ops_other_arm64.go

     1  // Copyright 2026 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  //go:build goexperiment.simd
     6  
     7  package archsimd
     8  
     9  // LoadInt8sPart loads an Int8s vector from memory pointed to by s and returns the number of
    10  // loaded elements.
    11  // Asm: Emulated, CPU Feature: SVE
    12  func LoadInt8sPart(s []int8) (Int8s, int) {
    13  	return loadInt8sMasked(&s[0], Mask8sFromCount(len(s))), min(len(s), vl())
    14  }
    15  
    16  // Asm: ZLD1B,
    17  func loadInt8sMasked(s *int8, m Mask8s) Int8s
    18  
    19  // StorePart stores x to memory pointed to by s and returns the number of
    20  // stored elements.
    21  // Asm: Emulated, CPU Feature: SVE
    22  func (x Int8s) StorePart(s []int8) int {
    23  	x.storeMasked(&s[0], Mask8sFromCount(len(s)))
    24  	return min(len(s), vl())
    25  }
    26  
    27  // Asm: ZST1B, CPU Feature: SVE
    28  func (x Int8s) storeMasked(s *int8, m Mask8s)
    29  
    30  // Mask8sFromCount creates a Mask8s that enables elements from index 0 up to x, exclusive.
    31  // Asm: PWHILELT, CPU Feature: SVE
    32  func Mask8sFromCount(x int) Mask8s
    33  
    34  // Greater returns a mask whose elements indicate whether x > y.
    35  // Asm: ZCMPGT, CPU Feature: SVE
    36  func (x Int8s) Greater(y Int8s) Mask8s
    37  
    38  // Masked returns the result vector of x masked by m, where m is active,
    39  // or 0 otherwise.
    40  // Asm: Emulated, CPU Feature: SVE
    41  func (x Int8s) Masked(m Mask8s) Int8s {
    42  	var zero Int8s
    43  	return x.IfElse(zero, m)
    44  }
    45  
    46  // IfElse returns x but with elements set to y where mask is false.
    47  // Asm: ZSEL, CPU Feature: SVE
    48  func (x Int8s) IfElse(y Int8s, m Mask8s) Int8s
    49  
    50  // Add adds two Int8s vectors element-wise.
    51  // Asm: ZADD, CPU Feature: SVE
    52  func (x Int8s) Add(y Int8s) Int8s
    53
View as plain text