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

Wycheproof fixes

Browse Source 
Curve25519: reduce to less than mod at end of curve25519, optimize to
not use imulq when andq can be used instead.
x86_64 Karatsuba Mul: Don't put add carry into overflow and then
continue adding zero. Overflow is lost.
This commit is contained in:
Sean Parkinson
2022-04-11 15:09:40 +10:00
parent cd1344bbed
commit 2995c96f11
3 changed files with 264 additions and 249 deletions
Download Patch File Download Diff File Expand all files Collapse all files

497
wolfcrypt/src/fe_x25519_asm.S
View File

File diff suppressed because it is too large Load Diff

8
wolfcrypt/src/sp_x86_64_asm.S
View File

@ -4677,7 +4677,6 @@ _sp_2048_mul_32:
addq $0x80, %r15
# Add
movq (%r15), %rax
xorq %r9, %r9
addq (%r11), %rax
movq 8(%r15), %rcx
movq %rax, (%r15)
@ -4728,7 +4727,6 @@ _sp_2048_mul_32:
movq %rax, 120(%r15)
adcq 128(%r11), %rcx
movq %rcx, 128(%r15)
adcq $0x00, %r9
# Add to zero
movq 136(%r11), %rax
adcq $0x00, %rax
@ -5356,7 +5354,6 @@ _sp_2048_mul_avx2_32:
addq $0x80, %r15
# Add
movq (%r15), %rax
xorq %r9, %r9
addq (%r11), %rax
movq 8(%r15), %rcx
movq %rax, (%r15)
@ -5407,7 +5404,6 @@ _sp_2048_mul_avx2_32:
movq %rax, 120(%r15)
adcq 128(%r11), %rcx
movq %rcx, 128(%r15)
adcq $0x00, %r9
# Add to zero
movq 136(%r11), %rax
adcq $0x00, %rax
@ -43508,7 +43504,6 @@ _sp_4096_mul_64:
addq $0x100, %r15
# Add
movq (%r15), %rax
xorq %r9, %r9
addq (%r11), %rax
movq 8(%r15), %rcx
movq %rax, (%r15)
@ -43607,7 +43602,6 @@ _sp_4096_mul_64:
movq %rcx, 248(%r15)
adcq 256(%r11), %r8
movq %r8, 256(%r15)
adcq $0x00, %r9
# Add to zero
movq 264(%r11), %rax
adcq $0x00, %rax
@ -44763,7 +44757,6 @@ _sp_4096_mul_avx2_64:
addq $0x100, %r15
# Add
movq (%r15), %rax
xorq %r9, %r9
addq (%r11), %rax
movq 8(%r15), %rcx
movq %rax, (%r15)
@ -44862,7 +44855,6 @@ _sp_4096_mul_avx2_64:
movq %rcx, 248(%r15)
adcq 256(%r11), %r8
movq %r8, 256(%r15)
adcq $0x00, %r9
# Add to zero
movq 264(%r11), %rax
adcq $0x00, %rax

8
wolfcrypt/src/sp_x86_64_asm.asm
View File

@ -4585,7 +4585,6 @@ ENDIF
add rsi, 128
; Add
mov rax, QWORD PTR [rsi]
xor r11, r11
add rax, QWORD PTR [r13]
mov r9, QWORD PTR [rsi+8]
mov QWORD PTR [rsi], rax
@ -4636,7 +4635,6 @@ ENDIF
mov QWORD PTR [rsi+120], rax
adc r9, QWORD PTR [r13+128]
mov QWORD PTR [rsi+128], r9
adc r11, 0
; Add to zero
mov rax, QWORD PTR [r13+136]
adc rax, 0
@ -5245,7 +5243,6 @@ ENDIF
add rsi, 128
; Add
mov rax, QWORD PTR [rsi]
xor r11, r11
add rax, QWORD PTR [r13]
mov r9, QWORD PTR [rsi+8]
mov QWORD PTR [rsi], rax
@ -5296,7 +5293,6 @@ ENDIF
mov QWORD PTR [rsi+120], rax
adc r9, QWORD PTR [r13+128]
mov QWORD PTR [rsi+128], r9
adc r11, 0
; Add to zero
mov rax, QWORD PTR [r13+136]
adc rax, 0
@ -42510,7 +42506,6 @@ ENDIF
add rsi, 256
; Add
mov rax, QWORD PTR [rsi]
xor r11, r11
add rax, QWORD PTR [r13]
mov r9, QWORD PTR [rsi+8]
mov QWORD PTR [rsi], rax
@ -42609,7 +42604,6 @@ ENDIF
mov QWORD PTR [rsi+248], r9
adc r10, QWORD PTR [r13+256]
mov QWORD PTR [rsi+256], r10
adc r11, 0
; Add to zero
mov rax, QWORD PTR [r13+264]
adc rax, 0
@ -43746,7 +43740,6 @@ ENDIF
add rsi, 256
; Add
mov rax, QWORD PTR [rsi]
xor r11, r11
add rax, QWORD PTR [r13]
mov r9, QWORD PTR [rsi+8]
mov QWORD PTR [rsi], rax
@ -43845,7 +43838,6 @@ ENDIF
mov QWORD PTR [rsi+248], r9
adc r10, QWORD PTR [r13+256]
mov QWORD PTR [rsi+256], r10
adc r11, 0
; Add to zero
mov rax, QWORD PTR [r13+264]
adc rax, 0