mirror of
https://github.com/zerotier/ZeroTierOne.git
synced 2024-12-22 06:17:48 +00:00
143 lines
2.8 KiB
ArmAsm
143 lines
2.8 KiB
ArmAsm
|
|
||
|
# qhasm: int64 rp
|
||
|
|
||
|
# qhasm: int64 xp
|
||
|
|
||
|
# qhasm: int64 yp
|
||
|
|
||
|
# qhasm: input rp
|
||
|
|
||
|
# qhasm: input xp
|
||
|
|
||
|
# qhasm: input yp
|
||
|
|
||
|
# qhasm: int64 r0
|
||
|
|
||
|
# qhasm: int64 r1
|
||
|
|
||
|
# qhasm: int64 r2
|
||
|
|
||
|
# qhasm: int64 r3
|
||
|
|
||
|
# qhasm: int64 t0
|
||
|
|
||
|
# qhasm: int64 t1
|
||
|
|
||
|
# qhasm: int64 t2
|
||
|
|
||
|
# qhasm: int64 t3
|
||
|
|
||
|
# qhasm: int64 caller1
|
||
|
|
||
|
# qhasm: int64 caller2
|
||
|
|
||
|
# qhasm: int64 caller3
|
||
|
|
||
|
# qhasm: int64 caller4
|
||
|
|
||
|
# qhasm: int64 caller5
|
||
|
|
||
|
# qhasm: int64 caller6
|
||
|
|
||
|
# qhasm: int64 caller7
|
||
|
|
||
|
# qhasm: caller caller1
|
||
|
|
||
|
# qhasm: caller caller2
|
||
|
|
||
|
# qhasm: caller caller3
|
||
|
|
||
|
# qhasm: caller caller4
|
||
|
|
||
|
# qhasm: caller caller5
|
||
|
|
||
|
# qhasm: caller caller6
|
||
|
|
||
|
# qhasm: caller caller7
|
||
|
|
||
|
# qhasm: stack64 caller4_stack
|
||
|
|
||
|
# qhasm: stack64 caller5_stack
|
||
|
|
||
|
# qhasm: stack64 caller6_stack
|
||
|
|
||
|
# qhasm: stack64 caller7_stack
|
||
|
|
||
|
# qhasm: enter crypto_sign_ed25519_amd64_64_sc25519_sub_nored
|
||
|
.text
|
||
|
.p2align 5
|
||
|
.globl _crypto_sign_ed25519_amd64_64_sc25519_sub_nored
|
||
|
.globl crypto_sign_ed25519_amd64_64_sc25519_sub_nored
|
||
|
_crypto_sign_ed25519_amd64_64_sc25519_sub_nored:
|
||
|
crypto_sign_ed25519_amd64_64_sc25519_sub_nored:
|
||
|
mov %rsp,%r11
|
||
|
and $31,%r11
|
||
|
add $0,%r11
|
||
|
sub %r11,%rsp
|
||
|
|
||
|
# qhasm: r0 = *(uint64 *)(xp + 0)
|
||
|
# asm 1: movq 0(<xp=int64#2),>r0=int64#4
|
||
|
# asm 2: movq 0(<xp=%rsi),>r0=%rcx
|
||
|
movq 0(%rsi),%rcx
|
||
|
|
||
|
# qhasm: r1 = *(uint64 *)(xp + 8)
|
||
|
# asm 1: movq 8(<xp=int64#2),>r1=int64#5
|
||
|
# asm 2: movq 8(<xp=%rsi),>r1=%r8
|
||
|
movq 8(%rsi),%r8
|
||
|
|
||
|
# qhasm: r2 = *(uint64 *)(xp + 16)
|
||
|
# asm 1: movq 16(<xp=int64#2),>r2=int64#6
|
||
|
# asm 2: movq 16(<xp=%rsi),>r2=%r9
|
||
|
movq 16(%rsi),%r9
|
||
|
|
||
|
# qhasm: r3 = *(uint64 *)(xp + 24)
|
||
|
# asm 1: movq 24(<xp=int64#2),>r3=int64#2
|
||
|
# asm 2: movq 24(<xp=%rsi),>r3=%rsi
|
||
|
movq 24(%rsi),%rsi
|
||
|
|
||
|
# qhasm: carry? r0 -= *(uint64 *)(yp + 0)
|
||
|
# asm 1: subq 0(<yp=int64#3),<r0=int64#4
|
||
|
# asm 2: subq 0(<yp=%rdx),<r0=%rcx
|
||
|
subq 0(%rdx),%rcx
|
||
|
|
||
|
# qhasm: carry? r1 -= *(uint64 *)(yp + 8) - carry
|
||
|
# asm 1: sbbq 8(<yp=int64#3),<r1=int64#5
|
||
|
# asm 2: sbbq 8(<yp=%rdx),<r1=%r8
|
||
|
sbbq 8(%rdx),%r8
|
||
|
|
||
|
# qhasm: carry? r2 -= *(uint64 *)(yp + 16) - carry
|
||
|
# asm 1: sbbq 16(<yp=int64#3),<r2=int64#6
|
||
|
# asm 2: sbbq 16(<yp=%rdx),<r2=%r9
|
||
|
sbbq 16(%rdx),%r9
|
||
|
|
||
|
# qhasm: r3 -= *(uint64 *)(yp + 24) - carry
|
||
|
# asm 1: sbbq 24(<yp=int64#3),<r3=int64#2
|
||
|
# asm 2: sbbq 24(<yp=%rdx),<r3=%rsi
|
||
|
sbbq 24(%rdx),%rsi
|
||
|
|
||
|
# qhasm: *(uint64 *)(rp + 0) = r0
|
||
|
# asm 1: movq <r0=int64#4,0(<rp=int64#1)
|
||
|
# asm 2: movq <r0=%rcx,0(<rp=%rdi)
|
||
|
movq %rcx,0(%rdi)
|
||
|
|
||
|
# qhasm: *(uint64 *)(rp + 8) = r1
|
||
|
# asm 1: movq <r1=int64#5,8(<rp=int64#1)
|
||
|
# asm 2: movq <r1=%r8,8(<rp=%rdi)
|
||
|
movq %r8,8(%rdi)
|
||
|
|
||
|
# qhasm: *(uint64 *)(rp + 16) = r2
|
||
|
# asm 1: movq <r2=int64#6,16(<rp=int64#1)
|
||
|
# asm 2: movq <r2=%r9,16(<rp=%rdi)
|
||
|
movq %r9,16(%rdi)
|
||
|
|
||
|
# qhasm: *(uint64 *)(rp + 24) = r3
|
||
|
# asm 1: movq <r3=int64#2,24(<rp=int64#1)
|
||
|
# asm 2: movq <r3=%rsi,24(<rp=%rdi)
|
||
|
movq %rsi,24(%rdi)
|
||
|
|
||
|
# qhasm: leave
|
||
|
add %r11,%rsp
|
||
|
mov %rdi,%rax
|
||
|
mov %rsi,%rdx
|
||
|
ret
|