diff mbox

Fwd: [PING 2][PATCH] libgcc: Add CFI directives to the soft floating point support code for ARM

Message ID CAOKbPbZpE6RZ91+j5O_fV6VqNNkweRysEu32zcxAY42urA8eYA@mail.gmail.com
State New
Headers show

Commit Message

Martin Galvan May 12, 2015, 1:01 p.m. UTC
On Tue, May 12, 2015 at 5:49 AM, Ramana Radhakrishnan
<ramana.radhakrishnan@arm.com> wrote:
> That's what I mean when I say email clients "munged it" : email clients and
> / or some popular email servers appear to end up munging white spaces and
> patches don't apply cleanly.
>
> So, no it doesn't work - once you've sent it through your email client /
> server. I am unable to apply the patch as it stands today either taking the
> raw text from the gcc-patches archive or from your email message in my
> inbox. It's not like line endings and things have been munged but every
> whitespace / tab is in trouble here.
>
> Please send it back as an attachment if you want me to apply it.

Oh, I see! Sorry for that, I thought the problem was on my side.
Here's the patch, it's the cfi.patch file.

Comments

Ramana Radhakrishnan May 13, 2015, 4:37 p.m. UTC | #1
On 12/05/15 14:01, Martin Galvan wrote:
> On Tue, May 12, 2015 at 5:49 AM, Ramana Radhakrishnan
> <ramana.radhakrishnan@arm.com> wrote:
>> That's what I mean when I say email clients "munged it" : email clients and
>> / or some popular email servers appear to end up munging white spaces and
>> patches don't apply cleanly.
>>
>> So, no it doesn't work - once you've sent it through your email client /
>> server. I am unable to apply the patch as it stands today either taking the
>> raw text from the gcc-patches archive or from your email message in my
>> inbox. It's not like line endings and things have been munged but every
>> whitespace / tab is in trouble here.
>>
>> Please send it back as an attachment if you want me to apply it.
>
> Oh, I see! Sorry for that, I thought the problem was on my side.
> Here's the patch, it's the cfi.patch file.
>


I'm not sure what's going on here and couldn't figure out what was going 
wrong in the 20 minutes I spent on it just now.

Applying the patch downloaded from 2 different inboxes still gives me 
the same result.

$> patch -p1 --dry-run < /work/cfi.patch
checking file libgcc/config/arm/ieee754-df.S
Hunk #2 FAILED at 57.
Hunk #3 succeeded at 70 with fuzz 2.
Hunk #4 FAILED at 86.
Hunk #5 FAILED at 153.
Hunk #6 FAILED at 418.
Hunk #7 FAILED at 425.
Hunk #8 FAILED at 440.
Hunk #9 FAILED at 462.
Hunk #10 FAILED at 485.
Hunk #11 FAILED at 555.
Hunk #12 FAILED at 566.
Hunk #13 FAILED at 601.
Hunk #14 FAILED at 653.
Hunk #15 FAILED at 720.
Hunk #16 FAILED at 868.
Hunk #17 FAILED at 1057.
Hunk #18 FAILED at 1068.
Hunk #19 FAILED at 1082.
Hunk #20 FAILED at 1090.
Hunk #21 FAILED at 1122.
Hunk #22 FAILED at 1133.
Hunk #23 succeeded at 1145 with fuzz 2.
Hunk #24 FAILED at 1155.
Hunk #25 FAILED at 1168.
Hunk #26 FAILED at 1228.
Hunk #27 succeeded at 1236 with fuzz 2.
Hunk #28 FAILED at 1254.
Hunk #29 succeeded at 1263 with fuzz 2.
Hunk #30 FAILED at 1297.
Hunk #31 succeeded at 1306 with fuzz 2.
Hunk #32 FAILED at 1336.
Hunk #33 succeeded at 1345 with fuzz 2.
Hunk #34 FAILED at 1410.
27 out of 34 hunks FAILED
checking file libgcc/config/arm/ieee754-sf.S
Hunk #1 FAILED at 31.
Hunk #2 succeeded at 49 with fuzz 2.
Hunk #3 FAILED at 285.
Hunk #4 FAILED at 293.
Hunk #5 FAILED at 317.
Hunk #6 succeeded at 324 with fuzz 2.
Hunk #7 FAILED at 411.
Hunk #8 succeeded at 422 with fuzz 2.
Hunk #9 FAILED at 457.
Hunk #10 FAILED at 468.
Hunk #11 FAILED at 621.
Hunk #12 FAILED at 761.
Hunk #13 FAILED at 785.
Hunk #14 FAILED at 799.
Hunk #15 FAILED at 807.
Hunk #16 FAILED at 826.
Hunk #17 FAILED at 835.
Hunk #18 succeeded at 847 with fuzz 2.
Hunk #19 FAILED at 860.
Hunk #20 FAILED at 869.
Hunk #21 FAILED at 927.
Hunk #22 succeeded at 935 with fuzz 2.
Hunk #23 FAILED at 952.
Hunk #24 succeeded at 961 with fuzz 2.
Hunk #25 FAILED at 995.
Hunk #26 succeeded at 1004 with fuzz 2.
Hunk #27 FAILED at 1034.
20 out of 27 hunks FAILED
checking file libgcc/config/arm/lib1funcs.S



> @@ -1149,12 +1250,16 @@ ARM_FUNC_START aeabi_cdrcmple
>      mov r3, ip
>      b   6f
>
> -ARM_FUNC_START aeabi_cdcmpeq
> +; ARM_FUNC_START aeabi_cdcmpeq
>  ARM_FUNC_ALIAS aeabi_cdcmple aeabi_cdcmpeq


There appears  to be a stray `;' there. How has this been tested ?

