mirror of
https://github.com/AuxXxilium/linux_dsm_epyc7002.git
synced 2024-12-16 05:47:35 +07:00
4e07dba7cb
Replaced libgcc functions with asm optimized implementation. Signed-off-by: Michal Simek <monstr@monstr.eu>
47 lines
839 B
ArmAsm
47 lines
839 B
ArmAsm
#include <linux/linkage.h>
|
|
|
|
/*
|
|
* Multiply operation for 32 bit integers.
|
|
* Input : Operand1 in Reg r5
|
|
* Operand2 in Reg r6
|
|
* Output: Result [op1 * op2] in Reg r3
|
|
*/
|
|
.text
|
|
.globl __mulsi3
|
|
.type __mulsi3, @function
|
|
.ent __mulsi3
|
|
|
|
__mulsi3:
|
|
.frame r1, 0, r15
|
|
add r3, r0, r0
|
|
beqi r5, result_is_zero /* multiply by zero */
|
|
beqi r6, result_is_zero /* multiply by zero */
|
|
bgeid r5, r5_pos
|
|
xor r4, r5, r6 /* get the sign of the result */
|
|
rsubi r5, r5, 0 /* make r5 positive */
|
|
r5_pos:
|
|
bgei r6, r6_pos
|
|
rsubi r6, r6, 0 /* make r6 positive */
|
|
r6_pos:
|
|
bri l1
|
|
l2:
|
|
add r5, r5, r5
|
|
l1:
|
|
srl r6, r6
|
|
addc r7, r0, r0
|
|
beqi r7, l2
|
|
bneid r6, l2
|
|
add r3, r3, r5
|
|
blti r4, negateresult
|
|
rtsd r15, 8
|
|
nop
|
|
negateresult:
|
|
rtsd r15, 8
|
|
rsub r3, r3, r0
|
|
result_is_zero:
|
|
rtsd r15, 8
|
|
addi r3, r0, 0
|
|
|
|
.size __mulsi3, . - __mulsi3
|
|
.end __mulsi3
|