mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-05 22:22:27 +00:00
This commit is contained in:
Ding, Yi
381
include/ck_tile/ops/rmsnorm2d/kernel/rmsnorm2d_fwd_kernel.hpp
Normal file
381
include/ck_tile/ops/rmsnorm2d/kernel/rmsnorm2d_fwd_kernel.hpp
Normal file
@@ -0,0 +1,381 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/common.hpp"
|
||||
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_traits.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
// host side args
|
||||
struct Rmsnorm2dFwdHostArgs
|
||||
{
|
||||
const void* p_x; // [m ,n], input, fp16/bf16
|
||||
const void* p_x_residual; // [m ,n], shortcut input, prec same as input, nullptr if not used
|
||||
const void* p_sm_scale; // [1 ,n], smooth scale input, fp32, nullptr if not used
|
||||
const void* p_gamma; // [1, n], gamma, prec same as input
|
||||
|
||||
void* p_y; // [m, n], output, fp16/bf16
|
||||
void* p_y_residual; // [m, n], shortcut output, prec same as input, nullptr if not used
|
||||
void* p_y_scale; // [m, 1], output a dynamic quant per row, nullptr if not used
|
||||
void* p_invRms; // [m, 1], output inv-rms, prec same as input, nullptr if not used
|
||||
void* p_y_unquant; // [m, n], output result before quant, nullptr if not used
|
||||
|
||||
float epsilon;
|
||||
|
||||
index_t m;
|
||||
index_t n;
|
||||
index_t x_stride; // x row_stride
|
||||
index_t xr_stride; // x residule row stride
|
||||
index_t y_stride; // y row stride
|
||||
index_t yr_stride; // y residule row stride
|
||||
};
|
||||
|
||||
// TODO: Extract some type to wrapper class
|
||||
template <typename Pipeline_, typename Epilogue_>
|
||||
struct Rmsnorm2dFwd
|
||||
{
|
||||
using Pipeline = remove_cvref_t<Pipeline_>;
|
||||
using Epilogue = remove_cvref_t<Epilogue_>;
|
||||
using Problem = typename Pipeline::Problem;
|
||||
|
||||
using XDataType = remove_cvref_t<typename Problem::XDataType>;
|
||||
using GammaDataType = remove_cvref_t<typename Problem::GammaDataType>;
|
||||
using ComputeDataType = remove_cvref_t<typename Problem::ComputeDataType>;
|
||||
using YDataType = remove_cvref_t<typename Problem::YDataType>;
|
||||
using InvRmsDataType = remove_cvref_t<typename Problem::InvRmsDataType>;
|
||||
using SmoothScaleDataType = remove_cvref_t<typename Problem::SmoothScaleDataType>;
|
||||
using YScaleDataType = remove_cvref_t<typename Problem::YScaleDataType>;
|
||||
using UnquantYDataType = remove_cvref_t<typename Problem::UnquantYDataType>;
|
||||
|
||||
// for simplicity, shortcut input/output type is same as X
|
||||
using XResidualDataType = XDataType;
|
||||
using YResidualDataType = XDataType;
|
||||
|
||||
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, null_type>;
|
||||
static constexpr bool kSaveInvRms = Problem::Traits::kSaveInvRms;
|
||||
static constexpr bool kSaveUnquant = Problem::Traits::kSaveUnquant;
|
||||
|
||||
static constexpr index_t Block_M = Problem::BlockShape::Block_M;
|
||||
static constexpr index_t Block_N = Problem::BlockShape::Block_N;
|
||||
static constexpr bool kPadM = false; // always no need to pad along M
|
||||
static constexpr bool kPadN = Problem::Traits::kPadN;
|
||||
static constexpr bool kTwoPass = Problem::Traits::kTwoPass;
|
||||
static constexpr auto kFusedAdd = Problem::Traits::kFusedAdd;
|
||||
static constexpr auto kFusedQuant = Problem::Traits::kFusedQuant;
|
||||
static constexpr auto kUseModelSensitiveRMSNorm = Problem::Traits::kUseModelSensitiveRMSNorm;
|
||||
|
||||
static constexpr index_t ThreadPerWarp_N = Problem::BlockShape::ThreadPerWarp_N;
|
||||
static constexpr index_t Vector_N = Problem::BlockShape::Vector_N;
|
||||
static constexpr index_t Repeat_N = Problem::BlockShape::Repeat_N;
|
||||
static constexpr index_t kBlockSize = Problem::BlockShape::BlockSize;
|
||||
|
||||
static constexpr auto I0 = number<0>{};
|
||||
static constexpr auto I1 = number<1>{};
|
||||
|
||||
struct Kargs
|
||||
{
|
||||
const void* p_x;
|
||||
const void* p_x_residual;
|
||||
const void* p_sm_scale;
|
||||
const void* p_gamma;
|
||||
|
||||
void* p_y;
|
||||
void* p_y_residual;
|
||||
void* p_y_scale;
|
||||
void* p_invRms;
|
||||
void* p_y_unquant;
|
||||
|
||||
float epsilon;
|
||||
|
||||
index_t m;
|
||||
index_t n;
|
||||
index_t x_stride; // x row_stride
|
||||
index_t xr_stride; // x residule row stride
|
||||
index_t y_stride; // y row stride
|
||||
index_t yr_stride; // y residule row stride
|
||||
};
|
||||
using Hargs = Rmsnorm2dFwdHostArgs;
|
||||
|
||||
CK_TILE_HOST static constexpr Kargs MakeKargs(const Hargs& hargs)
|
||||
{
|
||||
return Kargs{hargs.p_x,
|
||||
hargs.p_x_residual,
|
||||
hargs.p_sm_scale,
|
||||
hargs.p_gamma,
|
||||
hargs.p_y,
|
||||
hargs.p_y_residual,
|
||||
hargs.p_y_scale,
|
||||
hargs.p_invRms,
|
||||
hargs.p_y_unquant,
|
||||
hargs.epsilon,
|
||||
hargs.m,
|
||||
hargs.n,
|
||||
hargs.x_stride,
|
||||
hargs.xr_stride,
|
||||
hargs.y_stride,
|
||||
hargs.yr_stride};
|
||||
}
|
||||
|
||||
CK_TILE_HOST static constexpr auto GridSize(const Hargs& hargs)
|
||||
{
|
||||
return dim3(integer_divide_ceil(hargs.m, Block_M));
|
||||
}
|
||||
|
||||
CK_TILE_HOST static constexpr auto BlockSize()
|
||||
{
|
||||
return is_wave32() ? Problem::BlockShape::template GetBlockSize<true>()
|
||||
: Problem::BlockShape::template GetBlockSize<false>();
|
||||
}
|
||||
|
||||
// clang-format off
|
||||
template <typename T> struct t2s;
|
||||
template <> struct t2s<float> { static constexpr const char * name = "fp32"; };
|
||||
template <> struct t2s<ck_tile::fp16_t> { static constexpr const char * name = "fp16"; };
|
||||
template <> struct t2s<ck_tile::bf16_t> { static constexpr const char * name = "bf16"; };
|
||||
template <> struct t2s<ck_tile::fp8_t> { static constexpr const char * name = "fp8"; };
|
||||
template <> struct t2s<ck_tile::bf8_t> { static constexpr const char * name = "bf8"; };
|
||||
template <> struct t2s<ck_tile::int8_t> { static constexpr const char * name = "int8"; };
