c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-11-17 10:01:01 +03:00
Code Issues Packages Projects Releases Wiki Activity
a926aab902
mirror_zfs/module/icp/asm-ppc64/blake3/b3_ppc64le_sse2.S
Tino Reichardt 985c33b132
Introduce BLAKE3 checksums as an OpenZFS feature 
This commit adds BLAKE3 checksums to OpenZFS, it has similar
performance to Edon-R, but without the caveats around the latter.

Homepage of BLAKE3: https://github.com/BLAKE3-team/BLAKE3
Wikipedia: https://en.wikipedia.org/wiki/BLAKE_(hash_function)#BLAKE3

Short description of Wikipedia:

  BLAKE3 is a cryptographic hash function based on Bao and BLAKE2,
  created by Jack O'Connor, Jean-Philippe Aumasson, Samuel Neves, and
  Zooko Wilcox-O'Hearn. It was announced on January 9, 2020, at Real
  World Crypto. BLAKE3 is a single algorithm with many desirable
  features (parallelism, XOF, KDF, PRF and MAC), in contrast to BLAKE
  and BLAKE2, which are algorithm families with multiple variants.
  BLAKE3 has a binary tree structure, so it supports a practically
  unlimited degree of parallelism (both SIMD and multithreading) given
  enough input. The official Rust and C implementations are
  dual-licensed as public domain (CC0) and the Apache License.

Along with adding the BLAKE3 hash into the OpenZFS infrastructure a
new benchmarking file called chksum_bench was introduced.  When read
it reports the speed of the available checksum functions.

On Linux: cat /proc/spl/kstat/zfs/chksum_bench
On FreeBSD: sysctl kstat.zfs.misc.chksum_bench

This is an example output of an i3-1005G1 test system with Debian 11:

implementation      1k      4k     16k     64k    256k      1m      4m
edonr-generic     1196    1602    1761    1749    1762    1759    1751
skein-generic      546     591     608     615     619     612     616
sha256-generic     240     300     316     314     304     285     276
sha512-generic     353     441     467     476     472     467     426
blake3-generic     308     313     313     313     312     313     312
blake3-sse2        402    1289    1423    1446    1432    1458    1413
blake3-sse41       427    1470    1625    1704    1679    1607    1629
blake3-avx2        428    1920    3095    3343    3356    3318    3204
blake3-avx512      473    2687    4905    5836    5844    5643    5374

Output on Debian 5.10.0-10-amd64 system: (Ryzen 7 5800X)

implementation      1k      4k     16k     64k    256k      1m      4m
edonr-generic     1840    2458    2665    2719    2711    2723    2693
skein-generic      870     966     996     992    1003    1005    1009
sha256-generic     415     442     453     455     457     457     457
sha512-generic     608     690     711     718     719     720     721
blake3-generic     301     313     311     309     309     310     310
blake3-sse2        343    1865    2124    2188    2180    2181    2186
blake3-sse41       364    2091    2396    2509    2463    2482    2488
blake3-avx2        365    2590    4399    4971    4915    4802    4764

Output on Debian 5.10.0-9-powerpc64le system: (POWER 9)

implementation      1k      4k     16k     64k    256k      1m      4m
edonr-generic     1213    1703    1889    1918    1957    1902    1907
skein-generic      434     492     520     522     511     525     525
sha256-generic     167     183     187     188     188     187     188
sha512-generic     186     216     222     221     225     224     224
blake3-generic     153     152     154     153     151     153     153
blake3-sse2        391    1170    1366    1406    1428    1426    1414
blake3-sse41       352    1049    1212    1174    1262    1258    1259

Output on Debian 5.10.0-11-arm64 system: (Pi400)

implementation      1k      4k     16k     64k    256k      1m      4m
edonr-generic      487     603     629     639     643     641     641
skein-generic      271     299     303     308     309     309     307
sha256-generic     117     127     128     130     130     129     130
sha512-generic     145     165     170     172     173     174     175
blake3-generic      81      29      71      89      89      89      89
blake3-sse2        112     323     368     379     380     371     374
blake3-sse41       101     315     357     368     369     364     360

Structurally, the new code is mainly split into these parts:
- 1x cross platform generic c variant: blake3_generic.c
- 4x assembly for X86-64 (SSE2, SSE4.1, AVX2, AVX512)
- 2x assembly for ARMv8 (NEON converted from SSE2)
- 2x assembly for PPC64-LE (POWER8 converted from SSE2)
- one file for switching between the implementations

Note the PPC64 assembly requires the VSX instruction set and the
kfpu_begin() / kfpu_end() calls on PowerPC were updated accordingly.

