Files
composable_kernel/example/ck_tile/01_fmha/utils.hpp
Po Yen Chen c156989298 [CK_TILE] Add PagedAttention kernels (#1387)
* Use dictionary to config all the functions

* Add init codegen logic for fmha fwd appendkv

* Call HIP_CHECK_ERROR() macro to get real source info

* Setup meaningfull arguments

* Sync kernel name with the codegen

* Add knew/vnew tensors to the kernel argument

* Fix wrong K values after appending

* Fix vnew append errro

* Extract common logics

* Fix Vnew tile dstr for row major case

* Conditionally add fwd_splitkv API in fmha_fwd example

* Conditionally add call to fmha_fwd_splitkv()

* Remove "EXAMPLE_" prefix of cmake variables

* Regsiter API handlers automatically

* Early return if 0 < s_k_new is not supported

* Show message if we are ignoring option

* Unify CMakeLists.txt coding style

* Set num_splits=1 if split-kv is not supported

* Add length/stride getters for HostTensor

* Add RoPE example utilities

* Add reference_rotary_position_embedding() (not implemented)

* Finish reference_rotary_position_embedding() impl

* Fix typo of HostTensor<>::get_length()

* Fix compilation errors

* Fix wrong answer when interleaved=false

* Fix wrong answer when interleaved=true

* Append K/V in the host verification code

* Simplify K appending logics

* Simplify v_host_ref definition

* Reduce input/output dimensions

* Rename function: add "batched" prefix

* Apply RoPE on host side

* Rename RoPE utility function

* Fix wrong tensor size

* Avoid invoking deprecated method 'find_module'

* Pass RoPE kernel args

* Create Rotary Cos/Sin tile windows in kernel

* Add compute data type alias for RoPE

* Randomly generate seqlen_knew if needed

* Fix seqlen_knew enabling check logic

* Add minimum seqlen_k to generate compliance kvcache

* Fix compilation error in debug mode

* Fix wrong boundaries

* Fix wrong seqlen_k for kvcache

* Rename variables used in distributio encoding

* Fix rotary cos/sin tensor/tile size

* Add constraint to the rotary_dim option

* Remove unused inner namespace

* Add dram distribution for rotary_cos/rotary_sin (interleaved)

* Only apply interleaved RoPE on Knew for now

* Fix wrong thread starting offset

* Instantiate multiple kernels for RoPE approaches

* Clean-up pipeline

* Fix error in RoPE host reference

* Handle RoPE half-rotated logics

* Support 8x rotary_dim under half-rotated RoPE

* Add comment

* Apply elementwise function to the loaded tiles

* Unify parameter/variable naming style

* Remove constness from q_ptr

* Add code blocks for q_tile

* Apply RoPE to q_tile

* Remove debug print code in kernel

* Fix wrong knew/vnew appending positions

* Use better naming for tile indices

* Add make_tile_window() for adding distribution only

* Skip code if # of block is more than needed

* Move thread locating logics into policy

* Remove always true static_assert()

* Rename header

* Rename RotaryEmbeddingEnum

* Extract rotary embedding logic out

* Re-order parameters

* Align naming of some tile size constants

* Rename more tile size constants

* Fix wrong grid size

* Fix wrong shape of knew_host/vnew_host

* Fix wrong index into knew_host/vnew_host

* Fix wrong rotary_cos/rotary_sin memory size for Q

* Extract Q/Knew vector size to helper methods

* Use different rotary_cos/rotary_sin distr for Q/Knew

* Update host/device specifiers

* Fix wrong data type for Q rotary_cos/rotary_sin

* Remove RoPEComputeDataType type alias

* Shift rotary_cos/rotary_sin by cache_seqlen_k

* Add comment for why I just 't' for all padding flags

* Align commit message to the real comment

* Fix wrong pipeline

* Rename utility function

* Disable host verification if API not exist

* Fix wrong rope key for fp8 pipeline

* Allow only apply RoPE on Q (without append KV)

* Add append-kv smoke tests

* Remove debug statements

* Remove more debug statements

* Re-arrange the 'set +x' command

* Remove no-longer used method in pipeline

* Add missing init code

* Refine pipeline padding settings

* Enlarge rotary_dim limit (8 -> 16)

* Enlarge KPerThread for rotary_interleaved=false

* Update rotary_dim range in smoke_test_fwd.sh

* Add template argument 'kIsPagedKV' for splitkv kernels

* Launch splitkv kernel if given page_block_size

* Fix wrong kernel name

* Fix seqlen_k_min for pre-fill case (1 -> 0)

