linux_dsm_epyc7002/arch/microblaze/lib/umodsi3.S
Michal Simek 4e07dba7cb microblaze: Add libgcc function directly to kernel
Replaced libgcc functions with asm optimized implementation.

Signed-off-by: Michal Simek <monstr@monstr.eu>
2010-10-21 15:51:42 +10:00

87 lines
1.9 KiB
ArmAsm

#include <linux/linkage.h>
/*
* Unsigned modulo operation for 32 bit integers.
* Input : op1 in Reg r5
* op2 in Reg r6
* Output: op1 mod op2 in Reg r3
*/
.text
.globl __umodsi3
.type __umodsi3, @function
.ent __umodsi3
__umodsi3:
.frame r1, 0, r15
addik r1, r1, -12
swi r29, r1, 0
swi r30, r1, 4
swi r31, r1, 8
beqi r6, div_by_zero /* div_by_zero - division error */
beqid r5, result_is_zero /* result is zero */
addik r3, r0, 0 /* clear div */
addik r30, r0, 0 /* clear mod */
addik r29, r0, 32 /* initialize the loop count */
/* check if r6 and r5 are equal /* if yes, return 0 */
rsub r18, r5, r6
beqi r18, return_here
/* check if (uns)r6 is greater than (uns)r5. in that case, just return r5 */
xor r18, r5, r6
bgeid r18, 16
addik r3, r5, 0
blti r6, return_here
bri $lcheckr6
rsub r18, r5, r6 /* microblazecmp */
bgti r18, return_here
/* if r6 [bit 31] is set, then return result as r5-r6 */
$lcheckr6:
bgtid r6, div0
addik r3, r0, 0
addik r18, r0, 0x7fffffff
and r5, r5, r18
and r6, r6, r18
brid return_here
rsub r3, r6, r5
/* first part: try to find the first '1' in the r5 */
div0:
blti r5, div2
div1:
add r5, r5, r5 /* left shift logical r5 */
bgeid r5, div1
addik r29, r29, -1
div2:
/* left shift logical r5 get the '1' into the carry */
add r5, r5, r5
addc r3, r3, r3 /* move that bit into the mod register */
rsub r31, r6, r3 /* try to subtract (r3 a r6) */
blti r31, mod_too_small
/* move the r31 to mod since the result was positive */
or r3, r0, r31
addik r30, r30, 1
mod_too_small:
addik r29, r29, -1
beqi r29, loop_end
add r30, r30, r30 /* shift in the '1' into div */
bri div2 /* div2 */
loop_end:
bri return_here
div_by_zero:
result_is_zero:
or r3, r0, r0 /* set result to 0 */
return_here:
/* restore values of csrs and that of r3 and the divisor and the dividend */
lwi r29, r1, 0
lwi r30, r1, 4
lwi r31, r1, 8
rtsd r15, 8
addik r1, r1, 12
.size __umodsi3, . - __umodsi3
.end __umodsi3