Skip to content

GitLab

Unverified Commit 6bbb94f4 authored by Rostyslav Geyyer's avatar Rostyslav Geyyer Committed by GitHub
Browse files

Merge branch 'develop' into lwpck-1815

parents 4c850c90 a32b1bc6
Hide whitespace changes
Inline Side-by-side
Showing with 386 additions and 7 deletions
include/ck_tile/ops/welford/thread/thread_welford.hpp 0 → 100644
View file @ 6bbb94f4
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
namespace ck_tile {
template <typename ComputeDataType_, typename XDataType_>
struct ThreadWelford
{
using XDataType = remove_cvref_t<XDataType_>;
using ComputeDataType = remove_cvref_t<ComputeDataType_>;
template <typename T>
CK_TILE_DEVICE void Update(T& mean, T& var, T x)
{
if(ck_tile::isnan(x))
{
mean = x;
var = x;
}
else
{
T delta = x - mean;
mean += delta / cur_count_;
T delta2 = x - mean;
var += delta * delta2;
}
}
// [CAUSION] - max_count_ is to deal with the padding problem
// max_count_ is depend on caller, eg: naive and splitN welford will have different
// calculation of max_count_
CK_TILE_DEVICE constexpr ThreadWelford(int max_count) : cur_count_(0), max_count_(max_count) {}
template <typename XDistributedTensor_,
typename MeanDistributedTensor_,
typename VarDistributedTensor_>
CK_TILE_DEVICE void operator()(const XDistributedTensor_& x_tensor,
MeanDistributedTensor_& mean_tensor,
VarDistributedTensor_& var_tensor)
{
constexpr auto I0 = number<0>{};
constexpr auto I1 = number<1>{};
constexpr auto spans = XDistributedTensor_::get_distributed_spans();
sweep_tile_span(spans[I1], [&](auto dstr_idx_i1) {
if(cur_count_ < max_count_)
{
++cur_count_;
sweep_tile_span(spans[I0], [&](auto dstr_idx_i0) {
constexpr auto in_dstr_idx = make_tuple(dstr_idx_i0, dstr_idx_i1);
constexpr auto out_dstr_idx = make_tuple(dstr_idx_i0);
auto x = ck_tile::type_convert<ComputeDataType>(x_tensor[in_dstr_idx]);
Update(mean_tensor(out_dstr_idx), var_tensor(out_dstr_idx), x);
});
}
});
}
template <typename XDistributedTensor_>
CK_TILE_DEVICE static auto MakeInitialMeanVarDistributedTensor()
{
static_assert(std::is_same_v<XDataType, typename XDistributedTensor_::DataType>, "wrong!");
constexpr auto reduce_dims = sequence<1>{};
constexpr auto dstr =
make_static_tile_distribution(detail::make_reduce_tile_distribution_encoding(
XDistributedTensor_::get_tile_distribution()
.get_static_tile_distribution_encoding(),
reduce_dims));
auto tensor = make_static_distributed_tensor<ComputeDataType>(dstr);
clear_tile(tensor);
return tensor;
}
template <typename XDistributedTensor_>
CK_TILE_DEVICE auto operator()(const XDistributedTensor_& x_tensor)
{
auto mean_tensor = MakeInitialMeanVarDistributedTensor<XDistributedTensor_>();
auto var_tensor = MakeInitialMeanVarDistributedTensor<XDistributedTensor_>();
(*this)(x_tensor, mean_tensor, var_tensor);
return ck_tile::make_tuple(mean_tensor, var_tensor);
}
int cur_count_;
int max_count_;
};
} // namespace ck_tile
include/ck_tile/ops/welford/warp/warp_welford.hpp 0 → 100644
View file @ 6bbb94f4
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include "ck_tile/core.hpp"
namespace ck_tile {
template <typename ComputeDataType_, bool BroadcastLane = true, bool GetActualVariance = true>
struct WarpMergeWelford
{
using ComputeDataType = remove_cvref_t<ComputeDataType_>;
template <typename T>
CK_TILE_DEVICE static void
Merge(T& mean_a, T& var_a, int& count_a, T mean_b, T var_b, int count_b)
{
int count = count_a + count_b;
