ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-24 14:54:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix build wip

Browse Source
This commit is contained in:
carlushuang
2024-02-29 22:27:31 +00:00
parent f69356b1d7
commit fbd25cea35
43 changed files with 469 additions and 336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
include/ck_tile/core/numeric/bfloat16.hpp
View File

@@ -6,6 +6,7 @@
#include "ck_tile/core/numeric/arithmetic.hpp"
#include "ck_tile/core/numeric/half.hpp"
#include "ck_tile/core/numeric/integral_constant.hpp"
#include "ck_tile/core/utility/limits.hpp"
#include <stdint.h>
#pragma once
@@ -32,7 +33,7 @@ struct alignas(2) bfloat16_t
raw_type data;
CK_TILE_HOST_DEVICE
static bfloat16_t bit_cast(raw_type x)
static constexpr bfloat16_t bit_cast(raw_type x)
{
bfloat16_t y;
y.data = x;

43
include/ck_tile/core/numeric/float8.hpp
View File

@@ -8,6 +8,7 @@
#include "ck_tile/core/numeric/arithmetic.hpp"
#include "ck_tile/core/numeric/half.hpp"
#include "ck_tile/core/numeric/integral_constant.hpp"
#include "ck_tile/core/utility/limits.hpp"
#include <stdint.h>
#include <type_traits>
@@ -62,7 +63,7 @@ struct alignas(1) float8_e4m3_t
raw_type data;
CK_TILE_HOST_DEVICE
static float8_e4m3_t bit_cast(raw_type x)
static constexpr float8_e4m3_t bit_cast(raw_type x)
{
float8_e4m3_t y;
y.data = x;
@@ -70,37 +71,40 @@ struct alignas(1) float8_e4m3_t
}
// constructor
float8_e4m3_t() = default;
constexpr float8_e4m3_t() : data() {}
// construct from float
CK_TILE_HOST_DEVICE
explicit float8_e4m3_t(const float& x) { data = float_to_fp8_raw(x); }
explicit constexpr float8_e4m3_t(const float& x) { data = float_to_fp8_raw(x); }
// construct from int
CK_TILE_HOST_DEVICE
explicit float8_e4m3_t(const int& x) { data = float_to_fp8_raw(static_cast<float>(x)); }
explicit constexpr float8_e4m3_t(const int& x)
{
data = float_to_fp8_raw(static_cast<float>(x));
}
// construct from unsigned int
CK_TILE_HOST_DEVICE
explicit float8_e4m3_t(const unsigned int& x)
explicit constexpr float8_e4m3_t(const unsigned int& x)
{
data = float_to_fp8_raw(static_cast<float>(x));
}
// cast to float
CK_TILE_HOST_DEVICE
explicit operator float() const { return fp8_to_float_raw(data); }
explicit constexpr operator float() const { return fp8_to_float_raw(data); }
// cast to int
CK_TILE_HOST_DEVICE
explicit operator int() const { return static_cast<int>(fp8_to_float_raw(data)); }
explicit constexpr operator int() const { return static_cast<int>(fp8_to_float_raw(data)); }
// internal access
CK_TILE_HOST_DEVICE
raw_type& get() { return data; }
constexpr raw_type& get() { return data; }
CK_TILE_HOST_DEVICE
raw_type get() const { return data; }
constexpr raw_type get() const { return data; }
};
struct alignas(1) float8_e5m2_t
@@ -116,7 +120,7 @@ struct alignas(1) float8_e5m2_t
raw_type data;
CK_TILE_HOST_DEVICE
static float8_e5m2_t bit_cast(raw_type x)
static constexpr float8_e5m2_t bit_cast(raw_type x)
{
float8_e5m2_t y;
y.data = x;
@@ -124,37 +128,40 @@ struct alignas(1) float8_e5m2_t
}
// constructor
float8_e5m2_t() = default;
constexpr float8_e5m2_t() : data() {}
// construct from float
CK_TILE_HOST_DEVICE
explicit float8_e5m2_t(const float& x) { data = float_to_bf8_raw(x); }
explicit constexpr float8_e5m2_t(const float& x) { data = float_to_bf8_raw(x); }
// construct from int
CK_TILE_HOST_DEVICE
explicit float8_e5m2_t(const int& x) { data = float_to_bf8_raw(static_cast<float>(x)); }
explicit constexpr float8_e5m2_t(const int& x)
{
data = float_to_bf8_raw(static_cast<float>(x));
}
// construct from unsigned int
CK_TILE_HOST_DEVICE
explicit float8_e5m2_t(const unsigned int& x)
explicit constexpr float8_e5m2_t(const unsigned int& x)
{
data = float_to_bf8_raw(static_cast<float>(x));
}
// cast to float
CK_TILE_HOST_DEVICE
explicit operator float() const { return bf8_to_float_raw(data); }
explicit constexpr constexpr operator float() const { return bf8_to_float_raw(data); }
// cast to int
CK_TILE_HOST_DEVICE
explicit operator int() const { return static_cast<int>(bf8_to_float_raw(data)); }
explicit constexpr operator int() const { return static_cast<int>(bf8_to_float_raw(data)); }
// internal access
CK_TILE_HOST_DEVICE
raw_type& get() { return data; }
constexpr raw_type& get() { return data; }
CK_TILE_HOST_DEVICE
raw_type get() const { return data; }
constexpr raw_type get() const { return data; }
};
// below is sw fp8 conversion, not utilizing hw instruction

