ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] layernorm support fused-quant/fused-add (#1604)

Browse Source 
* add prenorm/postnorm support, refactor using generate.py

* update README

* update README

* fix format

* update some description and fix format

* update format

* format

* use non-raw for loading

* format and update n4096

* dynamic-quant ready

* update readme

* support fused dynamic-quant

* update fused-quant, with smooth

* update README

* update args

* update some based on comment
This commit is contained in:
carlushuang
2024-10-31 14:54:53 +08:00
committed by GitHub
parent 9a8a52130d
commit c3a4800c5f
61 changed files with 1790 additions and 766 deletions
Download Patch File Download Diff File Expand all files Collapse all files

104
include/ck_tile/core/numeric/int8.hpp Normal file
View File

@@ -0,0 +1,104 @@
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
#include "ck_tile/core/config.hpp"
#include "ck_tile/core/numeric/half.hpp"
#include "ck_tile/core/numeric/integral_constant.hpp"
#include "ck_tile/core/numeric/math.hpp"
#include "ck_tile/core/numeric/numeric.hpp"
#include "ck_tile/core/utility/bit_cast.hpp"
#include "ck_tile/core/utility/random.hpp"
#include <stdint.h>
#include <type_traits>
#pragma once
namespace ck_tile {
// use int8_t directly for int8 arithemetic
// here one can use ck_tile::int8_t to access original int8_t
using int8_t = int8_t;
// limits
template <class T>
struct numeric;
template <>
struct numeric<int8_t>
{
// minimum finite value, or minimum positive normalized value for float
CK_TILE_HOST_DEVICE static constexpr int8_t min() { return int8_t(-128); }
// minumum finite value
CK_TILE_HOST_DEVICE static constexpr int8_t lowest() { return int8_t(-128); }
// maximum finite value
CK_TILE_HOST_DEVICE static constexpr int8_t max() { return int8_t(127); }
// difference between 1.0 and next value representable by float
CK_TILE_HOST_DEVICE static constexpr int8_t epsilon()
{
return 1; // not used
}
CK_TILE_HOST_DEVICE static constexpr int8_t round_error()
{
return 1; // not used
}
// positive infinity value
CK_TILE_HOST_DEVICE static constexpr int8_t infinity()
{
return 1; // not used
}
// quiet NaN
CK_TILE_HOST_DEVICE static constexpr int8_t quiet_NaN()
{
return 1; // not used
}
// signaling NaN
CK_TILE_HOST_DEVICE static constexpr int8_t signaling_NaN()
{
return 1; // not used
}
// smallest positive subnormal value
CK_TILE_HOST_DEVICE static constexpr int8_t denorm_min()
{
return 1; // not used
}
CK_TILE_HOST_DEVICE static constexpr int8_t zero() { return 0; }
};
#if 0
template <typename T>
struct numeric_traits;
template <>
struct numeric_traits<int8_t>
{
static constexpr int exp = 5;
static constexpr int mant = 10;
static constexpr int bias = 15;
static constexpr uint16_t nan_mask = 0x7C00;
static constexpr uint16_t head_mask = 0xFC00;
static constexpr uint16_t mant_mask = 0x3FF;
static constexpr uint16_t exp_mask = 0x1F;
static constexpr uint32_t Inf = 0x7C00;
static constexpr uint32_t NegInf = 0xFC00;
static constexpr uint32_t NaN = 0x7C01;
static constexpr uint32_t Neg0 = 0x8000;
using bitwise_type = uint16_t;
};
#endif
CK_TILE_HOST_DEVICE
constexpr float int8_to_float(const int8_t& x) { return static_cast<float>(x); }
CK_TILE_HOST_DEVICE
constexpr int8_t float_to_int8(const float& x) { return static_cast<int8_t>(x); }
} // namespace ck_tile

4
include/ck_tile/core/numeric/type_convert.hpp
View File

@@ -10,6 +10,7 @@
#include "ck_tile/core/numeric/half.hpp"
#include "ck_tile/core/numeric/bfloat16.hpp"
#include "ck_tile/core/numeric/float8.hpp"
#include "ck_tile/core/numeric/int8.hpp"
namespace ck_tile {
@@ -60,6 +61,9 @@ CK_TILE_TYPE_CONVERT(bf16_t, bf16, float, float)
CK_TILE_TYPE_CONVERT(fp8_t, fp8, float, float)
CK_TILE_TYPE_CONVERT(bf8_t, bf8, float, float)
CK_TILE_TYPE_CONVERT(float, float, int8_t, int8)
CK_TILE_TYPE_CONVERT(int8_t, int8, float, float)
#undef CK_TILE_TYPE_CONVERT
#endif