HIP: Heterogenous-computing Interface for Portability
|
hip_bf16.h provides struct for __hip_bfloat16 types More...
#include <hip/amd_detail/amd_hip_common.h>
#include "amd_hip_vector_types.h"
#include "device_library_decls.h"
#include "math_fwd.h"
#include <climits>
Go to the source code of this file.
Classes | |
struct | __hip_bfloat16 |
Struct to represent a 16 bit brain floating point number. More... | |
struct | __hip_bfloat162 |
Struct to represent two 16 bit brain floating point numbers. More... | |
Macros | |
#define | __HOST_DEVICE__ __host__ __device__ |
Functions | |
__HOST_DEVICE__ float | __bfloat162float (__hip_bfloat16 a) |
Converts bfloat16 to float. | |
__HOST_DEVICE__ __hip_bfloat16 | __float2bfloat16 (float f) |
Converts float to bfloat16. | |
__HOST_DEVICE__ float2 | __bfloat1622float2 (const __hip_bfloat162 a) |
Converts and moves bfloat162 to float2. | |
__device__ __hip_bfloat162 | __bfloat162bfloat162 (const __hip_bfloat16 a) |
Moves bfloat16 value to bfloat162. | |
__device__ short int | __bfloat16_as_short (const __hip_bfloat16 h) |
Reinterprets bits in a __hip_bfloat16 as a signed short integer. | |
__device__ unsigned short int | __bfloat16_as_ushort (const __hip_bfloat16 h) |
Reinterprets bits in a __hip_bfloat16 as an unsigned signed short integer. | |
__HOST_DEVICE__ __hip_bfloat16 | __double2bfloat16 (const double a) |
Convert double to __hip_bfloat16. | |
__HOST_DEVICE__ __hip_bfloat162 | __float22bfloat162_rn (const float2 a) |
Convert float2 to __hip_bfloat162. | |
__device__ __hip_bfloat162 | __halves2bfloat162 (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Combine two __hip_bfloat16 to __hip_bfloat162. | |
__device__ __hip_bfloat16 | __high2bfloat16 (const __hip_bfloat162 a) |
Returns high 16 bits of __hip_bfloat162. | |
__device__ __hip_bfloat162 | __high2bfloat162 (const __hip_bfloat162 a) |
Returns high 16 bits of __hip_bfloat162. | |
__HOST_DEVICE__ float | __high2float (const __hip_bfloat162 a) |
Converts high 16 bits of __hip_bfloat162 to float and returns the result. | |
__device__ __hip_bfloat162 | __highs2bfloat162 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Extracts high 16 bits from each and combines them. | |
__device__ __hip_bfloat16 | __low2bfloat16 (const __hip_bfloat162 a) |
Returns low 16 bits of __hip_bfloat162. | |
__device__ __hip_bfloat162 | __low2bfloat162 (const __hip_bfloat162 a) |
Returns low 16 bits of __hip_bfloat162. | |
__HOST_DEVICE__ float | __low2float (const __hip_bfloat162 a) |
Converts low 16 bits of __hip_bfloat162 to float and returns the result. | |
__device__ __hip_bfloat162 | __lowhigh2highlow (const __hip_bfloat162 a) |
Swaps both halves. | |
__device__ __hip_bfloat162 | __lows2bfloat162 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Extracts low 16 bits from each and combines them. | |
__device__ __hip_bfloat16 | __short_as_bfloat16 (const short int a) |
Reinterprets short int into a bfloat16. | |
__device__ __hip_bfloat16 | __ushort_as_bfloat16 (const unsigned short int a) |
Reinterprets unsigned short int into a bfloat16. | |
__device__ __hip_bfloat16 | __hadd (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Adds two bfloat16 values. | |
__device__ __hip_bfloat16 | __hsub (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Subtracts two bfloat16 values. | |
__device__ __hip_bfloat16 | __hdiv (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Divides two bfloat16 values. | |
__device__ __hip_bfloat16 | __hfma (const __hip_bfloat16 a, const __hip_bfloat16 b, const __hip_bfloat16 c) |
Performs FMA of given bfloat16 values. | |
__device__ __hip_bfloat16 | __hmul (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Multiplies two bfloat16 values. | |
__device__ __hip_bfloat16 | __hneg (const __hip_bfloat16 a) |
Negate a bfloat16 value. | |
__device__ __hip_bfloat16 | __habs (const __hip_bfloat16 a) |
Returns absolute of a bfloat16. | |
