From: Tobias Klauser <tklauser@distanz.ch>
Date: Mon, 19 Oct 2020 09:09:10 +0000 (+0200)
Subject: internal/bytealg: add assembly implementation of Count/CountString for riscv64
X-Git-Tag: go1.16beta1~679
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=19f6422e005ed3f4b1a9f6850d382519673a3c87;p=gostls13.git

internal/bytealg: add assembly implementation of Count/CountString for riscv64

Simple single-byte loop count for now, to be further improved in future
CLs.

Benchmark on linux/riscv64 (HiFive Unleashed):

name               old time/op    new time/op     delta
CountSingle/10-4      190ns ± 1%      145ns ± 1%  -23.66%  (p=0.000 n=10+9)
CountSingle/32-4      422ns ± 1%      268ns ± 0%  -36.43%  (p=0.000 n=10+7)
CountSingle/4K-4     43.3µs ± 0%     23.8µs ± 0%  -45.09%  (p=0.000 n=8+10)
CountSingle/4M-4     54.2ms ± 1%     33.3ms ± 1%  -38.48%  (p=0.000 n=10+10)
CountSingle/64M-4     1.52s ± 1%      1.20s ± 1%  -21.20%  (p=0.000 n=9+9)

name               old speed      new speed       delta
CountSingle/10-4   52.7MB/s ± 1%   69.1MB/s ± 1%  +31.03%  (p=0.000 n=10+9)
CountSingle/32-4   75.9MB/s ± 1%  119.5MB/s ± 0%  +57.34%  (p=0.000 n=10+8)
CountSingle/4K-4   94.6MB/s ± 0%  172.2MB/s ± 0%  +82.10%  (p=0.000 n=8+10)
CountSingle/4M-4   77.4MB/s ± 1%  125.8MB/s ± 1%  +62.54%  (p=0.000 n=10+10)
CountSingle/64M-4  44.2MB/s ± 1%   56.1MB/s ± 1%  +26.91%  (p=0.000 n=9+9)

Change-Id: I2a6bd50d22d5f598517bb3c5a50066c54280cac5
Reviewed-on: https://go-review.googlesource.com/c/go/+/263541
Trust: Tobias Klauser <tobias.klauser@gmail.com>
Run-TryBot: Tobias Klauser <tobias.klauser@gmail.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Joel Sing <joel@sing.id.au>
---

diff --git a/src/internal/bytealg/count_generic.go b/src/internal/bytealg/count_generic.go
index 7cc1d50312..5575e81ab8 100644
--- a/src/internal/bytealg/count_generic.go
+++ b/src/internal/bytealg/count_generic.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !amd64,!arm,!arm64,!ppc64le,!ppc64,!s390x
+// +build !amd64,!arm,!arm64,!ppc64le,!ppc64,!riscv64,!s390x
 
 package bytealg
 
diff --git a/src/internal/bytealg/count_native.go b/src/internal/bytealg/count_native.go
index 0448fca9e8..b1ff1d265a 100644
--- a/src/internal/bytealg/count_native.go
+++ b/src/internal/bytealg/count_native.go
@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build amd64 arm arm64 ppc64le ppc64 s390x
+// +build amd64 arm arm64 ppc64le ppc64 riscv64 s390x
 
 package bytealg
 
diff --git a/src/internal/bytealg/count_riscv64.s b/src/internal/bytealg/count_riscv64.s
new file mode 100644
index 0000000000..3f4eb23286
--- /dev/null
+++ b/src/internal/bytealg/count_riscv64.s
@@ -0,0 +1,44 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "go_asm.h"
+#include "textflag.h"
+
+TEXT ·Count(SB),NOSPLIT,$0-40
+	MOV	b_base+0(FP), A1
+	MOV	b_len+8(FP), A2
+	MOVBU	c+24(FP), A3	// byte to count
+	MOV	ZERO, A4	// count
+	ADD	A1, A2		// end
+
+loop:
+	BEQ	A1, A2, done
+	MOVBU	(A1), A5
+	ADD	$1, A1
+	BNE	A3, A5, loop
+	ADD	$1, A4
+	JMP	loop
+
+done:
+	MOV	A4, ret+32(FP)
+	RET
+
+TEXT ·CountString(SB),NOSPLIT,$0-32
+	MOV	s_base+0(FP), A1
+	MOV	s_len+8(FP), A2
+	MOVBU	c+16(FP), A3	// byte to count
+	MOV	ZERO, A4	// count
+	ADD	A1, A2		// end
+
+loop:
+	BEQ	A1, A2, done
+	MOVBU	(A1), A5
+	ADD	$1, A1
+	BNE	A3, A5, loop
+	ADD	$1, A4
+	JMP	loop
+
+done:
+	MOV	A4, ret+24(FP)
+	RET