356 lines
4.9 KiB
ArmAsm
356 lines
4.9 KiB
ArmAsm
.file "floyd-2.sy"
|
|
.option pic
|
|
.attribute arch, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0"
|
|
.attribute unaligned_access, 0
|
|
.attribute stack_align, 16
|
|
.text
|
|
.align 1
|
|
.globl getvalue
|
|
.type getvalue, @function
|
|
getvalue:
|
|
or a5,a2,a3
|
|
blt a5,zero,.L5
|
|
bge a2,a1,.L5
|
|
ble a1,a3,.L5
|
|
mulw a2,a2,a1
|
|
addw a2,a2,a3
|
|
slli a2,a2,2
|
|
add a0,a0,a2
|
|
lw a0,0(a0)
|
|
ret
|
|
.L5:
|
|
li a0,-1
|
|
ret
|
|
.size getvalue, .-getvalue
|
|
.align 1
|
|
.globl mem_move
|
|
.type mem_move, @function
|
|
mem_move:
|
|
ble a2,zero,.L7
|
|
addiw a5,a2,-1
|
|
li a4,8
|
|
sext.w a7,a2
|
|
bleu a5,a4,.L9
|
|
or a5,a0,a1
|
|
andi a5,a5,7
|
|
bne a5,zero,.L9
|
|
addi a5,a0,4
|
|
beq a1,a5,.L9
|
|
srliw a6,a7,1
|
|
slli a6,a6,3
|
|
mv a5,a0
|
|
mv a4,a1
|
|
add a6,a6,a0
|
|
.L10:
|
|
ld a3,0(a5)
|
|
addi a5,a5,8
|
|
addi a4,a4,8
|
|
sd a3,-8(a4)
|
|
bne a5,a6,.L10
|
|
andi a5,a2,-2
|
|
beq a7,a5,.L7
|
|
slli a5,a5,32
|
|
srli a5,a5,30
|
|
add a0,a0,a5
|
|
lw a4,0(a0)
|
|
add a1,a1,a5
|
|
sw a4,0(a1)
|
|
ret
|
|
.L9:
|
|
slli a5,a2,2
|
|
add a5,a0,a5
|
|
.L12:
|
|
lw a4,0(a0)
|
|
addi a0,a0,4
|
|
addi a1,a1,4
|
|
sw a4,-4(a1)
|
|
bne a0,a5,.L12
|
|
.L7:
|
|
ret
|
|
.size mem_move, .-mem_move
|
|
.align 1
|
|
.globl floyd
|
|
.type floyd, @function
|
|
floyd:
|
|
mv t2,a1
|
|
ble a2,zero,.L69
|
|
addi sp,sp,-32
|
|
sd s0,24(sp)
|
|
lla s0,temp
|
|
sd s1,16(sp)
|
|
sd s2,8(sp)
|
|
sd s3,0(sp)
|
|
slli t4,a2,2
|
|
mv a1,s0
|
|
mv a6,s0
|
|
li t3,0
|
|
.L26:
|
|
mv a4,a6
|
|
mv a5,a0
|
|
li a7,0
|
|
.L25:
|
|
lw a3,0(a5)
|
|
addiw t0,a7,1
|
|
mv t6,a7
|
|
sw a3,0(a4)
|
|
sext.w a7,t0
|
|
add a5,a5,t4
|
|
add a4,a4,t4
|
|
bne a2,a7,.L25
|
|
addiw a5,t3,1
|
|
addi a0,a0,4
|
|
addi a6,a6,4
|
|
beq t6,t3,.L70
|
|
mv t3,a5
|
|
j .L26
|
|
.L69:
|
|
mulw a2,a2,a2
|
|
lla a1,temp
|
|
beq a2,zero,.L58
|
|
addiw a5,a2,-1
|
|
li a4,7
|
|
sext.w a6,a2
|
|
bleu a5,a4,.L60
|
|
andi a5,t2,7
|
|
bne a5,zero,.L60
|
|
lla a5,temp+4
|
|
