[PATCH] sbc: bitstream packing optimization for encoder.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi All,

Appears that bitstream packing is also one of the most CPU hungry
operations in SBC encoder, which did not get proper attention yet.
Could somebody review this patch and provide feedback/comments?

Benchmarks were done by running the following script:

./sbcenc -s 8 big_buck_bunny_480p_mono.au > /dev/null
./sbcenc -s 8 big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -s 8 -j big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -s 4 big_buck_bunny_480p_mono.au > /dev/null
./sbcenc -s 4 big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -s 4 -j big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -b 128 -s 8 big_buck_bunny_480p_mono.au > /dev/null
./sbcenc -b 128 -s 8 big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -b 128 -s 8 -j big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -b 64 -s 4 big_buck_bunny_480p_mono.au > /dev/null
./sbcenc -b 64 -s 4 big_buck_bunny_480p_stereo.au > /dev/null
./sbcenc -b 64 -s 4 -j big_buck_bunny_480p_stereo.au > /dev/null

Times from ARM device (ARM11):

before patch:
real    9m 42.98s
user    8m 51.06s
sys     0m 39.50s

after patch:
real    6m 35.55s
user    5m 44.09s
sys     0m 40.43s

Times from x86 PC (Intel Core2):

before patch:
real    0m49.437s
user    0m45.267s
sys     0m3.708s

after patch:
real    0m27.606s
user    0m23.869s
sys     0m3.656s

 
Best regards,
Siarhei Siamashka
>From cec91fe56d834109db6280330085832c79ded6b6 Mon Sep 17 00:00:00 2001
From: Siarhei Siamashka <siarhei.siamashka@xxxxxxxxx>
Date: Thu, 11 Dec 2008 21:21:28 +0200
Subject: [PATCH] sbc: bitstream writing optimization for encoder.

SBC encoder performance improvement up to 1.5x for ARM11
and almost twice faster for Intel Core2 in some cases.

Signed-off-by: Siarhei Siamashka <siarhei.siamashka@xxxxxxxxx>
---
 sbc/sbc.c |   65 ++++++++++++++++++++++++++++++++++--------------------------
 1 files changed, 37 insertions(+), 28 deletions(-)

diff --git a/sbc/sbc.c b/sbc/sbc.c
index 7072673..9b676e0 100644
--- a/sbc/sbc.c
+++ b/sbc/sbc.c
@@ -909,6 +909,29 @@ static int sbc_analyze_audio(struct sbc_encoder_state *state,
 	}
 }
 
+/* Supplementary bitstream writing macros for 'sbc_pack_frame' */
+
+#define PUT_BITS(v, n)\
+	{\
+		bits_cache = (v) | (bits_cache << (n));\
+		bits_count += (n);\
+		if (bits_count >= 16) {\
+			bits_count -= 8;\
+			*data_ptr++ = (uint8_t)(bits_cache >> bits_count);\
+			bits_count -= 8;\
+			*data_ptr++ = (uint8_t)(bits_cache >> bits_count);\
+		}\
+	} while (0);\
+
+#define FLUSH_BITS()\
+	while (bits_count >= 8) {\
+		bits_count -= 8;\
+		*data_ptr++ = (uint8_t)(bits_cache >> bits_count);\
+	}\
+	if (bits_count > 0) {\
+	    *data_ptr++ = (uint8_t)(bits_cache << (8 - bits_count));\
+	}\
+
 /*
  * Packs the SBC frame from frame into the memory at data. At most len
  * bytes will be used, should more memory be needed an appropriate
@@ -926,14 +949,18 @@ static int sbc_analyze_audio(struct sbc_encoder_state *state,
 
 static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 {
-	int produced;
+	/* Bitstream writer starts from the fourth byte */
+	uint8_t *data_ptr = data + 4;
+	uint32_t bits_cache = 0;
+	uint32_t bits_count = 0;
+
 	/* Will copy the header parts for CRC-8 calculation here */
 	uint8_t crc_header[11] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
 	int crc_pos = 0;
 
 	uint16_t audio_sample;
 