Reviewed-by: Felix Dörre <felix@dogcraft.de>
Reviewed-by: Ahelenia Ziemiańska <nabijaczleweli@nabijaczleweli.xyz>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tino Reichardt <milky-zfs@mcmilk.de>
Co-authored-by: Rich Ercolani <rincebrain@gmail.com>
Closes #10058
Closes #12918
2022-06-08 15:55:57 -07:00

2824 lines
47 KiB
ArmAsm
Raw Blame History

/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
/*
* Based on BLAKE3 v1.3.1, https://github.com/BLAKE3-team/BLAKE3
* Copyright (c) 2019-2022 Samuel Neves and Matthew Krupcale
* Copyright (c) 2022 Tino Reichardt <milky-zfs@mcmilk.de>
*
* This is converted assembly: SSE2 -> POWER8 PPC64 Little Endian
* Used tools: SIMDe https://github.com/simd-everywhere/simde
*/
#if (defined(__PPC64__) && defined(__LITTLE_ENDIAN__))
.text
.abiversion 2
.section .rodata.cst16,"aM",@progbits,16
.p2align 4
.LCPI0_0:
.byte 29
.byte 28
.byte 31
.byte 30
.byte 25
.byte 24
.byte 27
.byte 26
.byte 21
.byte 20
.byte 23
.byte 22
.byte 17
.byte 16
.byte 19
.byte 18
.LCPI0_1:
.long 1779033703
.long 3144134277
.long 1013904242
.long 2773480762
.LCPI0_2:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 19
.byte 18
.byte 17
.byte 16
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI0_3:
.byte 19
.byte 18
.byte 17
.byte 16
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI0_4:
.byte 23
.byte 22
.byte 21
.byte 20
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI0_5:
.byte 23
.byte 22
.byte 21
.byte 20
.byte 23
.byte 22
.byte 21
.byte 20
.byte 3
.byte 2
.byte 1
.byte 0
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI0_6:
.short 1
.short 2
.short 4
.short 8
.short 16
.short 32
.short 64
.short 128
.LCPI0_7:
.short 0
.short 0
.short 4
.short 8
.short 0
.short 0
.short 64
.short 128
.LCPI0_8:
.byte 19
.byte 18
.byte 17
.byte 16
.byte 19
.byte 18
.byte 17
.byte 16
.byte 31
.byte 30
.byte 29
.byte 28
.byte 31
.byte 30
.byte 29
.byte 28
.LCPI0_9:
.short 0
.short 0
.short 0
.short 0
.short 0
.short 0
.short 64
.short 128
.LCPI0_10:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 7
.byte 6
.byte 5
.byte 4
.byte 3
.byte 2
.byte 1
.byte 0
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI0_11:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 23
.byte 22
.byte 21
.byte 20
.byte 19
.byte 18
.byte 17
.byte 16
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI0_12:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 15
.byte 14
.byte 13
.byte 12