45
include/ck_tile/core/numeric/half.hpp
View File

@@ -3,26 +3,29 @@
#include "ck_tile/core/config.hpp"
#include "ck_tile/core/utility/bit_cast.hpp"
#include "ck_tile/core/utility/limits.hpp"
#include <hip/hip_fp16.h>
#pragma once
namespace ck_tile {
CK_TILE_HOST_DEVICE
float fp16_to_float_hip(const _Float16& x);
using fp16_hip_t = __half; // most of hip internal function use this type
CK_TILE_HOST_DEVICE
_Float16 float_to_fp16_hip(const float& x);
float fp16_to_float_hip(const fp16_hip_t& x);
// HIP use _Float16 as interchangable data type for float16
CK_TILE_HOST_DEVICE
fp16_hip_t float_to_fp16_hip(const float& x);
// HIP use fp16_hip_t as interchangable data type for float16
struct alignas(2) half_t
{
using raw_type = uint16_t;
raw_type data;
CK_TILE_HOST_DEVICE
static half_t bit_cast(raw_type x)
static constexpr half_t bit_cast(raw_type x)
{
half_t y;
y.data = x;
@@ -30,56 +33,62 @@ struct alignas(2) half_t
}
CK_TILE_HOST_DEVICE
_Float16 to_fp16() const { return reinterpret_cast<const raw_type&>(data); }
constexpr fp16_hip_t to_fp16() const { return ck_tile::bit_cast<fp16_hip_t>(data); }
// constructor
half_t() = default;
constexpr half_t() : data() {}
// construct from HIP half
CK_TILE_HOST_DEVICE
explicit half_t(const _Float16& x) : data(reinterpret_cast<const raw_type&>(x)) {}
explicit constexpr half_t(const fp16_hip_t& x) : data(ck_tile::bit_cast<raw_type>(x)) {}
// construct from float
CK_TILE_HOST_DEVICE
explicit half_t(const float& x) : half_t(float_to_fp16_hip(x)) {}
explicit constexpr half_t(const float& x) : half_t(float_to_fp16_hip(x)) {}
// construct from int
CK_TILE_HOST_DEVICE
explicit half_t(const int& x) : half_t(__int2half_rn(x)) {}
explicit constexpr half_t(const int& x) : half_t(static_cast<fp16_hip_t>(__int2half_rn(x))) {}
// construct from unsigned int
CK_TILE_HOST_DEVICE
explicit half_t(const unsigned int& x) : half_t(__uint2half_rn(x)) {}
explicit constexpr half_t(const unsigned int& x)
: half_t(static_cast<fp16_hip_t>(__uint2half_rn(x)))
{
}
// cast to float
CK_TILE_HOST_DEVICE
explicit operator float() const { return fp16_to_float_hip(to_fp16()); }
explicit constexpr operator float() const { return fp16_to_float_hip(to_fp16()); }
// cast to int
CK_TILE_HOST_DEVICE
explicit operator int() const { return static_cast<int>(fp16_to_float_hip(to_fp16())); }
explicit constexpr operator int() const
{
return static_cast<int>(fp16_to_float_hip(to_fp16()));
}
// internal access
CK_TILE_HOST_DEVICE
raw_type& get() { return data; }
constexpr raw_type& get() { return data; }
CK_TILE_HOST_DEVICE
raw_type get() const { return data; }
constexpr raw_type get() const { return data; }
};
// conversions
CK_TILE_HOST_DEVICE
float fp16_to_float_hip(const _Float16& x)
float fp16_to_float_hip(const fp16_hip_t& x)
{
// return __half2float(x);
return static_cast<float>(x);
}
CK_TILE_HOST_DEVICE
_Float16 float_to_fp16_hip(const float& x)
fp16_hip_t float_to_fp16_hip(const float& x)
{
// return __float2half(x);
return static_cast<_Float16>(x);
return static_cast<fp16_hip_t>(x);
}
CK_TILE_HOST_DEVICE

3
include/ck_tile/core/numeric/math.hpp
View File

@@ -6,6 +6,7 @@
#include "ck_tile/core/config.hpp"
#include "ck_tile/core/numeric/integer.hpp"
#include "ck_tile/core/numeric/integral_constant.hpp"
#include "ck_tile/core/utility/bit_cast.hpp"
#include <type_traits>
#include <stdint.h>
@@ -290,7 +291,7 @@ float abs(const float& x)
CK_TILE_HOST_DEVICE
bool isnan(const float& x)
{
uint32_t xx = reinterpret_cast<const uint32_t&>(x);
uint32_t xx = bit_cast<uint32_t>(x);
return (xx & 0x7fffffff) > 0x7F800000;
}