>
>      @ The status-returning routines are required to preserve all
>      @ registers except ip, lr, and cpsr.
>  6:  do_push {r0, lr}
> +    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8.
> +    .cfi_rel_offset r0, 0     @ Previous r0 is saved at sp.
> +    .cfi_rel_offset lr, 4     @ Previous lr is saved at sp + 4.
> +
>      ARM_CALL cmpdf2
>      @ Set the Z flag correctly, and the C flag unconditionally.
>      cmp r0, #0


Can you please generate a diff using svn diff from a pristine checkout 
of the sources, please and resend it as an attachment, making sure you 
test this properly with a regression run.

regards
Ramana
Ramana Radhakrishnan May 13, 2015, 4:58 p.m. UTC | #2
On 13/05/15 17:37, Ramana Radhakrishnan wrote:
>
>
> On 12/05/15 14:01, Martin Galvan wrote:
>> On Tue, May 12, 2015 at 5:49 AM, Ramana Radhakrishnan
>> <ramana.radhakrishnan@arm.com> wrote:
>>> That's what I mean when I say email clients "munged it" : email
>>> clients and
>>> / or some popular email servers appear to end up munging white spaces
>>> and
>>> patches don't apply cleanly.
>>>
>>> So, no it doesn't work - once you've sent it through your email client /
>>> server. I am unable to apply the patch as it stands today either
>>> taking the
>>> raw text from the gcc-patches archive or from your email message in my
>>> inbox. It's not like line endings and things have been munged but every
>>> whitespace / tab is in trouble here.
>>>
>>> Please send it back as an attachment if you want me to apply it.
>>
>> Oh, I see! Sorry for that, I thought the problem was on my side.
>> Here's the patch, it's the cfi.patch file.
>>
>
>
> I'm not sure what's going on here and couldn't figure out what was going
> wrong in the 20 minutes I spent on it just now.
>


Bah, I must be going blind.

>
> @@ -53,11 +57,13 @@
>
>  ARM_FUNC_START negdf2
>  ARM_FUNC_ALIAS aeabi_dneg negdf2
> +    CFI_START_FUNCTION
>
>      @ flip sign bit
>      eor xh, xh, #0x80000000

Either the mail server you are using or the source repository you have 
has replaced tabs with spaces, the eor here has a tab in the sources 
between eor and xh. Thus the patch is malformed according to the 
sources. I suspect that's the reason why all the other hunks are not 
applying.

Before reposting -

0. Please take care of the issue I mentioned in the previous email .

1. Please mail the patch so created to yourself on a different account 
and applying the patch you receive on pristine sources. Identify a mail 
server that you can use to send emails to lists that don't munge patches 
in this manner.

2. Please test such a tree and ensure no regressions in a run of the GCC 
testsuite.

3. Post the patch back to gcc-patches@gcc.gnu.org from the mail server 
that doesn't munge patches in this manner. I'll review it again then.

Thanks,
Ramana




>      RET
>
> +    CFI_END_FUNCTION
>      FUNC_END aeabi_dneg
>      FUNC_END negdf2







> Applying the patch downloaded from 2 different inboxes still gives me
> the same result.
>
> $> patch -p1 --dry-run < /work/cfi.patch
> checking file libgcc/config/arm/ieee754-df.S
> Hunk #2 FAILED at 57.
> Hunk #3 succeeded at 70 with fuzz 2.
> Hunk #4 FAILED at 86.
> Hunk #5 FAILED at 153.
> Hunk #6 FAILED at 418.
> Hunk #7 FAILED at 425.
> Hunk #8 FAILED at 440.
> Hunk #9 FAILED at 462.
> Hunk #10 FAILED at 485.
> Hunk #11 FAILED at 555.
> Hunk #12 FAILED at 566.
> Hunk #13 FAILED at 601.
> Hunk #14 FAILED at 653.
> Hunk #15 FAILED at 720.
> Hunk #16 FAILED at 868.
> Hunk #17 FAILED at 1057.
> Hunk #18 FAILED at 1068.
> Hunk #19 FAILED at 1082.
> Hunk #20 FAILED at 1090.
> Hunk #21 FAILED at 1122.
> Hunk #22 FAILED at 1133.
> Hunk #23 succeeded at 1145 with fuzz 2.
> Hunk #24 FAILED at 1155.
> Hunk #25 FAILED at 1168.
> Hunk #26 FAILED at 1228.
> Hunk #27 succeeded at 1236 with fuzz 2.
> Hunk #28 FAILED at 1254.
> Hunk #29 succeeded at 1263 with fuzz 2.
> Hunk #30 FAILED at 1297.
> Hunk #31 succeeded at 1306 with fuzz 2.
> Hunk #32 FAILED at 1336.
> Hunk #33 succeeded at 1345 with fuzz 2.
> Hunk #34 FAILED at 1410.
> 27 out of 34 hunks FAILED
> checking file libgcc/config/arm/ieee754-sf.S
> Hunk #1 FAILED at 31.
> Hunk #2 succeeded at 49 with fuzz 2.
> Hunk #3 FAILED at 285.
> Hunk #4 FAILED at 293.
> Hunk #5 FAILED at 317.
> Hunk #6 succeeded at 324 with fuzz 2.
> Hunk #7 FAILED at 411.
> Hunk #8 succeeded at 422 with fuzz 2.
> Hunk #9 FAILED at 457.
> Hunk #10 FAILED at 468.
> Hunk #11 FAILED at 621.
> Hunk #12 FAILED at 761.
> Hunk #13 FAILED at 785.
> Hunk #14 FAILED at 799.
> Hunk #15 FAILED at 807.
> Hunk #16 FAILED at 826.
> Hunk #17 FAILED at 835.
> Hunk #18 succeeded at 847 with fuzz 2.
> Hunk #19 FAILED at 860.
> Hunk #20 FAILED at 869.
> Hunk #21 FAILED at 927.
> Hunk #22 succeeded at 935 with fuzz 2.
> Hunk #23 FAILED at 952.
> Hunk #24 succeeded at 961 with fuzz 2.
> Hunk #25 FAILED at 995.
> Hunk #26 succeeded at 1004 with fuzz 2.
> Hunk #27 FAILED at 1034.
> 20 out of 27 hunks FAILED
> checking file libgcc/config/arm/lib1funcs.S
>
>
>
>> @@ -1149,12 +1250,16 @@ ARM_FUNC_START aeabi_cdrcmple
>>      mov r3, ip
>>      b   6f
>>
>> -ARM_FUNC_START aeabi_cdcmpeq
>> +; ARM_FUNC_START aeabi_cdcmpeq
>>  ARM_FUNC_ALIAS aeabi_cdcmple aeabi_cdcmpeq
>
>
> There appears  to be a stray `;' there. How has this been tested ?
>
>>
>>      @ The status-returning routines are required to preserve all
>>      @ registers except ip, lr, and cpsr.
>>  6:  do_push {r0, lr}
>> +    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8.
>> +    .cfi_rel_offset r0, 0     @ Previous r0 is saved at sp.
>> +    .cfi_rel_offset lr, 4     @ Previous lr is saved at sp + 4.
>> +
>>      ARM_CALL cmpdf2
>>      @ Set the Z flag correctly, and the C flag unconditionally.
>>      cmp r0, #0
>
>
> Can you please generate a diff using svn diff from a pristine checkout
> of the sources, please and resend it as an attachment, making sure you
> test this properly with a regression run.
>
> regards
> Ramana
>
Martin Galvan May 13, 2015, 5:07 p.m. UTC | #3
On Wed, May 13, 2015 at 1:58 PM, Ramana Radhakrishnan
<ramana.radhakrishnan@arm.com> wrote:
> On 13/05/15 17:37, Ramana Radhakrishnan wrote:
>> I'm not sure what's going on here and couldn't figure out what was going
>> wrong in the 20 minutes I spent on it just now.
>>
> Bah, I must be going blind.
>
>>
>> @@ -53,11 +57,13 @@
>>
>>  ARM_FUNC_START negdf2
>>  ARM_FUNC_ALIAS aeabi_dneg negdf2
>> +    CFI_START_FUNCTION
>>
>>      @ flip sign bit
>>      eor xh, xh, #0x80000000
>
>
> Either the mail server you are using or the source repository you have has
> replaced tabs with spaces, the eor here has a tab in the sources between eor
> and xh. Thus the patch is malformed according to the sources. I suspect
> that's the reason why all the other hunks are not applying.