.byte 31
.byte 30
.byte 29
.byte 28
.LCPI0_13:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 15
.byte 14
.byte 13
.byte 12
.byte 11
.byte 10
.byte 9
.byte 8
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI0_14:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.byte 23
.byte 22
.byte 21
.byte 20
.text
.globl zfs_blake3_compress_in_place_sse2
.p2align 2
.type zfs_blake3_compress_in_place_sse2,@function
zfs_blake3_compress_in_place_sse2:
.Lfunc_begin0:
.cfi_startproc
.Lfunc_gep0:
addis 2, 12, .TOC.-.Lfunc_gep0@ha
addi 2, 2, .TOC.-.Lfunc_gep0@l
.Lfunc_lep0:
.localentry zfs_blake3_compress_in_place_sse2, .Lfunc_lep0-.Lfunc_gep0
li 8, -64
mtvsrd 35, 5
li 5, 16
lfdx 0, 0, 4
vspltisw 12, 9
stxvd2x 60, 1, 8
li 8, -48
mtvsrd 36, 7
lfd 2, 16(4)
stxvd2x 61, 1, 8
li 8, -32
lfd 1, 8(4)
mtvsrwz 37, 6
rldicl 6, 6, 32, 32
addis 7, 2, .LCPI0_2@toc@ha
stxvd2x 62, 1, 8
li 8, -16
addi 7, 7, .LCPI0_2@toc@l
stxvd2x 63, 1, 8
li 8, 0
lvx 9, 0, 7
li 7, 48
mtvsrd 34, 8
xxmrghd 32, 1, 0
lxvd2x 0, 0, 3
lxvd2x 1, 3, 5
lfd 3, 24(4)
addis 8, 2, .LCPI0_5@toc@ha
vmrghb 3, 2, 3
addi 8, 8, .LCPI0_5@toc@l
vmrghb 4, 2, 4
vspltb 2, 2, 7
xxmrghd 33, 3, 2
vpkudum 7, 1, 0
vmrglh 3, 2, 3
vmrglh 2, 2, 4
mtvsrwz 36, 6
addis 6, 2, .LCPI0_0@toc@ha
addi 6, 6, .LCPI0_0@toc@l
vperm 10, 1, 0, 9
vmrghw 4, 4, 5
xxswapd 37, 1
lxvd2x 1, 4, 7
addis 7, 2, .LCPI0_8@toc@ha
addi 7, 7, .LCPI0_8@toc@l
vmrglw 2, 2, 3
xxswapd 35, 0
xxswapd 41, 1
xxspltd 62, 42, 1
vadduwm 3, 7, 3
vadduwm 6, 3, 5
xxmrgld 36, 34, 36
lvx 2, 0, 6
addis 6, 2, .LCPI0_1@toc@ha
addi 6, 6, .LCPI0_1@toc@l
xxlxor 35, 38, 36
lvx 4, 0, 6
li 6, 32
lxvd2x 0, 4, 6
addis 4, 2, .LCPI0_3@toc@ha
addis 6, 2, .LCPI0_7@toc@ha
vperm 8, 3, 3, 2
vspltisw 3, 10
addi 4, 4, .LCPI0_3@toc@l
addi 6, 6, .LCPI0_7@toc@l
vadduwm 3, 3, 3
vadduwm 11, 8, 4
xxlxor 36, 43, 37
vadduwm 5, 6, 10
vrlw 0, 4, 3
vspltisw 4, 12
vadduwm 4, 4, 4
vadduwm 1, 0, 5
xxlxor 37, 33, 40
xxswapd 40, 0
vrlw 6, 5, 4
vspltisw 5, -16
vpkudum 13, 9, 8
vsubuwm 5, 12, 5
lvx 12, 0, 4
addis 4, 2, .LCPI0_4@toc@ha
addi 4, 4, .LCPI0_4@toc@l
vadduwm 11, 6, 11
xxswapd 0, 38
vadduwm 1, 1, 13
xxsldwi 50, 45, 45, 1
xxlxor 32, 43, 32
xxsldwi 43, 43, 43, 3
xxsldwi 33, 33, 33, 1
vperm 12, 8, 9, 12
vrlw 0, 0, 5
vadduwm 1, 0, 1
xxlxor 38, 33, 0
vadduwm 1, 1, 12
vperm 6, 6, 6, 2
vadduwm 15, 6, 11
lvx 11, 0, 4
addis 4, 2, .LCPI0_6@toc@ha
addi 4, 4, .LCPI0_6@toc@l
xxlxor 32, 47, 32
lvx 17, 0, 4
addis 4, 2, .LCPI0_9@toc@ha
vperm 14, 10, 7, 11
