run(x_grid_desc_m_k,
gamma_grid_desc_m_k,
......@@ -46,7 +46,7 @@ __global__ void kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
p_gamma_global,
p_beta_global,
p_y_global,
acc_elementwise_op);
y_elementwise_op);
};
} // namespace ck
......@@ -58,9 +58,9 @@ namespace device {
template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename AccDataType,
typename ConputeDataType,
typename YDataType,
typename AccElementwiseOperation,
typename YElementwiseOperation,
index_t Rank,
index_t NumReduceDim,
index_t BlockSize,
......@@ -78,9 +78,9 @@ template <typename XDataType,
struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ConputeDataType,
YDataType,
AccElementwiseOperation,
YElementwiseOperation,
Rank,
NumReduceDim>
{
......@@ -172,8 +172,8 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
AccElementwiseOperation,
ConputeDataType,
YElementwiseOperation,
GridDesc_M_K,
BlockSize,
MThreadClusterSize,
......@@ -194,8 +194,8 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
AccElementwiseOperation,
ConputeDataType,
YElementwiseOperation,
GridDesc_M_K,
BlockSize,
MThreadClusterSize,
......@@ -220,7 +220,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
const std::vector<index_t> betaStrides,
const std::vector<index_t> yStrides,
const std::vector<index_t> reduceDims,
AccElementwiseOperation acc_elementwise_op,
YElementwiseOperation y_elementwise_op,
double epsilon,
const XDataType* p_x,
const GammaDataType* p_gamma,
......@@ -230,9 +230,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
p_gamma_(p_gamma),
p_beta_(p_beta),
p_y_(p_y),
acc_elementwise_op_(acc_elementwise_op)
y_elementwise_op_(y_elementwise_op)
{
epsilon_ = static_cast<AccDataType>(epsilon);
epsilon_ = static_cast<ConputeDataType>(epsilon);
Lengths_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(lengths, reduceDims);
xStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(xStrides, reduceDims);
......@@ -265,7 +265,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
x_grid_desc_m_k_.GetLength(Number<1>{}) <= KThreadClusterSize * KThreadSliceSize;
}
AccDataType epsilon_;
ConputeDataType epsilon_;
const XDataType* p_x_;
const GammaDataType* p_gamma_;
......@@ -278,7 +278,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
std::vector<index_t> betaStrides_;
std::vector<index_t> yStrides_;
AccElementwiseOperation acc_elementwise_op_;
YElementwiseOperation y_elementwise_op_;
int blkGroupSize_;
int numBlockTileIteration_;
......@@ -301,16 +301,16 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
AccElementwiseOperation,
ConputeDataType,
YElementwiseOperation,
GridDesc_M_K>
: kernel_normalization<GridwiseReduceLayernormGeneric,
XDataType,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
AccElementwiseOperation,
ConputeDataType,
YElementwiseOperation,
GridDesc_M_K>;
float avg_time = 0;
......@@ -329,7 +329,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
arg.p_gamma_,
arg.p_beta_,
arg.p_y_,
arg.acc_elementwise_op_);
arg.y_elementwise_op_);
return (avg_time);
};
......@@ -429,7 +429,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
void* p_y,
void* p_saveMean,
void* p_saveInvVar,
AccElementwiseOperation acc_elementwise_op) override
YElementwiseOperation y_elementwise_op) override
{
// TODO
// Optional cache of the intermediate results (mean and InvVariance) during the
......@@ -443,7 +443,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
betaStrides,
yStrides,
reduceDims,
acc_elementwise_op,
y_elementwise_op,
epsilon,
static_cast<const XDataType*>(p_x),
static_cast<const GammaDataType*>(p_gamma),
......