__device__ __hip_bfloat162 | __h2div (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Divides bfloat162 values. | |
__device__ __hip_bfloat162 | __habs2 (const __hip_bfloat162 a) |
Returns absolute of a bfloat162. | |
__device__ __hip_bfloat162 | __hadd2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Adds two bfloat162 values. | |
__device__ __hip_bfloat162 | __hfma2 (const __hip_bfloat162 a, const __hip_bfloat162 b, const __hip_bfloat162 c) |
Performs FMA of given bfloat162 values. | |
__device__ __hip_bfloat162 | __hmul2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Multiplies two bfloat162 values. | |
__device__ __hip_bfloat162 | __hneg2 (const __hip_bfloat162 a) |
Converts a bfloat162 into negative. | |
__device__ __hip_bfloat162 | __hsub2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Subtracts two bfloat162 values. | |
__device__ bool | __heq (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values. | |
__device__ bool | __hequ (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered equal. | |
__device__ bool | __hgt (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - greater than. | |
__device__ bool | __hgtu (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered greater than. | |
__device__ bool | __hge (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - greater than equal. | |
__device__ bool | __hgeu (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered greater than equal. | |
__device__ bool | __hne (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - not equal. | |
__device__ bool | __hneu (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered not equal. | |
__device__ __hip_bfloat16 | __hmax (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - return max. | |
__device__ __hip_bfloat16 | __hmin (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - return min. | |
__device__ bool | __hlt (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - less than operator. | |
__device__ bool | __hltu (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered less than. | |
__device__ bool | __hle (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - less than. | |
__device__ bool | __hleu (const __hip_bfloat16 a, const __hip_bfloat16 b) |
Compare two bfloat162 values - unordered less than equal. | |
__device__ int | __hisinf (const __hip_bfloat16 a) |
Checks if number is inf. | |
__device__ bool | __hisnan (const __hip_bfloat16 a) |
Checks if number is nan. | |
__device__ bool | __hbeq2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Checks if two numbers are equal. | |
__device__ bool | __hbequ2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Checks if two numbers are equal - unordered. | |
__device__ bool | __hbge2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a >= b. | |
__device__ bool | __hbgeu2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a >= b - unordered. | |
__device__ bool | __hbgt2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a > b. | |
__device__ bool | __hbgtu2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a > b - unordered. | |
__device__ bool | __hble2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a <= b. | |
__device__ bool | __hbleu2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a <= b - unordered. | |
__device__ bool | __hblt2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a < b. | |
__device__ bool | __hbltu2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a < b - unordered. | |
__device__ bool | __hbne2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a != b. | |
__device__ bool | __hbneu2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a != b. | |
__device__ __hip_bfloat162 | __heq2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a != b, returns 1.0 if equal, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hge2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a >= b, returns 1.0 if greater than equal, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hgt2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a > b, returns 1.0 if greater than equal, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hisnan2 (const __hip_bfloat162 a) |
Check for a is NaN, returns 1.0 if NaN, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hle2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a <= b, returns 1.0 if greater than equal, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hlt2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Check for a < b, returns 1.0 if greater than equal, otherwise 0.0. | |
