Skip to content

GitLab

Commit d0c80b12 authored by shengnxu's avatar shengnxu
Browse files

fix more issues, current status, inline asm using more register than available

parent a759277d
Show whitespace changes
Inline Side-by-side
Showing with 143 additions and 68 deletions
example/ck_tile/15_fused_moe/instances/fused_moegemm_api.cpp
View file @ d0c80b12
......@@ -37,3 +37,4 @@ float fused_moegemm(fused_moegemm_traits t, fused_moegemm_args a, const ck_tile:
// clang-format on
return r;
}
example/ck_tile/15_fused_moe/instances/fused_moegemm_api_internal.hpp
View file @ d0c80b12
......@@ -5,11 +5,26 @@
#include "fused_moegemm_api_traits.hpp"
#include "ck_tile/ops/fused_moe.hpp"
#include "fused_moegemm_api.cpp"
#include <iostream>
template <ck_tile::index_t... Is>
using S = ck_tile::sequence<Is...>;
template<typename dtype, typename problem>
struct PipelineDispatch;
template<typename problem> struct PipelineDispatch<ck_tile::int8_t , problem> {
using type = ck_tile::FusedMoeGemmPipeline_FlatmmUk_int8<problem>;
};
template<typename problem> struct PipelineDispatch<ck_tile::bf16_t, problem> {
using type = ck_tile::FusedMoeGemmPipeline_FlatmmUk<problem>;
};
template<typename problem> struct PipelineDispatch<ck_tile::fp16_t, problem> {
using type = ck_tile::FusedMoeGemmPipeline_FlatmmUk<problem>;
};
// do not the define of this tepmlate function inside the _api.cpp, otherwise will block make -j
template <typename Ts_>
float fused_moegemm_(const ck_tile::stream_config& s, fused_moegemm_args a)
......@@ -38,8 +53,8 @@ float fused_moegemm_(const ck_tile::stream_config& s, fused_moegemm_args a)
f_traits>;
// using f_pipeline = ck_tile::FusedMoeGemmPipeline_FlatmmEx<f_problem>;
using f_pipeline = ck_tile::FusedMoeGemmPipeline_FlatmmUk_int8<
>;
using f_pipeline = typename PipelineDispatch<typename Ts_::ADataType, f_problem>::type;
using f_partitioner = ck_tile::FusedMoeGemmTilePartitioner_Linear<f_shape>;
using f_kernel = ck_tile::FusedMoeGemmKernel<f_partitioner, f_pipeline, void>;
......
example/ck_tile/15_fused_moe/instances/fused_moegemm_int8_m32.cpp 0 → 100644
View file @ d0c80b12
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <ck_tile/core.hpp>
#include "fused_moegemm.hpp"
#include "fused_moegemm_api_traits.hpp"
#include "fused_moegemm_api_internal.hpp"
// clang-format off
template float fused_moegemm_<
fmoe_<ck_tile::int8_t, ck_tile::int8_t, ck_tile::bf16_t, float, float, float, float, S<32, 512, 256, 256>, S<1, 4, 1>, S<16, 16, 64>, 1, 1>
>(const ck_tile::stream_config& s, fused_moegemm_args a);
// clang-format on
example/ck_tile/15_fused_moe/main.cpp
View file @ d0c80b12
......@@ -204,7 +204,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
using ADataType = typename TypeConfig::ADataType;
using GDataType = typename TypeConfig::GDataType;
using DDataType = typename TypeConfig::DDataType;
using AccDataType = typename TypeConfig::AccDataType;
// using AccDataType = typename TypeConfig::AccDataType;
using ODataType = typename TypeConfig::ODataType;
using AScaleDataType = typename TypeConfig::AScaleDataType;
using GScaleDataType = typename TypeConfig::GScaleDataType;
......@@ -218,12 +218,12 @@ bool run(const ck_tile::ArgParser& arg_parser)
ck_tile::HostTensor<GDataType> g_host({experts, shared_intermediate_size_0, hidden_size});
ck_tile::HostTensor<DDataType> d_host({experts, hidden_size, shared_intermediate_size_1});
ck_tile::HostTensor<ODataType> o_host({tokens, hidden_size}, {stride, 1});
if (fused_quant == 1)
{
// if (fused_quant == 1)
// {
ck_tile::HostTensor<AScaleDataType> sa_host({tokens, topk});
} else{
ck_tile::HostTensor<AScaleDataType> sa_host({tokens});
}
// } else{
