* [Bluez-devel] SBC optimizations
@ 2008-02-21 22:29 Cidorvan Leite
2008-02-22 1:58 ` Brad Midgley
0 siblings, 1 reply; 5+ messages in thread
From: Cidorvan Leite @ 2008-02-21 22:29 UTC (permalink / raw)
To: bluez-devel
[-- Attachment #1: Type: text/plain, Size: 190 bytes --]
Hi!
I replaced the 64bits multiplies by 32bits and fixed the volume, now
the encode and decode are >10% faster.
Unfortunately, I couldn't keep the same volume for decode with 4 subbands...
[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: sbc.patch --]
[-- Type: text/x-patch; name=sbc.patch, Size: 4342 bytes --]
Index: sbc.c
===================================================================
RCS file: /cvsroot/bluez/utils/sbc/sbc.c,v
retrieving revision 1.61
diff -u -p -r1.61 sbc.c
--- sbc.c 19 Feb 2008 19:47:25 -0000 1.61
+++ sbc.c 21 Feb 2008 22:08:56 -0000
@@ -551,7 +551,7 @@ static inline void sbc_synthesize_four(s
struct sbc_frame *frame, int ch, int blk)
{
int i, j, k, idx;
- sbc_extended_t res;
+ sbc_fixed_t res;
for (i = 0; i < 8; i++) {
/* Shifting */
@@ -592,7 +592,7 @@ static inline void sbc_synthesize_eight(
struct sbc_frame *frame, int ch, int blk)
{
int i, j, k, idx;
- sbc_extended_t res;
+ sbc_fixed_t res;
for (i = 0; i < 16; i++) {
/* Shifting */
@@ -667,8 +667,7 @@ static void sbc_encoder_init(struct sbc_
static inline void _sbc_analyze_four(const int32_t *in, int32_t *out)
{
- sbc_fixed_t t[8];
- sbc_extended_t s[5];
+ sbc_fixed_t t[8], s[5];
t[0] = SCALE4_STAGE1( /* Q8 */
MULA(_sbc_proto_4[0], in[8] - in[32], /* Q18 */
@@ -752,8 +751,7 @@ static inline void sbc_analyze_four(stru
static inline void _sbc_analyze_eight(const int32_t *in, int32_t *out)
{
- sbc_fixed_t t[8];
- sbc_extended_t s[8];
+ sbc_fixed_t t[8], s[8];
t[0] = SCALE8_STAGE1( /* Q10 */
MULA(_sbc_proto_8[0], (in[16] - in[64]), /* Q18 = Q18 * Q0 */
Index: sbc_math.h
===================================================================
RCS file: /cvsroot/bluez/utils/sbc/sbc_math.h,v
retrieving revision 1.17
diff -u -p -r1.17 sbc_math.h
--- sbc_math.h 15 Feb 2008 18:06:32 -0000 1.17
+++ sbc_math.h 21 Feb 2008 22:08:56 -0000
@@ -28,49 +28,45 @@
always be correct and every compiler *should* generate optimal code */
#define ASR(val, bits) ((-2 >> 1 == -1) ? \
((int32_t)(val)) >> (bits) : ((int32_t) (val)) / (1 << (bits)))
-#define ASR_64(val, bits) ((-2 >> 1 == -1) ? \
- ((long long)(val)) >> (bits) : ((long long) (val)) / (1 << (bits)))
#define SCALE_PROTO4_TBL 15
#define SCALE_ANA4_TBL 16
-#define SCALE_PROTO8_TBL 15
-#define SCALE_ANA8_TBL 16
-#define SCALE_SPROTO4_TBL 16
-#define SCALE_SPROTO8_TBL 16
-#define SCALE_NPROTO4_TBL 10
-#define SCALE_NPROTO8_TBL 12
-#define SCALE_SAMPLES 14
+#define SCALE_PROTO8_TBL 16
+#define SCALE_ANA8_TBL 17
+#define SCALE_SPROTO4_TBL 15
+#define SCALE_SPROTO8_TBL 14
+#define SCALE_NPROTO4_TBL 13
+#define SCALE_NPROTO8_TBL 11
#define SCALE4_STAGE1_BITS 16
-#define SCALE4_STAGE2_BITS 16
-#define SCALE4_STAGED1_BITS 14
-#define SCALE4_STAGED2_BITS 14
-#define SCALE8_STAGE1_BITS 16
-#define SCALE8_STAGE2_BITS 16
-#define SCALE8_STAGED1_BITS 14
-#define SCALE8_STAGED2_BITS 14
+#define SCALE4_STAGE2_BITS 15
+#define SCALE4_STAGED1_BITS 12
+#define SCALE4_STAGED2_BITS 16
+#define SCALE8_STAGE1_BITS 15
+#define SCALE8_STAGE2_BITS 15
+#define SCALE8_STAGED1_BITS 15
+#define SCALE8_STAGED2_BITS 16