|
||||
// clang-format on
|
||||
|
||||
// in byte
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize() { return Pipeline::GetSmemSize(); }
|
||||
|
||||
CK_TILE_HOST static std::string GetName()
|
||||
{
|
||||
#define _SS_ std::string
|
||||
#define _TS_ std::to_string
|
||||
// clang-format off
|
||||
using S_ = typename Problem::BlockShape;
|
||||
auto surfix = [&] () {
|
||||
std::string n;
|
||||
if (kFusedAdd != Rmsnorm2dFusedAddEnum::NO_ADD) n += _SS_("_") + Rmsnorm2dFusedAddEnumName<kFusedAdd>::name;
|
||||
if (kFusedQuant != Rmsnorm2dFusedQuantEnum::NO_SWEEP) n += _SS_("_") + Rmsnorm2dFusedQuantEnumName<kFusedQuant>::name;
|
||||
if (kPadN) n += "_pn";
|
||||
if (kSaveInvRms) n += "_rms";
|
||||
if (kTwoPass) n += "_2p";
|
||||
if (kUseModelSensitiveRMSNorm == Rmsnorm2dSensitiveEnum::NO_SPECIFIC_MODEL) n += "_nsm";
|
||||
else if (kUseModelSensitiveRMSNorm == Rmsnorm2dSensitiveEnum::T5_MODEL_LIKE) n += "_t5ml";
|
||||
return n; }();
|
||||
|
||||
auto prec_str = [&] () {
|
||||
std::string base_str = _SS_(t2s<XDataType>::name);
|
||||
if (!std::is_same_v<XDataType, YDataType>) {
|
||||
base_str += _SS_("_") + _SS_(t2s<YDataType>::name);
|
||||
}
|
||||
if (kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT) {
|
||||
base_str += _SS_("_sx") + _SS_(t2s<SmoothScaleDataType>::name);
|
||||
base_str += _SS_("_sy") + _SS_(t2s<YScaleDataType>::name);
|
||||
}
|
||||
if (kFusedQuant == Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT) {
|
||||
base_str += _SS_("_sy") + _SS_(t2s<YScaleDataType>::name);
|
||||
}
|
||||
return base_str;
|
||||
}();
|
||||
|
||||
return _SS_("rmsnorm2d_fwd_") + _SS_(prec_str) + "_" +
|
||||
_TS_(S_::Block_M) + "x" + _TS_(S_::Block_N) + "_" + _TS_(S_::WarpPerBlock_M) + "x" + _TS_(S_::WarpPerBlock_N) + "_" +
|
||||
_TS_(S_::Warp_M) + "x" + _TS_(S_::Warp_N) + "_" + _TS_(S_::Vector_M) + "x" + _TS_(S_::Vector_N) + "_" +
|
||||
_SS_(Pipeline::name) + surfix;
|
||||
// clang-format on
|
||||
#undef _SS_
|
||||
#undef _TS_
|
||||
}
|
||||
|
||||
CK_TILE_DEVICE void operator()(Kargs kargs) const
|
||||
{
|
||||
const auto iM = get_block_id() * Block_M;
|
||||
|
||||
const auto x_window = [&]() {
|
||||
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<const XDataType*>(kargs.p_x),
|
||||
make_tuple(kargs.m, kargs.n),
|
||||
make_tuple(kargs.x_stride, 1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
const auto tmp2_ = pad_tensor_view(
|
||||
tmp_, make_tuple(number<Block_M>{}, number<Block_N>{}), sequence<kPadM, kPadN>{});
|
||||
return make_tile_window(
|
||||
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
|
||||
}();
|
||||
|
||||
const auto x_residual_window = [&]() {
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD ||
|
||||
kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<const XResidualDataType*>(kargs.p_x_residual),
|
||||
make_tuple(kargs.m, kargs.n),
|
||||
make_tuple(kargs.xr_stride, 1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
const auto tmp2_ = pad_tensor_view(tmp_,
|
||||
make_tuple(number<Block_M>{}, number<Block_N>{}),
|
||||
sequence<kPadM, kPadN>{});
|
||||
return make_tile_window(
|
||||
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
|
||||
}
|
||||
else
|
||||
{
|
||||
return make_null_tile_window(make_tuple(number<Block_M>{}, number<Block_N>{}));
|
||||
}
|
||||
}();
|
||||
|
||||
const auto gamma_window = [&]() {
|
||||
const auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<const GammaDataType*>(kargs.p_gamma),
|
||||
make_tuple(kargs.n),
|
||||
make_tuple(1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
const auto tmp2_ =
|
||||
pad_tensor_view(tmp_, make_tuple(number<Block_N>{}), sequence<kPadN>{});
|
||||
|
||||
return make_tile_window(tmp2_, make_tuple(number<Block_N>{}), {0});
|
||||
}();
|
||||
|
||||
auto y_window = [&]() {
|
||||
auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<YDataType*>(kargs.p_y),
|
||||
make_tuple(kargs.m, kargs.n),
|
||||
make_tuple(kargs.y_stride, 1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
auto tmp2_ = pad_tensor_view(
|
||||
tmp_, make_tuple(number<Block_M>{}, number<Block_N>{}), sequence<kPadM, kPadN>{});
|
||||
return make_tile_window(
|
||||
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
|
||||
}();
|
||||
|
||||
auto y_residual_window = [&]() {
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<YResidualDataType*>(kargs.p_y_residual),
|
||||
make_tuple(kargs.m, kargs.n),
|
||||
make_tuple(kargs.yr_stride, 1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
auto tmp2_ = pad_tensor_view(tmp_,
|
||||
make_tuple(number<Block_M>{}, number<Block_N>{}),
|
||||
sequence<kPadM, kPadN>{});
|
||||
return make_tile_window(
|
||||
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
|
||||
}
|
||||
else
|
||||
{
|
||||
return make_null_tile_window(make_tuple(number<Block_M>{}, number<Block_N>{}));
|
||||
}
|
||||
}();
|
||||
|
||||
auto inv_rms_window = [&]() {
|
||||
if constexpr(kSaveInvRms)
|
||||
{
|
||||
const auto inv_rms_m = [&]() {
|
||||
const auto inv_rms_dram_naive =
|
||||
make_naive_tensor_view_packed<address_space_enum::global>(
|
||||
static_cast<InvRmsDataType*>(kargs.p_invRms),
|
||||
make_tuple(kargs.m),
|
||||
number<1>{});
|
||||
|
||||
return pad_tensor_view(
|
||||
inv_rms_dram_naive, make_tuple(number<Block_M>{}), sequence<kPadM>{});
|
||||
}();
|
||||
return make_tile_window(inv_rms_m, make_tuple(number<Block_M>{}), {iM});
|
||||
}
|
||||
else
|
||||
return make_null_tile_window(make_tuple(number<Block_M>{}));
|
||||
}();
|
||||
|
||||
auto sm_scale_window = [&]() {
|
||||
if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT)
|
||||
{
|
||||
const auto win_ = [&]() {
|
||||
const auto tmp_0_ = make_naive_tensor_view_packed<address_space_enum::global>(
|
||||
static_cast<const SmoothScaleDataType*>(kargs.p_sm_scale),
|
||||
make_tuple(kargs.n),
|
||||
number<Vector_N>{});
|
||||
|
||||
return pad_tensor_view(tmp_0_,
|
||||