T count_ = type_convert<T>(count);
T count_a_ = type_convert<T>(count_a);
T count_b_ = type_convert<T>(count_b);
T count_b_over_count = count == 0 ? type_convert<T>(0) : count_b_ / count_;
T delta = mean_b - mean_a;
mean_a += delta * count_b_over_count;
var_a += var_b + delta * delta * count_a_ * count_b_over_count;
count_a = count;
}
template <typename MeanDistributedTensor_, typename VarDistributedTensor_>
CK_TILE_DEVICE void
operator()(MeanDistributedTensor_& mean_tensor, VarDistributedTensor_& var_tensor, int& count)
{
using Dstr = typename MeanDistributedTensor_::StaticTileDistribution;
using DstrEncode = typename Dstr::DstrEncode;
using DstrEncodeDetail = typename DstrEncode::detail;
static_assert(std::is_same_v<Dstr, typename VarDistributedTensor_::StaticTileDistribution>,
"wrong!");
constexpr index_t NDimP = Dstr::get_num_of_dimension_p();
constexpr index_t NDimR = Dstr::get_num_of_dimension_r();
constexpr index_t idim_p_lane = NDimP - 1;
const auto ps_idx = make_array<index_t>(get_warp_id(), get_lane_id());
const auto rs_idx =
mean_tensor.get_tile_distribution().calculate_rs_index_from_ps_index(ps_idx);
constexpr index_t thread_buf_size = MeanDistributedTensor_::get_thread_buffer_size();
static_assert(thread_buf_size == VarDistributedTensor_::get_thread_buffer_size());
const int original_count = count;
// loop over thread data
static_for<0, thread_buf_size, 1>{}([&](auto i) {
auto v_local_mean = mean_tensor.get_thread_buffer()[i];
auto v_local_var = var_tensor.get_thread_buffer()[i];
auto v_local_count = original_count;
// cross-lane reduce for replication
// only reduce on R dimension correspond to lane
// (lane id maps to this R dimension)
static_for<0, NDimR, 1>{}([&](auto idim_r) {
// FIXME: nasty to use does_p_own_r_
if constexpr(DstrEncodeDetail::does_p_own_r_[idim_p_lane][idim_r])
{
constexpr index_t r_length = DstrEncode::rs_lengths_[idim_r];
constexpr index_t lid_over_rid_derivative =
DstrEncodeDetail::ps_over_rs_derivative_[idim_p_lane][idim_r];
static_assert(is_power_of_two_integer(r_length),
"wrong! only support power of 2 reduction");
constexpr index_t nstage = integer_log2_floor(r_length);
// reduction sweep forward
static_for<0, nstage, 1>{}([&](auto istage) {
constexpr index_t lid_delta =
lid_over_rid_derivative * (1 << (nstage - istage - 1));
// pull data from remote lane
const auto v_remote_mean = warp_shuffle_down(v_local_mean, lid_delta);
const auto v_remote_var = warp_shuffle_down(v_local_var, lid_delta);
const auto v_remote_count = warp_shuffle_down(v_local_count, lid_delta);
// welford merge
Merge(v_local_mean,
v_local_var,
v_local_count,
v_remote_mean,
v_remote_var,
v_remote_count);
});
}
});
// cross-lane broadcast for replication
// only broadcast on R dimension correspond to lane
// (lane id maps to this R dimension)
if constexpr(BroadcastLane)
{
static_for<0, NDimR, 1>{}([&](auto idim_r) {
// FIXME: nasty to use does_p_own_r_
if constexpr(DstrEncodeDetail::does_p_own_r_[idim_p_lane][idim_r])
{
const index_t r_id = rs_idx[idim_r];
constexpr index_t r_length = DstrEncode::rs_lengths_[idim_r];
constexpr index_t lid_over_rid_derivative =
DstrEncodeDetail::ps_over_rs_derivative_[NDimP - 1][idim_r];
static_assert(is_power_of_two_integer(r_length),
"wrong! only support power of 2 reduction");
constexpr index_t nstage = integer_log2_floor(r_length);
// broadcast sweep backward
static_for<0, nstage, 1>{}([&](auto istage) {
// do I hold reduced data?