* Add copy_const<> type trait

* Add another make_tile_window()

* Introduce 'TileWindowNavigator' types

* Simplify TileWindowNavigator interfaces

* Fix tile window navigation bugs

* Disable calling fmha_fwd()

* Remove ununnecessary data members

* Simplify more make_tile_window() overloads

* Move V tile through TileWindowNavigator

* Fix uneven split checking logic

* Move code after decide seqlen_q/seqlen_k

* Make sure we always start reading complete tile

* Use 128 as minimus page_block_size

* Fix wrong origin for bias

* Add batch_stride_k/batch_stride_v in group mode

* Unify origin

* Add missing kernel arguments for group mode

* Add paged-kv codegen logic for appendkv kernels

* Add block_table kernel args for appendkv kernel

* Add tile navigators to the appendkv kernel

* Fix wrong tensor descriptor lengths

* Pass re-created tile window to pipeline

* Fix wrong strides for appendkv kernel

* Allow transit tile_window to another page-block

* Handle cross-page-block write

* Donot perform write again if already in last page-block

* Always add fmha_fwd() api

* Add missing group mode argument

* Remove debug macro usages

* Rename option s_k_new to s_knew

* Separate splitkv/non-splitkv args/traits

* Remove fmha_fwd_dispatch()

* Fix compilation errors

* Remove dropout code in splitkv kernel

* Allow problem types without define kHasDropout attr

* Use generic lambda to init traits objects

* Separate more non-splitkv & splitkv traits/args

* Display more info for specific kernels

* Show more detailed warning message

* Rename 'max_num_blocks' to 'max_num_page_blocks'

* Remove no-longer used pipeline files

* Wrap code by #if directives

* Move functors to the begining of validation code

* Use generic lambda to init all the api traits/args

* Fix wrong seqlen for kvcache

* Add missing comment

* Rename TileWindowNavigator to PageBlockNavigator

* Only expose necessary methods (not attributes)

* Re-order pipeline paremeters

* Refine smoke_test_fwd.sh

* Fix wrong arugment count

* Make tile window directly via PageBlockNavigator

* Remove unused template paremeter

* Remove group mode from appendkv kernel

* Fix skcheck logic

* Fix wrong syntax in skcheck expr

* Use meaningful options in smoke test

* Remove options

* Fix formatting

* Fix more format

* Re-organize bash functions

* Pass cache_batch_idx to kernels

* Support cache_batch_idx in example

* Fix compilation error

* Add more appendkv test

* Add more case for appendkv

* Fix unexisted attribute

* Remove 0 < seqlen_knew constraint

* Clarify the case in warning message

* Remove macro checking

* Force batch mode when invoking appendkv & splitkv apis

* Fix mode overriding logics

* Fix wrong parameter name

* Randomize seqlen_k if use kvcache

* Use randomized seqlen_k for kvcache

* Avoid using too small rotary_cos & rotary_sin

* Rename parameter

* Add seqlen_q & seqlen_k rules

* Add comment

* Add more comments

* Fix compilation errors

* Fix typo in comment

* Remove type argument

* Avoid seqlen_k=0 for kvcache

* Revert "Avoid seqlen_k=0 for kvcache"

This reverts commit 21c4df89e4.

* Fix wrong uneven split checking logics

* Only randomize kvcache seqlen_k if 1 < batch

* Return earlier if split is empty

* Revert "Only randomize kvcache seqlen_k if 1 < batch"

This reverts commit b9a4ab0d7e.

* Re-order seqlen_k_start adjustment logics

* Fix compilation errors

* Re-format script

* Find executable from folder automatically

* Fix kvcache seqlen_k generating logic

* Make comment more clear

* Fix wrong knew/vew appending logic on host

* Add s_barrier to sync threads

* Revert "Add s_barrier to sync threads"

This reverts commit d3f550f30c.