// ck_tile::HostTensor<AScaleDataType> sa_host({tokens});
// }
ck_tile::HostTensor<GScaleDataType> sg_host({experts, shared_intermediate_size_0});
ck_tile::HostTensor<DScaleDataType> sd_host({experts, shared_intermediate_size_1});
ck_tile::HostTensor<YSmoothScaleDataType> sy_host({experts, shared_intermediate_size_1}); // smooth-quant
......@@ -425,7 +425,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
experts,
block_m);
ck_tile::reference_fused_moe<AccDataType, ck_tile::element_wise::Gelu>(
ck_tile::reference_fused_moe<float, ck_tile::element_wise::Gelu>(
a_host,
g_host,
d_host,
......@@ -535,7 +535,7 @@ bool run(const ck_tile::ArgParser& arg_parser)
if(do_validation)
{
ck_tile::reference_fused_moe<AccDataType, ck_tile::element_wise::Gelu>(
ck_tile::reference_fused_moe<float, ck_tile::element_wise::Gelu>(
a_host,
g_host,
d_host,
......@@ -555,7 +555,8 @@ bool run(const ck_tile::ArgParser& arg_parser)
hidden_size,
shared_intermediate_size_0,
topk,
gate_only);
gate_only,
1);
auto o_dev = o_buf.ToHost<ODataType>();
// o_dev.savetxt("gpu-out.txt", "float");
......@@ -604,6 +605,13 @@ int main(int argc, char* argv[])
? 0
: -2;
}
else if(prec_i == "int8" && prec_w == "int8" && prec_o == "bf16" && prec_kw == "fp32")
{
return run<ck_tile::int8_t, ck_tile::int8_t, ck_tile::bf16_t, float, float, float, float>(
arg_parser)
? 0
: -2;
}
return -3;
}
include/ck_tile/host/reference/reference_fused_moe.hpp
View file @ d0c80b12
......@@ -107,10 +107,9 @@ void reference_fused_moe(
return;
ck_tile::index_t i_expert = sorted_expert_ids_host.mData[i_tile];
ck_tile::index_t i_token = sorted_token_ids_host.mData[i_flatten];
ck_tile::index_t i_weight_idx;
ck_tile::index_t i_weight_idx = i_token >> 24;
if(fquant == 1)
{
i_weight_idx = i_token >> 24;
i_token = i_token & 0xffffff;
}
if (i_token >= tokens)
......
include/ck_tile/ops/flatmm/block/flatmm_32x512x256_1x4x1_16x16x64_int8.hpp
View file @ d0c80b12
......@@ -245,9 +245,9 @@ struct Flatmm_32x512x256_1x4x1_16x16x64_int8 : public Flatmm_32x512x256_1x4x1_16
// TODO: need paired with tile_window_linear!
// TODO: need call init_raw() before call this function!
template <typename AQRes, typename DQRes, typename GQRes, typename SMQRes, typename ARes, typename ACoords, typename BRes, typename BCoords>
template <typename AToken_id, typename AQRes, typename DQRes, typename GQRes, typename SMQRes, typename ARes, typename ACoords, typename BRes, typename BCoords>
CK_TILE_DEVICE auto
operator()( index_t row_ids_a_,
operator()( const AToken_id& row_ids_a_,
const AQRes& res_aq,
const DQRes& res_dq,
const GQRes& res_gq,
......@@ -263,6 +263,7 @@ struct Flatmm_32x512x256_1x4x1_16x16x64_int8 : public Flatmm_32x512x256_1x4x1_16
{
static_assert(ACoords::size() == Block_M * Block_K / BlockSize / 4 /*2x per dword*/); // 8
static_assert(BCoords::size() == Repeat_N);
static_assert(AToken_id::size() == Repeat_M);
auto a_sst = make_tile_window(
make_tensor_view<address_space_enum::lds>(
......@@ -371,6 +372,8 @@ struct Flatmm_32x512x256_1x4x1_16x16x64_int8 : public Flatmm_32x512x256_1x4x1_16
register int v_z62 asm("v190") = 0;
register int v_z63 asm("v191") = 0;
index_t temp0 = static_cast<index_t>(row_ids_a_[number<0>{}]);
index_t temp1 = static_cast<index_t>(row_ids_a_[number<1>{}]);
// B nr->kr
#pragma clang diagnostic push
......@@ -397,7 +400,8 @@ struct Flatmm_32x512x256_1x4x1_16x16x64_int8 : public Flatmm_32x512x256_1x4x1_16
// [v_acc_13]"+v"(v_acc[13]),
// [v_acc_14]"+v"(v_acc[14]),
// [v_acc_15]"+v"(v_acc[15]),
[v_token_id]"+v"(row_ids_a_),
[v_token_id0]"+v"(temp0),
[v_token_id1]"+v"(temp1),
[s_mem_]"+r"(smem)
: [s_res_aq0]"s"(res_aq[0]),
[s_res_aq1]"s"(res_aq[1]),
......
