Skip to content

ARM32/Thumb2: generated asm fixes #10725

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
SparkiDev wants to merge 1 commit into
base: master
Choose a base branch
from
Open

ARM32/Thumb2: generated asm fixes #10725

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
229 changes: 170 additions & 59 deletions wolfcrypt/src/port/arm/armv8-32-aes-asm.S
View file
Open in desktop

Large diffs are not rendered by default.

229 changes: 170 additions & 59 deletions wolfcrypt/src/port/arm/armv8-32-aes-asm_c.c
View file
Open in desktop

Large diffs are not rendered by default.

34 changes: 11 additions & 23 deletions wolfcrypt/src/port/arm/armv8-32-curve25519.S
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,8 +155,7 @@ fe_add_sub_op:
#endif
# Sub
sbcs r10, r4, r6
sbcs r11, r5, r7
sbc lr, lr, lr
sbc r11, r5, r7
# Add
subs r12, r12, #1
adcs r8, r4, r6
Expand Down Expand Up @@ -222,12 +221,9 @@ fe_add_sub_op:
#else
strd r8, r9, [r0, #24]
#endif
# Multiply -modulus by underflow
lsl r3, lr, #1
mvn lr, #18
orr r3, r3, r11, lsr #31
mul lr, r3, lr
# Sub -x*modulus (if overflow)
# Add -modulus on underflow
mov lr, #19
and lr, lr, r11, asr #31
ldm r1, {r4, r5, r6, r7, r8, r9}
subs r4, r4, lr
sbcs r5, r5, #0
Expand Down Expand Up @@ -263,12 +259,9 @@ fe_sub_op:
sbcs r10, r2, r10
sbcs r11, r3, r11
sbcs r12, r4, r12
sbcs lr, r5, lr
sbc r3, r3, r3
mvn r2, #18
lsl r3, r3, #1
orr r3, r3, lr, lsr #31
mul r2, r3, r2
sbc lr, r5, lr
mov r2, #19
and r2, r2, lr, asr #31
subs r6, r6, r2
sbcs r7, r7, #0
sbcs r8, r8, #0
Expand Down Expand Up @@ -312,13 +305,9 @@ fe_add_op:
adcs r10, r2, r10
adcs r11, r3, r11
adcs r12, r4, r12
mov r3, #0
adcs lr, r5, lr
adc r3, r3, #0
adc lr, r5, lr
mov r2, #19
lsl r3, r3, #1
orr r3, r3, lr, lsr #31
mul r2, r3, r2
and r2, r2, lr, asr #31
adds r6, r6, r2
adcs r7, r7, #0
adcs r8, r8, #0
Expand Down Expand Up @@ -575,6 +564,7 @@ fe_isnonzero:
fe_isnegative:
push {r4, r5, lr}
ldm r0!, {r2, r3, r4, r5}
and r12, r2, #1
adds r1, r2, #19
adcs r1, r3, #0
adcs r1, r4, #0
Expand All @@ -583,11 +573,9 @@ fe_isnegative:
adcs r1, r2, #0
adcs r1, r3, #0
adcs r1, r4, #0
ldr r2, [r0, #-16]
adc r1, r5, #0
and r0, r2, #1
lsr r1, r1, #31
eor r0, r0, r1
eor r0, r12, r1
pop {r4, r5, pc}
.size fe_isnegative,.-fe_isnegative
#if defined(HAVE_ED25519_MAKE_KEY) || defined(HAVE_ED25519_SIGN) || defined(WOLFSSL_CURVE25519_USE_ED25519)
Expand Down
36 changes: 12 additions & 24 deletions wolfcrypt/src/port/arm/armv8-32-curve25519_c.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,8 +198,7 @@ WC_OMIT_FRAME_POINTER void fe_add_sub_op()
#endif
/* Sub */
"sbcs r10, r4, r6\n\t"
"sbcs r11, r5, r7\n\t"
"sbc lr, lr, lr\n\t"
"sbc r11, r5, r7\n\t"
/* Add */
"subs r12, r12, #1\n\t"
"adcs r8, r4, r6\n\t"
Expand Down Expand Up @@ -265,12 +264,9 @@ WC_OMIT_FRAME_POINTER void fe_add_sub_op()
#else
"strd r8, r9, [r0, #24]\n\t"
#endif
/* Multiply -modulus by underflow */
"lsl r3, lr, #1\n\t"
"mvn lr, #18\n\t"
"orr r3, r3, r11, lsr #31\n\t"
"mul lr, r3, lr\n\t"
/* Sub -x*modulus (if overflow) */
/* Add -modulus on underflow */
"mov lr, #19\n\t"
"and lr, lr, r11, asr #31\n\t"
"ldm r1, {r4, r5, r6, r7, r8, r9}\n\t"
"subs r4, r4, lr\n\t"
"sbcs r5, r5, #0\n\t"
Expand Down Expand Up @@ -319,12 +315,9 @@ WC_OMIT_FRAME_POINTER void fe_sub_op()
"sbcs r10, r2, r10\n\t"
"sbcs r11, r3, r11\n\t"
"sbcs r12, r4, r12\n\t"
"sbcs lr, r5, lr\n\t"
"sbc r3, r3, r3\n\t"
"mvn r2, #18\n\t"
"lsl r3, r3, #1\n\t"
"orr r3, r3, lr, lsr #31\n\t"
"mul r2, r3, r2\n\t"
"sbc lr, r5, lr\n\t"
"mov r2, #19\n\t"
"and r2, r2, lr, asr #31\n\t"
"subs r6, r6, r2\n\t"
"sbcs r7, r7, #0\n\t"
"sbcs r8, r8, #0\n\t"
Expand Down Expand Up @@ -398,13 +391,9 @@ WC_OMIT_FRAME_POINTER void fe_add_op()
"adcs r10, r2, r10\n\t"
"adcs r11, r3, r11\n\t"
"adcs r12, r4, r12\n\t"
"mov r3, #0\n\t"
"adcs lr, r5, lr\n\t"
"adc r3, r3, #0\n\t"
"adc lr, r5, lr\n\t"
"mov r2, #19\n\t"
"lsl r3, r3, #1\n\t"
"orr r3, r3, lr, lsr #31\n\t"
"mul r2, r3, r2\n\t"
"and r2, r2, lr, asr #31\n\t"
"adds r6, r6, r2\n\t"
"adcs r7, r7, #0\n\t"
"adcs r8, r8, #0\n\t"
Expand Down Expand Up @@ -797,6 +786,7 @@ WC_OMIT_FRAME_POINTER int fe_isnegative(const fe a)

