DPDK patches and discussions
 help / color / mirror / Atom feed
From: Kamil Chalupnik <kamilx.chalupnik@intel.com>
To: dev@dpdk.org
Cc: amr.mokhtar@intel.com, KamilX Chalupnik <kamilx.chalupnik@intel.com>
Subject: [dpdk-dev] [PATCH v2 02/14] baseband/turbo_sw: memory copying optimized or removed
Date: Wed,  9 May 2018 16:17:56 +0200	[thread overview]
Message-ID: <20180509141756.23604-1-kamilx.chalupnik@intel.com> (raw)
In-Reply-To: <20180426133008.12388-4-kamilx.chalupnik@intel.com>

From: KamilX Chalupnik <kamilx.chalupnik@intel.com>

Optimization of Turbo Software driver by limiting or changing
usage of memory copying function.

Signed-off-by: Kamil Chalupnik <kamilx.chalupnik@intel.com>
---
 drivers/baseband/turbo_sw/bbdev_turbo_software.c | 100 +++++++++++++++--------
 1 file changed, 64 insertions(+), 36 deletions(-)

diff --git a/drivers/baseband/turbo_sw/bbdev_turbo_software.c b/drivers/baseband/turbo_sw/bbdev_turbo_software.c
index 26b8560..35d2827 100644
--- a/drivers/baseband/turbo_sw/bbdev_turbo_software.c
+++ b/drivers/baseband/turbo_sw/bbdev_turbo_software.c
@@ -452,7 +452,7 @@ is_dec_input_valid(int32_t k_idx, int16_t kw, int16_t in_length)
 
 static inline void
 process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
-		uint8_t cb_idx, uint8_t c, uint16_t k, uint16_t ncb,
+		uint8_t r, uint8_t c, uint16_t k, uint16_t ncb,
 		uint32_t e, struct rte_mbuf *m_in, struct rte_mbuf *m_out,
 		uint16_t in_offset, uint16_t out_offset, uint16_t total_left)
 {
@@ -460,6 +460,7 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 	int16_t k_idx;
 	uint16_t m;
 	uint8_t *in, *out0, *out1, *out2, *tmp_out, *rm_out;
+	uint64_t first_3_bytes = 0;
 	struct rte_bbdev_op_turbo_enc *enc = &op->turbo_enc;
 	struct bblib_crc_request crc_req;
 	struct bblib_crc_response crc_resp;
@@ -479,16 +480,25 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
 			return;
 		}
-		/* copy the input to the temporary buffer to be able to extend
-		 * it by 3 CRC bytes
-		 */
-		rte_memcpy(q->enc_in, in, (k - 24) >> 3);
 		crc_req.data = in;
 		crc_req.len = (k - 24) >> 3;
-		crc_resp.data = q->enc_in;
-		bblib_lte_crc24a_gen(&crc_req, &crc_resp);
+		/* Check if there is a room for CRC bits. If not use
+		 * the temporary buffer.
+		 */
+		if (rte_pktmbuf_append(m_in, 3) == NULL) {
+			rte_memcpy(q->enc_in, in, (k - 24) >> 3);
+			in = q->enc_in;
+		} else {
+			/* Store 3 first bytes of next CB as they will be
+			 * overwritten by CRC bytes. If it is the last CB then
+			 * there is no point to store 3 next bytes and this
+			 * if..else branch will be omitted.
+			 */
+			first_3_bytes = *((uint64_t *)&in[(k - 32) >> 3]);
+		}
 
-		in = q->enc_in;
+		crc_resp.data = in;
+		bblib_lte_crc24a_gen(&crc_req, &crc_resp);
 	} else if (enc->op_flags & RTE_BBDEV_TURBO_CRC_24B_ATTACH) {
 		/* CRC24B */
 		ret = is_enc_input_valid(k - 24, k_idx, total_left);
@@ -496,16 +506,25 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
 			return;
 		}
-		/* copy the input to the temporary buffer to be able to extend
-		 * it by 3 CRC bytes
-		 */
-		rte_memcpy(q->enc_in, in, (k - 24) >> 3);
 		crc_req.data = in;
 		crc_req.len = (k - 24) >> 3;
-		crc_resp.data = q->enc_in;
-		bblib_lte_crc24b_gen(&crc_req, &crc_resp);
+		/* Check if there is a room for CRC bits. If this is the last
+		 * CB in TB. If not use temporary buffer.
+		 */
+		if ((c - r == 1) && (rte_pktmbuf_append(m_in, 3) == NULL)) {
+			rte_memcpy(q->enc_in, in, (k - 24) >> 3);
+			in = q->enc_in;
+		} else if (c - r > 1) {
+			/* Store 3 first bytes of next CB as they will be
+			 * overwritten by CRC bytes. If it is the last CB then
+			 * there is no point to store 3 next bytes and this
+			 * if..else branch will be omitted.
+			 */
+			first_3_bytes = *((uint64_t *)&in[(k - 32) >> 3]);
+		}
 
-		in = q->enc_in;
+		crc_resp.data = in;
+		bblib_lte_crc24b_gen(&crc_req, &crc_resp);
 	} else {
 		ret = is_enc_input_valid(k, k_idx, total_left);
 		if (ret != 0) {
@@ -519,10 +538,32 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 	/* Each bit layer output from turbo encoder is (k+4) bits long, i.e.
 	 * input length + 4 tail bits. That's (k/8) + 1 bytes after rounding up.
 	 * So dst_data's length should be 3*(k/8) + 3 bytes.
+	 * In Rate-matching bypass case outputs pointers passed to encoder
+	 * (out0, out1 and out2) can directly point to addresses of output from
+	 * turbo_enc entity.
 	 */
-	out0 = q->enc_out;
-	out1 = RTE_PTR_ADD(out0, (k >> 3) + 1);
-	out2 = RTE_PTR_ADD(out1, (k >> 3) + 1);
+	if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH) {
+		out0 = q->enc_out;
+		out1 = RTE_PTR_ADD(out0, (k >> 3) + 1);
+		out2 = RTE_PTR_ADD(out1, (k >> 3) + 1);
+	} else {
+		out0 = (uint8_t *)rte_pktmbuf_append(m_out, (k >> 3) * 3 + 2);
+		if (out0 == NULL) {
+			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
+			rte_bbdev_log(ERR,
+					"Too little space in output mbuf");
+			return;
+		}
+		enc->output.length += (k >> 3) * 3 + 2;
+		/* rte_bbdev_op_data.offset can be different than the
+		 * offset of the appended bytes
+		 */
+		out0 = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
+		out1 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
+				out_offset + (k >> 3) + 1);
+		out2 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
+				out_offset + 2 * ((k >> 3) + 1));
+	}
 
 	turbo_req.case_id = k_idx;
 	turbo_req.input_win = in;
@@ -536,6 +577,10 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 		return;
 	}
 
+	/* Restore 3 first bytes of next CB if they were overwritten by CRC*/
+	if (first_3_bytes != 0)
+		*((uint64_t *)&in[(k - 32) >> 3]) = first_3_bytes;
+
 	/* Rate-matching */
 	if (enc->op_flags & RTE_BBDEV_TURBO_RATE_MATCH) {
 		/* get output data starting address */
@@ -552,7 +597,7 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 		rm_out = rte_pktmbuf_mtod_offset(m_out, uint8_t *, out_offset);
 
 		/* index of current code block */
-		rm_req.r = cb_idx;
+		rm_req.r = r;
 		/* total number of code block */
 		rm_req.C = c;
 		/* For DL - 1, UL - 0 */
@@ -613,23 +658,6 @@ process_enc_cb(struct turbo_sw_queue *q, struct rte_bbdev_enc_op *op,
 			tmp_out++;
 		}
 		*tmp_out = 0;
-
-		/* copy shifted output to turbo_enc entity */
-		out0 = (uint8_t *)rte_pktmbuf_append(m_out,
-				(k >> 3) * 3 + 2);
-		if (out0 == NULL) {
-			op->status |= 1 << RTE_BBDEV_DATA_ERROR;
-			rte_bbdev_log(ERR,
-					"Too little space in output mbuf");
-			return;
-		}
-		enc->output.length += (k >> 3) * 3 + 2;
-		/* rte_bbdev_op_data.offset can be different than the
-		 * offset of the appended bytes
-		 */
-		out0 = rte_pktmbuf_mtod_offset(m_out, uint8_t *,
-				out_offset);
-		rte_memcpy(out0, q->enc_out, (k >> 3) * 3 + 2);
 	}
 }
 
-- 
2.5.5

  parent reply	other threads:[~2018-05-09 14:18 UTC|newest]

Thread overview: 37+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-04-26 13:29 [dpdk-dev] [PATCH 01/13] baseband/turbo_sw: update DPDK to work with FlexRAN 1.4.0 Kamil Chalupnik
2018-04-26 13:29 ` [dpdk-dev] [PATCH 02/13] doc/turbo_sw: update Wireless Baseband Device documentation Kamil Chalupnik
2018-05-07 13:37   ` De Lara Guarch, Pablo
2018-05-09 14:46   ` [dpdk-dev] [PATCH v2 12/14] " Kamil Chalupnik
2018-04-26 13:29 ` [dpdk-dev] [PATCH 03/13] doc/bbdev: dynamic lib support Kamil Chalupnik
2018-05-09 14:51   ` [dpdk-dev] [PATCH v2 13/14] " Kamil Chalupnik
2018-04-26 13:29 ` [dpdk-dev] [PATCH 04/13] baseband/turbo_sw: memcpy changed or removed from driver Kamil Chalupnik
2018-05-07 12:26   ` De Lara Guarch, Pablo
2018-05-09 14:17   ` Kamil Chalupnik [this message]
2018-04-26 13:29 ` [dpdk-dev] [PATCH 05/13] baseband/turbo_sw: scalling input LLR to range [-16 16] Kamil Chalupnik
2018-05-07 12:50   ` De Lara Guarch, Pablo
2018-05-09 14:23   ` [dpdk-dev] [PATCH v2 04/14] baseband/turbo_sw: scalling likelihood ratio (LLR) input Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 06/13] baseband/turbo_sw: increase internal buffers Kamil Chalupnik
2018-05-09 14:25   ` [dpdk-dev] [PATCH v2 05/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 07/13] baseband/turbo_sw: support for optional CRC overlap Kamil Chalupnik
2018-05-09 14:28   ` [dpdk-dev] [PATCH v2 06/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 08/13] app/bbdev: update test vectors names Kamil Chalupnik
2018-05-07 13:15   ` De Lara Guarch, Pablo
2018-05-09 14:37   ` [dpdk-dev] [PATCH v2 09/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 09/13] bbdev: measure offload cost Kamil Chalupnik
2018-05-07 13:29   ` De Lara Guarch, Pablo
     [not found]     ` <EEA9FF629BF25B47BD67ADE995041EE23D0352A7@IRSMSX103.ger.corp.intel.com>
2018-05-08  9:08       ` De Lara Guarch, Pablo
     [not found]         ` <EEA9FF629BF25B47BD67ADE995041EE23D035350@IRSMSX103.ger.corp.intel.com>
2018-05-08 10:16           ` De Lara Guarch, Pablo
2018-05-09 14:30   ` [dpdk-dev] [PATCH v2 07/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 10/13] doc: update tests and usage of test app description Kamil Chalupnik
2018-05-09 14:55   ` [dpdk-dev] [PATCH v2 14/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 11/13] app/bbdev: added new test vectors Kamil Chalupnik
2018-05-09 14:39   ` [dpdk-dev] [PATCH v2 10/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 12/13] bbdev: split queue groups Kamil Chalupnik
2018-05-09 14:35   ` [dpdk-dev] [PATCH v2 08/14] " Kamil Chalupnik
2018-04-26 13:30 ` [dpdk-dev] [PATCH 13/13] app/bbdev: improve readability of test application Kamil Chalupnik
2018-05-09 14:42   ` [dpdk-dev] [PATCH v2 11/14] " Kamil Chalupnik
2018-05-09 18:55     ` De Lara Guarch, Pablo
2018-04-26 13:30 ` [dpdk-dev] [PATCH 00/13] Documentation and Turbo Software Baseband Device Update Kamil Chalupnik
2018-05-09 14:14 ` [dpdk-dev] [PATCH v2 01/14] baseband/turbo_sw: update DPDK to work with FlexRAN 1.4.0 Kamil Chalupnik
2018-05-09 14:14   ` [dpdk-dev] [PATCH v2 00/14] Documentation and Turbo Software Baseband Device Update Kamil Chalupnik
2018-05-09 19:21     ` De Lara Guarch, Pablo

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20180509141756.23604-1-kamilx.chalupnik@intel.com \
    --to=kamilx.chalupnik@intel.com \
    --cc=amr.mokhtar@intel.com \
    --cc=dev@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).