addi 4, 4, .LCPI0_9@toc@l
vrlw 0, 0, 3
vadduwm 1, 0, 1
xxlxor 38, 33, 38
vrlw 6, 6, 4
vadduwm 8, 6, 15
xxswapd 0, 38
lvx 6, 0, 8
xxlxor 32, 40, 32
xxsldwi 40, 40, 40, 1
vperm 13, 12, 18, 6
vrlw 9, 0, 5
vadduwm 0, 1, 14
lvx 1, 0, 7
xxsldwi 46, 46, 46, 3
xxsldwi 32, 32, 32, 3
vperm 7, 7, 7, 1
vadduwm 15, 9, 0
xxlxor 32, 47, 0
vperm 16, 0, 0, 2
lvx 0, 0, 6
addis 6, 2, .LCPI0_10@toc@ha
vcmpequh 0, 0, 17
vadduwm 19, 16, 8
xxlxor 40, 51, 41
xxsel 45, 39, 45, 32
vrlw 31, 8, 3
lvx 8, 0, 4
addis 4, 2, .LCPI0_11@toc@ha
addi 4, 4, .LCPI0_11@toc@l
vcmpequh 7, 8, 17
vadduwm 8, 15, 13
vadduwm 15, 31, 8
lvx 8, 0, 4
addi 4, 6, .LCPI0_10@toc@l
lvx 17, 0, 4
addis 4, 2, .LCPI0_12@toc@ha
xxlxor 41, 47, 48
xxsldwi 47, 47, 47, 1
addi 4, 4, .LCPI0_12@toc@l
xxlnor 48, 39, 39
vrlw 29, 9, 4
vperm 9, 16, 16, 8
xxland 48, 50, 39
vperm 17, 30, 12, 17
vperm 16, 16, 16, 8
vmrghw 12, 12, 10
lvx 10, 0, 4
addis 4, 2, .LCPI0_13@toc@ha
vadduwm 19, 29, 19
addi 4, 4, .LCPI0_13@toc@l
xxlxor 63, 51, 63
xxsldwi 51, 51, 51, 3
xxland 0, 49, 41
vrlw 17, 31, 5
xxlor 48, 0, 48
xxswapd 0, 61
vperm 18, 12, 18, 10
vadduwm 15, 15, 16
xxland 60, 48, 39
vadduwm 15, 17, 15
vperm 28, 28, 28, 8
xxlxor 63, 47, 0
vadduwm 15, 15, 18
vperm 31, 31, 31, 2
vperm 30, 18, 16, 6
vadduwm 19, 31, 19
xxlxor 44, 51, 49
vrlw 12, 12, 3
vadduwm 15, 12, 15
xxlxor 49, 47, 63
vperm 31, 13, 14, 11
vrlw 17, 17, 4
vperm 14, 14, 14, 1
vadduwm 15, 15, 31
vadduwm 19, 17, 19
xxswapd 0, 49
xxsldwi 47, 47, 47, 3
xxsel 46, 46, 62, 32
xxlxor 44, 51, 44
xxsldwi 51, 51, 51, 1
vrlw 12, 12, 5
vadduwm 15, 12, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 19, 17, 19
xxlxor 44, 51, 44
vrlw 29, 12, 3
vadduwm 12, 15, 14
vadduwm 15, 29, 12
lvx 12, 0, 4
addis 4, 2, .LCPI0_14@toc@ha
addi 4, 4, .LCPI0_14@toc@l
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
vperm 30, 13, 18, 12
vrlw 17, 17, 4
vmrghw 13, 18, 13
xxland 0, 62, 41
vadduwm 19, 17, 19
vperm 16, 13, 16, 10
xxlxor 61, 51, 61
xxsldwi 50, 51, 51, 3
xxsldwi 51, 63, 63, 3
vrlw 30, 29, 5
xxlor 61, 60, 0
xxswapd 0, 49
vperm 31, 14, 19, 11
vadduwm 15, 15, 29
vperm 19, 19, 19, 1
vadduwm 15, 30, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 16
vperm 17, 17, 17, 2
vadduwm 18, 17, 18
xxlxor 45, 50, 62
vperm 30, 16, 29, 6
vrlw 13, 13, 3
vadduwm 15, 13, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 31
xxsldwi 63, 63, 63, 3
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 18, 17, 18
xxswapd 0, 49
xxlxor 45, 50, 45
