mxf8_utils.hpp File Reference

mxf8_utils.hpp File Reference#

Composable Kernel: mxf8_utils.hpp File Reference
Namespaces | Macros | Functions
mxf8_utils.hpp File Reference
#include "ck/utility/numeric_limits.hpp"
#include "ck/utility/mxfp_utils.hpp"

Go to the source code of this file.

Namespaces

namespace  ck
namespace  ck::fp8_impl

Macros

#define CK_MX_FP8_CVT_FAST_PATH   0

Functions

template<typename Y, typename X>
__host__ __device__ constexpr Y ck::mxf8_convert_sr (X x, float scale)
template<typename Y, typename X>
__host__ __device__ constexpr Y ck::mxf8_convert_rne (X x, float scale)
template<>
__host__ __device__ f8_ocp_t ck::mxf8_convert_rne< f8_ocp_t, float > (float x, float scale)
template<>
__host__ __device__ bf8_ocp_t ck::mxf8_convert_rne< bf8_ocp_t, float > (float x, float scale)
template<>
__host__ __device__ f8x2_ocp_t ck::mxf8_convert_rne< f8x2_ocp_t, float2_t > (float2_t x, float scale)
template<>
__host__ __device__ bf8x2_ocp_t ck::mxf8_convert_rne< bf8x2_ocp_t, float2_t > (float2_t x, float scale)
template<>
__host__ __device__ f8x16_ocp_t ck::mxf8_convert_rne< f8x16_ocp_t, float16_t > (float16_t x, float scale)
template<>
__host__ __device__ bf8x16_ocp_t ck::mxf8_convert_rne< bf8x16_ocp_t, float16_t > (float16_t x, float scale)
template<>
__host__ __device__ f8x32_ocp_t ck::mxf8_convert_rne< f8x32_ocp_t, float32_t > (float32_t x, float scale)
template<>
__host__ __device__ bf8x32_ocp_t ck::mxf8_convert_rne< bf8x32_ocp_t, float32_t > (float32_t x, float scale)
template<>
__host__ __device__ f8_ocp_t ck::mxf8_convert_sr< f8_ocp_t, float > (float x, float scale)
template<>
__host__ __device__ bf8_ocp_t ck::mxf8_convert_sr< bf8_ocp_t, float > (float x, float scale)
template<>
__host__ __device__ f8x2_ocp_t ck::mxf8_convert_sr< f8x2_ocp_t, float2_t > (float2_t x, float scale)
template<>
__host__ __device__ bf8x2_ocp_t ck::mxf8_convert_sr< bf8x2_ocp_t, float2_t > (float2_t x, float scale)
template<>
__host__ __device__ f8x16_ocp_t ck::mxf8_convert_sr< f8x16_ocp_t, float16_t > (float16_t x, float scale)
template<>
__host__ __device__ bf8x16_ocp_t ck::mxf8_convert_sr< bf8x16_ocp_t, float16_t > (float16_t x, float scale)
template<>
__host__ __device__ f8x32_ocp_t ck::mxf8_convert_sr< f8x32_ocp_t, float32_t > (float32_t x, float scale)
template<>
__host__ __device__ bf8x32_ocp_t ck::mxf8_convert_sr< bf8x32_ocp_t, float32_t > (float32_t x, float scale)

Macro Definition Documentation

◆ CK_MX_FP8_CVT_FAST_PATH

#define CK_MX_FP8_CVT_FAST_PATH   0