mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-05 22:22:27 +00:00
upgrade from clang-format-12 to clang-format-18 (#2568)
* upgrade to clang-format-18 * update to clang-format-18 in pre-commit-config
This commit is contained in:
Illia Silin
@@ -25,27 +25,27 @@ float fused_moe(fused_moe_traits t, fused_moe_args a, const ck_tile::stream_conf
|
||||
}();
|
||||
|
||||
auto t0 = fused_moesorting_trait{"int32", "fp32", t.local_expert_masking};
|
||||
auto a0 = fused_moesorting_args
|
||||
{
|
||||
a.topk_ids_ptr, // const void* p_topk_ids;
|
||||
a.topk_weight_ptr, // const void* p_weights;
|
||||
a.local_expert_mask_ptr, // const void* p_local_expert_mask;
|
||||
a.local_tokens,
|
||||
a.sorted_token_ids_ptr, // void* p_sorted_token_ids;
|
||||
a.sorted_weight_ptr, // void* p_sorted_weights;
|
||||
a.sorted_expert_ids_ptr, // void* p_sorted_expert_ids;
|
||||
a.num_sorted_tiles_ptr, // void* p_total_tokens_post_pad;
|
||||
a.o_ptr, // void* p_moe_buf;
|
||||
a.ws_ptr, // void* p_ws;
|
||||
a.num_tokens, // index_t tokens;
|
||||
a.block_m, // index_t unit_size;
|
||||
a.num_experts, // index_t num_experts;
|
||||
a.topk, // index_t topk;
|
||||
auto a0 = fused_moesorting_args{
|
||||
a.topk_ids_ptr, // const void* p_topk_ids;
|
||||
a.topk_weight_ptr, // const void* p_weights;
|
||||
a.local_expert_mask_ptr, // const void* p_local_expert_mask;
|
||||
a.local_tokens,
|
||||
a.sorted_token_ids_ptr, // void* p_sorted_token_ids;
|
||||
a.sorted_weight_ptr, // void* p_sorted_weights;
|
||||
a.sorted_expert_ids_ptr, // void* p_sorted_expert_ids;
|
||||
a.num_sorted_tiles_ptr, // void* p_total_tokens_post_pad;
|
||||
a.o_ptr, // void* p_moe_buf;
|
||||
a.ws_ptr, // void* p_ws;
|
||||
a.num_tokens, // index_t tokens;
|
||||
a.block_m, // index_t unit_size;
|
||||
a.num_experts, // index_t num_experts;
|
||||
a.topk, // index_t topk;
|
||||
#if MOE_SORTING_FMOE_2D_BUF
|
||||
a.stride_token, o_data_bytes,
|
||||
a.stride_token,
|
||||
o_data_bytes,
|
||||
#else
|
||||
static_cast<ck_tile::long_index_t>(a.num_tokens) *
|
||||
a.stride_token* o_data_bytes // index_t moe_buf_bytes;
|
||||
static_cast<ck_tile::long_index_t>(a.num_tokens) * a.stride_token *
|
||||
o_data_bytes // index_t moe_buf_bytes;
|
||||
#endif
|
||||
};
|
||||
|
||||
|
||||
@@ -16,11 +16,11 @@ float fused_moegemm_(const ck_tile::stream_config& s, fused_moegemm_args a)
|
||||
{
|
||||
using f_traits = ck_tile::FusedMoeGemmTraits<Ts_::GateOnly, Ts_::FusedQuant == 1, 1 /*atomic*/>;
|
||||
using f_shape = ck_tile::FusedMoeGemmShape<typename Ts_::BlockTile_0,
|
||||
typename Ts_::WarpPerBlock_0,
|
||||
typename Ts_::WarpTile_0,
|
||||
typename Ts_::BlockTile_1,
|
||||
typename Ts_::WarpPerBlock_0,
|
||||
typename Ts_::WarpTile_0>;
|
||||
typename Ts_::WarpPerBlock_0,
|
||||
typename Ts_::WarpTile_0,
|
||||
typename Ts_::BlockTile_1,
|
||||
typename Ts_::WarpPerBlock_0,
|
||||
typename Ts_::WarpTile_0>;
|
||||
|
||||
constexpr auto get_activation_ = []() {
|
||||
if constexpr(Ts_::Activation == 0)
|
||||
|
||||
@@ -40,11 +40,11 @@
|
||||
constexpr bool local_expert_masking = local_expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemEx<index_t, \
|
||||
ms_weight_type, \
|
||||
sub_token_tile, \
|
||||
sub_token_onshot, \
|
||||
local_expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
sub_token_tile, \
|
||||
sub_token_onshot, \
|
||||
local_expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingKernel<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
@@ -204,11 +204,11 @@ float fused_moesorting(fused_moesorting_trait t, fused_moesorting_args a, ck_til
|
||||
constexpr bool expert_masking = expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemMp<ms_index_t, \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingMultiPhaseKernel_P0<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
@@ -222,11 +222,11 @@ float fused_moesorting(fused_moesorting_trait t, fused_moesorting_args a, ck_til
|
||||
constexpr bool expert_masking = expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemMp<ms_index_t, \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingMultiPhaseKernel_P1<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
@@ -240,11 +240,11 @@ float fused_moesorting(fused_moesorting_trait t, fused_moesorting_args a, ck_til
|
||||
constexpr bool expert_masking = expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemMp<ms_index_t, \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingMultiPhaseKernel_P2<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
@@ -258,11 +258,11 @@ float fused_moesorting(fused_moesorting_trait t, fused_moesorting_args a, ck_til
|
||||
constexpr bool expert_masking = expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemMp<ms_index_t, \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingMultiPhaseKernel_P3<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
@@ -277,11 +277,11 @@ float fused_moesorting(fused_moesorting_trait t, fused_moesorting_args a, ck_til
|
||||
constexpr bool expert_masking = expert_masking_; \
|
||||
constexpr bool local_token = local_token_; \
|
||||
using ms_problem = ck_tile::MoeSortingProblemMp<ms_index_t, \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
ms_weight_type, \
|
||||
mesh_type_, \
|
||||
unroll_num, \
|
||||
expert_masking, \
|
||||
local_token>; \
|
||||
using kernel = ck_tile::MoeSortingMultiPhaseKernel_P23<ms_problem>; \
|
||||
auto kargs = kernel::MakeKargs(a); \
|
||||
const dim3 grids = kernel::GridSize(a); \
|
||||
|
||||
@@ -218,8 +218,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
|
||||
return std::string(", st:") + std::to_string(stride);
|
||||
}();
|
||||
|
||||
std::cout << "[" << api_str << "|" << prec_str << "]"
|
||||
<< " t:" << tokens;
|
||||
std::cout << "[" << api_str << "|" << prec_str << "]" << " t:" << tokens;
|
||||
|
||||
if(is_local_token)
|
||||
{
|
||||
|
||||
Reference in New Issue
Block a user