beq t2,a5,.L60
|
|
srliw a3,a6,1
|
|
slli a3,a3,3
|
|
mv a5,t2
|
|
lla a4,temp
|
|
add a3,a3,t2
|
|
.L63:
|
|
ld a0,0(a4)
|
|
addi a5,a5,8
|
|
addi a4,a4,8
|
|
sd a0,-8(a5)
|
|
bne a5,a3,.L63
|
|
andi a5,a2,-2
|
|
beq a5,a6,.L58
|
|
slli a5,a5,32
|
|
srli a5,a5,30
|
|
add a1,a5,a1
|
|
lw a4,0(a1)
|
|
add a5,t2,a5
|
|
sw a4,0(a5)
|
|
ret
|
|
.L70:
|
|
sext.w t3,t3
|
|
li s2,0
|
|
li s3,0
|
|
.L27:
|
|
subw s1,t3,t6
|
|
mv a6,s2
|
|
mv a0,s0
|
|
li t5,0
|
|
.L35:
|
|
mv a2,s1
|
|
lw a3,0(a0)
|
|
sext.w a4,a2
|
|
slli a4,a4,2
|
|
add a4,a4,a1
|
|
blt a3,zero,.L28
|
|
.L71:
|
|
lw a4,0(a4)
|
|
addw a5,a2,a6
|
|
slli a5,a5,2
|
|
add a5,a5,a1
|
|
addw a3,a3,a4
|
|
blt a4,zero,.L29
|
|
lw a4,0(a5)
|
|
sext.w t1,a3
|
|
blt a4,zero,.L68
|
|
bgt a4,t1,.L68
|
|
.L29:
|
|
beq a2,t3,.L28
|
|
addiw a2,a2,1
|
|
lw a3,0(a0)
|
|
sext.w a4,a2
|
|
slli a4,a4,2
|
|
add a4,a4,a1
|
|
bge a3,zero,.L71
|
|
.L28:
|
|
addiw a5,t5,1
|
|
add a0,a0,t4
|
|
addw a6,a7,a6
|
|
beq t5,t6,.L32
|
|
mv t5,a5
|
|
j .L35
|
|
.L68:
|
|
sw a3,0(a5)
|
|
j .L29
|
|
.L32:
|
|
addiw a5,s3,1
|
|
addw t3,a7,t3
|
|
addi s0,s0,4
|
|
subw s2,s2,a7
|
|
beq s3,t6,.L34
|
|
mv s3,a5
|
|
j .L27
|
|
.L34:
|
|
mulw a2,t0,t0
|
|
li a4,7
|
|
addiw a5,a2,-1
|
|
sext.w a6,a2
|
|
bleu a5,a4,.L37
|
|
andi a5,t2,7
|
|
bne a5,zero,.L37
|
|
lla a5,temp+4
|
|
beq t2,a5,.L37
|
|
srliw a3,a6,1
|
|
slli a3,a3,3
|
|
mv a5,t2
|
|
lla a4,temp
|
|
add a3,a3,t2
|
|
.L38:
|
|
ld a0,0(a4)
|
|
addi a5,a5,8
|
|
addi a4,a4,8
|
|
sd a0,-8(a5)
|
|
bne a5,a3,.L38
|
|
andi a5,a2,-2
|
|
beq a5,a6,.L22
|
|
slli a5,a5,32
|
|
srli a5,a5,30
|
|
add a1,a5,a1
|
|
lw a4,0(a1)
|
|
add a5,t2,a5
|
|
sw a4,0(a5)
|
|
.L22:
|
|
ld s0,24(sp)
|
|
ld s1,16(sp)
|
|
ld s2,8(sp)
|
|
ld s3,0(sp)
|
|
addi sp,sp,32
|
|
jr ra
|
|
.L37:
|
|
slli a2,a2,2
|
|
lla a4,temp
|
|
add a5,t2,a2
|
|
.L42:
|
|
lw a3,0(a4)
|
|
addi t2,t2,4
|
|
addi a4,a4,4
|
|
sw a3,-4(t2)
|
|
bne a5,t2,.L42
|
|
ld s0,24(sp)
|
|
ld s1,16(sp)
|
|
ld s2,8(sp)
|
|
ld s3,0(sp)
|
|
addi sp,sp,32
|
|
jr ra
|
|
.L60:
|
|
slli a2,a2,2
|
|
lla a4,temp
|
|