xxsldwi 50, 50, 50, 1
vrlw 13, 13, 5
vadduwm 15, 13, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 18, 17, 18
xxlxor 45, 50, 45
vrlw 28, 13, 3
xxsel 45, 51, 62, 32
xxland 51, 61, 39
vperm 30, 14, 16, 12
vadduwm 15, 15, 13
vperm 19, 19, 19, 8
vmrghw 14, 16, 14
vadduwm 15, 28, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 62, 41
vrlw 17, 17, 4
xxlor 51, 51, 0
vadduwm 15, 15, 19
vadduwm 18, 17, 18
xxswapd 0, 49
xxlxor 60, 50, 60
xxsldwi 48, 50, 50, 3
vperm 18, 14, 29, 10
vrlw 30, 28, 5
vperm 29, 18, 19, 6
vadduwm 15, 30, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 18
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 46, 48, 62
vperm 30, 13, 31, 11
vrlw 14, 14, 3
vperm 31, 31, 31, 1
vadduwm 15, 14, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 30
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 16, 17, 16
xxswapd 0, 49
xxlxor 46, 48, 46
xxsldwi 48, 48, 48, 1
vrlw 14, 14, 5
vadduwm 15, 14, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 46, 48, 46
vrlw 28, 14, 3
xxsel 46, 63, 61, 32
xxland 63, 51, 39
vperm 29, 13, 18, 12
vadduwm 15, 15, 14
vperm 31, 31, 31, 8
vmrghw 13, 18, 13
vadduwm 15, 28, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 61, 41
vrlw 17, 17, 4
xxlor 63, 63, 0
vperm 13, 13, 19, 10
xxsldwi 51, 62, 62, 3
vadduwm 15, 15, 31
vperm 30, 14, 19, 11
vadduwm 16, 17, 16
xxswapd 0, 49
xxlxor 60, 48, 60
xxsldwi 48, 48, 48, 3
vrlw 29, 28, 5
vadduwm 15, 29, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 13
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 50, 48, 61
vrlw 18, 18, 3
vadduwm 15, 18, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 30
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 11, 17, 16
xxswapd 0, 49
xxlxor 48, 43, 50
xxsldwi 43, 43, 43, 1
vperm 18, 19, 19, 1
vrlw 16, 16, 5
vperm 19, 13, 31, 6
vadduwm 15, 16, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 29, 17, 11
xxlxor 43, 61, 48
vrlw 16, 11, 3
xxsel 43, 50, 51, 32
xxland 50, 63, 39
vperm 19, 14, 13, 12
vadduwm 15, 15, 11
vperm 18, 18, 18, 8
vmrghw 13, 13, 14
vadduwm 15, 16, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 51, 41
lvx 19, 0, 4
vrlw 17, 17, 4
xxlor 50, 50, 0
vperm 13, 13, 31, 10
xxsldwi 63, 62, 62, 3
vadduwm 15, 15, 18
vperm 19, 11, 31, 19
vadduwm 29, 17, 29
xxswapd 0, 49
vperm 1, 31, 31, 1
xxlxor 48, 61, 48
xxsldwi 46, 61, 61, 3
vperm 6, 13, 18, 6
vrlw 16, 16, 5
xxsel 32, 33, 38, 32
xxland 38, 50, 39
vadduwm 15, 16, 15
vperm 7, 11, 13, 12