__device__ __hip_bfloat162 | __hmax2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Returns max of two elements. | |
__device__ __hip_bfloat162 | __hmin2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Returns min of two elements. | |
__device__ __hip_bfloat162 | __hne2 (const __hip_bfloat162 a, const __hip_bfloat162 b) |
Checks for not equal to. | |
__device__ __hip_bfloat16 | hceil (const __hip_bfloat16 h) |
Calculate ceil of bfloat16. | |
__device__ __hip_bfloat16 | hcos (const __hip_bfloat16 h) |
Calculate cosine of bfloat16. | |
__device__ __hip_bfloat16 | hexp (const __hip_bfloat16 h) |
Calculate exponential of bfloat16. | |
__device__ __hip_bfloat16 | hexp10 (const __hip_bfloat16 h) |
Calculate exponential 10 of bfloat16. | |
__device__ __hip_bfloat16 | hexp2 (const __hip_bfloat16 h) |
Calculate exponential 2 of bfloat16. | |
__device__ __hip_bfloat16 | hfloor (const __hip_bfloat16 h) |
Calculate floor of bfloat16. | |
__device__ __hip_bfloat16 | hlog (const __hip_bfloat16 h) |
Calculate natural log of bfloat16. | |
__device__ __hip_bfloat16 | hlog10 (const __hip_bfloat16 h) |
Calculate log 10 of bfloat16. | |
__device__ __hip_bfloat16 | hlog2 (const __hip_bfloat16 h) |
Calculate log 2 of bfloat16. | |
__device__ __hip_bfloat16 | hrcp (const __hip_bfloat16 h) |
Calculate reciprocal. | |
__device__ __hip_bfloat16 | hrint (const __hip_bfloat16 h) |
Round to nearest int. | |
__device__ __hip_bfloat16 | hrsqrt (const __hip_bfloat16 h) |
Reciprocal square root. | |
__device__ __hip_bfloat16 | hsin (const __hip_bfloat16 h) |
Calculate sin of bfloat16. | |
__device__ __hip_bfloat16 | hsqrt (const __hip_bfloat16 h) |
Calculate sqrt of bfloat16. | |
__device__ __hip_bfloat16 | htrunc (const __hip_bfloat16 h) |
Calculate truncate of bfloat16. | |
__device__ __hip_bfloat162 | h2ceil (const __hip_bfloat162 h) |
Calculate ceil of bfloat162. | |
__device__ __hip_bfloat162 | h2cos (const __hip_bfloat162 h) |
Calculate cosine of bfloat162. | |
__device__ __hip_bfloat162 | h2exp (const __hip_bfloat162 h) |
Calculate exponential of bfloat162. | |
__device__ __hip_bfloat162 | h2exp10 (const __hip_bfloat162 h) |
Calculate exponential 10 of bfloat162. | |
__device__ __hip_bfloat162 | h2exp2 (const __hip_bfloat162 h) |
Calculate exponential 2 of bfloat162. | |
__device__ __hip_bfloat162 | h2floor (const __hip_bfloat162 h) |
Calculate floor of bfloat162. | |
__device__ __hip_bfloat162 | h2log (const __hip_bfloat162 h) |
Calculate natural log of bfloat162. | |
__device__ __hip_bfloat162 | h2log10 (const __hip_bfloat162 h) |
Calculate log 10 of bfloat162. | |
__device__ __hip_bfloat162 | h2log2 (const __hip_bfloat162 h) |
Calculate log 2 of bfloat162. | |
__device__ __hip_bfloat162 | h2rcp (const __hip_bfloat162 h) |
Calculate vector reciprocal. | |
__device__ __hip_bfloat162 | h2rint (const __hip_bfloat162 h) |
Calculate vector round to nearest int. | |
__device__ __hip_bfloat162 | h2rsqrt (const __hip_bfloat162 h) |
Calculate vector reciprocal square root. | |
__device__ __hip_bfloat162 | h2sin (const __hip_bfloat162 h) |
Calculate sin of bfloat162. | |
__device__ __hip_bfloat162 | h2sqrt (const __hip_bfloat162 h) |
Calculate sqrt of bfloat162. | |
__device__ __hip_bfloat162 | h2trunc (const __hip_bfloat162 h) |
Calculate truncate of bfloat162. | |
hip_bf16.h provides struct for __hip_bfloat16 types
MIT License
Copyright (c) 2019 - 2023 Advanced Micro Devices, Inc. All rights reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.