That makes sense. I downloaded the gcc sources from the git mirror at
github-- perhaps that's what's munging the patch?

> Before reposting -
>
> 0. Please take care of the issue I mentioned in the previous email .

Will do. I honestly have no idea how that semicolon ended there, since
I tried a dry run of the file I sent you and it didn't complain.
Perhaps I accidentally typed it in before closing the file, I don't
know.

> 1. Please mail the patch so created to yourself on a different account and
> applying the patch you receive on pristine sources. Identify a mail server
> that you can use to send emails to lists that don't munge patches in this
> manner.

I used git send-email through gmail when contributing to various
projects such as gdb, and so far haven't had any problems like this
one.

> 2. Please test such a tree and ensure no regressions in a run of the GCC
> testsuite.
>
> 3. Post the patch back to gcc-patches@gcc.gnu.org from the mail server that
> doesn't munge patches in this manner. I'll review it again then.

Will do. Thanks a lot!
Marek Polacek May 13, 2015, 5:13 p.m. UTC | #4
On Wed, May 13, 2015 at 02:07:34PM -0300, Martin Galvan wrote:
> That makes sense. I downloaded the gcc sources from the git mirror at
> github-- perhaps that's what's munging the patch?

That mirror seems to be out-of-date.  I suggest using
<https://gcc.gnu.org/git/?p=gcc.git;a=summary>.

	Marek
diff mbox

Patch

diff --git a/libgcc/config/arm/ieee754-df.S b/libgcc/config/arm/ieee754-df.S
index f75630b..d1f9066 100644
--- a/libgcc/config/arm/ieee754-df.S
+++ b/libgcc/config/arm/ieee754-df.S
@@ -33,8 +33,12 @@ 
  * Only the default rounding mode is intended for best performances.
  * Exceptions aren't supported yet, but that can be added quite easily
  * if necessary without impacting performances.
+ *
+ * In the CFI related comments, 'previousOffset' refers to the previous offset
+ * from sp used to compute the CFA.
  */

+    .cfi_sections .debug_frame

 #ifndef __ARMEB__
 #define xl r0
@@ -53,11 +57,13 @@ 

 ARM_FUNC_START negdf2
 ARM_FUNC_ALIAS aeabi_dneg negdf2
+    CFI_START_FUNCTION

     @ flip sign bit
     eor xh, xh, #0x80000000
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_dneg
     FUNC_END negdf2

@@ -66,6 +72,7 @@  ARM_FUNC_ALIAS aeabi_dneg negdf2
 #ifdef L_arm_addsubdf3

 ARM_FUNC_START aeabi_drsub
+    CFI_START_FUNCTION

     eor xh, xh, #0x80000000 @ flip sign bit of first arg
     b   1f
@@ -81,7 +88,11 @@  ARM_FUNC_ALIAS aeabi_dsub subdf3
 ARM_FUNC_START adddf3
 ARM_FUNC_ALIAS aeabi_dadd adddf3

-1:  do_push {r4, r5, lr}
+1:  do_push {r4, r5, lr}        @ sp -= 12
+    .cfi_adjust_cfa_offset 12   @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0       @ Registers are saved from sp to sp + 8
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8

     @ Look for zeroes, equal values, INF, or NAN.
     shift1  lsl, r4, xh, #1
@@ -148,6 +159,11 @@  ARM_FUNC_ALIAS aeabi_dadd adddf3
     @ Since this is not common case, rescale them off line.
     teq r4, r5
     beq LSYM(Lad_d)
+
+@ CFI note: we're lucky that the branches to Lad_* that appear after this function
+@ have a CFI state that's exactly the same as the one we're in at this
+@ point. Otherwise the CFI would change to a different state after the branch,
+@ which would be disastrous for backtracing.
 LSYM(Lad_x):