typedef int32_t sbc_fixed_t;
-typedef long long sbc_extended_t;
-#define SCALE4_STAGE1(src) ASR_64(src, SCALE4_STAGE1_BITS)
-#define SCALE4_STAGE2(src) ASR_64(src, SCALE4_STAGE2_BITS)
-#define SCALE4_STAGED1(src) ASR_64(src, SCALE4_STAGED1_BITS)
-#define SCALE4_STAGED2(src) ASR_64(src, SCALE4_STAGED2_BITS)
-#define SCALE8_STAGE1(src) ASR_64(src, SCALE8_STAGE1_BITS)
-#define SCALE8_STAGE2(src) ASR_64(src, SCALE8_STAGE2_BITS)
-#define SCALE8_STAGED1(src) ASR_64(src, SCALE8_STAGED1_BITS)
-#define SCALE8_STAGED2(src) ASR_64(src, SCALE8_STAGED2_BITS)
+#define SCALE4_STAGE1(src) ASR(src, SCALE4_STAGE1_BITS)
+#define SCALE4_STAGE2(src) ASR(src, SCALE4_STAGE2_BITS)
+#define SCALE4_STAGED1(src) ASR(src, SCALE4_STAGED1_BITS)
+#define SCALE4_STAGED2(src) ASR(src, SCALE4_STAGED2_BITS)
+#define SCALE8_STAGE1(src) ASR(src, SCALE8_STAGE1_BITS)
+#define SCALE8_STAGE2(src) ASR(src, SCALE8_STAGE2_BITS)
+#define SCALE8_STAGED1(src) ASR(src, SCALE8_STAGED1_BITS)
+#define SCALE8_STAGED2(src) ASR(src, SCALE8_STAGED2_BITS)
#define SBC_FIXED_0(val) { val = 0; }
-#define MUL(a, b) ((sbc_extended_t)(a) * (b))
+#define MUL(a, b) ((a) * (b))
#ifdef __arm__
#define MULA(a, b, res) ({ \
- long long tmp = res; \
+ int tmp = res; \
__asm__( \
- "smlal %Q0, %R0, %2, %3" \
+ "mla %0, %2, %3, %0" \
: "=&r" (tmp) \
: "0" (tmp), "r" (a), "r" (b)); \
tmp; })
#else
-#define MULA(a, b, res) ((sbc_extended_t)(a) * (b) + (res))
+#define MULA(a, b, res) ((a) * (b) + (res))
#endif
[-- Attachment #3: Type: text/plain, Size: 228 bytes --]
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
[-- Attachment #4: Type: text/plain, Size: 164 bytes --]
_______________________________________________
Bluez-devel mailing list
Bluez-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Bluez-devel] SBC optimizations
2008-02-21 22:29 [Bluez-devel] SBC optimizations Cidorvan Leite
@ 2008-02-22 1:58 ` Brad Midgley
2008-02-22 14:00 ` Cidorvan Leite
0 siblings, 1 reply; 5+ messages in thread
From: Brad Midgley @ 2008-02-22 1:58 UTC (permalink / raw)
To: BlueZ development
Cidorvan
How significant is the volume drop in 4 subbands? Should we just leave
the decoder at 64 bit mults so we don't needlessly suffer quality
issues on faster platforms?
Brad
On Thu, Feb 21, 2008 at 3:29 PM, Cidorvan Leite <cidorvan@gmail.com> wrote:
> Hi!
>
> I replaced the 64bits multiplies by 32bits and fixed the volume, now
> the encode and decode are >10% faster.
> Unfortunately, I couldn't keep the same volume for decode with 4 subbands...
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Bluez-devel mailing list
> Bluez-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bluez-devel
>
>
--
Brad
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bluez-devel mailing list
Bluez-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Bluez-devel] SBC optimizations
2008-02-22 1:58 ` Brad Midgley
@ 2008-02-22 14:00 ` Cidorvan Leite
2008-02-22 14:19 ` Cidorvan Leite
0 siblings, 1 reply; 5+ messages in thread
From: Cidorvan Leite @ 2008-02-22 14:00 UTC (permalink / raw)
To: BlueZ development
Hi Brad!
The volume drop is not significant.
Comparing the versions, I didn't see any change in quality with my
headphone, so I think that we can change to 32bits.
By the way, I am not sure if last version of decode with 4 subbands is
correct about overflows.
