public inbox for linux-snps-arc@lists.infradead.org
 help / color / mirror / Atom feed
From: Sergey Matyukevich <geomatsi@gmail.com>
To: linux-snps-arc@lists.infradead.org
Cc: Vineet Gupta <vgupta@kernel.org>,
	Vladimir Isaev <isaev@synopsys.com>,
	Sergey Matyukevich <geomatsi@gmail.com>,
	Sergey Matyukevich <sergey.matyukevich@synopsys.com>
Subject: [RFC PATCH 07/13] ARCv2: memset: rewrite using double load/stores
Date: Tue, 22 Feb 2022 17:15:00 +0300	[thread overview]
Message-ID: <20220222141506.4003433-8-geomatsi@gmail.com> (raw)
In-Reply-To: <20220222141506.4003433-1-geomatsi@gmail.com>

From: Vineet Gupta <vgupta@kernel.org>

Signed-off-by: Vineet Gupta <vgupta@kernel.org>
---
 arch/arc/lib/memset-archs.S | 112 ++++++++++++++----------------------
 1 file changed, 43 insertions(+), 69 deletions(-)

diff --git a/arch/arc/lib/memset-archs.S b/arch/arc/lib/memset-archs.S
index 330e22f7cf3c..a9a0ccef761d 100644
--- a/arch/arc/lib/memset-archs.S
+++ b/arch/arc/lib/memset-archs.S
@@ -5,6 +5,7 @@
 
 #include <linux/linkage.h>
 #include <asm/cache.h>
+#include <asm/assembler.h>
 
 /*
  * The memset implementation below is optimized to use prefetchw and prealloc
@@ -55,7 +56,7 @@ ENTRY_CFI(memset)
 1:
 #endif
 
-;;; Destination is aligned
+	; promote memset pattern from char to int (double actually for STD)
 	and	r1, r1, 0xFF
 	asl	r4, r1, 8
 	or	r4, r4, r1
@@ -63,75 +64,48 @@ ENTRY_CFI(memset)
 	or	r5, r5, r4
 	mov	r4, r5
 
-	sub3	lp_count, r2, 8
-	cmp     r2, 64
-	bmsk.hi	r2, r2, 5
-	mov.ls	lp_count, 0
-	add3.hi	r2, r2, 8
-
-;;; Convert len to Dwords, unfold x8
-	lsr.f	lp_count, lp_count, 6
-
-	lpnz	@.Lset64bytes
-	;; LOOP START
-	PREALLOC_INSTR	r3, 64	; alloc next line w/o fetching
-
-#ifdef CONFIG_ARC_HAS_LL64
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-#else
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-#endif
-.Lset64bytes:
-
-	lsr.f	lp_count, r2, 5 ;Last remaining  max 124 bytes
-	lpnz	.Lset32bytes
-	;; LOOP START
-#ifdef CONFIG_ARC_HAS_LL64
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-	std.ab	r4, [r3, 8]
-#else
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-	st.ab	r4, [r3, 4]
-#endif
-.Lset32bytes:
-
-	and.f	lp_count, r2, 0x1F ;Last remaining 31 bytes
-.Lsmallchunk:
-	lpnz	.Lcopy3bytes
-	;; LOOP START
+	; Loop #a:
+	; - Updates 1 cache line worth data (64 bytes) per iteration
+	; - PREALLOC the next line.
+	;
+	; = Only entered if at least 2 lines worth of work (i.e. >= 128 bytes),
+	;   else PREALLOC for next can "bleed" past end of buffer, causing data
+	;   corruption issue if that line is owned by some other core.
+	; = Last 64 bytes (even for min 128 bytes work) are NOT done here to
+	;   avoid PREALLOC issue
+
+	sub     r6, r2, 64
+	cmp	r2, 64
+	bmsk.hi	r2, r2, 5	; trailing 63 bytes
+	mov.ls	r6, 0
+	add.hi	r2, r2, 64	; line skipped in loop below
+
+	lsr.f	lp_count, r6, 6
+	lpnz	2f
+	PREALLOCR r3, 64
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+2:
+	; Loop #b: Remaining 32 / 64 bytes
+	lsr.f	lp_count, r2, 5
+	lpnz	.Lbyteloop
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+	ST64.ab	r4, r3, 8
+
+.Lbyteloop:
+	; Loop #c: straggler 31 bytes
+	and.f	lp_count, r2, 0x1F
+	lpnz	4f
 	stb.ab	r1, [r3, 1]
-.Lcopy3bytes:
-
+4:
 	j	[blink]
 
 END_CFI(memset)
-- 
2.25.1


_______________________________________________
linux-snps-arc mailing list
linux-snps-arc@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-snps-arc

  parent reply	other threads:[~2022-02-22 14:15 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-02-22 14:14 [RFC PATCH 00/13] ARC: handle the lack of ZOL support Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 01/13] ARC: uaccess: elide unaligned handling if hardware supports Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 02/13] ARC: Kconfig: introduce option to disable ZOL Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 03/13] ARC: uaccess: drop CC_OPTIMIZE_FOR_SIZE Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 04/13] ARC: uaccess: elide ZOL, use double load/stores Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 05/13] ARCv2: memset: don't prefetch for len == 0 which happens a lot Sergey Matyukevich
2022-02-22 14:14 ` [RFC PATCH 06/13] ARCv2: memset: elide unaligned handling if hardware supports Sergey Matyukevich
2022-02-22 14:15 ` Sergey Matyukevich [this message]
2022-02-22 14:15 ` [RFC PATCH 08/13] ARC: string: use generic C code if no ZOL support Sergey Matyukevich
2022-02-22 14:15 ` [RFC PATCH 09/13] ARC: delay: elide ZOL Sergey Matyukevich
2022-02-22 14:15 ` [RFC PATCH 10/13] ARC: checksum: " Sergey Matyukevich
2022-02-22 14:15 ` [RFC PATCH 11/13] ARC: head: " Sergey Matyukevich
2022-02-22 14:15 ` [RFC PATCH 12/13] ARC: build: inhibit ZOL generation by compiler Sergey Matyukevich
2022-02-22 14:15 ` [RFC PATCH 13/13] ARC: pt_regs: handle the case when ZOL is not supported Sergey Matyukevich
2022-02-28  2:09 ` [RFC PATCH 00/13] ARC: handle the lack of ZOL support Vineet Gupta
2022-03-03 19:22   ` Sergey Matyukevich
2022-03-23 10:09   ` [RFC PATCH 00/13] ARC: handle the lack of ZOL supporty Sergey Matyukevich

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=20220222141506.4003433-8-geomatsi@gmail.com \
    --to=geomatsi@gmail.com \
    --cc=isaev@synopsys.com \
    --cc=linux-snps-arc@lists.infradead.org \
    --cc=sergey.matyukevich@synopsys.com \
    --cc=vgupta@kernel.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