     @ Compensate for the exponent overlapping the mantissa MSB added later
@@ -413,6 +429,7 @@  LSYM(Lad_i):
     orrne   xh, xh, #0x00080000 @ quiet NAN
     RETLDM  "r4, r5"

+    CFI_END_FUNCTION
     FUNC_END aeabi_dsub
     FUNC_END subdf3
     FUNC_END aeabi_dadd
@@ -420,12 +437,19 @@  LSYM(Lad_i):

 ARM_FUNC_START floatunsidf
 ARM_FUNC_ALIAS aeabi_ui2d floatunsidf
+    CFI_START_FUNCTION

     teq r0, #0
     do_it   eq, t
     moveq   r1, #0
     RETc(eq)
-    do_push {r4, r5, lr}
+
+    do_push {r4, r5, lr}        @ sp -= 12
+    .cfi_adjust_cfa_offset 12   @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0       @ Registers are saved from sp + 0 to sp + 8.
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8
+
     mov r4, #0x400      @ initial exponent
     add r4, r4, #(52-1 - 1)
     mov r5, #0          @ sign bit is 0
@@ -435,17 +459,25 @@  ARM_FUNC_ALIAS aeabi_ui2d floatunsidf
     mov xh, #0
     b   LSYM(Lad_l)

+    CFI_END_FUNCTION
     FUNC_END aeabi_ui2d
     FUNC_END floatunsidf

 ARM_FUNC_START floatsidf
 ARM_FUNC_ALIAS aeabi_i2d floatsidf
+    CFI_START_FUNCTION

     teq r0, #0
     do_it   eq, t
     moveq   r1, #0
     RETc(eq)
-    do_push {r4, r5, lr}
+
+    do_push {r4, r5, lr}        @ sp -= 12
+    .cfi_adjust_cfa_offset 12   @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0       @ Registers are saved from sp + 0 to sp + 8.
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8
+
     mov r4, #0x400      @ initial exponent
     add r4, r4, #(52-1 - 1)
     ands    r5, r0, #0x80000000 @ sign bit in r5
@@ -457,11 +489,13 @@  ARM_FUNC_ALIAS aeabi_i2d floatsidf
     mov xh, #0
     b   LSYM(Lad_l)

+    CFI_END_FUNCTION
     FUNC_END aeabi_i2d
     FUNC_END floatsidf

 ARM_FUNC_START extendsfdf2
 ARM_FUNC_ALIAS aeabi_f2d extendsfdf2
+    CFI_START_FUNCTION

     movs    r2, r0, lsl #1      @ toss sign bit
     mov xh, r2, asr #3      @ stretch exponent
@@ -480,34 +514,54 @@  ARM_FUNC_ALIAS aeabi_f2d extendsfdf2

     @ value was denormalized.  We can normalize it now.
     do_push {r4, r5, lr}
+    .cfi_adjust_cfa_offset 12   @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0       @ Registers are saved from sp + 0 to sp + 8.
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8
+
     mov r4, #0x380      @ setup corresponding exponent
     and r5, xh, #0x80000000 @ move sign bit in r5
     bic xh, xh, #0x80000000
     b   LSYM(Lad_l)

+    CFI_END_FUNCTION
     FUNC_END aeabi_f2d
     FUNC_END extendsfdf2

 ARM_FUNC_START floatundidf
 ARM_FUNC_ALIAS aeabi_ul2d floatundidf
+    CFI_START_FUNCTION
+    .cfi_remember_state        @ Save the current CFA state.

     orrs    r2, r0, r1
     do_it   eq
     RETc(eq)

-    do_push {r4, r5, lr}
+    do_push {r4, r5, lr}       @ sp -= 12
+    .cfi_adjust_cfa_offset 12  @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0      @ Registers are saved from sp + 0 to sp + 8
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8

     mov r5, #0
     b   2f

 ARM_FUNC_START floatdidf
 ARM_FUNC_ALIAS aeabi_l2d floatdidf
+    .cfi_restore_state
+    @ Restore the CFI state we saved above. If we didn't do this then the
+    @ following instructions would have the CFI state that was set by the
+    @ offset adjustments made in floatundidf.

     orrs    r2, r0, r1
     do_it   eq
     RETc(eq)

-    do_push {r4, r5, lr}
+    do_push {r4, r5, lr}       @ sp -= 12
+    .cfi_adjust_cfa_offset 12  @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r4, 0      @ Registers are saved from sp to sp + 8
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset lr, 8

     ands    r5, ah, #0x80000000 @ sign bit in r5
     bpl 2f
@@ -550,6 +604,7 @@  ARM_FUNC_ALIAS aeabi_l2d floatdidf
     add r4, r4, r2
     b   LSYM(Lad_p)

+    CFI_END_FUNCTION
     FUNC_END floatdidf
     FUNC_END aeabi_l2d
     FUNC_END floatundidf
@@ -561,7 +616,14 @@  ARM_FUNC_ALIAS aeabi_l2d floatdidf

 ARM_FUNC_START muldf3
 ARM_FUNC_ALIAS aeabi_dmul muldf3
-    do_push {r4, r5, r6, lr}
+    CFI_START_FUNCTION
+
+    do_push {r4, r5, r6, lr}    @ sp -= 16
+    .cfi_adjust_cfa_offset 16   @ CFA is now sp + previousOffset + 16
+    .cfi_rel_offset r4, 0       @ Registers are saved from sp to sp + 12.
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset r6, 8
+    .cfi_rel_offset lr, 12

     @ Mask out exponents, trap any zero/denormal/INF/NAN.
     mov ip, #0xff
@@ -596,7 +658,16 @@  ARM_FUNC_ALIAS aeabi_dmul muldf3
     and   r6, r6, #0x80000000

