feedc0de/wolfssl
1
0
Fork 0
forked from wolfSSL/wolfssl
Code Pull Requests Activity

Merge pull request #2362 from SparkiDev/chacha20_asm_fix

Browse Source 
Fix chacha20 AVX1 assembly to not use register RBX
This commit is contained in:
toddouska
2019-07-18 08:43:44 -07:00
committed by GitHub
parent b73c53471c 7fe4521083
commit 45abd67bd1
1 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
wolfcrypt/src/chacha_asm.S
View File

@@ -975,24 +975,24 @@ L_chacha20_avx1_partial_crypt_start:
vmovdqu %xmm3, 48(%r10) vmovdqu %xmm3, 48(%r10)
addl $0x01, 48(%rdi) addl $0x01, 48(%rdi)
movl %ecx, %r8d movl %ecx, %r8d
xorq %rbx, %rbx xorq %r11, %r11
andl $7, %r8d andl $7, %r8d
jz L_chacha20_avx1_partial_start64 jz L_chacha20_avx1_partial_start64
L_chacha20_avx1_partial_start8: L_chacha20_avx1_partial_start8:
movzbl (%r10,%rbx,1), %eax movzbl (%r10,%r11,1), %eax
xorb (%rsi,%rbx,1), %al xorb (%rsi,%r11,1), %al
movb %al, (%rdx,%rbx,1) movb %al, (%rdx,%r11,1)
incl %ebx incl %r11d
cmpl %r8d, %ebx cmpl %r8d, %r11d
jne L_chacha20_avx1_partial_start8 jne L_chacha20_avx1_partial_start8
je L_chacha20_avx1_partial_end64 je L_chacha20_avx1_partial_end64
L_chacha20_avx1_partial_start64: L_chacha20_avx1_partial_start64:
movq (%r10,%rbx,1), %rax movq (%r10,%r11,1), %rax
xorq (%rsi,%rbx,1), %rax xorq (%rsi,%r11,1), %rax
movq %rax, (%rdx,%rbx,1) movq %rax, (%rdx,%r11,1)
addl $8, %ebx addl $8, %r11d
L_chacha20_avx1_partial_end64: L_chacha20_avx1_partial_end64:
cmpl %ecx, %ebx cmpl %ecx, %r11d
jne L_chacha20_avx1_partial_start64 jne L_chacha20_avx1_partial_start64
L_chacha20_avx1_partial_done: L_chacha20_avx1_partial_done:
addq $0x190, %rsp addq $0x190, %rsp