ARM:VFPv3:enable {d16-d31} access
Subsystems (2)
A message that is not in the archive (sent off-list or not yet received): Tarun Kanti DebBarma <[email protected]
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
From: Tarun Kanti Debbarma <a0876346@a0876346-desktop.(none)>
This patch attempts to fix two related problems:
(1) vfp_get_double(), vfp_put_double() functions have VFPv3 specific implementation guarded within CONFIG_VFPv3 macro. The intent is to access {d16-d31} additional registers provided in VFPv3. However, it still wrongly refers to {d0-d15}. This has been corrected.
(2) While compiling the above changes, it produced compilation error because arch/arm/vfp/Makefile had the -mfpu=vfp option which could not recognize VFPv3 registers. This has been corrected to -mfpu=vfp3. With this option we are also able to compile pre-VFPv3 code.
--- arch/arm/vfp/Makefile | 2 +- arch/arm/vfp/vfphw.S | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-) mode change 100644 => 100755 arch/arm/vfp/vfphw.S
arch/arm/vfp/Makefile +1 -1
@@ -7,7 +7,7 @@ # EXTRA_CFLAGS := -DDEBUG # EXTRA_AFLAGS := -DDEBUG -KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp) +KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=vfp3 -mfloat-abi=softfp) LDFLAGS +=--no-warn-mismatch obj-y += vfp.o
arch/arm/vfp/vfphw.S +4 -4
@@ -254,8 +254,8 @@ ENTRY(vfp_get_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr mov pc, lr .org 1b + 8 .endr
@@ -276,8 +276,8 @@ ENTRY(vfp_put_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mcrr p11, 3, r1, r2, c\dr @ fmdrr r1, r2, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmdrr d\dr, r1, r2 mov pc, lr .org 1b + 8 .endr
Signature · 1 line
1.6.0.4
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
(Including ARM mailing list)
From: Tarun Kanti Debbarma <a0876346@a0876346-desktop.(none)>
This patch attempts to fix two related problems:
(1) vfp_get_double(), vfp_put_double() functions have VFPv3 specific implementation guarded within CONFIG_VFPv3 macro. The intent is to access {d16-d31} additional registers provided in VFPv3. However, it still wrongly refers to {d0-d15}. This has been corrected.
(2) While compiling the above changes, it produced compilation error because arch/arm/vfp/Makefile had the -mfpu=vfp option which could not recognize VFPv3 registers. This has been corrected to -mfpu=vfp3. With this option we are also able to compile pre-VFPv3 code.
--- arch/arm/vfp/Makefile | 2 +- arch/arm/vfp/vfphw.S | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-) mode change 100644 => 100755 arch/arm/vfp/vfphw.S
arch/arm/vfp/Makefile +1 -1
@@ -7,7 +7,7 @@ # EXTRA_CFLAGS := -DDEBUG # EXTRA_AFLAGS := -DDEBUG -KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp) +KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=vfp3 -mfloat-abi=softfp) LDFLAGS +=--no-warn-mismatch obj-y += vfp.o
arch/arm/vfp/vfphw.S +4 -4
@@ -254,8 +254,8 @@ ENTRY(vfp_get_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr mov pc, lr .org 1b + 8 .endr
@@ -276,8 +276,8 @@ ENTRY(vfp_put_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mcrr p11, 3, r1, r2, c\dr @ fmdrr r1, r2, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmdrr d\dr, r1, r2 mov pc, lr .org 1b + 8 .endr
Signature · 1 line
1.6.0.4
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Quoted 10 lines · Tarun Kanti
(Including ARM mailing list)
From: Tarun Kanti Debbarma <a0876346@a0876346-desktop.(none)>
This patch attempts to fix two related problems:
(1)
vfp_get_double(),vfp_put_double()functions have VFPv3 specific implementation guarded withinCONFIG_VFPv3macro. The intent is to access {d16-d31} additional registers provided in VFPv3. However, it still wrongly refers to {d0-d15}. This has been corrected.
This does not seem to be the case, see below.
Quoted 32 lines
(2) While compiling the above changes, it produced compilation error because
arch/arm/vfp/Makefilehad the -mfpu=vfp option which could not recognize VFPv3 registers. This has been corrected to -mfpu=vfp3. With this option we are also able to compile pre-VFPv3 code.Signed-off-by Tarun Kanti Debbarma <[email protected]>--- arch/arm/vfp/Makefile | 2 +- arch/arm/vfp/vfphw.S | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-) mode change 100644 => 100755 arch/arm/vfp/vfphw.Sarch/arm/vfp/Makefile +1
-2@@ -7,7 +7,7 @@ # EXTRA_CFLAGS := -DDEBUG # EXTRA_AFLAGS := -DDEBUG -KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp) +KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=vfp3 -mfloat-abi=softfp) LDFLAGS +=--no-warn-mismatch obj-y += vfp.oarch/arm/vfp/vfphw.S +0
-0
arch/arm/vfp/vfphw.S ENTRY(vfp_get_double) · line 258
Quoted 7 lines
arch/arm/vfp/vfphw.S +2
-2@@ -254,8 +254,8 @@ ENTRY(vfp_get_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr
The existing code does exactly the same, but avoids the need to specify -mfpu=vfp3 option. It's quite easy to verify:
/**** test.S *******/
test:
mrrc p11, 3, r0, r1, c0
fmrrd r0, r1, d16/*******************/
gcc -c test.S
objdump -d test.o
Disassembly of section .text:
00000000 <test>:
0: ec510b30 vmov r0, r1, d16 4: ec510b30 vmov r0, r1, d16
The comment in the existing code is a bit misleading though.