include/ck_tile/ops/flatmm/block/flatmm_sn_32x256x512_1x4x1_16x16x64_int8.hpp
View file @ d0c80b12
......@@ -98,7 +98,7 @@ struct FlatmmSn_32x256x512_1x4x1_16x16x64_int8 : public FlatmmSn_32x256x512_1x4x
index_t tile_offset_half_b, //splited load alone K in to 2 part
index_t tile_offset_o)
{
static_assert(BCoords::size() == 8); // 8
static_assert(BCoords::size() == 4); // 8
static_assert(OCoords::size() == 8);
const index_t tile_stride_b_bytes = tile_offset_b * sizeof(BDataType);
......@@ -238,11 +238,6 @@ struct FlatmmSn_32x256x512_1x4x1_16x16x64_int8 : public FlatmmSn_32x256x512_1x4x
[v_os_b1]"v"(static_cast<index_t>(cached_coords_b[number<1>{}] * sizeof(BDataType))),
[v_os_b2]"v"(static_cast<index_t>(cached_coords_b[number<2>{}] * sizeof(BDataType))),
[v_os_b3]"v"(static_cast<index_t>(cached_coords_b[number<3>{}] * sizeof(BDataType))),
[v_os_b4]"v"(static_cast<index_t>(cached_coords_b[number<4>{}] * sizeof(BDataType))),
[v_os_b5]"v"(static_cast<index_t>(cached_coords_b[number<5>{}] * sizeof(BDataType))),
[v_os_b6]"v"(static_cast<index_t>(cached_coords_b[number<6>{}] * sizeof(BDataType))),
[v_os_b7]"v"(static_cast<index_t>(cached_coords_b[number<7>{}] * sizeof(BDataType))),
[s_tile_os_o]"s"(tile_stride_o_bytes),
[s_tile_os_b_half]"s"(tile_offset_half_b_bytes),
[s_tile_os_b]"s"(tile_stride_b_bytes),
......@@ -393,11 +388,6 @@ struct FlatmmSn_32x256x512_1x4x1_16x16x64_int8 : public FlatmmSn_32x256x512_1x4x
[v_os_b1]"v"(static_cast<index_t>(cached_coords_b[number<1>{}] * sizeof(BDataType))),
[v_os_b2]"v"(static_cast<index_t>(cached_coords_b[number<2>{}] * sizeof(BDataType))),
[v_os_b3]"v"(static_cast<index_t>(cached_coords_b[number<3>{}] * sizeof(BDataType))),
[v_os_b4]"v"(static_cast<index_t>(cached_coords_b[number<4>{}] * sizeof(BDataType))),
[v_os_b5]"v"(static_cast<index_t>(cached_coords_b[number<5>{}] * sizeof(BDataType))),
[v_os_b6]"v"(static_cast<index_t>(cached_coords_b[number<6>{}] * sizeof(BDataType))),
[v_os_b7]"v"(static_cast<index_t>(cached_coords_b[number<7>{}] * sizeof(BDataType))),
[s_tile_os_o]"s"(tile_stride_o_bytes),
[s_tile_os_b_half]"s"(tile_offset_half_b_bytes),
[s_tile_os_b]"s"(tile_stride_b_bytes),
......