include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp
View file @ 9d2280d6
......@@ -16,8 +16,8 @@ template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename YDataType,
typename AccDataType,
typename AccElementwiseOperation,
typename ComputeDataType,
typename YElementwiseOperation,
typename GridDesc_M_K,
index_t BlockSize,
index_t MThreadClusterSize,
......@@ -70,9 +70,9 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
using ThreadwiseWelford =
ThreadwiseWelford<AccDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
ThreadwiseWelford<ComputeDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
using BlockwiseWelford = BlockwiseWelford<AccDataType,
using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
BlockSize,
ThreadClusterLengths_M_K,
ThreadClusterArrangeOrder>;
......@@ -115,12 +115,12 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
const GridDesc_M_K& beta_grid_desc_m_k,
const GridDesc_M_K& y_grid_desc_m_k,
index_t num_k_block_tile_iteration,
AccDataType epsilon,
ComputeDataType epsilon,
const XDataType* const __restrict__ p_x_global,
const GammaDataType* const __restrict__ p_gamma_global,
const BetaDataType* const __restrict__ p_beta_global,
YDataType* const __restrict__ p_y_global,
const AccElementwiseOperation acc_elementwise_op)
const YElementwiseOperation y_elementwise_op)
{
if constexpr(SweepOnce)
{
......@@ -133,7 +133,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto x_thread_buf = generate_tuple(
[&](auto) {
return StaticBuffer<AddressSpaceEnum::Vgpr,
AccDataType,
ComputeDataType,
MThreadSliceSize * XSrcVectorSize,
true>{};
},
......@@ -142,7 +142,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto gamma_thread_buf = generate_tuple(
[&](auto) {
return StaticBuffer<AddressSpaceEnum::Vgpr,
AccDataType,
ComputeDataType,
MThreadSliceSize * GammaSrcVectorSize,
true>{};
},
......@@ -151,7 +151,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto beta_thread_buf = generate_tuple(
[&](auto) {
return StaticBuffer<AddressSpaceEnum::Vgpr,
AccDataType,
ComputeDataType,
MThreadSliceSize * BetaSrcVectorSize,
true>{};
},
......@@ -160,14 +160,16 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto y_thread_buf = generate_tuple(
[&](auto) {
return StaticBuffer<AddressSpaceEnum::Vgpr,
AccDataType,
ComputeDataType,
MThreadSliceSize * YDstVectorSize,
true>{};
},
Number<ThreadBufferNumber>{});
StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> mean_thread_buf;
StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> var_thread_buf;
StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
mean_thread_buf;
StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
var_thread_buf;
const index_t thread_local_id = get_thread_local_1d_id();
const index_t block_global_id = get_block_1d_id();
......@@ -179,7 +181,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
const auto thread_k_cluster_id = thread_cluster_idx[I1];
auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
AccDataType,
ComputeDataType,
GridDesc_M_K,
decltype(thread_buffer_desc_m_k),
ThreadBufferLengths_M_K,
......@@ -195,7 +197,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto threadwise_gamma_load =
ThreadwiseTensorSliceTransfer_v2<GammaDataType,
AccDataType,
ComputeDataType,
GridDesc_M_K,
decltype(thread_buffer_desc_m_k),
ThreadBufferLengths_M_K,
......@@ -211,7 +213,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
auto threadwise_beta_load =
ThreadwiseTensorSliceTransfer_v2<BetaDataType,
AccDataType,
ComputeDataType,
GridDesc_M_K,
decltype(thread_buffer_desc_m_k),
ThreadBufferLengths_M_K,
......@@ -226,11 +228,11 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
thread_k_cluster_id * BetaSrcVectorSize));
auto threadwise_y_store =
ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
YDataType,
decltype(thread_buffer_desc_m_k),
GridDesc_M_K,
AccElementwiseOperation,
YElementwiseOperation,
ThreadBufferLengths_M_K,
ThreadBufferDimAccessOrder,
YDstVectorDim,
......@@ -242,7 +244,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
make_multi_index(block_global_id * M_BlockTileSize +
thread_m_cluster_id * MThreadSliceSize,
thread_k_cluster_id * YDstVectorSize),
acc_elementwise_op);
y_elementwise_op);
constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
constexpr auto thread_copy_bwd_step_m_k =
......@@ -261,8 +263,8 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
threadwise_welford.max_count_ = GetKPerThread(x_grid_desc_m_k, thread_k_cluster_id);
static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
mean_thread_buf(I) = type_convert<AccDataType>(0.0f);
var_thread_buf(I) = type_convert<AccDataType>(0.0f);
mean_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
var_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
});
// Separate sweep once and sweep twice pipeline
......
library/src/tensor_operation_instance/gpu/normalization/device_normalization_f16_instance.cpp
View file @ 9d2280d6
......@@ -21,7 +21,7 @@ template <typename OutElementwise, index_t Rank, index_t Reduce>
// clang-format off
using device_normalization_f16_instances =
std::tuple <
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
......
library/src/tensor_operation_instance/gpu/normalization/device_normalization_f32_instance.cpp
View file @ 9d2280d6
......@@ -19,7 +19,7 @@ using Pass = ck::tensor_operation::element_wise::PassThrough;
template <typename OutElementwise, index_t Rank, index_t Reduce>
using device_layernorm_f32_instances = std::tuple<
// clang-format off
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
......
