Adding SVML support to AVX512 targets

This commit is contained in:
Dmitry Babokin
2016-11-30 05:27:10 +03:00
parent f04a04a7e3
commit ff298f21b7

View File

@@ -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)