38 lines
918 B
ArmAsm
38 lines
918 B
ArmAsm
|
;; SPDX-License-Identifier: GPL-2.0-or-later
|
||
|
|
;; Copyright (C) 2010 Texas Instruments Incorporated
|
||
|
|
;; Contributed by Mark Salter <msalter@redhat.com>.
|
||
|
|
;;
|
||
|
|
|
||
|
|
#include <linux/linkage.h>
|
||
|
|
|
||
|
|
;; uint64_t __c6xabi_mpyll(uint64_t x, uint64_t y)
|
||
|
|
;;
|
||
|
|
;; 64x64 multiply
|
||
|
|
;; First compute partial results using 32-bit parts of x and y:
|
||
|
|
;;
|
||
|
|
;; b63 b32 b31 b0
|
||
|
|
;; -----------------------------
|
||
|
|
;; | 1 | 0 |
|
||
|
|
;; -----------------------------
|
||
|
|
;;
|
||
|
|
;; P0 = X0*Y0
|
||
|
|
;; P1 = X0*Y1 + X1*Y0
|
||
|
|
;; P2 = X1*Y1
|
||
|
|
;;
|
||
|
|
;; result = (P2 << 64) + (P1 << 32) + P0
|
||
|
|
;;
|
||
|
|
;; Since the result is also 64-bit, we can skip the P2 term.
|
||
|
|
|
||
|
|
.text
|
||
|
|
ENTRY(__c6xabi_mpyll)
|
||
|
|
mpy32u .m1x A4,B4,A1:A0 ; X0*Y0
|
||
|
|
b .s2 B3
|
||
|
|
|| mpy32u .m2x B5,A4,B1:B0 ; X0*Y1 (don't need upper 32-bits)
|
||
|
|
|| mpy32u .m1x A5,B4,A3:A2 ; X1*Y0 (don't need upper 32-bits)
|
||
|
|
nop
|
||
|
|
nop
|
||
|
|
mv .s1 A0,A4
|
||
|
|
add .l1x A2,B0,A5
|
||
|
|
add .s1 A1,A5,A5
|
||
|
|
ENDPROC(__c6xabi_mpyll)
|