profiler/include/profiler/profile_layernorm_impl.hpp
View file @ 9d2280d6
......@@ -19,7 +19,7 @@ namespace profiler {
template <typename XDataType,
typename GammaDataType,
typename BetaDataType,
typename AccDataType,
typename ComputeDataType,
typename YDataType,
index_t Rank>
bool profile_layernorm_impl(int do_verification,
......@@ -86,7 +86,7 @@ bool profile_layernorm_impl(int do_verification,
using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
......@@ -109,7 +109,7 @@ bool profile_layernorm_impl(int do_verification,
GammaDataType,
BetaDataType,
YDataType,
AccDataType,
ComputeDataType,
PassThrough,
Rank,
NumReduceDim>;
......
test/normalization/test_groupnorm_fp16.cpp
View file @ 9d2280d6
......@@ -15,7 +15,7 @@ class TestGroupnorm : public ::testing::Test
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 AccDataType = std::tuple_element_t<3, Tuple>;
using ComputeDataType = std::tuple_element_t<3, Tuple>;
using YDataType = std::tuple_element_t<4, Tuple>;
void Run()
......@@ -36,7 +36,7 @@ class TestGroupnorm : public ::testing::Test
ck::profiler::profile_groupnorm_impl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType>(true, 2, false, false, length);
EXPECT_TRUE(success);
}
......@@ -44,7 +44,7 @@ class TestGroupnorm : public ::testing::Test
};
using KernelTypes = ::testing::Types<
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
std::tuple<F16, F16, F16, F32, F16>>;
TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
......
test/normalization/test_groupnorm_fp32.cpp
View file @ 9d2280d6
......@@ -15,7 +15,7 @@ class TestGroupnorm : public ::testing::Test
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 AccDataType = std::tuple_element_t<3, Tuple>;
using ComputeDataType = std::tuple_element_t<3, Tuple>;
using YDataType = std::tuple_element_t<4, Tuple>;
void Run()
......@@ -34,7 +34,7 @@ class TestGroupnorm : public ::testing::Test
ck::profiler::profile_groupnorm_impl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType>(true, 2, false, false, length);
EXPECT_TRUE(success);
}
......@@ -42,7 +42,7 @@ class TestGroupnorm : public ::testing::Test
};
using KernelTypes = ::testing::Types<
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
std::tuple<F32, F32, F32, F32, F32>>;
TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
......
test/normalization/test_layernorm2d_fp16.cpp
View file @ 9d2280d6
......@@ -15,7 +15,7 @@ class TestLayernorm2d : public ::testing::Test
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 AccDataType = std::tuple_element_t<3, Tuple>;
using ComputeDataType = std::tuple_element_t<3, Tuple>;
using YDataType = std::tuple_element_t<4, Tuple>;
void Run()
......@@ -29,7 +29,7 @@ class TestLayernorm2d : public ::testing::Test
bool success = ck::profiler::profile_layernorm_impl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType,
2>(true, 2, false, false, length);
EXPECT_TRUE(success);
......@@ -38,7 +38,7 @@ class TestLayernorm2d : public ::testing::Test
};
using KernelTypes = ::testing::Types<
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
std::tuple<F16, F16, F16, F32, F16>>;
TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
......
test/normalization/test_layernorm2d_fp32.cpp
View file @ 9d2280d6
......@@ -15,7 +15,7 @@ class TestLayernorm2d : public ::testing::Test
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 AccDataType = std::tuple_element_t<3, Tuple>;
using ComputeDataType = std::tuple_element_t<3, Tuple>;
using YDataType = std::tuple_element_t<4, Tuple>;
void Run()
......@@ -29,7 +29,7 @@ class TestLayernorm2d : public ::testing::Test
bool success = ck::profiler::profile_layernorm_impl<XDataType,
GammaDataType,
BetaDataType,
AccDataType,
ComputeDataType,
YDataType,
2>(true, 2, false, false, length);
EXPECT_TRUE(success);
......@@ -38,7 +38,7 @@ class TestLayernorm2d : public ::testing::Test
};
using KernelTypes = ::testing::Types<
// XDataType, GammaDataType, BetaDataType, AccDataType, YDataType>
// XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
std::tuple<F32, F32, F32, F32, F32>>;
TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
......
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