const bool do_i_hold_reduced_data = r_id < (1 << istage);
constexpr index_t lid_delta = lid_over_rid_derivative * (1 << istage);
// pull data from remote lane
const auto v_remote_mean = warp_shuffle_up(v_local_mean, lid_delta);
const auto v_remote_var = warp_shuffle_up(v_local_var, lid_delta);
const auto v_remote_count = warp_shuffle_up(v_local_count, lid_delta);
// decide whether to update local data with remote data
v_local_mean = do_i_hold_reduced_data ? v_local_mean : v_remote_mean;
v_local_var = do_i_hold_reduced_data ? v_local_var : v_remote_var;
v_local_count = do_i_hold_reduced_data ? v_local_count : v_remote_count;
});
}
});
}
mean_tensor.get_thread_buffer()(i) = v_local_mean;
if constexpr(GetActualVariance)
var_tensor.get_thread_buffer()(i) = v_local_var / v_local_count;
else
var_tensor.get_thread_buffer()(i) = v_local_var;
count = v_local_count;
});
}
};
} // namespace ck_tile
library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_outelementop_instance.hpp
View file @ 6bbb94f4
......@@ -73,6 +73,43 @@ using device_grouped_conv_fwd_xdl_outelementop_f8_instances = std::tuple<
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
typename DsLayout,
typename ELayout,
ConvolutionForwardSpecialization ConvSpec,
typename OutElementOp>
using device_grouped_conv_fwd_xdl_outelementop_bf8_instances = std::tuple<
// clang-format off
//########################################| NumDim| A| B| Ds| E| AData| BData| AccData| CShuffle| Ds| EData| A| B| CDE| ConvForward| 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|
//########################################| Spatial| Layout| Layout| Layout| Layout| Type| Type| Type| DataType| DataType| Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| 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| |
//########################################| | | | | | | | | | | | 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| |
//########################################| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | |
#if(defined(CK_ENABLE_FP8) && defined(CK_ENABLE_BF8))
// generic instance
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, BF8>,
// instances for small conv.K and conv.C
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 1, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 1, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 256, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 256, 32, 8, 8, 32, 32, 2, 4, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 128, 32, 8, 8, 32, 32, 4, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 64, 128, 32, 8, 8, 32, 32, 2, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 64, 32, 8, 8, 32, 32, 2, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 128, 64, 32, 8, 8, 32, 32, 2, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 256, 64, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 128, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 32, 1, 4>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 128, 32, 128, 32, 8, 8, 32, 32, 1, 2, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 8>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 64, 32, 32, 8, 8, 32, 32, 2, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BF8>,
DeviceGroupedConvFwdMultipleABD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout, DsLayout,ELayout, BF8, BF8, F32, F32, Tuple<>, F8, PassThrough, PassThrough, OutElementOp, ConvSpec, GemmMNKPadding, 1, 64, 32, 64, 32, 8, 8, 32, 32, 1, 2, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, S<4, 16, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 8, 8, 1, 1, 1, S<1, 16, 1, 4>, 8, BF8>
#endif
// clang-format on
>;
template <index_t NDimSpatial,
typename ALayout,
typename BLayout,
......