include/ck_tile/ops/flatmm/block/uk/flatmm_uk_gfx9_32x512x256_1x1x1_16x16x32_int8.inc
View file @ d0c80b12
......@@ -78,18 +78,17 @@
" v_mov_b32 v53, 0x00007fff \n"
" s_waitcnt 0x0000 \n"
";---------------------------------------------- \n"
" v_mov_b32 %[v_token_id], %[v_token_id] \n"
" v_lshrrev_b32 v54, 24, %[v_token_id] \n"
" v_mul_i32_i24 v54, s66, v54 \n"
" v_and_b32 v55, 0x00ffffff, %[v_token_id] \n"
" v_add_u32 %[v_token_id], v54, v55 \n"
" v_lshrrev_b32 v54, 24, v7 \n"
" v_mul_i32_i24 v54, s66, v54 \n"
" v_and_b32 v55, 0x00ffffff, v7 \n"
" v_lshrrev_b32 v54, 24, %[v_token_id0] \n"
" v_mul_i32_i24 v54, s66, v54 \n"
" v_and_b32 v55, 0x00ffffff, %[v_token_id0] \n"
" v_add_u32 v6, v54, v55 \n"
" v_lshrrev_b32 v54, 24, %[v_token_id1] \n"
" v_mul_i32_i24 v54, s66, v54 \n"
" v_and_b32 v55, 0x00ffffff, %[v_token_id1] \n"
" v_add_u32 v7, v54, v55 \n"
" v_lshlrev_b32 %[v_token_id], 2, %[v_token_id] \n"
" v_lshlrev_b32 v6, 2, v6 \n"
" v_lshlrev_b32 v7, 2, v7 \n"
" buffer_load_dword v14, %[v_token_id], s[28:31], 0 offen \n"
" buffer_load_dword v14, v6, s[28:31], 0 offen \n"
" buffer_load_dword v15, v7, s[28:31], 0 offen \n"
" buffer_load_dword v16, v10, s[32:35], 0 offen \n"
" buffer_load_dword v17, v11, s[32:35], 0 offen \n"
......
include/ck_tile/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_policy.hpp
View file @ d0c80b12
......@@ -804,6 +804,13 @@ struct FusedMoeGemmPipelineFlatmmPolicy
{
return Flatmm_32x512x128_1x4x1_16x16x32_FP16{};
}
else if constexpr(std::is_same_v<typename Problem::ADataType, ck_tile::int8_t> &&
std::is_same_v<typename Problem::GDataType, ck_tile::int8_t> &&
S_::Block_M0 == 32 && S_::Block_N0 == 512 && S_::Block_K0 == 256 &&
S_::Warp_M0 == 16 && S_::Warp_N0 == 16 && S_::Warp_K0 == 64)
{
return Flatmm_32x512x256_1x4x1_16x16x64_int8{};
}
}
template <typename Problem>
......@@ -851,6 +858,20 @@ struct FusedMoeGemmPipelineFlatmmPolicy
// return FlatmmSn_32x128x512_1x4x1_16x16x32_FP16{};
return FlatmmSn_32x128x512_1x4x1_16x16x32_FP16_itl{};
}
else if constexpr(std::is_same_v<typename Problem::YDataType, ck_tile::int8_t> &&
std::is_same_v<typename Problem::DDataType, ck_tile::int8_t> &&
S_::Block_M1 == 32 && S_::Block_N1 == 256 && S_::Block_K1 == 512 &&
S_::Warp_M0 == 16 && S_::Warp_N0 == 16 && S_::Warp_K0 == 64 &&
T_::PipeInterleave == false)
{
return FlatmmSn_32x256x512_1x4x1_16x16x64_int8{};
// return FlatmmSn_32x128x512_1x4x1_16x16x32_FP16_itl{};
}
else
{
return FlatmmSn_32x256x512_1x4x1_16x16x64_int8{};
// return FlatmmSn_32x128x512_1x4x1_16x16x32_FP16_itl{};
}
}
};
} // namespace ck_tile
include/ck_tile/ops/fused_moe/pipeline/fused_moegemm_pipeline_flatmm_uk_int8.hpp
View file @ d0c80b12
......@@ -116,7 +116,19 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
return coords;
}
CK_TILE_DEVICE auto GetRowCoords_A_mma(index_t base_offset)
{
// constexpr index_t KLans = 2;
constexpr index_t MLans = 16;
constexpr index_t MRepeat = BlockShape::Repeat_M1;
auto base_coord = threadIdx.x % MLans + base_offset;
array<index_t, MRepeat> coords;
static_for<0, MRepeat, 1>{}([&](auto i) { coords.at(i) = base_coord + i * MLans; });
return coords;
}
template <typename ROW_COORDS>
CK_TILE_DEVICE auto GetRowID(const ROW_COORDS coords, const IndexDataType* sorted_token_ids_ptr)
