mirror of
https://github.com/zeldaret/oot.git
synced 2024-11-11 03:39:59 +00:00
7334ffa373
* Format all handwritten asm and document some * Use c preprocessor for constants * Fix * Fix PI_STATUS_ERROR, some label improvements * Avoid hi/lo for constants * Some more comments * Properly mark functions as functions and their sizes * Fix merge * Improvements * Review suggestions, rework procedure start/end macros to be more like libreultra * Move IPL3 symbol definitions into ipl3.s * Fix undefined_syms, add include and language guards to asm.h and fix the comment in gbi.h * Consistent hex capitalization, add some MIPS builtin defines to CC_CHECK to behave properly * Add -no-pad-sections assembler option and clean up alignment in gu files and bzero * Further suggestions and improvements * Matrix conversion function clarifications * Fix passing AVOID_UB to gcc * Suggestions * Suggestions, global interrupt mask improvements * Further suggestions, interrupt mask comments * Comments fixes, rdb.h * Switch from # comments to // comments, remove unnecesary .set gp=64 directives * Further review suggestions * Missed one
93 lines
1.8 KiB
ArmAsm
93 lines
1.8 KiB
ArmAsm
#include "ultra64/asm.h"
|
|
|
|
.set noat
|
|
.set noreorder
|
|
|
|
.section .text
|
|
|
|
.balign 16
|
|
|
|
LEAF(bcmp)
|
|
slti $at, $a2, 0x10
|
|
bnez $at, bytecmp
|
|
xor $v0, $a0, $a1
|
|
andi $v0, $v0, 3
|
|
bnez $v0, unaligncmp
|
|
negu $t8, $a0
|
|
andi $t8, $t8, 3
|
|
beqz $t8, wordcmp
|
|
subu $a2, $a2, $t8
|
|
move $v0, $v1
|
|
lwl $v0, ($a0)
|
|
lwl $v1, ($a1)
|
|
addu $a0, $a0, $t8
|
|
addu $a1, $a1, $t8
|
|
bne $v0, $v1, cmpne
|
|
wordcmp:
|
|
li $at, ~3
|
|
and $a3, $a2, $at
|
|
beqz $a3, bytecmp
|
|
subu $a2, $a2, $a3
|
|
addu $a3, $a3, $a0
|
|
lw $v0, ($a0)
|
|
1:
|
|
lw $v1, ($a1)
|
|
addiu $a0, $a0, 4
|
|
addiu $a1, $a1, 4
|
|
bne $v0, $v1, cmpne
|
|
nop
|
|
bnel $a0, $a3, 1b
|
|
lw $v0, ($a0)
|
|
b bytecmp
|
|
nop
|
|
unaligncmp:
|
|
negu $a3, $a1
|
|
andi $a3, $a3, 3
|
|
beqz $a3, partaligncmp
|
|
subu $a2, $a2, $a3
|
|
addu $a3, $a3, $a0
|
|
lbu $v0, ($a0)
|
|
1:
|
|
lbu $v1, ($a1)
|
|
addiu $a0, $a0, 1
|
|
addiu $a1, $a1, 1
|
|
bne $v0, $v1, cmpne
|
|
nop
|
|
bnel $a0, $a3, 1b
|
|
lbu $v0, ($a0)
|
|
partaligncmp:
|
|
li $at, ~3
|
|
and $a3, $a2, $at
|
|
beqz $a3, bytecmp
|
|
subu $a2, $a2, $a3
|
|
addu $a3, $a3, $a0
|
|
lwl $v0, ($a0)
|
|
1:
|
|
lw $v1, ($a1)
|
|
lwr $v0, 3($a0)
|
|
addiu $a0, $a0, 4
|
|
addiu $a1, $a1, 4
|
|
bne $v0, $v1, cmpne
|
|
nop
|
|
bnel $a0, $a3, 1b
|
|
lwl $v0, ($a0)
|
|
bytecmp:
|
|
blez $a2, cmpdone
|
|
addu $a3, $a2, $a0
|
|
lbu $v0, ($a0)
|
|
1:
|
|
lbu $v1, ($a1)
|
|
addiu $a0, $a0, 1
|
|
addiu $a1, $a1, 1
|
|
bne $v0, $v1, cmpne
|
|
nop
|
|
bnel $a0, $a3, 1b
|
|
lbu $v0, ($a0)
|
|
cmpdone:
|
|
jr $ra
|
|
move $v0, $zero
|
|
|
|
cmpne:
|
|
jr $ra
|
|
li $v0, 1
|
|
END(bcmp)
|