make_tuple(number<Block_N>{}),
|
||||
sequence<false>{}); // sm_scale no need pad
|
||||
}();
|
||||
return make_tile_window(win_, make_tuple(number<Block_N>{}), {0});
|
||||
}
|
||||
else
|
||||
{
|
||||
return make_null_tile_window(make_tuple(number<Block_N>{}));
|
||||
}
|
||||
}();
|
||||
|
||||
auto y_scale_window = [&]() {
|
||||
if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT ||
|
||||
kFusedQuant == Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT)
|
||||
{
|
||||
const auto win_ = [&]() {
|
||||
const auto tmp_0_ = make_naive_tensor_view_packed<address_space_enum::global>(
|
||||
static_cast<YScaleDataType*>(kargs.p_y_scale),
|
||||
make_tuple(kargs.m),
|
||||
number<1>{});
|
||||
|
||||
return pad_tensor_view(
|
||||
tmp_0_, make_tuple(number<Block_M>{}), sequence<kPadM>{});
|
||||
}();
|
||||
return make_tile_window(win_, make_tuple(number<Block_M>{}), {iM});
|
||||
}
|
||||
else
|
||||
{
|
||||
return make_null_tile_window(make_tuple(number<Block_M>{}));
|
||||
}
|
||||
}();
|
||||
|
||||
auto unquant_y_window = [&]() {
|
||||
if constexpr((kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT ||
|
||||
kFusedQuant == Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT) &&
|
||||
kSaveUnquant)
|
||||
{
|
||||
auto tmp_ = make_naive_tensor_view<address_space_enum::global>(
|
||||
static_cast<UnquantYDataType*>(kargs.p_y_unquant),
|
||||
make_tuple(kargs.m, kargs.n),
|
||||
make_tuple(kargs.y_stride, 1),
|
||||
number<Vector_N>{},
|
||||
number<1>{});
|
||||
|
||||
auto tmp2_ = pad_tensor_view(tmp_,
|
||||
make_tuple(number<Block_M>{}, number<Block_N>{}),
|
||||
sequence<kPadM, kPadN>{});
|
||||
return make_tile_window(
|
||||
tmp2_, make_tuple(number<Block_M>{}, number<Block_N>{}), {iM, 0});
|
||||
}
|
||||
else
|
||||
{
|
||||
return make_null_tile_window(make_tuple(number<Block_M>{}, number<Block_N>{}));
|
||||
}
|
||||
}();
|
||||
|
||||
__shared__ char smem[GetSmemSize()];
|
||||
|
||||
Pipeline{}(x_window,
|
||||
x_residual_window,
|
||||
gamma_window,
|
||||
y_window,
|
||||
y_residual_window,
|
||||
inv_rms_window,
|
||||
sm_scale_window,
|
||||
y_scale_window,
|
||||
unquant_y_window,
|
||||
static_cast<const ComputeDataType>(kargs.epsilon),
|
||||
kargs.n,
|
||||
smem,
|
||||
Epilogue{});
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,95 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/reduce/block/block_reduce2d_problem.hpp"
|
||||
#include "ck_tile/ops/reduce/block/block_reduce2d.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
struct Rmsnorm2dFwdPipelineDefaultPolicy
|
||||
{
|
||||
template <typename Problem>
|
||||
CK_TILE_DEVICE static constexpr auto MakeXBlockTileDistribution()
|
||||
{
|
||||
using S = typename Problem::BlockShape;
|
||||
|
||||
return make_static_tile_distribution(
|
||||
tile_distribution_encoding<
|
||||
sequence<>,
|
||||
tuple<sequence<S::Repeat_M, S::WarpPerBlock_M, S::ThreadPerWarp_M, S::Vector_M>,
|
||||
sequence<S::Repeat_N, S::WarpPerBlock_N, S::ThreadPerWarp_N, S::Vector_N>>,
|
||||
tuple<sequence<1, 2>, sequence<1, 2>>,
|
||||
tuple<sequence<1, 1>, sequence<2, 2>>,
|
||||
sequence<1, 1, 2, 2>,
|
||||
sequence<0, 3, 0, 3>>{});
|
||||
}
|
||||
|
||||
template <typename Problem>
|
||||
CK_TILE_DEVICE static constexpr auto MakeGammaBlockTileDistribution()
|
||||
{
|
||||
using S = typename Problem::BlockShape;
|
||||
|
||||
return make_static_tile_distribution(
|
||||
tile_distribution_encoding<
|
||||
sequence<S::WarpPerBlock_M, S::ThreadPerWarp_M>,
|
||||
tuple<sequence<S::Repeat_N, S::WarpPerBlock_N, S::ThreadPerWarp_N, S::Vector_N>>,
|
||||
tuple<sequence<0, 1>, sequence<0, 1>>,
|
||||
tuple<sequence<0, 1>, sequence<1, 2>>,
|
||||
sequence<1, 1>,
|
||||
sequence<0, 3>>{});
|
||||
}
|
||||
|
||||
template <typename Problem>
|
||||
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2d()
|
||||
{
|
||||
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
|
||||
typename Problem::ComputeDataType,
|
||||
typename Problem::BlockShape>;
|
||||
return BlockReduce2d<P_>{};
|
||||
}
|
||||
|
||||
template <typename Problem>
|
||||
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2dSync()
|
||||
{
|
||||
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
|
||||
typename Problem::ComputeDataType,
|
||||
typename Problem::BlockShape>;
|
||||
return BlockReduce2dSync<P_>{};
|
||||
}
|
||||
|
||||
template <typename Problem>
|
||||
CK_TILE_HOST_DEVICE static constexpr auto GetBlockReduce2dCrossWarpSync()
|
||||
{
|
||||
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
|
||||
typename Problem::ComputeDataType,
|
||||
typename Problem::BlockShape>;
|
||||
return BlockReduce2dCrossWarpSync<P_>{};
|
||||
}
|
||||
|
||||
template <typename Problem>
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
|
||||
{
|
||||
if constexpr(Problem::kNeedCrossWarpSync)
|
||||
{
|
||||
using P_ = BlockReduce2dProblem<typename Problem::ComputeDataType,
|
||||
typename Problem::ComputeDataType,
|
||||
typename Problem::BlockShape>;
|
||||
|
||||
using block_reduce2d = BlockReduce2d<P_>;
|
||||
using x_block_tile =
|
||||
decltype(make_static_distributed_tensor<typename Problem::ComputeDataType>(
|
||||
MakeXBlockTileDistribution<Problem>()));
|
||||
using y_block_tile = decltype(block_reduce2d::template MakeYBlockTile<x_block_tile>());
|
||||
|
||||
return GetBlockReduce2dCrossWarpSync<Problem>().template GetSmemSize<y_block_tile>();
|
||||
}
|
||||
else
|
||||
{
|
||||
return 1; // zero size arrays are an extension
|
||||
}
|
||||
}
|
||||
};
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,226 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
|
||||
#include <string>
|
||||
#include <type_traits>
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
/**
|
||||
* @brief This T5Pass implements the RMSNorm2d forward pipeline as a variant
|
||||
* based on Rmsnorm2dFwdPipelineOnePass and Rmsnorm2dFwdPipelineTwoPass using a T5 model-like
|
||||
* method.