{
......@@ -178,7 +190,6 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
index_t expert_stride_0 = shared_intermediate_size_0 * kargs.hidden_size;
index_t expert_stride_1 = shared_intermediate_size_1 * kargs.hidden_size;
/////////////
index_t a_scale_expert_stride_0 = kargs.hidden_size;
index_t g_scale_expert_stride_0 = shared_intermediate_size_0;
index_t smq_scale_expert_stride_0 = shared_intermediate_size_0;
index_t d_scale_expert_stride_1 = kargs.hidden_size;
......@@ -192,13 +203,25 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
BlockShape::Block_Kr1); // intermediate_tile_id * Block_N / (N in W)
auto row_coords_a = GetRowCoords_A(sorted_tile_id * BlockShape::Block_M0);
auto row_coords_a_mma = GetRowCoords_A_mma(sorted_tile_id * BlockShape::Block_M0);
auto row_ids_a = GetRowID(
row_coords_a, reinterpret_cast<const IndexDataType*>(kargs.sorted_token_ids_ptr));
auto token_id = row_ids_a & 0xffffff;
auto row_ids_a_mma = GetRowID(
row_coords_a_mma, reinterpret_cast<const IndexDataType*>(kargs.sorted_token_ids_ptr));
auto token_id = generate_tuple(
[&](auto i) {
return (row_ids_a[i]) &0xffffff;
},
number<row_ids_a.size()>{});
// auto token_id_mma = generate_tuple(
// [&](auto i) {
// return (row_ids_a_mma[i]) &0xffffff;
// },
// number<row_ids_a_mma.size()>{});
//addr in fact
auto a_coords = generate_tuple(
[&](auto i) {
return (token_id[i]) * kargs.stride_token +
return (row_ids_a[i]) * kargs.stride_token +
threadIdx.x % (BlockShape::Block_K0 / kAlignmentA) * kAlignmentA;
},
number<row_ids_a.size()>{});
......@@ -208,7 +231,7 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
//////aq
auto aq_win = [&]() {
const AScaleDataType* aq_ptr = reinterpret_cast<const AScaleDataType*>(kargs.a_scale_ptr);
auto aq_view_ = make_naive_tensor_view<address_space_enum::global>(
auto aq_view_ = make_naive_tensor_view_packed<address_space_enum::global>(
aq_ptr,
make_tuple(kargs.num_tokens * kargs.topk),
number<1>{});
......@@ -249,7 +272,7 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
static_cast<long_index_t>(expert_id) * g_scale_expert_stride_0 +
intermediate_tile_id * BlockShape::Block_N0;
// const GDataType* g_ptr = reinterpret_cast<const GScaleDataType*>(kargs.g_scale_ptr);//remember to add expert id for inline
auto gq_view_ = make_naive_tensor_view<address_space_enum::global>(
auto gq_view_ = make_naive_tensor_view_packed<address_space_enum::global>(
gq_ptr,
make_tuple(shared_intermediate_size_1),
number<1>{});
......@@ -264,7 +287,7 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
static_cast<long_index_t>(expert_id) * smq_scale_expert_stride_0 +
intermediate_tile_id * BlockShape::Block_N0;
// const GDataType* g_ptr = reinterpret_cast<const GScaleDataType*>(kargs.g_scale_ptr);//remember to add expert id for inline
auto smq_view_ = make_naive_tensor_view<address_space_enum::global>(
auto smq_view_ = make_naive_tensor_view_packed<address_space_enum::global>(
smq_ptr,
make_tuple(shared_intermediate_size_1),
number<1>{});
......@@ -303,7 +326,7 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
const DScaleDataType* g_ptr = reinterpret_cast<const DScaleDataType*>(kargs.d_scale_ptr) +