* Support only using 1 row of rotary_cos/rotary_sin

* Rotate Q in different way

* Unify tensor view creation logics

* Fix wrong argument

* Add mask to switch how we use the rotary_cos/sin

* Move attr from traits to problem

* Move has_mask to fmha_fwd_appendkv_args

* Support use uint32_t as SAD operand in Alibi<>

* Use sad_u32() in splitkv kernels

* Store tensor views in PageBlockNavigator

* Use stored tensor view to update tile windows

* Enlarge tensor view size

* Remove debug code

* Fix wrong tensor view size

* Wrap tensor view into PageBlockNavigator

* Add DataType member to PageBlockNavigator

* Remove unnecessary member functions

* Refind macro use

* Fix typo

* Add blank line between directives and actual code

* Re-format files

* Remove type in comment

---------

Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
2024-08-28 20:50:43 +08:00

256 lines
8.7 KiB
C++

// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <algorithm>
#include <cstdint>
#include <cstdlib>
#include <functional>
#include <optional>
#include <ostream>
#include <sstream>
#include <string>
#include <tuple>
#include <utility>
#include <vector>
#include "ck_tile/core/container/span.hpp"
enum class mode_enum
{
batch = 0,
group
};
std::ostream& operator<<(std::ostream& stream, mode_enum mode)
{
return stream << (mode == mode_enum::batch ? "batch" : "group");
}
std::vector<int32_t> to_seqstarts(ck_tile::span<const int32_t> seqlens)
{
std::vector<int32_t> seqstarts = {0};
for(int32_t seqlen : seqlens)
{
seqstarts.push_back(seqstarts.back() + seqlen);
}
assert(seqstarts.size() == seqlens.size() + 1);
return seqstarts;
}
std::vector<int32_t> generate_seqlens(unsigned count,
int32_t seqlen_avg,
int32_t seqlen_min = -1, // if not negative, clamp min
int32_t seqlen_max = -1, // if not negative, clamp max
std::optional<unsigned> seed = std::nullopt)
{
assert(0 < count);
seqlen_min = (0 < seqlen_min ? seqlen_min : 1);
seqlen_max = (0 < seqlen_max ? seqlen_max : std::numeric_limits<int32_t>::max());
assert(seqlen_min <= seqlen_max);
std::vector<int32_t> seqlens(count, std::clamp(seqlen_avg, seqlen_min, seqlen_max));
if(1 < count)
{
using size_type = std::vector<int32_t>::size_type;
std::mt19937 random_engine(seed.has_value() ? *seed : std::random_device{}());
std::uniform_int_distribution<size_type> idx_dist(0, count - 1);
auto next_idx = std::bind(idx_dist, std::ref(random_engine));
std::uniform_int_distribution<size_type> step_dist(1, count - 1);
auto next_step = std::bind(step_dist, std::ref(random_engine));
for(unsigned repeat = seqlen_avg * (count / 2); 0 < repeat; --repeat)
{
const size_type to_decrease = next_idx();
// make sure each elements of seqlens is always greater than seqlen_min
if(seqlens[to_decrease] == seqlen_min)
{
continue;
}
const size_type to_increase = (to_decrease + next_step()) % count;
if(seqlens[to_increase] >= seqlen_max)
{
continue;
}
--seqlens[to_decrease];
++seqlens[to_increase];
}
}
return seqlens;
}
// return random integer generated uniformly in range [low, high]
template <typename Int = int>
auto randint(Int low, Int high, std::optional<unsigned> seed = std::nullopt)
-> std::enable_if_t<std::is_integral_v<Int>, Int>
{
std::mt19937 engine(seed.has_value() ? *seed : std::random_device{}());
std::uniform_int_distribution<Int> dist(low, high);
return dist(engine);
}
// return random integers generated uniformly in range [low, high]
template <typename Int, typename ForwardIterator>
auto randints(ForwardIterator first,
ForwardIterator last,
Int low,
Int high,
std::optional<unsigned> seed = std::nullopt)
-> std::enable_if_t<std::is_integral_v<Int>>
{
std::mt19937 engine(seed.has_value() ? *seed : std::random_device{}());
std::uniform_int_distribution<Int> dist(low, high);
std::generate(first, last, [&] { return dist(engine); });
}
/*
* decode the seqlen string from cmdline
* example (assume batch=3)
* q_val=1,2,3 k_val=4,5,6 -> OK