|
||||
*
|
||||
* The T5 model, developed by Google, is a transformer-based architecture designed to perform
|
||||
* a variety of NLP tasks. The T5-like approach employed here is characterized by how RMS
|
||||
* normalization is handled, particularly where intermediate values are cast to BF16. This aims to
|
||||
* achieve a similar value distribution to that produced by the VLLM hip implementation, thereby
|
||||
* enhancing model accuracy.
|
||||
*
|
||||
* Note: While this implementation improves precision and can reduce discrepancies with VLLM, it is
|
||||
* not guaranteed to eliminate all differences or ensure uniform outcomes across every use case.
|
||||
*
|
||||
* This implementation is a variant based on the original one-pass and two-pass approaches,
|
||||
* allowing for both fused and non-fused add operations.
|
||||
*/
|
||||
|
||||
template <typename Problem_, typename Policy_ = Rmsnorm2dFwdPipelineDefaultPolicy>
|
||||
struct Rmsnorm2dFwdPipelineModelSensitiveT5Pass
|
||||
{
|
||||
using Problem = ck_tile::remove_cvref_t<Problem_>;
|
||||
using Policy = ck_tile::remove_cvref_t<Policy_>;
|
||||
|
||||
using XDataType = ck_tile::remove_cvref_t<typename Problem::XDataType>;
|
||||
using GammaDataType = ck_tile::remove_cvref_t<typename Problem::GammaDataType>;
|
||||
using ComputeDataType = ck_tile::remove_cvref_t<typename Problem::ComputeDataType>;
|
||||
using YDataType = ck_tile::remove_cvref_t<typename Problem::YDataType>;
|
||||
using InvRmsDataType = ck_tile::remove_cvref_t<typename Problem::InvRmsDataType>;
|
||||
|
||||
using XResidualDataType = XDataType;
|
||||
using YResidualDataType = XDataType;
|
||||
|
||||
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, ck_tile::null_type>;
|
||||
static constexpr bool kSaveInvRms = Problem::Traits::kSaveInvRms;
|
||||
static constexpr bool kSaveUnquant = Problem::Traits::kSaveUnquant;
|
||||
|
||||
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
|
||||
static constexpr bool kPadM = false; // TODO - BlockRmsnorm2dFwdProblem::kPadM
|
||||
static constexpr bool kPadN = Problem::Traits::kPadN;
|
||||
static constexpr auto kFusedAdd = Problem::Traits::kFusedAdd;
|
||||
static constexpr auto kFusedQuant = Problem::Traits::kFusedQuant;
|
||||
|
||||
static constexpr const char* name = []() {
|
||||
if constexpr(kNeedCrossWarpSync)
|
||||
return "bpr_op"; // block per row
|
||||
else
|
||||
return "wpr_op"; // warp per row
|
||||
}();
|
||||
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
|
||||
{
|
||||
return Policy::template GetSmemSize<Problem>();
|
||||
}
|
||||
|
||||
template <typename XWindow,
|
||||
typename XResidualWindow,
|
||||
typename GammaWindow,
|
||||
typename YWindow,
|
||||
typename YResidualWindow,
|
||||
typename InvRmsWindow,
|
||||
typename SmoothScaleWindow,
|
||||
typename YScaleWindow,
|
||||
typename UnquantYWindow,
|
||||
typename Epilogue>
|
||||
CK_TILE_DEVICE auto operator()(const XWindow& x_window_,
|
||||
const XResidualWindow& x_residual_window_,
|
||||
const GammaWindow& gamma_window_,
|
||||
YWindow& y_window_,
|
||||
const YResidualWindow& y_residual_window_,
|
||||
InvRmsWindow& inv_rms_window,
|
||||
const SmoothScaleWindow& sm_scale_window_,
|
||||
YScaleWindow& y_scale_window_,
|
||||
UnquantYWindow& unquant_y_window,
|
||||
ComputeDataType epsilon,
|
||||
ck_tile::index_t row_size,
|
||||
void* smem,
|
||||
Epilogue) const
|
||||
{
|
||||
const auto x_window =
|
||||
make_tile_window(x_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
const auto gamma_window = make_tile_window(
|
||||
gamma_window_, Policy::template MakeGammaBlockTileDistribution<Problem>());
|
||||
const auto x_residual_window = make_tile_window(
|
||||
x_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
auto y_residual_window = make_tile_window(
|
||||
y_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
|
||||
auto reduce_square_sum_func = ReduceOp::SquareAdd{};
|
||||
auto reduce_sum_func = ReduceOp::Add{};
|
||||
auto block_reduce2d = Policy::template GetBlockReduce2d<Problem>();
|
||||
auto block_reduce2d_sync = Policy::template GetBlockReduce2dSync<Problem>();
|
||||
auto block_reduce2d_cross_warp_sync =
|
||||
Policy::template GetBlockReduce2dCrossWarpSync<Problem>();
|
||||
|
||||
auto x = load_tile(x_window);
|
||||
auto x_resi = load_tile(x_residual_window);
|
||||
|
||||
// load gamma (TODO: support no gamma?)