     @ Well, no way to make it shorter without the umull instruction.
-    stmfd   sp!, {r6, r7, r8, r9, sl, fp}
+    stmfd   sp!, {r6, r7, r8, r9, sl, fp}   @ sp -= 24
+    .cfi_remember_state         @ Save the current CFI state.
+    .cfi_adjust_cfa_offset 24   @ CFA is now sp + previousOffset + 24.
+    .cfi_rel_offset r6, 0       @ Registers are saved from sp to sp + 20.
+    .cfi_rel_offset r7, 4
+    .cfi_rel_offset r8, 8
+    .cfi_rel_offset r9, 12
+    .cfi_rel_offset sl, 16
+    .cfi_rel_offset fp, 20
+
     mov r7, xl, lsr #16
     mov r8, yl, lsr #16
     mov r9, xh, lsr #16
@@ -648,8 +719,8 @@  ARM_FUNC_ALIAS aeabi_dmul muldf3
     mul fp, xh, yh
     adcs    r5, r5, fp
     adc r6, r6, #0
-    ldmfd   sp!, {yl, r7, r8, r9, sl, fp}
-
+    ldmfd   sp!, {yl, r7, r8, r9, sl, fp}   @ sp += 24
+    .cfi_restore_state   @ Restore the previous CFI state.
 #else

     @ Here is the actual multiplication.
@@ -715,7 +786,6 @@  LSYM(Lml_1):
     orr xh, xh, #0x00100000
     mov lr, #0
     subs    r4, r4, #1
-
 LSYM(Lml_u):
     @ Overflow?
     bgt LSYM(Lml_o)
@@ -863,13 +933,20 @@  LSYM(Lml_n):
     orr xh, xh, #0x00f80000
     RETLDM  "r4, r5, r6"

+    CFI_END_FUNCTION
     FUNC_END aeabi_dmul
     FUNC_END muldf3

 ARM_FUNC_START divdf3
 ARM_FUNC_ALIAS aeabi_ddiv divdf3
+    CFI_START_FUNCTION

     do_push {r4, r5, r6, lr}
+    .cfi_adjust_cfa_offset 16
+    .cfi_rel_offset r4, 0
+    .cfi_rel_offset r5, 4
+    .cfi_rel_offset r6, 8
+    .cfi_rel_offset lr, 12

     @ Mask out exponents, trap any zero/denormal/INF/NAN.
     mov ip, #0xff
@@ -1052,6 +1129,7 @@  LSYM(Ldv_s):
     bne LSYM(Lml_z)     @ 0 / <non_zero> -> 0
     b   LSYM(Lml_n)     @ 0 / 0 -> NAN

+    CFI_END_FUNCTION
     FUNC_END aeabi_ddiv
     FUNC_END divdf3

@@ -1063,6 +1141,7 @@  LSYM(Ldv_s):

 ARM_FUNC_START gtdf2
 ARM_FUNC_ALIAS gedf2 gtdf2
+    CFI_START_FUNCTION
     mov ip, #-1
     b   1f

@@ -1077,6 +1156,10 @@  ARM_FUNC_ALIAS eqdf2 cmpdf2
     mov ip, #1          @ how should we specify unordered here?

 1:  str ip, [sp, #-4]!
+    .cfi_adjust_cfa_offset 4        @ CFA is now sp + previousOffset + 4.
+    @ We're not adding CFI for ip as it's pushed into the stack only because
+    @ it may be popped off later as a return value (i.e. we're not preserving
+    @ it anyways).

     @ Trap any INF/NAN first.
     mov ip, xh, lsl #1
@@ -1085,10 +1168,18 @@  ARM_FUNC_ALIAS eqdf2 cmpdf2
     do_it   ne
     COND(mvn,s,ne)  ip, ip, asr #21
     beq 3f
+    .cfi_remember_state
+    @ Save the current CFI state. This is done because the branch is conditional,
+    @ and if we don't take it we'll issue a .cfi_adjust_cfa_offset and return.
+    @ If we do take it, however, the .cfi_adjust_cfa_offset from the non-branch
+    @ code will affect the branch code as well. To avoid this we'll restore
+    @ the current state before executing the branch code.

     @ Test for equality.
     @ Note that 0.0 is equal to -0.0.
 2:  add sp, sp, #4
+    .cfi_adjust_cfa_offset -4       @ CFA is now sp + previousOffset.
+
     orrs    ip, xl, xh, lsl #1  @ if x == 0.0 or -0.0
     do_it   eq, e
     COND(orr,s,eq)  ip, yl, yh, lsl #1  @ and y == 0.0 or -0.0
@@ -1117,8 +1208,13 @@  ARM_FUNC_ALIAS eqdf2 cmpdf2
     orr r0, r0, #1
     RET

-    @ Look for a NAN.
-3:  mov ip, xh, lsl #1
+3:  @ Look for a NAN.
+
+    @ Restore the previous CFI state (i.e. keep the CFI state as it was
+    @ before the branch).
+    .cfi_restore_state
+
+    mov ip, xh, lsl #1
     mvns    ip, ip, asr #21
     bne 4f
     orrs    ip, xl, xh, lsl #12
@@ -1128,9 +1224,13 @@  ARM_FUNC_ALIAS eqdf2 cmpdf2
     bne 2b
     orrs    ip, yl, yh, lsl #12
     beq 2b          @ y is not NAN
+
 5:  ldr r0, [sp], #4        @ unordered return code
+    .cfi_adjust_cfa_offset -4       @ CFA is now sp + previousOffset.
+
     RET

+    CFI_END_FUNCTION
     FUNC_END gedf2
     FUNC_END gtdf2
     FUNC_END ledf2
@@ -1140,6 +1240,7 @@  ARM_FUNC_ALIAS eqdf2 cmpdf2
     FUNC_END cmpdf2

 ARM_FUNC_START aeabi_cdrcmple
+    CFI_START_FUNCTION

     mov ip, r0
     mov r0, r2
@@ -1149,12 +1250,16 @@  ARM_FUNC_START aeabi_cdrcmple
     mov r3, ip
     b   6f

-ARM_FUNC_START aeabi_cdcmpeq
+; ARM_FUNC_START aeabi_cdcmpeq
 ARM_FUNC_ALIAS aeabi_cdcmple aeabi_cdcmpeq