xxlxor 49, 47, 0
vadduwm 15, 15, 13
vperm 17, 17, 17, 2
vperm 6, 6, 6, 8
vadduwm 14, 17, 14
xxlxor 48, 46, 48
vrlw 16, 16, 3
vadduwm 15, 16, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 3
vrlw 17, 17, 4
vadduwm 15, 15, 19
vadduwm 14, 17, 14
xxswapd 0, 49
xxlxor 48, 46, 48
xxsldwi 46, 46, 46, 1
vrlw 16, 16, 5
vadduwm 15, 16, 15
xxlxor 49, 47, 0
vadduwm 0, 15, 0
vperm 17, 17, 17, 2
xxland 0, 39, 41
xxlor 38, 38, 0
vadduwm 14, 17, 14
xxlxor 48, 46, 48
vrlw 16, 16, 3
vadduwm 0, 16, 0
xxlxor 33, 32, 49
xxsldwi 32, 32, 32, 1
vrlw 1, 1, 4
vadduwm 0, 0, 6
vadduwm 8, 1, 14
xxswapd 0, 33
xxlxor 44, 40, 48
xxsldwi 38, 40, 40, 3
vrlw 7, 12, 5
vadduwm 0, 7, 0
xxlxor 33, 32, 0
vperm 2, 1, 1, 2
vmrghw 1, 13, 11
vadduwm 6, 2, 6
vperm 1, 1, 18, 10
xxlxor 39, 38, 39
vrlw 3, 7, 3
vadduwm 0, 0, 1
vadduwm 0, 3, 0
xxlxor 34, 32, 34
xxsldwi 0, 32, 32, 3
vrlw 2, 2, 4
vadduwm 4, 2, 6
xxswapd 2, 34
xxlxor 35, 36, 35
xxsldwi 1, 36, 36, 1
vrlw 3, 3, 5
xxlxor 0, 1, 0
xxswapd 0, 0
xxlxor 1, 35, 2
stxvd2x 0, 0, 3
xxswapd 1, 1
stxvd2x 1, 3, 5
li 3, -16
lxvd2x 63, 1, 3
li 3, -32
lxvd2x 62, 1, 3
li 3, -48
lxvd2x 61, 1, 3
li 3, -64
lxvd2x 60, 1, 3
blr
.long 0
.quad 0
.Lfunc_end0:
.size zfs_blake3_compress_in_place_sse2, .Lfunc_end0-.Lfunc_begin0
.cfi_endproc
.section .rodata.cst16,"aM",@progbits,16
.p2align 4
.LCPI1_0:
.byte 29
.byte 28
.byte 31
.byte 30
.byte 25
.byte 24
.byte 27
.byte 26
.byte 21
.byte 20
.byte 23
.byte 22
.byte 17
.byte 16
.byte 19
.byte 18
.LCPI1_1:
.long 1779033703
.long 3144134277
.long 1013904242
.long 2773480762
.LCPI1_2:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 19
.byte 18
.byte 17
.byte 16
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI1_3:
.byte 19
.byte 18
.byte 17
.byte 16
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI1_4:
.byte 23
.byte 22
.byte 21
.byte 20
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI1_5:
.byte 23
.byte 22
.byte 21
.byte 20
.byte 23
.byte 22
.byte 21
.byte 20
.byte 3
.byte 2
.byte 1
.byte 0
.byte 3
.byte 2
.byte 1
.byte 0
.LCPI1_6:
.short 1
.short 2
.short 4
.short 8
.short 16
.short 32
.short 64
.short 128
.LCPI1_7:
.short 0
.short 0
.short 4
.short 8
.short 0
.short 0
.short 64
.short 128
.LCPI1_8:
.byte 19
.byte 18
.byte 17
.byte 16
.byte 19
.byte 18
.byte 17
.byte 16
.byte 31
.byte 30
.byte 29
.byte 28
.byte 31
.byte 30
.byte 29
.byte 28
.LCPI1_9:
.short 0
.short 0
.short 0
.short 0
.short 0
.short 0
.short 64
.short 128
