Adding SVML support to AVX512 targets
This commit is contained in:
@@ -617,16 +617,16 @@ define i64 @__popcnt_int64(i64) nounwind readonly alwaysinline {
|
||||
}
|
||||
ctlztz()
|
||||
|
||||
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
||||
; FIXME: need either to wire these up to the 8-wide SVML entrypoints,
|
||||
; or, use the macro to call the 4-wide ones twice with our 8-wide
|
||||
; vectors...
|
||||
|
||||
;; TODO: should we use masked versions of SVML functions?
|
||||
;; svml
|
||||
|
||||
include(`svml.m4')
|
||||
svml_stubs(float,f,WIDTH)
|
||||
svml_stubs(double,d,WIDTH)
|
||||
svml_declare(float,f16,16)
|
||||
svml_define(float,f16,16,f)
|
||||
|
||||
;; double precision
|
||||
svml_declare(double,8,8)
|
||||
svml_define_x(double,8,8,d,16)
|
||||
|
||||
|
||||
|
||||
|
||||
Reference in New Issue
Block a user