|
||||
const auto gamma = load_tile(gamma_window);
|
||||
|
||||
auto acc = cast_tile<ComputeDataType>(x);
|
||||
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD ||
|
||||
kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
[[maybe_unused]] auto pre_out =
|
||||
make_static_distributed_tensor<YResidualDataType>(x.get_tile_distribution());
|
||||
|
||||
sweep_tile(x_resi, [&](auto idx) {
|
||||
// compute x = x_resi + x
|
||||
acc(idx) = type_convert<ComputeDataType>(x_resi(idx)) + acc(idx);
|
||||
|
||||
// To make norm input align with residual output
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
if constexpr(std::is_same_v<YResidualDataType, ck_tile::bf16_t>)
|
||||
{
|
||||
pre_out(idx) = float_to_bf16<bf16_rounding_mode::standard>(acc(idx));
|
||||
}
|
||||
else
|
||||
{
|
||||
pre_out(idx) = type_convert<YResidualDataType>(acc(idx));
|
||||
}
|
||||
acc(idx) = type_convert<ComputeDataType>(pre_out(idx));
|
||||
}
|
||||
});
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
store_tile(y_residual_window, pre_out);
|
||||
}
|
||||
}
|
||||
|
||||
// compute mean square each-thread->cross-lane->cross-warp
|
||||
auto square_sum = block_reduce2d.template MakeYBlockTile<decltype(acc)>();
|
||||
set_tile(square_sum, 0);
|
||||
if constexpr((Problem::BlockShape::Repeat_N * Problem::BlockShape::Vector_N) % 2 == 0)
|
||||
{
|
||||
sweep_tile(
|
||||
acc,
|
||||
[&](auto idx_0, auto idx_1) {
|
||||
square_sum(idx_0) += acc[idx_0] * acc[idx_0] + acc[idx_1] * acc[idx_1];
|
||||
},
|
||||
sequence<1, 2>{});
|
||||
}
|
||||
else
|
||||
{
|
||||
square_sum = block_reduce2d(acc,
|
||||
reduce_square_sum_func.GetIdentityValue<ComputeDataType>(),
|
||||
reduce_square_sum_func);
|
||||
}
|
||||
block_reduce2d_sync(square_sum, reduce_sum_func);
|
||||
block_reduce2d_cross_warp_sync(square_sum, smem, reduce_sum_func);
|
||||
|
||||
// compute inv-rms
|
||||
auto inv_rms = tile_elementwise_in(
|
||||
[&](const auto& v_) { return rsqrtf(v_ / row_size + epsilon); }, square_sum);
|
||||
|
||||
if constexpr(kSaveInvRms)
|
||||
store_tile(inv_rms_window, cast_tile<InvRmsDataType>(inv_rms));
|
||||
|
||||
// rmsnorm computation
|
||||
auto rmsn = make_static_distributed_tensor<ComputeDataType>(x.get_tile_distribution());
|
||||
sweep_tile(rmsn, [&, inv_rms_ = inv_rms](auto idx) {
|
||||
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
|
||||
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
|
||||
|
||||
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
|
||||
|
||||
if constexpr(std::is_same_v<XDataType, ck_tile::bf16_t>)
|
||||
{
|
||||
const auto tmp = acc[idx] * inv_rms_[i_idx];
|
||||
const auto tmp_bf16 = float_to_bf16<bf16_rounding_mode::standard>(tmp);
|
||||
const auto rmsn_ = type_convert<ComputeDataType>(tmp_bf16) * gamma_;
|
||||
rmsn(idx) = rmsn_;
|
||||
}
|
||||
else
|
||||
{
|
||||
const auto tmp = type_convert<XDataType>(acc[idx] * inv_rms_[i_idx]);
|
||||
const auto rmsn_ = type_convert<ComputeDataType>(tmp) * gamma_;
|
||||
rmsn(idx) = rmsn_;
|
||||
}
|
||||
});
|
||||
|
||||
if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT)
|
||||
{
|
||||
if constexpr(kSaveUnquant)
|
||||
{
|
||||
Epilogue{}(
|
||||
unquant_y_window, y_window_, sm_scale_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, sm_scale_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
}
|
||||
else if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT)
|
||||
{
|
||||
if constexpr(kSaveUnquant)
|
||||
{
|
||||
Epilogue{}(unquant_y_window, y_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, rmsn, nullptr);
|
||||
}
|
||||
}
|
||||
};
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,167 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
|
||||
#include <string>
|
||||
#include <type_traits>
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
template <typename Problem_, typename Policy_ = Rmsnorm2dFwdPipelineDefaultPolicy>
|
||||
struct Rmsnorm2dFwdPipelineOnePass
|
||||
{
|
||||
using Problem = ck_tile::remove_cvref_t<Problem_>;
|
||||
using Policy = ck_tile::remove_cvref_t<Policy_>;
|
||||
|
||||
using XDataType = ck_tile::remove_cvref_t<typename Problem::XDataType>;
|
||||
using GammaDataType = ck_tile::remove_cvref_t<typename Problem::GammaDataType>;
|
||||
using ComputeDataType = ck_tile::remove_cvref_t<typename Problem::ComputeDataType>;
|
||||
using YDataType = ck_tile::remove_cvref_t<typename Problem::YDataType>;
|
||||
using InvRmsDataType = ck_tile::remove_cvref_t<typename Problem::InvRmsDataType>;
|
||||
|
||||
using XResidualDataType = XDataType;
|
||||
using YResidualDataType = XDataType;
|
||||
|
||||
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, ck_tile::null_type>;
|
||||
static constexpr bool kSaveInvRms = Problem::Traits::kSaveInvRms;
|
||||
static constexpr bool kSaveUnquant = Problem::Traits::kSaveUnquant;
|
||||
|
||||
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
|
||||
static constexpr bool kPadM = false; // TODO - BlockRmsnorm2dFwdProblem::kPadM
|
||||
static constexpr bool kPadN = Problem::Traits::kPadN;
|
||||
static constexpr auto kFusedAdd = Problem::Traits::kFusedAdd;
|
||||
static constexpr auto kFusedQuant = Problem::Traits::kFusedQuant;
|
||||
|
||||
static constexpr const char* name = []() {
|
||||
if constexpr(kNeedCrossWarpSync)
|
||||
return "bpr_op"; // block per row
|
||||
else
|
||||
return "wpr_op"; // warp per row
|
||||
}();
|
||||
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
|
||||
{
|
||||
return Policy::template GetSmemSize<Problem>();
|
||||
}
|
||||
|
||||
template <typename XWindow,
|
||||
typename XResidualWindow,
|
||||
typename GammaWindow,
|
||||
typename YWindow,
|
||||
typename YResidualWindow,
|
||||
typename InvRmsWindow,
|
||||
typename SmoothScaleWindow,
|
||||
typename YScaleWindow,
|
||||
typename UnquantYWindow,
|
||||
typename Epilogue>
|
||||
CK_TILE_DEVICE auto operator()(const XWindow& x_window_,
|
||||
const XResidualWindow& x_residual_window_,
|
||||
const GammaWindow& gamma_window_,
|
||||
YWindow& y_window_,
|
||||
const YResidualWindow& y_residual_window_,
|
||||
InvRmsWindow& inv_rms_window,
|
||||
const SmoothScaleWindow& sm_scale_window_,
|
||||
YScaleWindow& y_scale_window_,
|
||||
UnquantYWindow& unquant_y_window,
|
||||
ComputeDataType epsilon,
|
||||
ck_tile::index_t row_size,
|
||||
void* smem,
|
||||
Epilogue) const
|
||||
{
|
||||
const auto x_window =
|
||||
make_tile_window(x_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
const auto gamma_window = make_tile_window(
|
||||
gamma_window_, Policy::template MakeGammaBlockTileDistribution<Problem>());
|
||||
const auto x_residual_window = make_tile_window(
|
||||
x_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
auto y_residual_window = make_tile_window(
|
||||
y_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
|
||||
auto reduce_square_sum_func = ReduceOp::SquareAdd{};
|
||||
auto reduce_sum_func = ReduceOp::Add{};
|
||||
auto block_reduce2d = Policy::template GetBlockReduce2d<Problem>();
|
||||
auto block_reduce2d_sync = Policy::template GetBlockReduce2dSync<Problem>();
|
||||
auto block_reduce2d_cross_warp_sync =
|
||||
Policy::template GetBlockReduce2dCrossWarpSync<Problem>();
|
||||
|
||||
auto x = load_tile(x_window);
|
||||
auto x_resi = load_tile(x_residual_window);
|
||||
|
||||
// load gamma (TODO: support no gamma?)