Signature · 2 lines
Best regards, Siarhei Siamashka
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Hello Siarhei Siamashka,
Thanks for the inputs. Please see my comments.
-----Original Message-----
From: Siarhei Siamashka [mailto:siarhei.siamashka at gmail.com]
Sent: Tuesday, May 25, 2010 5:57 PM
To: DebBarma, Tarun Kanti
Cc: linux-omap at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH] ARM:VFPv3:enable {d16-d31} access
Quoted 10 lines · Tarun Kanti
(Including ARM mailing list)
From: Tarun Kanti Debbarma <a0876346@a0876346-desktop.(none)>
This patch attempts to fix two related problems:
(1)
vfp_get_double(),vfp_put_double()functions have VFPv3 specific implementation guarded withinCONFIG_VFPv3macro. The intent is to access {d16-d31} additional registers provided in VFPv3. However, it still wrongly refers to {d0-d15}. This has been corrected.
This does not seem to be the case, see below.
Are you saying following line generates code for {d16-d31}?
.irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
Quoted 32 lines
(2) While compiling the above changes, it produced compilation error because
arch/arm/vfp/Makefilehad the -mfpu=vfp option which could not recognize VFPv3 registers. This has been corrected to -mfpu=vfp3. With this option we are also able to compile pre-VFPv3 code.Signed-off-by Tarun Kanti Debbarma <[email protected]>--- arch/arm/vfp/Makefile | 2 +- arch/arm/vfp/vfphw.S | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-) mode change 100644 => 100755 arch/arm/vfp/vfphw.Sarch/arm/vfp/Makefile +1
-2@@ -7,7 +7,7 @@ # EXTRA_CFLAGS := -DDEBUG # EXTRA_AFLAGS := -DDEBUG -KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp) +KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=vfp3 -mfloat-abi=softfp) LDFLAGS +=--no-warn-mismatch obj-y += vfp.oarch/arm/vfp/vfphw.S +0
-0
arch/arm/vfp/vfphw.S ENTRY(vfp_get_double) · line 258
Quoted 7 lines
arch/arm/vfp/vfphw.S +2
-2@@ -254,8 +254,8 @@ ENTRY(vfp_get_double) .endr #ifdef CONFIG_VFPv3 @ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr
The existing code does exactly the same, but avoids the need to specify -mfpu=vfp3 option. It's quite easy to verify:
/**** test.S *******/
test:
mrrc p11, 3, r0, r1, c0
fmrrd r0, r1, d16/*******************/
gcc -c test.S
objdump -d test.o
Disassembly of section .text:
00000000 <test>:
0: ec510b30 vmov r0, r1, d16 4: ec510b30 vmov r0, r1, d16
The comment in the existing code is a bit misleading though.
I get following compilation error when:
-fmpu=vfp3 option used with
mrrc p11, 3, r0, r1, c\dr instruction
arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c16' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c17' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c18' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c19' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c20' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c21' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c22' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c23' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c24' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c25' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c26' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c27' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c28' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c29' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c30' arch/arm/vfp/vfphw.S:234: Error: co-processor register expected -- mrrc p11,3,r0,r1,c31'
Is this behavior expected?
However I do not get the same when I use:
fmrrd r0, r1, d\dr
Question: Is there any disadvantage of using: fmrrd r0, r1, d\dr over the other one?
Thanks once again for your inputs.
Signature · 2 lines
Best regards, Siarhei Siamashka
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
In reply to DebBarma, Tarun Kantiarch/arm/vfp/vfphw.S ENTRY(vfp_get_double) · line 258
Quoted 6 lines · Tarun Kanti
#ifdef
CONFIG_VFPv3arch/arm/vfp/vfphw.S +2
-2@ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr
The existing code is correct.
For every fmrrd instruction, there is a corresponding mrrc version which assembles to exactly the same opcode.
mrrc instructions take:
1. Co-processor number, range 0-15.
2. Opcode number N, range 0-15.
3. Destination register 1, range 0-15.
4. Destination register 2, range 0-15.
5. Co-processor register number R, range 0-15.
For fmrrd encodings, the first 16 registers are encoded using N=1 with R=0 to 15. The second 16 registers are encoded using N=3 with R=0 to 15. Specifying a co-processor register number greater than 15 is illegal, hence why the 'irp' specifies the numbers 0 to 15.
If we look at the instruction encodings, for MRRC:
3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
| cond |1 1 0 0 0 1 0 1| Rn | Rm |CP Num | N | R |
For FMRRD:
3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
| cond |1 1 0 0 0 1 0 1| Rn | Rm |1 0 1 1 0 0 M 1| R |
where "M" and "R" together define the register.