__asm__ __volatile__ (
"ldm %[a]!, {r2, r3, r4, r5}\n\t"
"and r12, r2, #1\n\t"
"adds r1, r2, #19\n\t"
"adcs r1, r3, #0\n\t"
"adcs r1, r4, #0\n\t"
Expand All @@ -805,19 +795,17 @@ WC_OMIT_FRAME_POINTER int fe_isnegative(const fe a)
"adcs r1, r2, #0\n\t"
"adcs r1, r3, #0\n\t"
"adcs r1, r4, #0\n\t"
"ldr r2, [%[a], #-16]\n\t"
"adc r1, r5, #0\n\t"
"and %[a], r2, #1\n\t"
"lsr r1, r1, #31\n\t"
"eor %[a], %[a], r1\n\t"
"eor %[a], r12, r1\n\t"
#ifndef WOLFSSL_NO_VAR_ASSIGN_REG
: [a] "+r" (a)
:
#else
:
: [a] "r" (a)
#endif /* !WOLFSSL_NO_VAR_ASSIGN_REG */
: "memory", "cc", "r1", "r2", "r3", "r4", "r5"
: "memory", "cc", "r1", "r2", "r3", "r4", "r5", "r12"
);
return (word32)(size_t)a;
}
Expand Down
Loading
Loading