library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward_convscale.hpp
View file @ 6bbb94f4
......@@ -39,7 +39,22 @@ void add_device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_f8_instanc
F8>>>& instances);
#endif
#if defined(CK_ENABLE_FP8) && defined(CK_ENABLE_BF8)
#if(defined(CK_ENABLE_FP8) && defined(CK_ENABLE_BF8))
void add_device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_bf8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
BF8,
BF8,
ck::Tuple<>,
F8,
PassThrough,
PassThrough,
ConvScale,
BF8>>>& instances);
void add_device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_f8_bf8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
......@@ -115,7 +130,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
op_ptrs);
}
#endif
#if defined(CK_ENABLE_FP8) && defined(CK_ENABLE_BF8)
#if(defined(CK_ENABLE_FP8) && defined(CK_ENABLE_BF8))
if constexpr(is_same_v<InDataType, BF8> && is_same_v<WeiDataType, BF8> &&
is_same_v<OutDataType, F8> && is_same_v<AComputeType, BF8> &&
is_same_v<BComputeType, BF8>)
{
add_device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_bf8_instances(
op_ptrs);
}
if constexpr(is_same_v<InDataType, f8_t> && is_same_v<WeiDataType, bf8_t> &&
is_same_v<OutDataType, f8_t> && is_same_v<AComputeType, f8_t> &&
is_same_v<BComputeType, bf8_t>)
......
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_convscale/CMakeLists.txt
View file @ 6bbb94f4
# ONLY XDL_KERNELS
set(GROUPED_CONV3D_FWD_CONVSCALE
xdl/device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_f8_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_bf8_instance.cpp
xdl/device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_f8_bf8_instance.cpp)
add_instance_library(device_grouped_conv3d_fwd_convscale_instance ${GROUPED_CONV3D_FWD_CONVSCALE})
library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd_convscale/xdl/device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_bf8_instance.cpp 0 → 100644
View file @ 6bbb94f4
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_outelementop_instance.hpp"
#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
namespace instance {
using ConvScale = ck::tensor_operation::element_wise::ConvScale;
void add_device_grouped_conv3d_fwd_xdl_convscale_ndhwgc_gkzyxc_ndhwgk_bf8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleABD<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
BF8,
BF8,
ck::Tuple<>,
F8,
PassThrough,
PassThrough,
ConvScale,
BF8>>>& instances)
{
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_outelementop_bf8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ConvFwdDefault,
ConvScale>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_outelementop_bf8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ConvFwd1x1P0,
ConvScale>{});
add_device_operation_instances(
instances,
device_grouped_conv_fwd_xdl_outelementop_bf8_instances<3,
NDHWGC,
GKZYXC,
ck::Tuple<>,
NDHWGK,
ConvFwd1x1S1P0,
ConvScale>{});
}
} // namespace instance
} // namespace device
} // namespace tensor_operation
} // namespace ck
test/grouped_convnd_fwd/test_grouped_convnd_fwd_multi_ab_interface.cpp
View file @ 6bbb94f4
// SPDX-License-Identifier: MIT
// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
......@@ -207,7 +207,7 @@ TEST_F(TestGroupedConvndFwdMultiAInterface, MultiA)
std::array<const void*, NumAs> as{nullptr, nullptr};
const void* b = nullptr;
EXPECT_TRUE(this->template Run(as, b));
EXPECT_TRUE(this->Run(as, b));
}
TEST_F(TestGroupedConvndFwdMultiBInterface, MultiB)
......@@ -215,7 +215,7 @@ TEST_F(TestGroupedConvndFwdMultiBInterface, MultiB)
const void* a = nullptr;
std::array<const void*, NumBs> bs{nullptr, nullptr};
EXPECT_TRUE(this->template Run(a, bs));
EXPECT_TRUE(this->Run(a, bs));
}
TEST_F(TestGroupedConvndFwdMultiABInterface, MultiAB)
......@@ -223,7 +223,7 @@ TEST_F(TestGroupedConvndFwdMultiABInterface, MultiAB)
std::array<const void*, NumAs> as{nullptr, nullptr};
std::array<const void*, NumBs> bs{nullptr, nullptr};
EXPECT_TRUE(this->template Run(as, bs));
EXPECT_TRUE(this->Run(as, bs));
}
TEST_F(TestGroupedConvndFwdInterface, SingleAB)
......@@ -231,5 +231,5 @@ TEST_F(TestGroupedConvndFwdInterface, SingleAB)
const void* a = nullptr;
const void* b = nullptr;
EXPECT_TRUE(this->template Run(a, b));
EXPECT_TRUE(this->Run(a, b));
}
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