From 41ebac8b7d7485a5396ae25ce2412cafcd03f1a2 Mon Sep 17 00:00:00 2001
From: liqiang <liqiang64@huawei.com>
Date: Thu, 2 Sep 2021 17:31:48 +0800
Subject: [PATCH] Accelerate Adler32 using arm64 SVE instructions

	This patch uses the SVE instruction set to rewrite the Adler32
	algorithm (checksum algorithm in libz). By dividing the data into
	blocks, a vector operation can complete a data block in parallel.

	Measured on a Taishan 1951 machine that supports 256bit width SVE,
	this algorithm is about 3~5 times faster than the algorithm implemented
	in C language in libz. The wider the bit width, the better the
	acceleration effect. Below are the results of my measured random
	data of 1M and 10M:

		[root@xxx adler32]# ./benchmark 1000000
		Libz alg: Time used:    608 us, 1644.7 Mb/s.
		SVE  alg: Time used:    166 us, 6024.1 Mb/s.

		[root@xxx adler32]# ./benchmark 10000000
		Libz alg: Time used:   6484 us, 1542.3 Mb/s.
		SVE  alg: Time used:   2034 us, 4916.4 Mb/s.

	On machines that support ARM64 sve instructions, this algorithm can
	effectively accelerate adler32, thereby achieving the effect of improving
	the performance of the basic compression algorithm libz.

	In the implementation of this patch, blocks can be of any size, so the
	algorithm can automatically adapt to SVE hardware with different bit
	widths without modifying the code.

Signed-off-by: liqiang <liqiang64@huawei.com>
---
 contrib/arm/adler32_sve.S | 129 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 129 insertions(+)
 create mode 100644 contrib/arm/adler32_sve.S