* q_val=1,2,3 -> OK, k same as q
* q_val=1,2 -> OK, q will rand remaining 1 element, k same as q
* q_val=1,2 k_val=4,5 -> OK, q/k will rand remaining 1 element
* q_val=1,2,3,4 -> OK, but ignore exceed one
*
* q_val=1,2 k_val=4,5,6 -> not OK, k must have same splits with q
* q_val=1,2 k_val=4 -> not OK, k must have same splits with q
*/
std::tuple<std::vector<ck_tile::index_t>,
std::vector<ck_tile::index_t>,
std::vector<ck_tile::index_t>>
decode_seqlen(mode_enum mode,
ck_tile::index_t batch,
std::string q_val,
std::string k_val,
std::string k_pad_val,
ck_tile::index_t seqlen_k_min = 0,
bool use_kvcache = false,
std::optional<unsigned> seed = std::nullopt)
{
#define _S2I_(str_) static_cast<ck_tile::index_t>(std::atoi((str_).c_str()))
if(mode == mode_enum::batch)
{
ck_tile::index_t q = _S2I_(q_val);
ck_tile::index_t k = _S2I_(k_val);
auto s_q = std::vector<ck_tile::index_t>(batch, q);
auto s_k = [&] {
const ck_tile::index_t seqlen_k_max = (k < 0 ? q : k);
std::vector<ck_tile::index_t> seqlen_ks(batch, seqlen_k_max);
if(1 < batch && use_kvcache)
{
// to keep the original s_k value, we always use seqlen_k_max in first batch
randints(std::next(seqlen_ks.begin()),
seqlen_ks.end(),
seqlen_k_min,
seqlen_k_max,
seed);
return seqlen_ks;
}
return seqlen_ks;
}();
auto s_kpad = std::vector<ck_tile::index_t>(batch, -1); // TODO: batch not support k_padding
// s_k should be greater than or equal to seqlen_k_min if provided
if(s_k.back() < seqlen_k_min)
{
std::ostringstream msg;
msg << __FILE__ << ":" << __LINE__ << ": seqlen_k (=" << s_k.back()
<< ") is less than minimum seqlen_k (=" << seqlen_k_min << ")";
throw std::runtime_error(msg.str());
}
return std::make_tuple(s_q, s_k, s_kpad);
}
else
{
ck_tile::index_t idx = 0;
std::string::size_type pos_q = 0;
std::string::size_type pos_k = 0;
std::string::size_type pos_kp = 0;
std::vector<ck_tile::index_t> s_q;
std::vector<ck_tile::index_t> s_k;
std::vector<ck_tile::index_t> s_kpad;
while(true)
{
auto found_q = q_val.find(',', pos_q);
auto found_k = k_val.find(',', pos_k);
auto found_kp = k_pad_val.find(',', pos_kp);
ck_tile::index_t q = _S2I_(
q_val.substr(pos_q, found_q == std::string::npos ? found_q : found_q - pos_q));
ck_tile::index_t k = _S2I_(
k_val.substr(pos_k, found_k == std::string::npos ? found_k : found_k - pos_k));
ck_tile::index_t kp = _S2I_(k_pad_val.substr(
pos_kp, found_kp == std::string::npos ? found_kp : found_kp - pos_kp));
s_q.push_back(q);
s_k.push_back(k < 0 ? q : k);
s_kpad.push_back(kp);
// s_k should be greater than or equal to seqlen_k_min
if(s_k.back() < seqlen_k_min)
{
std::ostringstream msg;
msg << __FILE__ << ":" << __LINE__ << ": seqlen_k (=" << s_k.back()
<< ") is less than minimum seqlen_k (=" << seqlen_k_min << ")";
throw std::runtime_error(msg.str());
}
idx++;
if(found_q == std::string::npos || idx >= batch)
{
break;
}
pos_q = found_q + 1;
pos_k = found_k == std::string::npos ? pos_k : found_k + 1;
pos_kp = found_kp == std::string::npos ? pos_kp : found_kp + 1;
}
if(idx < batch)
{
auto rem_q = generate_seqlens(batch - idx, s_q.back(), 1, s_kpad.back(), seed);
auto rem_k =
generate_seqlens(batch - idx, s_k.back(), seqlen_k_min, s_kpad.back(), seed);
s_q.insert(s_q.end(), rem_q.begin(), rem_q.end());
s_k.insert(s_k.end(), rem_k.begin(), rem_k.end());
s_kpad.insert(s_kpad.end(), batch - idx, s_kpad.back());
}
return std::make_tuple(s_q, s_k, s_kpad);
}
#undef _S2I_
}
int env_get_int(const char* var_name, int default_int)
{
char* v = getenv(var_name);
int r = default_int;
if(v)
r = std::atoi(v);
return r;
}
template <typename RandomAccessIterator, typename Int>
std::enable_if_t<std::is_integral_v<Int>> iota_shuffle(RandomAccessIterator first,
RandomAccessIterator last,
Int value,
std::optional<unsigned> seed = std::nullopt)
{
std::iota(first, last, value);
std::mt19937 engine(seed.has_value() ? *seed : std::random_device{}());
std::shuffle(first, last, engine);
}