|
||||
const auto gamma = load_tile(gamma_window);
|
||||
|
||||
auto acc = cast_tile<ComputeDataType>(x);
|
||||
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD ||
|
||||
kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
sweep_tile(x_resi, [&](auto idx) {
|
||||
// compute x = x_resi + x
|
||||
acc(idx) = type_convert<ComputeDataType>(x_resi(idx)) + acc(idx);
|
||||
});
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
store_tile(y_residual_window, cast_tile<YResidualDataType>(acc));
|
||||
}
|
||||
}
|
||||
|
||||
// compute mean square each-thread->cross-lane->cross-warp
|
||||
auto square_sum = block_reduce2d(acc,
|
||||
reduce_square_sum_func.GetIdentityValue<ComputeDataType>(),
|
||||
reduce_square_sum_func);
|
||||
block_reduce2d_sync(square_sum, reduce_sum_func);
|
||||
block_reduce2d_cross_warp_sync(square_sum, smem, reduce_sum_func);
|
||||
|
||||
// compute inv-rms
|
||||
auto inv_rms = tile_elementwise_in(
|
||||
[&](const auto& v_) { return rsqrtf(v_ / row_size + epsilon); }, square_sum);
|
||||
|
||||
if constexpr(kSaveInvRms)
|
||||
store_tile(inv_rms_window, cast_tile<InvRmsDataType>(inv_rms));
|
||||
|
||||
// rmsnorm computation
|
||||
auto rmsn = make_static_distributed_tensor<ComputeDataType>(x.get_tile_distribution());
|
||||
sweep_tile(rmsn, [&, inv_rms_ = inv_rms](auto idx) {
|
||||
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
|
||||
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
|
||||
|
||||
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
|
||||
|
||||
auto rmsn_ = acc[idx] * inv_rms_[i_idx] * gamma_;
|
||||
|
||||
rmsn(idx) = rmsn_;
|
||||
});
|
||||
|
||||
if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT)
|
||||
{
|
||||
if constexpr(kSaveUnquant)
|
||||
{
|
||||
Epilogue{}(
|
||||
unquant_y_window, y_window_, sm_scale_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, sm_scale_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
}
|
||||
else if constexpr(kFusedQuant == Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT)
|
||||
{
|
||||
if constexpr(kSaveUnquant)
|
||||
{
|
||||
Epilogue{}(unquant_y_window, y_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, y_scale_window_, rmsn, smem);
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
Epilogue{}(y_window_, rmsn, nullptr);
|
||||
}
|
||||
}
|
||||
};
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,38 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core/utility/type_traits.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
template <typename XDataType_,
|
||||
typename GammaDataType_,
|
||||
typename ComputeDataType_,
|
||||
typename YDataType_,
|
||||
typename InvRmsDataType_,
|
||||
typename UnquantYDataType_,
|
||||
typename SmoothScaleDataType_,
|
||||
typename YScaleDataType_,
|
||||
typename BlockShape_,
|
||||
typename Traits_>
|
||||
struct Rmsnorm2dFwdPipelineProblem
|
||||
{
|
||||
using XDataType = remove_cvref_t<XDataType_>;
|
||||
using GammaDataType = remove_cvref_t<GammaDataType_>;
|
||||
using ComputeDataType = remove_cvref_t<ComputeDataType_>;
|
||||
using YDataType = remove_cvref_t<YDataType_>;
|
||||
using InvRmsDataType = remove_cvref_t<InvRmsDataType_>;
|
||||
using UnquantYDataType = remove_cvref_t<UnquantYDataType_>;
|
||||
using SmoothScaleDataType = remove_cvref_t<SmoothScaleDataType_>;
|
||||
using YScaleDataType = remove_cvref_t<YScaleDataType_>;
|
||||
using BlockShape = remove_cvref_t<BlockShape_>;
|
||||
|
||||
static constexpr bool kNeedCrossLaneSync = BlockShape::ThreadPerWarp_N > 1;
|
||||
static constexpr bool kNeedCrossWarpSync = BlockShape::WarpPerBlock_N > 1;
|
||||
|
||||
using Traits = remove_cvref_t<Traits_>;
|
||||
};
|
||||
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,205 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core.hpp"
|
||||
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
|
||||
#include <string>
|
||||
#include <type_traits>
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
template <typename Problem_, typename Policy_ = Rmsnorm2dFwdPipelineDefaultPolicy>
|
||||
struct Rmsnorm2dFwdPipelineTwoPass
|
||||
{
|
||||
using Problem = ck_tile::remove_cvref_t<Problem_>;
|
||||
using Policy = ck_tile::remove_cvref_t<Policy_>;
|
||||
|
||||
using XDataType = ck_tile::remove_cvref_t<typename Problem::XDataType>;
|
||||
using GammaDataType = ck_tile::remove_cvref_t<typename Problem::GammaDataType>;
|
||||
using ComputeDataType = ck_tile::remove_cvref_t<typename Problem::ComputeDataType>;
|
||||
using YDataType = ck_tile::remove_cvref_t<typename Problem::YDataType>;
|
||||
using InvRmsDataType = ck_tile::remove_cvref_t<typename Problem::InvRmsDataType>;
|
||||
|
||||
using XResidualDataType = XDataType;
|
||||
using YResidualDataType = XDataType;
|
||||
|
||||
static constexpr bool kHasGamma = !std::is_same_v<GammaDataType, ck_tile::null_type>;
|
||||
static constexpr bool kSaveInvRms = Problem::Traits::kSaveInvRms;
|
||||
|
||||
static constexpr bool kNeedCrossWarpSync = Problem::kNeedCrossWarpSync;
|
||||
static constexpr bool kPadM = false; // TODO - BlockRmsnorm2dFwdProblem::kPadM
|
||||
static constexpr bool kPadN = Problem::Traits::kPadN;
|
||||
static constexpr auto kFusedAdd = Problem::Traits::kFusedAdd;
|
||||
static constexpr auto kFusedQuant = Problem::Traits::kFusedQuant;
|
||||
|
||||
static constexpr const char* name = []() {
|
||||
if constexpr(kNeedCrossWarpSync)
|
||||