     @ The status-returning routines are required to preserve all
     @ registers except ip, lr, and cpsr.
 6:  do_push {r0, lr}
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8.
+    .cfi_rel_offset r0, 0     @ Previous r0 is saved at sp.
+    .cfi_rel_offset lr, 4     @ Previous lr is saved at sp + 4.
+
     ARM_CALL cmpdf2
     @ Set the Z flag correctly, and the C flag unconditionally.
     cmp r0, #0
@@ -1162,59 +1267,86 @@  ARM_FUNC_ALIAS aeabi_cdcmple aeabi_cdcmpeq
     @ that the first operand was smaller than the second.
     do_it   mi
     cmnmi   r0, #0
+
     RETLDM  "r0"

+    CFI_END_FUNCTION
     FUNC_END aeabi_cdcmple
     FUNC_END aeabi_cdcmpeq
     FUNC_END aeabi_cdrcmple

 ARM_FUNC_START  aeabi_dcmpeq
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!        @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cdcmple
     do_it   eq, e
     moveq   r0, #1  @ Equal to.
     movne   r0, #0  @ Less than, greater than, or unordered.
+
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_dcmpeq

 ARM_FUNC_START  aeabi_dcmplt
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!        @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cdcmple
     do_it   cc, e
     movcc   r0, #1  @ Less than.
     movcs   r0, #0  @ Equal to, greater than, or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_dcmplt

 ARM_FUNC_START  aeabi_dcmple
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!        @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cdcmple
     do_it   ls, e
     movls   r0, #1  @ Less than or equal to.
     movhi   r0, #0  @ Greater than or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_dcmple

 ARM_FUNC_START  aeabi_dcmpge
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!        @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cdrcmple
     do_it   ls, e
     movls   r0, #1  @ Operand 2 is less than or equal to operand 1.
     movhi   r0, #0  @ Operand 2 greater than operand 1, or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_dcmpge

 ARM_FUNC_START  aeabi_dcmpgt
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!        @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cdrcmple
     do_it   cc, e
     movcc   r0, #1  @ Operand 2 is less than operand 1.
@@ -1222,6 +1354,7 @@  ARM_FUNC_START  aeabi_dcmpgt
             @ or they are unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_dcmpgt

 #endif /* L_cmpdf2 */
@@ -1230,6 +1363,7 @@  ARM_FUNC_START  aeabi_dcmpgt

 ARM_FUNC_START unorddf2
 ARM_FUNC_ALIAS aeabi_dcmpun unorddf2
+    .cfi_startproc

     mov ip, xh, lsl #1
     mvns    ip, ip, asr #21
@@ -1247,6 +1381,7 @@  ARM_FUNC_ALIAS aeabi_dcmpun unorddf2
 3:  mov r0, #1          @ arguments are unordered.
     RET

+    .cfi_endproc
     FUNC_END aeabi_dcmpun
     FUNC_END unorddf2

@@ -1256,6 +1391,7 @@  ARM_FUNC_ALIAS aeabi_dcmpun unorddf2

 ARM_FUNC_START fixdfsi
 ARM_FUNC_ALIAS aeabi_d2iz fixdfsi
+    CFI_START_FUNCTION

     @ check exponent range.
     mov r2, xh, lsl #1
@@ -1289,6 +1425,7 @@  ARM_FUNC_ALIAS aeabi_d2iz fixdfsi
 4:  mov r0, #0          @ How should we convert NAN?
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_d2iz
     FUNC_END fixdfsi

@@ -1298,6 +1435,7 @@  ARM_FUNC_ALIAS aeabi_d2iz fixdfsi

 ARM_FUNC_START fixunsdfsi
 ARM_FUNC_ALIAS aeabi_d2uiz fixunsdfsi
+    CFI_START_FUNCTION

     @ check exponent range.
     movs    r2, xh, lsl #1
@@ -1327,6 +1465,7 @@  ARM_FUNC_ALIAS aeabi_d2uiz fixunsdfsi
 4:  mov r0, #0          @ How should we convert NAN?
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_d2uiz
     FUNC_END fixunsdfsi

@@ -1336,6 +1475,7 @@  ARM_FUNC_ALIAS aeabi_d2uiz fixunsdfsi

 ARM_FUNC_START truncdfsf2
 ARM_FUNC_ALIAS aeabi_d2f truncdfsf2
+    CFI_START_FUNCTION

     @ check exponent range.
     mov r2, xh, lsl #1
@@ -1400,6 +1540,7 @@  ARM_FUNC_ALIAS aeabi_d2f truncdfsf2
     orr r0, r0, #0x00800000
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_d2f
     FUNC_END truncdfsf2

diff --git a/libgcc/config/arm/ieee754-sf.S b/libgcc/config/arm/ieee754-sf.S
index 3e76241..31f93a2 100644
--- a/libgcc/config/arm/ieee754-sf.S
+++ b/libgcc/config/arm/ieee754-sf.S
@@ -31,16 +31,21 @@ 
  * Only the default rounding mode is intended for best performances.
  * Exceptions aren't supported yet, but that can be added quite easily
  * if necessary without impacting performances.
+ *
+ * In the CFI related comments, 'previousOffset' refers to the previous offset
+ * from sp used to compute the CFA.
  */

 #ifdef L_arm_negsf2

 ARM_FUNC_START negsf2
 ARM_FUNC_ALIAS aeabi_fneg negsf2
+    CFI_START_FUNCTION

     eor r0, r0, #0x80000000 @ flip sign bit
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_fneg
     FUNC_END negsf2

@@ -49,6 +54,7 @@  ARM_FUNC_ALIAS aeabi_fneg negsf2
 #ifdef L_arm_addsubsf3

 ARM_FUNC_START aeabi_frsub
+    CFI_START_FUNCTION

     eor r0, r0, #0x80000000 @ flip sign bit of first arg
     b   1f
@@ -284,6 +290,7 @@  LSYM(Lad_i):
     orrne   r0, r0, #0x00400000 @ quiet NAN
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_frsub
     FUNC_END aeabi_fadd
     FUNC_END addsf3
@@ -292,6 +299,7 @@  LSYM(Lad_i):