add a5,t2,a2
|
|
.L66:
|
|
lw a3,0(a4)
|
|
addi t2,t2,4
|
|
addi a4,a4,4
|
|
sw a3,-4(t2)
|
|
bne a5,t2,.L66
|
|
ret
|
|
.L58:
|
|
ret
|
|
.size floyd, .-floyd
|
|
.section .text.startup,"ax",@progbits
|
|
.align 1
|
|
.globl main
|
|
.type main, @function
|
|
main:
|
|
addi sp,sp,-16
|
|
sd ra,8(sp)
|
|
sd s0,0(sp)
|
|
call getint@plt
|
|
mv s0,a0
|
|
lla a0,w
|
|
call getarray@plt
|
|
li a0,64
|
|
call _sysy_starttime@plt
|
|
mv a2,s0
|
|
lla a1,dst
|
|
lla a0,w
|
|
call floyd
|
|
li a0,66
|
|
call _sysy_stoptime@plt
|
|
mulw a0,s0,s0
|
|
lla a1,dst
|
|
call putarray@plt
|
|
ld ra,8(sp)
|
|
ld s0,0(sp)
|
|
li a0,0
|
|
addi sp,sp,16
|
|
jr ra
|
|
.size main, .-main
|
|
.globl c
|
|
.globl dst
|
|
.globl w
|
|
.globl temp
|
|
.globl _sysy_idx
|
|
.globl _sysy_us
|
|
.globl _sysy_s
|
|
.globl _sysy_m
|
|
.globl _sysy_h
|
|
.globl _sysy_l2
|
|
.globl _sysy_l1
|
|
.globl _sysy_end
|
|
.globl _sysy_start
|
|
.bss
|
|
.align 3
|
|
.type c, @object
|
|
.size c, 8388608
|
|
c:
|
|
.zero 8388608
|
|
.type dst, @object
|
|
.size dst, 8388608
|
|
dst:
|
|
.zero 8388608
|
|
.type w, @object
|
|
.size w, 8388608
|
|
w:
|
|
.zero 8388608
|
|
.type temp, @object
|
|
.size temp, 8388608
|
|
temp:
|
|
.zero 8388608
|
|
.type _sysy_idx, @object
|
|
.size _sysy_idx, 4
|
|
_sysy_idx:
|
|
.zero 4
|
|
.zero 4
|
|
.type _sysy_us, @object
|
|
.size _sysy_us, 4096
|
|
_sysy_us:
|
|
.zero 4096
|
|
.type _sysy_s, @object
|
|
.size _sysy_s, 4096
|
|
_sysy_s:
|
|
.zero 4096
|
|
.type _sysy_m, @object
|
|
.size _sysy_m, 4096
|
|
_sysy_m:
|
|
.zero 4096
|
|
.type _sysy_h, @object
|
|
.size _sysy_h, 4096
|
|
_sysy_h:
|
|
.zero 4096
|
|
.type _sysy_l2, @object
|
|
.size _sysy_l2, 4096
|
|
_sysy_l2:
|
|
.zero 4096
|
|
.type _sysy_l1, @object
|
|
.size _sysy_l1, 4096
|
|
_sysy_l1:
|
|
.zero 4096
|
|
.type _sysy_end, @object
|
|
.size _sysy_end, 16
|
|
_sysy_end:
|
|
.zero 16
|
|
.type _sysy_start, @object
|
|
.size _sysy_start, 16
|
|
_sysy_start:
|
|
.zero 16
|
|
.ident "GCC: (Debian 12.2.0-13) 12.2.0"
|
|
.section .note.GNU-stack,"",@progbits
|