diff --git a/contrib/arm/adler32_sve.S b/contrib/arm/adler32_sve.S
new file mode 100644
index 0000000..97c5930
--- /dev/null
+++ b/contrib/arm/adler32_sve.S
@@ -0,0 +1,129 @@
+/******************************************************************************
+ * Copyright (c) Huawei Technologies Co., Ltd. 2018-2020. All rights reserved.
+ * iSulad licensed under the Mulan PSL v2.
+ * You can use this software according to the terms and conditions of the Mulan PSL v2.
+ * You may obtain a copy of Mulan PSL v2 at:
+ *     http://license.coscl.org.cn/MulanPSL2
+ * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY OR FIT FOR A PARTICULAR
+ * PURPOSE.
+ * See the Mulan PSL v2 for more details.
+ * Author: liqiang
+ * Create: 2020-07-13
+ * Description: Use SVE instruction to optimize adler32 algorithm.
+ * Enhancement: 2020-10-13 
+                Automatically support different SVE vector length(128~2048). 
+ ******************************************************************************/
+
+.file "adler32_sve.S"
+.text
+.align 4
+
+//The supported sve vector length range is 128~2048 by this Adler_sequence
+.Adler_sequence:
+    .short 256,255,254,253,252,251,250,249,248,247,246,245,244,243,242,241,240,239,238,237,236,235,234,233,232,231,230,229,228,227,226,225,224,223,222,221,220,219,218,217,216,215,214,213,212,211,210,209,208,207,206,205,204,203,202,201,200,199,198,197,196,195,194,193,192,191,190,189,188,187,186,185,184,183,182,181,180,179,178,177,176,175,174,173,172,171,170,169,168,167,166,165,164,163,162,161,160,159,158,157,156,155,154,153,152,151,150,149,148,147,146,145,144,143,142,141,140,139,138,137,136,135,134,133,132,131,130,129,128,127,126,125,124,123,122,121,120,119,118,117,116,115,114,113,112,111,110,109,108,107,106,105,104,103,102,101,100,99,98,97,96,95,94,93,92,91,90,89,88,87,86,85,84,83,82,81,80,79,78,77,76,75,74,73,72,71,70,69,68,67,66,65,64,63,62,61,60,59,58,57,56,55,54,53,52,51,50,49,48,47,46,45,44,43,42,41,40,39,38,37,36,35,34,33,32,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1
+
+.global adler32_sve
+.type adler32_sve, %function
+adler32_sve:
+    // x0: unsigned long adler
+    // x1: const unsigned char *buf
+    // x2: unsigned long len
+    
+    // w10 : A = adler & 0xffff
+    // w11 : B = (adler >> 16) & 0xffff
+    // first byte A = 1, B = 0
+    and w10, w0, #0xffff
+    lsr w11, w0, #16
+    // less than and equal 63byte, jumper to normal proc
+    cmp x2, #0x3f
+    b.le Lnormal_proc
+    
+    // Get the length of the sve vector to x6.
+    mov x6, #0
+    addvl x6, x6, #1
+    adr x12, .Adler_sequence
+    ptrue p0.h
+
+    // Get the starting position of the required sequence.
+    mov x9, #256
+    sub x9, x9, x6
+    ld1h z24.h, p0/z, [x12, x9, lsl #1] // taps1 to z24.h
+    inch x9
+    ld1h z25.h, p0/z, [x12, x9, lsl #1] // taps2 to z25.h
+    // must bigger than 64byte 
+    ptrue p0.b
+    ptrue p1.h
+    mov x9, #0
+.align 4
+LBig_loop:
+    // x is SVE vector length.
+    // Bn = Bn-1 + An-1 * x + x * D1 + (x-1) * D2 + ... + 1 * Dx
+    // An = An-1 + D1 + D2 + D3 + ... + Dx
+    
+    .macro ADLER_BLOCK_32
+    ld1b z0.b, p0/z, [x1, x9]
+    
+    uaddv d20, p0, z0.b // D1 + D2 + ... + D32
+    mov x12, v20.2d[0]  // mov sum to w12
+    madd x11, x10, x6, x11 // Bn = An-1 * 32 + Bn-1
+
+    uunpklo z26.h, z0.b
+    uunpkhi z27.h, z0.b
+    mul z26.h, p1/m, z26.h, z24.h // x * D1 + (x-1) * D2 + ... + (x/2 + 1) * D(x/2)
+    mul z27.h, p1/m, z27.h, z25.h // (x/2) * D(x/2 + 1) + (x/2 - 1) * D(x/2 + 2) + ... + 1 * Dx
+
+    uaddv d21, p1, z26.h
+    uaddv d22, p1, z27.h
+    mov x13, v21.2d[0]
+    mov x14, v22.2d[0]
+
+    add x11, x13, x11
+    add x11, x14, x11     // Bn += x * D1 + (x-1) * D2 + ... + 1 * Dx
+    add x10, x12, x10     // An += D1 + D2 + ... + Dx
+    incb x9
+    .endm
+    mov x15, #4
+    ADLER_BLOCK_32
+    ADLER_BLOCK_32
+    ADLER_BLOCK_32
+    ADLER_BLOCK_32
+    
+    // calc = reg0 % 65521
+    .macro mod65521, reg0, reg1, reg2
+    mov w\reg1, #0x8071
+    mov w\reg2, #0xfff1
+    movk w\reg1, #0x8007, lsl #16
+    umull x\reg1, w\reg0, w\reg1
+    lsr x\reg1, x\reg1, #47
+    msub w\reg0, w\reg1, w\reg2, w\reg0
+    .endm
+    
+    mod65521 10, 14, 16
+    mod65521 11, 14, 16
+
+Lloop_cond:
+    mul x12, x6, x15
+    sub x2, x2, x12
+    cmp x2, x12
+    b.ge LBig_loop
+
+Lnormal_proc:
+    cmp x2, #0
+    b.eq Lret
+
+    ldrb w15, [x1, x9]
+    add x9, x9, #1
+    add x10, x15, x10
+    add x11, x10, x11
+    sub x2, x2, #1
+    b Lnormal_proc
+
+Lret:
+    mod65521 10, 14, 5
+    mod65521 11, 14, 5
+    lsl x11, x11, #16
+    orr x0, x10, x11
+    ret
+
+.size adler32_sve, .-adler32_sve
-- 
2.17.1