static_cast<long_index_t>(expert_id) * d_scale_expert_stride_1;
// const GDataType* g_ptr = reinterpret_cast<const GScaleDataType*>(kargs.d_scale_ptr)//remember to add expert_id as expert_idx
auto g_view_ = make_naive_tensor_view<address_space_enum::global>(
auto g_view_ = make_naive_tensor_view_packed<address_space_enum::global>(
g_ptr,
make_tuple(kargs.hidden_size),
number<1>{});
......@@ -323,12 +346,12 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
auto d_coords = [&]() {
constexpr index_t Nr_ = 4;
constexpr index_t Nw_ = 4;
constexpr index_t Kr0_ = 2;//no more need in int8, method changed, this will be handed in res_s
constexpr index_t Kr1_ = 4;
constexpr index_t Kl_ = 4;
// constexpr index_t Kr0_ = 2;//no more need in int8, method changed, this will be handed in res_s
// constexpr index_t Kr1_ = 4;
// constexpr index_t Kl_ = 4;
constexpr index_t Nl_ = 16;
constexpr index_t Kv_ = 16;
constexpr index_t W_ = Kl_ * Nl_ * Kv_;
// constexpr index_t W_ = Kl_ * Nl_ * Kv_;
//constexpr index_t num_offsets_ = Nr_ * Kr0_;
constexpr index_t num_offsets_ = Nr_ ;
index_t base_os_ = (threadIdx.x % 64) * Kv_ + (threadIdx.x / 64) *
......@@ -351,18 +374,18 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
number<row_ids_a.size()>{});
auto o_flags =
generate_tuple([&](auto i) { return cmp_lt_to_exec(token_id, kargs.num_tokens); },
generate_tuple([&](auto i) { return cmp_lt_to_exec(token_id[i], kargs.num_tokens); },
number<row_ids_a.size()>{});
auto bridge_sst_win = [&]() {
constexpr auto desc_ = Policy::template MakeBridgeLdsStoreForUKDesc<Problem>();
constexpr auto dist_ = Policy::template GetUK_0<Problem>().MakeCBlockDist();
return make_tile_window_linear(make_tensor_view<address_space_enum::lds>(
reinterpret_cast<YDataType*>(smem), desc_),
desc_.get_lengths(),
{0, 0},
dist_);
}();
// auto bridge_sst_win = [&]() {
// constexpr auto desc_ = Policy::template MakeBridgeLdsStoreForUKDesc<Problem>();
// constexpr auto dist_ = Policy::template GetUK_0<Problem>().MakeCBlockDist();
// return make_tile_window_linear(make_tensor_view<address_space_enum::lds>(
// reinterpret_cast<YDataType*>(smem), desc_),
// desc_.get_lengths(),
// {0, 0},
// dist_);
// }();
auto o_res =
make_wave_buffer_resource(reinterpret_cast<const ODataType*>(kargs.o_ptr),
kargs.num_tokens * kargs.stride_token * sizeof(ODataType));
......@@ -372,12 +395,13 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
row_coords_o, reinterpret_cast<const TopkWeightDataType*>(kargs.sorted_weight_ptr));
auto uk_0 = Policy::template GetUK_0<Problem>();
auto acc_0= uk_0(
row_ids_a,//fake token id, 2D index for X scale
// auto acc_0= uk_0(
uk_0(
row_ids_a_mma,//fake token id, 2D index for X scale
aq_res,
gq_res,
gq_res,
dq_res,
gq_res,
smq_res,
a_res,
a_coords,
g_res,
......@@ -415,7 +439,7 @@ struct FusedMoeGemmPipeline_FlatmmUk_int8
kargs.hidden_size, // total n number
w_scale,
BlockShape::Block_N1,
shared_intermediate_size_1 * Block_N1 - kr_1 * BlockShape::Block_W1, // along N
shared_intermediate_size_1 * BlockShape::Block_N1 - kr_1 * BlockShape::Block_W1, // along N
kr_1 * BlockShape::Block_W1,
BlockShape::Block_N1); // along N
}
......
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