As I said above, the existing code is correct. What problem are you actually trying to solve here?
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Russell,
-----Original Message-----
From: Russell King - ARM Linux [mailto:linux at arm.linux.org.uk]
Sent: Wednesday, May 26, 2010 12:28 AM
To: DebBarma, Tarun Kanti
Cc: linux-omap at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH] ARM:VFPv3:enable {d16-d31} access
arch/arm/vfp/vfphw.S ENTRY(vfp_get_double) · line 258
Quoted 6 lines · Tarun Kanti
#ifdef
CONFIG_VFPv3arch/arm/vfp/vfphw.S +2
-2@ d16 - d31 registers - .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15 -1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr + .irp dr,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31 +1: fmrrd r0, r1, d\dr
The existing code is correct.
For every fmrrd instruction, there is a corresponding mrrc version which assembles to exactly the same opcode.
mrrc instructions take:
1. Co-processor number, range 0-15.
2. Opcode number N, range 0-15.
3. Destination register 1, range 0-15.
4. Destination register 2, range 0-15.
5. Co-processor register number R, range 0-15.
For fmrrd encodings, the first 16 registers are encoded using N=1 with R=0 to 15. The second 16 registers are encoded using N=3 with R=0 to 15. Specifying a co-processor register number greater than 15 is illegal, hence why the 'irp' specifies the numbers 0 to 15.
If we look at the instruction encodings, for MRRC:
3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
| cond |1 1 0 0 0 1 0 1| Rn | Rm |CP Num | N | R |
For FMRRD:
3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
| cond |1 1 0 0 0 1 0 1| Rn | Rm |1 0 1 1 0 0 M 1| R |
where "M" and "R" together define the register.
As I said above, the existing code is correct. What problem are you actually trying to solve here?
######################
I have a test case which exercise all VFP general purpose registers by writing a known value and reading it back using vfp_put_double() and vfp_get_double() APIs.
long long d1=777, d2=0;
int i=31;
for (; i>=0; i--){
vfp_put_double(d1, i);
d2 = vfp_get_double(i);
printk("D%d read=%lld\n",i, (long long)d2);
d2 = 0.0;
}1) With the existing implementation I am not able to correctly write/read {d0-d15} but not the {d16-d31} set
2) With my changes I am able to write/read correctly.
Tarun
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Quoted 4 lines · Tarun Kanti
1) With the existing implementation I am not able to correctly write/read {d0-d15} but not the {d16-d31} set
2) With my changes I am able to write/read correctly.
The reason this happens is simple. In vfp_get_double(), we have:
1: fmrrd r0, r1, d\dr 1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr
but in vfp_put_double(), we have:
1: fmdrr d\dr, r0, r1 1: mcrr p11, 3, r1, r2, c\dr @ fmdrr r1, r2, d\dr
Note the different registers. Change "r1, r2" to "r0, r1" and it should work.
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Russell,
Quoted 18 lines
-----Original Message-----
From: Russell King - ARM Linux [mailto:linux at arm.linux.org.uk]
Sent: Thursday, May 27, 2010 1:40 AM
To: DebBarma, Tarun Kanti
Cc: linux-omap at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH] ARM:VFPv3:enable {d16-d31} accessQuoted 4 lines · Tarun Kanti
1) With the existing implementation I am not able to correctly write/read {d0-d15} but not the {d16-d31} set
2) With my changes I am able to write/read correctly.
The reason this happens is simple. In
vfp_get_double(), we have:1: fmrrd r0, r1, d\dr 1: mrrc p11, 3, r0, r1, c\dr @ fmrrd r0, r1, d\dr
Quoted 7 lines
but in
vfp_put_double(), we have:1: fmdrr d\dr, r0, r1 1: mcrr p11, 3, r1, r2, c\dr @ fmdrr r1, r2, d\drNote the different registers. Change "r1, r2" to "r0, r1" and it should work.
Yes, that's right. I figured that out yesterday and confirmed the test results. Thanks!
Can I go ahead with next version of the patch?
BR
Tarun
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Quoted 5 lines · Tarun Kanti
Quoted 2 lines
Note the different registers. Change "r1, r2" to "r0, r1" and it should work.
Yes, that's right. I figured that out yesterday and confirmed the test results. Thanks!
I've now committed a fix to change those registers. Are you ok with being credited with a Reported-by tag?
PATCH 1/1 ARM:VFPv3:enable {d16-d31} access
Quoted 9 lines
-----Original Message-----
From: Russell King - ARM Linux [mailto:linux at arm.linux.org.uk]
Sent: Thursday, May 27, 2010 3:00 PM
To: DebBarma, Tarun Kanti
Cc: linux-omap at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH] ARM:VFPv3:enable {d16-d31} accessOn Thu, May 27, 2010 at 10:07:40AM +0530, DebBarma, Tarun Kanti wrote:
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
Note the different registers. Change "r1, r2" to "r0, r1" and it should work.
Yes, that's right. I figured that out yesterday and confirmed the test results. Thanks!
I've now committed a fix to change those registers. Are you ok with being credited with a Reported-by tag?
That's fine!