-	int ch, sb, blk, bit;	/* channel, subband, block and bit counters */
+	int ch, sb, blk;	/* channel, subband, block and bit counters */
 	int bits[2][8];		/* bits distribution */
 	int levels[2][8];	/* levels are derived from that */
 
@@ -973,8 +1000,6 @@ static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 
 	/* Can't fill in crc yet */
 
-	produced = 32;
-
 	crc_header[0] = data[1];
 	crc_header[1] = data[2];
 	crc_pos = 16;
@@ -999,6 +1024,7 @@ static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 		u_int32_t scalefactor_j[2];
 		uint8_t scale_factor_j[2];
 
+		uint8_t joint = 0;
 		frame->joint = 0;
 
 		for (sb = 0; sb < frame->subbands - 1; sb++) {
@@ -1031,6 +1057,7 @@ static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 			if ((scalefactor[0][sb] + scalefactor[1][sb]) >
 					(scalefactor_j[0] + scalefactor_j[1]) ) {
 				/* use joint stereo for this subband */
+				joint |= 1 << (frame->subbands - 1 - sb);
 				frame->joint |= 1 << sb;
 				frame->scale_factor[0][sb] = scale_factor_j[0];
 				frame->scale_factor[1][sb] = scale_factor_j[1];
@@ -1045,24 +1072,16 @@ static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 			}
 		}
 
-		data[4] = 0;
-		for (sb = 0; sb < frame->subbands - 1; sb++)
-			data[4] |= ((frame->joint >> sb) & 0x01) << (frame->subbands - 1 - sb);
-
-		crc_header[crc_pos >> 3] = data[4];
-
-		produced += frame->subbands;
+		PUT_BITS(joint, frame->subbands);
+		crc_header[crc_pos >> 3] = joint;
 		crc_pos += frame->subbands;
 	}
 
 	for (ch = 0; ch < frame->channels; ch++) {
 		for (sb = 0; sb < frame->subbands; sb++) {
-			data[produced >> 3] <<= 4;
+			PUT_BITS(frame->scale_factor[ch][sb] & 0x0F, 4);
 			crc_header[crc_pos >> 3] <<= 4;
-			data[produced >> 3] |= frame->scale_factor[ch][sb] & 0x0F;
 			crc_header[crc_pos >> 3] |= frame->scale_factor[ch][sb] & 0x0F;
-
-			produced += 4;
 			crc_pos += 4;
 		}
 	}
@@ -1088,25 +1107,15 @@ static int sbc_pack_frame(uint8_t *data, struct sbc_frame *frame, size_t len)
 						(uint16_t) ((((frame->sb_sample_f[blk][ch][sb]*levels[ch][sb]) >>
 									(frame->scale_factor[ch][sb] + 1)) +
 								levels[ch][sb]) >> 1);
-					audio_sample <<= 16 - bits[ch][sb];
-					for (bit = 0; bit < bits[ch][sb]; bit++) {
-						data[produced >> 3] <<= 1;
-						if (audio_sample & 0x8000)
-							data[produced >> 3] |= 0x1;
-						audio_sample <<= 1;
-						produced++;
-					}
+					PUT_BITS(audio_sample, bits[ch][sb]);
 				}
 			}
 		}
 	}
 
-	/* align the last byte */
-	if (produced % 8) {
-		data[produced >> 3] <<= 8 - (produced % 8);
-	}
+	FLUSH_BITS();
 
-	return (produced + 7) >> 3;
+	return data_ptr - data;
 }
 
 struct sbc_priv {
-- 
1.5.6.5


[Index of Archives]     [Bluez Devel]     [Linux Wireless Networking]     [Linux Wireless Personal Area Networking]     [Linux ATH6KL]     [Linux USB Devel]     [Linux Media Drivers]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [Big List of Linux Books]

  Powered by Linux