return "bpr_tp"; // block per row
|
||||
else
|
||||
return "wpr_tp"; // warp per row
|
||||
}();
|
||||
|
||||
CK_TILE_HOST_DEVICE static constexpr index_t GetSmemSize()
|
||||
{
|
||||
return Policy::template GetSmemSize<Problem>();
|
||||
}
|
||||
|
||||
template <typename XWindow,
|
||||
typename XResidualWindow,
|
||||
typename GammaWindow,
|
||||
typename YWindow,
|
||||
typename YResidualWindow,
|
||||
typename InvRmsWindow,
|
||||
typename SmoothScaleWindow,
|
||||
typename YScaleWindow,
|
||||
typename UnquantYWindow,
|
||||
typename Epilogue>
|
||||
CK_TILE_DEVICE auto operator()(const XWindow& x_window_,
|
||||
const XResidualWindow& x_residual_window_,
|
||||
const GammaWindow& gamma_window_,
|
||||
YWindow& y_window,
|
||||
const YResidualWindow& y_residual_window_,
|
||||
InvRmsWindow& inv_rms_window,
|
||||
const SmoothScaleWindow& /*sm_scale_window_*/,
|
||||
YScaleWindow& /*y_scale_window*/,
|
||||
UnquantYWindow& /*unquant_y_window*/,
|
||||
ComputeDataType epsilon,
|
||||
ck_tile::index_t row_size,
|
||||
void* smem,
|
||||
Epilogue) const
|
||||
{
|
||||
auto x_window =
|
||||
make_tile_window(x_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
auto gamma_window = make_tile_window(
|
||||
gamma_window_, Policy::template MakeGammaBlockTileDistribution<Problem>());
|
||||
auto x_residual_window = make_tile_window(
|
||||
x_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
auto y_residual_window = make_tile_window(
|
||||
y_residual_window_, Policy::template MakeXBlockTileDistribution<Problem>());
|
||||
|
||||
// Problem::BlockShape
|
||||
static constexpr index_t Block_N = Problem::BlockShape::Block_N;
|
||||
index_t num_n_tile_iteration =
|
||||
amd_wave_read_first_lane(integer_divide_ceil(row_size, Block_N));
|
||||
|
||||
auto reduce_square_sum_func = ReduceOp::SquareAdd{};
|
||||
auto reduce_sum_func = ReduceOp::Add{};
|
||||
auto block_reduce2d = Policy::template GetBlockReduce2d<Problem>();
|
||||
auto block_reduce2d_sync = Policy::template GetBlockReduce2dSync<Problem>();
|
||||
auto block_reduce2d_cross_warp_sync =
|
||||
Policy::template GetBlockReduce2dCrossWarpSync<Problem>();
|
||||
|
||||
using ComputeTensorType = decltype(cast_tile<ComputeDataType>(load_tile(x_window)));
|
||||
auto square_sum = block_reduce2d.template MakeYBlockTile<ComputeTensorType>();
|
||||
set_tile(square_sum, reduce_square_sum_func.GetIdentityValue<ComputeDataType>());
|
||||
|
||||
for(int iN = amd_wave_read_first_lane(0); iN < num_n_tile_iteration; ++iN)
|
||||
{
|
||||
auto x = load_tile(x_window);
|
||||
auto x_resi = load_tile(x_residual_window);
|
||||
|
||||
move_tile_window(x_window, {0, Block_N});
|
||||
move_tile_window(x_residual_window, {0, Block_N});
|
||||
|
||||
auto acc = cast_tile<ComputeDataType>(x);
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD ||
|
||||
kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
sweep_tile(x_resi, [&](auto idx) {
|
||||
// compute x = x_resi + x
|
||||
acc(idx) = type_convert<ComputeDataType>(x_resi(idx)) + acc(idx);
|
||||
});
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
store_tile(y_residual_window, cast_tile<YResidualDataType>(acc));
|
||||
move_tile_window(y_residual_window, {0, Block_N});
|
||||
}
|
||||
}
|
||||
|
||||
block_reduce2d(acc, square_sum, reduce_square_sum_func);
|
||||
}
|
||||
|
||||
block_reduce2d_sync(square_sum, reduce_sum_func);
|
||||
block_reduce2d_cross_warp_sync(square_sum, smem, reduce_sum_func);
|
||||
|
||||
// compute inv-rms
|
||||
auto inv_rms = tile_elementwise_in(
|
||||
[&](const auto& v_) {
|
||||
return type_convert<ComputeDataType>(1.0f) / (sqrt(v_ / row_size + epsilon));
|
||||
},
|
||||
square_sum);
|
||||
|
||||
if constexpr(kSaveInvRms)
|
||||
store_tile(inv_rms_window, cast_tile<InvRmsDataType>(inv_rms));
|
||||
|
||||
// reverse read x to reuse cache
|
||||
ck_tile::index_t stride_to_right_most_window =
|
||||
row_size % Block_N == 0 ? row_size - Block_N : row_size - row_size % Block_N;
|
||||
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
move_tile_window(y_residual_window, {0, -Block_N});
|
||||
}
|
||||
else
|
||||
{
|
||||
move_tile_window(x_window, {0, -Block_N});
|
||||
move_tile_window(x_residual_window, {0, -Block_N});
|
||||
}
|
||||
move_tile_window(gamma_window, {stride_to_right_most_window});
|
||||
move_tile_window(y_window, {0, stride_to_right_most_window});
|
||||
|
||||
// rmsnorm computation
|
||||
for(int iN = amd_wave_read_first_lane(0); iN < num_n_tile_iteration; ++iN)
|
||||
{
|
||||
auto acc = make_static_distributed_tensor<ComputeDataType>(
|
||||
decltype(load_tile(x_window))::get_tile_distribution());
|
||||
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD_STORE)
|
||||
{
|
||||
acc = cast_tile<ComputeDataType>(load_tile(y_residual_window));
|
||||
move_tile_window(y_residual_window, {0, -Block_N});
|
||||
}
|
||||
else
|
||||
{
|
||||
acc = cast_tile<ComputeDataType>(load_tile(x_window));
|
||||
move_tile_window(x_window, {0, -Block_N});
|
||||
|
||||
if constexpr(kFusedAdd == Rmsnorm2dFusedAddEnum::PRE_ADD)
|
||||
{
|
||||
auto x_resi = load_tile(x_residual_window);
|
||||
sweep_tile(x_resi, [&](auto idx) {
|
||||
// compute x = x_resi + x
|
||||
acc(idx) = type_convert<ComputeDataType>(x_resi(idx)) + acc(idx);
|
||||
});
|
||||
move_tile_window(x_residual_window, {0, -Block_N});
|
||||
}
|
||||
}
|
||||
|
||||
// load gamma (TODO: support no gamma?)