 ARM_FUNC_START floatunsisf
 ARM_FUNC_ALIAS aeabi_ui2f floatunsisf
+    CFI_START_FUNCTION

     mov r3, #0
     b   1f
@@ -316,6 +324,7 @@  ARM_FUNC_ALIAS aeabi_i2f floatsisf
     mov al, #0
     b   2f

+    CFI_END_FUNCTION
     FUNC_END aeabi_i2f
     FUNC_END floatsisf
     FUNC_END aeabi_ui2f
@@ -323,6 +332,7 @@  ARM_FUNC_ALIAS aeabi_i2f floatsisf

 ARM_FUNC_START floatundisf
 ARM_FUNC_ALIAS aeabi_ul2f floatundisf
+    CFI_START_FUNCTION

     orrs    r2, r0, r1
     do_it   eq
@@ -409,6 +419,7 @@  ARM_FUNC_ALIAS aeabi_l2f floatdisf
     biceq   r0, r0, ip, lsr #31
     RET

+    CFI_END_FUNCTION
     FUNC_END floatdisf
     FUNC_END aeabi_l2f
     FUNC_END floatundisf
@@ -420,6 +431,7 @@  ARM_FUNC_ALIAS aeabi_l2f floatdisf

 ARM_FUNC_START mulsf3
 ARM_FUNC_ALIAS aeabi_fmul mulsf3
+    CFI_START_FUNCTION

     @ Mask out exponents, trap any zero/denormal/INF/NAN.
     mov ip, #0xff
@@ -454,7 +466,13 @@  LSYM(Lml_x):
     and r3, ip, #0x80000000

     @ Well, no way to make it shorter without the umull instruction.
-    do_push {r3, r4, r5}
+    do_push {r3, r4, r5}       @ sp -= 12
+    .cfi_remember_state        @ Save the current CFI state
+    .cfi_adjust_cfa_offset 12  @ CFA is now sp + previousOffset + 12
+    .cfi_rel_offset r3, 0      @ Registers are saved from sp to sp + 8
+    .cfi_rel_offset r4, 4
+    .cfi_rel_offset r5, 8
+
     mov r4, r0, lsr #16
     mov r5, r1, lsr #16
     bic r0, r0, r4, lsl #16
@@ -465,7 +483,8 @@  LSYM(Lml_x):
     mla r0, r4, r1, r0
     adds    r3, r3, r0, lsl #16
     adc r1, ip, r0, lsr #16
-    do_pop  {r0, r4, r5}
+    do_pop  {r0, r4, r5}       @ sp += 12
+    .cfi_restore_state         @ Restore the previous CFI state

 #else

@@ -618,11 +637,13 @@  LSYM(Lml_n):
     orr r0, r0, #0x00c00000
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_fmul
     FUNC_END mulsf3

 ARM_FUNC_START divsf3
 ARM_FUNC_ALIAS aeabi_fdiv divsf3
+    CFI_START_FUNCTION

     @ Mask out exponents, trap any zero/denormal/INF/NAN.
     mov ip, #0xff
@@ -758,6 +779,7 @@  LSYM(Ldv_s):
     bne LSYM(Lml_z)     @ 0 / <non_zero> -> 0
     b   LSYM(Lml_n)     @ 0 / 0 -> NAN

+    CFI_END_FUNCTION
     FUNC_END aeabi_fdiv
     FUNC_END divsf3

@@ -782,6 +804,7 @@  LSYM(Ldv_s):

 ARM_FUNC_START gtsf2
 ARM_FUNC_ALIAS gesf2 gtsf2
+    CFI_START_FUNCTION
     mov ip, #-1
     b   1f

@@ -796,6 +819,10 @@  ARM_FUNC_ALIAS eqsf2 cmpsf2
     mov ip, #1          @ how should we specify unordered here?

 1:  str ip, [sp, #-4]!
+    .cfi_adjust_cfa_offset 4  @ CFA is now sp + previousOffset + 4.
+    @ We're not adding CFI for ip as it's pushed into the stack only because
+    @ it may be popped off later as a return value (i.e. we're not preserving
+    @ it anyways).

     @ Trap any INF/NAN first.
     mov r2, r0, lsl #1
@@ -804,10 +831,18 @@  ARM_FUNC_ALIAS eqsf2 cmpsf2
     do_it   ne
     COND(mvn,s,ne)  ip, r3, asr #24
     beq 3f
+    .cfi_remember_state
+    @ Save the current CFI state. This is done because the branch is conditional,
+    @ and if we don't take it we'll issue a .cfi_adjust_cfa_offset and return.
+    @ If we do take it, however, the .cfi_adjust_cfa_offset from the non-branch
+    @ code will affect the branch code as well. To avoid this we'll restore
+    @ the current state before executing the branch code.

     @ Compare values.
     @ Note that 0.0 is equal to -0.0.
 2:  add sp, sp, #4
+    .cfi_adjust_cfa_offset -4       @ CFA is now sp + previousOffset.
+
     orrs    ip, r2, r3, lsr #1  @ test if both are 0, clear C flag
     do_it   ne
     teqne   r0, r1          @ if not 0 compare sign
@@ -823,8 +858,13 @@  ARM_FUNC_ALIAS eqsf2 cmpsf2
     orrne   r0, r0, #1
     RET

-    @ Look for a NAN.
-3:  mvns    ip, r2, asr #24
+3:  @ Look for a NAN.
+
+    @ Restore the previous CFI state (i.e. keep the CFI state as it was
+    @ before the branch).
+    .cfi_restore_state
+
+    mvns    ip, r2, asr #24
     bne 4f
     movs    ip, r0, lsl #9
     bne 5f          @ r0 is NAN
@@ -832,9 +872,12 @@  ARM_FUNC_ALIAS eqsf2 cmpsf2
     bne 2b
     movs    ip, r1, lsl #9
     beq 2b          @ r1 is not NAN
+
 5:  ldr r0, [sp], #4        @ return unordered code.
+    .cfi_adjust_cfa_offset -4       @ CFA is now sp + previousOffset.
     RET

