bfloat-8

Vocabulary

math.vectors.simd

Class description

A 128-bit sequence of eight BF16 values. Elements are Factor floats. Ordinary arithmetic computes in binary32 before rounding the stored result to BF16. Dedicated BF16 dot and matrix operations accumulate in float-4 with Arm's baseline BF16 computation semantics.

Definition

IN: math.vectors.simd

TUPLE: bfloat-8 < simd-128 ; final

Methods

USING: kernel math.vectors.simd sequences ;

M: bfloat-8 length drop 8 ; inline


USING: classes kernel math.vectors.simd sequences ;

M: bfloat-8 like
    drop dup \ bfloat-8 instance? [ >bfloat-8 ] unless ; inline


USING: kernel math.vectors.simd math.vectors.simd.private ;

M: bfloat-8 new-underlying drop \ bfloat-8 boa ; inline


USING: cpu.architecture generic kernel math.vectors.simd
math.vectors.simd.intrinsics math.vectors.simd.private
sequences.private ;

M: bfloat-8 nth-unsafe
    swap \ bfloat-8-rep
    [ (simd-select) ]
    [ M\ bfloat-8 nth-unsafe (call-next-method) ] vx->x-op ;
    inline


USING: kernel math.vectors.simd prettyprint.custom ;

M: bfloat-8 pprint-delims drop \ bfloat-8{ \ } ;


USING: accessors math.floats.small.c-types math.vectors.simd
math.vectors.simd.private sequences.private ;

M: bfloat-8 set-nth-unsafe
    underlying>> bfloat set-vector-element ; inline


USING: kernel math.floats.small.c-types math.vectors.simd ;

M: bfloat-8 simd-element-type drop bfloat ; inline


USING: cpu.architecture kernel math.vectors.simd ;

M: bfloat-8 simd-rep drop bfloat-8-rep ; inline


USING: kernel math.vectors.simd ;

M: bfloat-8 simd-with drop bfloat-8-with ; inline


USING: math math.vectors math.vectors.simd sequences ;

M: bfloat-8 v* [ * float>bits bits>float ] 2map ;


USING: math math.vectors math.vectors.simd sequences ;

M: bfloat-8 v+ [ + float>bits bits>float ] 2map ;


USING: math math.vectors math.vectors.simd sequences ;

M: bfloat-8 v- [ - float>bits bits>float ] 2map ;


USING: math math.vectors math.vectors.simd sequences ;

M: bfloat-8 v/ [ /f float>bits bits>float ] 2map ;


USING: math.libm math.vectors math.vectors.simd sequences ;

M: bfloat-8 vfma [ ffmaf ] 3map ;


USING: math.vectors math.vectors.simd ;

M: bfloat-8 vs* v* ;


USING: math.vectors math.vectors.simd ;

M: bfloat-8 vs+ v+ ;


USING: math.vectors math.vectors.simd ;

M: bfloat-8 vs- v- ;


USING: math math.libm math.vectors math.vectors.simd sequences ;

M: bfloat-8 vsqrt [ fsqrt float>bits bits>float ] map ;