|
||||
const auto gamma = load_tile(gamma_window);
|
||||
|
||||
// rmsnorm computation
|
||||
auto rmsn = make_static_distributed_tensor<ComputeDataType>(
|
||||
decltype(load_tile(x_window))::get_tile_distribution());
|
||||
sweep_tile(rmsn, [&, inv_rms_ = inv_rms](auto idx) {
|
||||
constexpr auto i_idx = make_tuple(idx[number<0>{}]);
|
||||
constexpr auto j_idx = make_tuple(idx[number<1>{}]);
|
||||
|
||||
const auto gamma_ = type_convert<ComputeDataType>(gamma[j_idx]);
|
||||
|
||||
auto rmsn_ = acc(idx) * inv_rms_[i_idx] * gamma_;
|
||||
|
||||
rmsn(idx) = rmsn_;
|
||||
});
|
||||
|
||||
static_assert(kFusedQuant == Rmsnorm2dFusedQuantEnum::NO_SWEEP);
|
||||
Epilogue{}(y_window, rmsn, nullptr);
|
||||
|
||||
move_tile_window(gamma_window, {-Block_N});
|
||||
move_tile_window(y_window, {0, -Block_N});
|
||||
}
|
||||
}
|
||||
};
|
||||
} // namespace ck_tile
|
||||
@@ -0,0 +1,73 @@
|
||||
// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
|
||||
// SPDX-License-Identifier: MIT
|
||||
|
||||
#pragma once
|
||||
|
||||
#include "ck_tile/core/utility/type_traits.hpp"
|
||||
|
||||
namespace ck_tile {
|
||||
|
||||
enum class Rmsnorm2dFusedAddEnum
|
||||
{
|
||||
NO_ADD = 0,
|
||||
// fused add before RMSNorm and store result to global
|
||||
PRE_ADD_STORE = 1,
|
||||
// fused add before RMSNorm, but not store result
|
||||
PRE_ADD = 2,
|
||||
};
|
||||
|
||||
// clang-format off
|
||||
template<Rmsnorm2dFusedAddEnum> struct Rmsnorm2dFusedAddEnumName;
|
||||
template<> struct Rmsnorm2dFusedAddEnumName<Rmsnorm2dFusedAddEnum::NO_ADD> { static constexpr const char * name = "no"; };
|
||||
template<> struct Rmsnorm2dFusedAddEnumName<Rmsnorm2dFusedAddEnum::PRE_ADD_STORE> { static constexpr const char * name = "pras"; };
|
||||
template<> struct Rmsnorm2dFusedAddEnumName<Rmsnorm2dFusedAddEnum::PRE_ADD> { static constexpr const char * name = "pra"; };
|
||||
// clang-format on
|
||||
|
||||
enum class Rmsnorm2dFusedQuantEnum
|
||||
{
|
||||
NO_SWEEP = 0,
|
||||
SMOOTH_DYNAMIC_QUANT = 1, // smooth oulier + rowwise quant, need input x-scale and store y_scale
|
||||
DYNAMIC_QUANT = 2, // rowwise quant, store out a y-scale
|
||||
};
|
||||
|
||||
// clang-format off
|
||||
template<Rmsnorm2dFusedQuantEnum> struct Rmsnorm2dFusedQuantEnumName;
|
||||
template<> struct Rmsnorm2dFusedQuantEnumName<Rmsnorm2dFusedQuantEnum::NO_SWEEP> { static constexpr const char * name = "no"; };
|
||||
template<> struct Rmsnorm2dFusedQuantEnumName<Rmsnorm2dFusedQuantEnum::DYNAMIC_QUANT> { static constexpr const char * name = "dqt"; };
|
||||
template<> struct Rmsnorm2dFusedQuantEnumName<Rmsnorm2dFusedQuantEnum::SMOOTH_DYNAMIC_QUANT> { static constexpr const char * name = "smdqt"; };
|
||||
// clang-format on
|
||||
|
||||
enum class Rmsnorm2dSensitiveEnum
|
||||
{
|
||||
NO_SPECIFIC_MODEL = 0,
|
||||
// T5-like model for RMSNorm. The T5 model, developed by Google, is a transformer-based
|
||||
// architecture designed for a variety of NLP tasks. This option mimics T5's approach to
|
||||
// RMSNorm, aiming to ensure similar value distributions and enhance accuracy.
|
||||
T5_MODEL_LIKE = 1,
|
||||
};
|
||||
|
||||
// clang-format off
|
||||
template<Rmsnorm2dSensitiveEnum> struct Rmsnorm2dSensitiveEnumName;
|
||||
template<> struct Rmsnorm2dSensitiveEnumName<Rmsnorm2dSensitiveEnum::NO_SPECIFIC_MODEL> { static constexpr const char * name = "nsm"; };
|
||||
template<> struct Rmsnorm2dSensitiveEnumName<Rmsnorm2dSensitiveEnum::T5_MODEL_LIKE> { static constexpr const char * name = "t5ml"; };
|
||||
// clang-format on
|
||||
|
||||
template <bool kPadN_,
|
||||
bool kSaveInvRms_,
|
||||
bool kSaveUnquant_,
|
||||
bool kTwoPass_,
|
||||
Rmsnorm2dFusedAddEnum kFusedAdd_,
|
||||
Rmsnorm2dFusedQuantEnum kFusedQuant_,
|
||||
Rmsnorm2dSensitiveEnum kUseModelSensitiveRMSNorm_>
|
||||
struct Rmsnorm2dFwdTraits
|
||||
{
|
||||
static constexpr bool kPadN = kPadN_;
|
||||
static constexpr bool kSaveInvRms = kSaveInvRms_;
|
||||
static constexpr bool kSaveUnquant = kSaveUnquant_;
|
||||
static constexpr bool kTwoPass = kTwoPass_;
|
||||
static constexpr Rmsnorm2dFusedAddEnum kFusedAdd = kFusedAdd_;
|
||||
static constexpr Rmsnorm2dFusedQuantEnum kFusedQuant = kFusedQuant_;
|
||||
static constexpr Rmsnorm2dSensitiveEnum kUseModelSensitiveRMSNorm = kUseModelSensitiveRMSNorm_;
|
||||
};
|
||||
|
||||
} // namespace ck_tile
|
||||
Reference in New Issue
Block a user