mirror of
https://github.com/hanzoai/wasm2go.git
synced 2026-10-30 15:06:48 +00:00
730057c146
Assembly overrides can now declare an arm64 body for FEAT_FHM, the f16-product / f32-accumulate multiply-adds (FMLAL, FMLAL2 and their by-element forms). The base package detects it (hw.optional.arm.FEAT_FHM on darwin, HWCAP_ASIMDFHM on linux) as CPUFHM, and the dispatch ladder checks it first: the arm64 levels are independent features rather than a ladder, so the order only decides which of several present bodies runs. Co-authored-by: Claude Fable 5.1 <[email protected]>
36 lines
918 B
Go
36 lines
918 B
Go
//go:build arm64 && darwin
|
|
|
|
package helpers
|
|
|
|
import "syscall"
|
|
|
|
// CPUDotProd reports FEAT_DotProd (the SDOT/UDOT instructions).
|
|
// Feature-gated splice bodies dispatch on it at runtime; the
|
|
// portable twin body runs when it is false.
|
|
var CPUDotProd = detectDotProd()
|
|
|
|
// CPUI8MM reports FEAT_I8MM (SMMLA/UMMLA); see the linux twin for
|
|
// why it is detected apart from CPUDotProd.
|
|
var CPUI8MM = detectI8MM()
|
|
|
|
// CPUFHM reports FEAT_FHM (FMLAL/FMLSL: f16 products accumulated in
|
|
// f32). Bodies that keep f16 operands unwidened dispatch on it.
|
|
var CPUFHM = detectFHM()
|
|
|
|
func detectDotProd() bool {
|
|
return sysctlFeature("hw.optional.arm.FEAT_DotProd")
|
|
}
|
|
|
|
func detectI8MM() bool {
|
|
return sysctlFeature("hw.optional.arm.FEAT_I8MM")
|
|
}
|
|
|
|
func detectFHM() bool {
|
|
return sysctlFeature("hw.optional.arm.FEAT_FHM")
|
|
}
|
|
|
|
func sysctlFeature(name string) bool {
|
|
v, err := syscall.SysctlUint32(name)
|
|
return err == nil && v != 0
|
|
}
|