On Thu, Feb 21, 2008 at 10:58 PM, Brad Midgley <bmidgley@gmail.com> wrote:
> Cidorvan
>
> How significant is the volume drop in 4 subbands? Should we just leave
> the decoder at 64 bit mults so we don't needlessly suffer quality
> issues on faster platforms?
>
> Brad
>
>
>
> On Thu, Feb 21, 2008 at 3:29 PM, Cidorvan Leite <cidorvan@gmail.com> wrote:
> > Hi!
> >
> > I replaced the 64bits multiplies by 32bits and fixed the volume, now
> > the encode and decode are >10% faster.
> > Unfortunately, I couldn't keep the same volume for decode with 4 subbands...
> >
> > -------------------------------------------------------------------------
> > This SF.net email is sponsored by: Microsoft
> > Defy all challenges. Microsoft(R) Visual Studio 2008.
> > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> > _______________________________________________
> > Bluez-devel mailing list
> > Bluez-devel@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/bluez-devel
> >
> >
>
>
>
> --
> Brad
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Bluez-devel mailing list
> Bluez-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bluez-devel
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bluez-devel mailing list
Bluez-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Bluez-devel] SBC optimizations
2008-02-22 14:00 ` Cidorvan Leite
@ 2008-02-22 14:19 ` Cidorvan Leite
2008-02-22 15:46 ` Brad Midgley
0 siblings, 1 reply; 5+ messages in thread
From: Cidorvan Leite @ 2008-02-22 14:19 UTC (permalink / raw)
To: BlueZ development
Hi Brad!
I would say:
"I am not sure if last version(64bits) of decode with 4 subbands is
correct about overflows."
On Fri, Feb 22, 2008 at 11:00 AM, Cidorvan Leite <cidorvan@gmail.com> wrote:
> Hi Brad!
>
> The volume drop is not significant.
> Comparing the versions, I didn't see any change in quality with my
> headphone, so I think that we can change to 32bits.
> By the way, I am not sure if last version of decode with 4 subbands is
> correct about overflows.
>
>
>
> On Thu, Feb 21, 2008 at 10:58 PM, Brad Midgley <bmidgley@gmail.com> wrote:
> > Cidorvan
> >
> > How significant is the volume drop in 4 subbands? Should we just leave
> > the decoder at 64 bit mults so we don't needlessly suffer quality
> > issues on faster platforms?
> >
> > Brad
> >
> >
> >
> > On Thu, Feb 21, 2008 at 3:29 PM, Cidorvan Leite <cidorvan@gmail.com> wrote:
> > > Hi!
> > >
> > > I replaced the 64bits multiplies by 32bits and fixed the volume, now
> > > the encode and decode are >10% faster.
> > > Unfortunately, I couldn't keep the same volume for decode with 4 subbands...
> > >
> > > -------------------------------------------------------------------------
> > > This SF.net email is sponsored by: Microsoft
> > > Defy all challenges. Microsoft(R) Visual Studio 2008.
> > > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> > > _______________________________________________
> > > Bluez-devel mailing list
> > > Bluez-devel@lists.sourceforge.net
> > > https://lists.sourceforge.net/lists/listinfo/bluez-devel
> > >
> > >
> >
> >
> >
> > --
> > Brad
> >
> > -------------------------------------------------------------------------
> > This SF.net email is sponsored by: Microsoft
> > Defy all challenges. Microsoft(R) Visual Studio 2008.
> > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> > _______________________________________________
> > Bluez-devel mailing list
> > Bluez-devel@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/bluez-devel
> >
>
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bluez-devel mailing list
Bluez-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Bluez-devel] SBC optimizations
2008-02-22 14:19 ` Cidorvan Leite
@ 2008-02-22 15:46 ` Brad Midgley
0 siblings, 0 replies; 5+ messages in thread
From: Brad Midgley @ 2008-02-22 15:46 UTC (permalink / raw)
To: BlueZ development
Cidorvan
> "I am not sure if last version(64bits) of decode with 4 subbands is
> correct about overflows."
Some volume reduction is ok, especially if it means we avoid another overflow.
Anyway, it looks like the commit is in.
Thanks again.
--
Brad
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bluez-devel mailing list
Bluez-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bluez-devel
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2008-02-22 15:46 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2008-02-21 22:29 [Bluez-devel] SBC optimizations Cidorvan Leite
2008-02-22 1:58 ` Brad Midgley
2008-02-22 14:00 ` Cidorvan Leite
2008-02-22 14:19 ` Cidorvan Leite
2008-02-22 15:46 ` Brad Midgley
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox