linux-rdma.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
To: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
	Mike Marciniszyn
	<mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>,
	Sebastian Sanchez
	<sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Subject: [PATCH v2 04/15] IB/hfi1: Get rid of divide when setting the tx request header
Date: Thu, 04 May 2017 05:14:16 -0700	[thread overview]
Message-ID: <20170504121415.32747.51843.stgit@scvm10.sc.intel.com> (raw)
In-Reply-To: <20170504120126.32747.45131.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>

From: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>

Div instructions show costly in profiles when
the tx request header is set. Using right shift
instead of a divide operation reduces the cycles
spent in the function that sets the tx request
header as shown in the profile. Use right shift
operation instead.

Profile before change:
43.24%  009
|
|--23.41%-- user_sdma_send_pkts
|          |
|          |--99.90%-- hfi1_user_sdma_process_requestAfter:

Profile after change:
45.75%  009
|
|--14.81%-- user_sdma_send_pkts
|          |
|          |--99.95%-- hfi1_user_sdma_process_request

Reviewed-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
 drivers/infiniband/hw/hfi1/user_sdma.c |   34 ++++++++++++++++----------------
 1 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c
index 0749689..8adb6df 100644
--- a/drivers/infiniband/hw/hfi1/user_sdma.c
+++ b/drivers/infiniband/hw/hfi1/user_sdma.c
@@ -143,7 +143,9 @@
 
 /* KDETH OM multipliers and switch over point */
 #define KDETH_OM_SMALL     4
+#define KDETH_OM_SMALL_SHIFT     2
 #define KDETH_OM_LARGE     64
+#define KDETH_OM_LARGE_SHIFT     6
 #define KDETH_OM_MAX_SIZE  (1 << ((KDETH_OM_LARGE / KDETH_OM_SMALL) + 1))
 
 /* Tx request flag bits */
@@ -229,12 +231,6 @@ struct user_sdma_request {
 	 */
 	u32 tidoffset;
 	/*
-	 * KDETH.OM
-	 * Remember this because the header template always sets it
-	 * to 0.
-	 */
-	u8 omfactor;
-	/*
 	 * We copy the iovs for this request (based on
 	 * info.iovcnt). These are only the data vectors
 	 */
@@ -1323,6 +1319,7 @@ static int set_txreq_header(struct user_sdma_request *req,
 {
 	struct hfi1_user_sdma_pkt_q *pq = req->pq;
 	struct hfi1_pkt_header *hdr = &tx->hdr;
+	u8 omfactor; /* KDETH.OM */
 	u16 pbclen;
 	int ret;
 	u32 tidval = 0, lrhlen = get_lrh_len(*hdr, pad_len(datalen));
@@ -1400,8 +1397,9 @@ static int set_txreq_header(struct user_sdma_request *req,
 			}
 			tidval = req->tids[req->tididx];
 		}
-		req->omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
-			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE : KDETH_OM_SMALL;
+		omfactor = EXP_TID_GET(tidval, LEN) * PAGE_SIZE >=
+			KDETH_OM_MAX_SIZE ? KDETH_OM_LARGE_SHIFT :
+			KDETH_OM_SMALL_SHIFT;
 		/* Set KDETH.TIDCtrl based on value for this TID. */
 		KDETH_SET(hdr->kdeth.ver_tid_offset, TIDCTRL,
 			  EXP_TID_GET(tidval, CTRL));
@@ -1416,12 +1414,12 @@ static int set_txreq_header(struct user_sdma_request *req,
 		 * transfer.
 		 */
 		SDMA_DBG(req, "TID offset %ubytes %uunits om%u",
-			 req->tidoffset, req->tidoffset / req->omfactor,
-			 req->omfactor != KDETH_OM_SMALL);
+			 req->tidoffset, req->tidoffset >> omfactor,
+			 omfactor != KDETH_OM_SMALL_SHIFT);
 		KDETH_SET(hdr->kdeth.ver_tid_offset, OFFSET,
-			  req->tidoffset / req->omfactor);
+			  req->tidoffset >> omfactor);
 		KDETH_SET(hdr->kdeth.ver_tid_offset, OM,
-			  req->omfactor != KDETH_OM_SMALL);
+			  omfactor != KDETH_OM_SMALL_SHIFT);
 	}
 done:
 	trace_hfi1_sdma_user_header(pq->dd, pq->ctxt, pq->subctxt,
@@ -1433,6 +1431,7 @@ static int set_txreq_header_ahg(struct user_sdma_request *req,
 				struct user_sdma_txreq *tx, u32 len)
 {
 	int diff = 0;
+	u8 omfactor; /* KDETH.OM */
 	struct hfi1_user_sdma_pkt_q *pq = req->pq;
 	struct hfi1_pkt_header *hdr = &req->hdr;
 	u16 pbclen = le16_to_cpu(hdr->pbc[0]);
@@ -1484,14 +1483,15 @@ static int set_txreq_header_ahg(struct user_sdma_request *req,
 			}
 			tidval = req->tids[req->tididx];
 		}
-		req->omfactor = ((EXP_TID_GET(tidval, LEN) *
+		omfactor = ((EXP_TID_GET(tidval, LEN) *
 				  PAGE_SIZE) >=
-				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE :
-			KDETH_OM_SMALL;
+				 KDETH_OM_MAX_SIZE) ? KDETH_OM_LARGE_SHIFT :
+				 KDETH_OM_SMALL_SHIFT;
 		/* KDETH.OM and KDETH.OFFSET (TID) */
 		AHG_HEADER_SET(req->ahg, diff, 7, 0, 16,
-			       ((!!(req->omfactor - KDETH_OM_SMALL)) << 15 |
-				((req->tidoffset / req->omfactor) & 0x7fff)));
+			       ((!!(omfactor - KDETH_OM_SMALL_SHIFT)) << 15 |
+				((req->tidoffset >> omfactor)
+				 & 0x7fff)));
 		/* KDETH.TIDCtrl, KDETH.TID, KDETH.Intr, KDETH.SH */
 		val = cpu_to_le16(((EXP_TID_GET(tidval, CTRL) & 0x3) << 10) |
 				   (EXP_TID_GET(tidval, IDX) & 0x3ff));

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

  parent reply	other threads:[~2017-05-04 12:14 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-05-04 12:13 [PATCH v2 00/15] IB/hfi1: hfi1 driver patches for-next Dennis Dalessandro
     [not found] ` <20170504120126.32747.45131.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-05-04 12:13   ` [PATCH v2 01/15] IB/hfi1: Fix checks for Offline transient state Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 02/15] IB/hfi1, IB/rdmavt: Move r_adefered to r_lock cache line Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 03/15] IB/hfi1: Fix yield logic in send engine Dennis Dalessandro
2017-05-04 12:14   ` Dennis Dalessandro [this message]
2017-05-04 12:14   ` [PATCH v2 05/15] IB/hfi1: Adjust default eager_buffer_size to 8MB Dennis Dalessandro
     [not found]     ` <20170504121421.32747.27378.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-05-04 12:42       ` Leon Romanovsky
2017-05-04 12:14   ` [PATCH v2 06/15] IB/hfi1: Return an error on memory allocation failure Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 07/15] IB/hfi1: Fix a subcontext memory leak Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 08/15] IB/hfi1: Name function prototype parameters Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 09/15] IB/hfi1: Use filedata rather than filepointer Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 10/15] IB/hfi1: Remove atomic operations for SDMA_REQ_HAVE_AHG bit Dennis Dalessandro
2017-05-04 12:14   ` [PATCH v2 11/15] IB/hfi1: Search shared contexts on the opened device, not all devices Dennis Dalessandro
2017-05-04 12:15   ` [PATCH v2 12/15] IB/hfi1: Correctly clear the pkey Dennis Dalessandro
2017-05-04 12:15   ` [PATCH v2 13/15] IB/hfi1: Clean up context initialization Dennis Dalessandro
2017-05-04 12:15   ` [PATCH v2 14/15] IB/hfi1: Fix an assign/ordering issue with shared context IDs Dennis Dalessandro
2017-05-04 12:15   ` [PATCH v2 15/15] IB/hfi1: Clean up on context initialization failure Dennis Dalessandro
2017-05-04 23:45   ` [PATCH v2 00/15] IB/hfi1: hfi1 driver patches for-next Doug Ledford

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=20170504121415.32747.51843.stgit@scvm10.sc.intel.com \
    --to=dennis.dalessandro-ral2jqcrhueavxtiumwx3w@public.gmane.org \
    --cc=dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org \
    --cc=linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org \
    --cc=mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org \
    --cc=sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.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).