+    CFI_END_FUNCTION
     FUNC_END gesf2
     FUNC_END gtsf2
     FUNC_END lesf2
@@ -844,6 +887,7 @@  ARM_FUNC_ALIAS eqsf2 cmpsf2
     FUNC_END cmpsf2

 ARM_FUNC_START aeabi_cfrcmple
+    CFI_START_FUNCTION

     mov ip, r0
     mov r0, r1
@@ -856,6 +900,13 @@  ARM_FUNC_ALIAS aeabi_cfcmple aeabi_cfcmpeq
     @ The status-returning routines are required to preserve all
     @ registers except ip, lr, and cpsr.
 6:  do_push {r0, r1, r2, r3, lr}
+    .cfi_adjust_cfa_offset 20  @ CFA is at sp + previousOffset + 20
+    .cfi_rel_offset r0, 0      @ Registers are saved from sp to sp + 16
+    .cfi_rel_offset r1, 4
+    .cfi_rel_offset r2, 8
+    .cfi_rel_offset r3, 12
+    .cfi_rel_offset lr, 16
+
     ARM_CALL cmpsf2
     @ Set the Z flag correctly, and the C flag unconditionally.
     cmp r0, #0
@@ -865,57 +916,82 @@  ARM_FUNC_ALIAS aeabi_cfcmple aeabi_cfcmpeq
     cmnmi   r0, #0
     RETLDM  "r0, r1, r2, r3"

+    CFI_END_FUNCTION
     FUNC_END aeabi_cfcmple
     FUNC_END aeabi_cfcmpeq
     FUNC_END aeabi_cfrcmple

 ARM_FUNC_START  aeabi_fcmpeq
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!    @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cfcmple
     do_it   eq, e
     moveq   r0, #1  @ Equal to.
     movne   r0, #0  @ Less than, greater than, or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmpeq

 ARM_FUNC_START  aeabi_fcmplt
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!    @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cfcmple
     do_it   cc, e
     movcc   r0, #1  @ Less than.
     movcs   r0, #0  @ Equal to, greater than, or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmplt

 ARM_FUNC_START  aeabi_fcmple
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!    @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cfcmple
     do_it   ls, e
     movls   r0, #1  @ Less than or equal to.
     movhi   r0, #0  @ Greater than or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmple

 ARM_FUNC_START  aeabi_fcmpge
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!    @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cfrcmple
     do_it   ls, e
     movls   r0, #1  @ Operand 2 is less than or equal to operand 1.
     movhi   r0, #0  @ Operand 2 greater than operand 1, or unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmpge

 ARM_FUNC_START  aeabi_fcmpgt
+    CFI_START_FUNCTION
+
+    str lr, [sp, #-8]!    @ sp -= 8
+    .cfi_adjust_cfa_offset 8  @ CFA is now sp + previousOffset + 8
+    .cfi_rel_offset lr, 0     @ lr is at sp

-    str lr, [sp, #-8]!
     ARM_CALL aeabi_cfrcmple
     do_it   cc, e
     movcc   r0, #1  @ Operand 2 is less than operand 1.
@@ -923,6 +999,7 @@  ARM_FUNC_START  aeabi_fcmpgt
             @ or they are unordered.
     RETLDM

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmpgt

 #endif /* L_cmpsf2 */
@@ -931,6 +1008,7 @@  ARM_FUNC_START  aeabi_fcmpgt

 ARM_FUNC_START unordsf2
 ARM_FUNC_ALIAS aeabi_fcmpun unordsf2
+    CFI_START_FUNCTION

     mov r2, r0, lsl #1
     mov r3, r1, lsl #1
@@ -947,6 +1025,7 @@  ARM_FUNC_ALIAS aeabi_fcmpun unordsf2
 3:  mov r0, #1          @ arguments are unordered.
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_fcmpun
     FUNC_END unordsf2

@@ -956,6 +1035,7 @@  ARM_FUNC_ALIAS aeabi_fcmpun unordsf2

 ARM_FUNC_START fixsfsi
 ARM_FUNC_ALIAS aeabi_f2iz fixsfsi
+    CFI_START_FUNCTION

     @ check exponent range.
     mov r2, r0, lsl #1
@@ -989,6 +1069,7 @@  ARM_FUNC_ALIAS aeabi_f2iz fixsfsi
 4:  mov r0, #0          @ What should we convert NAN to?
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_f2iz
     FUNC_END fixsfsi

@@ -998,6 +1079,7 @@  ARM_FUNC_ALIAS aeabi_f2iz fixsfsi

 ARM_FUNC_START fixunssfsi
 ARM_FUNC_ALIAS aeabi_f2uiz fixunssfsi
+    CFI_START_FUNCTION

     @ check exponent range.
     movs    r2, r0, lsl #1
@@ -1027,6 +1109,7 @@  ARM_FUNC_ALIAS aeabi_f2uiz fixunssfsi
 4:  mov r0, #0          @ What should we convert NAN to?
     RET

+    CFI_END_FUNCTION
     FUNC_END aeabi_f2uiz
     FUNC_END fixunssfsi

diff --git a/libgcc/config/arm/lib1funcs.S b/libgcc/config/arm/lib1funcs.S
index d2d0d20..4b63924 100644
--- a/libgcc/config/arm/lib1funcs.S
+++ b/libgcc/config/arm/lib1funcs.S
@@ -1965,6 +1965,16 @@  LSYM(Lchange_\register):

 #endif /* Arch supports thumb.  */

+.macro CFI_START_FUNCTION
+    .cfi_startproc
+    .cfi_remember_state
+.endm
+
+.macro CFI_END_FUNCTION
+    .cfi_restore_state
+    .cfi_endproc
+.endm
+
 #ifndef __symbian__
 #ifndef __ARM_ARCH_6M__
 #include "ieee754-df.S"