.LCPI1_10:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 7
.byte 6
.byte 5
.byte 4
.byte 3
.byte 2
.byte 1
.byte 0
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI1_11:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 23
.byte 22
.byte 21
.byte 20
.byte 19
.byte 18
.byte 17
.byte 16
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI1_12:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 15
.byte 14
.byte 13
.byte 12
.byte 31
.byte 30
.byte 29
.byte 28
.LCPI1_13:
.byte 31
.byte 30
.byte 29
.byte 28
.byte 15
.byte 14
.byte 13
.byte 12
.byte 11
.byte 10
.byte 9
.byte 8
.byte 27
.byte 26
.byte 25
.byte 24
.LCPI1_14:
.byte 27
.byte 26
.byte 25
.byte 24
.byte 11
.byte 10
.byte 9
.byte 8
.byte 3
.byte 2
.byte 1
.byte 0
.byte 23
.byte 22
.byte 21
.byte 20
.text
.globl zfs_blake3_compress_xof_sse2
.p2align 2
.type zfs_blake3_compress_xof_sse2,@function
zfs_blake3_compress_xof_sse2:
.Lfunc_begin1:
.cfi_startproc
.Lfunc_gep1:
addis 2, 12, .TOC.-.Lfunc_gep1@ha
addi 2, 2, .TOC.-.Lfunc_gep1@l
.Lfunc_lep1:
.localentry zfs_blake3_compress_xof_sse2, .Lfunc_lep1-.Lfunc_gep1
li 9, -80
mtvsrd 35, 5
li 5, 16
lfdx 0, 0, 4
addis 10, 2, .LCPI1_2@toc@ha
vspltisw 12, 9
std 30, -16(1)
addis 12, 2, .LCPI1_8@toc@ha
addis 30, 2, .LCPI1_5@toc@ha
addis 11, 2, .LCPI1_7@toc@ha
stxvd2x 60, 1, 9
li 9, -64
mtvsrd 36, 7
lfd 2, 16(4)
addi 10, 10, .LCPI1_2@toc@l
addi 12, 12, .LCPI1_8@toc@l
addi 11, 11, .LCPI1_7@toc@l
stxvd2x 61, 1, 9
li 9, -48
lfd 3, 24(4)
mtvsrwz 37, 6
rldicl 6, 6, 32, 32
lvx 9, 0, 10
stxvd2x 62, 1, 9
li 9, -32
li 10, 32
stxvd2x 63, 1, 9
li 9, 0
mtvsrd 34, 9
xxmrghd 33, 3, 2
lfd 1, 8(4)
vmrghb 3, 2, 3
vmrghb 4, 2, 4
vspltb 2, 2, 7
xxmrghd 32, 1, 0
lxvd2x 0, 0, 3
lxvd2x 1, 3, 5
vpkudum 7, 1, 0
vmrglh 3, 2, 3
vmrglh 2, 2, 4
mtvsrwz 36, 6
addis 6, 2, .LCPI1_0@toc@ha
addi 6, 6, .LCPI1_0@toc@l
vperm 10, 1, 0, 9
vmrghw 4, 4, 5
xxswapd 37, 1
vmrglw 2, 2, 3
xxswapd 35, 0
lxvd2x 0, 4, 10
xxspltd 62, 42, 1
vadduwm 3, 7, 3
vadduwm 6, 3, 5
xxmrgld 36, 34, 36
lvx 2, 0, 6
addis 6, 2, .LCPI1_1@toc@ha
addi 6, 6, .LCPI1_1@toc@l
xxlxor 35, 38, 36
lvx 4, 0, 6
li 6, 48
lxvd2x 1, 4, 6
addis 4, 2, .LCPI1_3@toc@ha
vperm 8, 3, 3, 2
vspltisw 3, 10
addi 4, 4, .LCPI1_3@toc@l
xxswapd 41, 1
vadduwm 3, 3, 3
vadduwm 11, 8, 4
xxlxor 36, 43, 37
vadduwm 5, 6, 10
vrlw 0, 4, 3
vspltisw 4, 12
vadduwm 4, 4, 4
vadduwm 1, 0, 5
xxlxor 37, 33, 40
xxswapd 40, 0
vrlw 6, 5, 4
vspltisw 5, -16
vpkudum 13, 9, 8
vsubuwm 5, 12, 5
lvx 12, 0, 4
addis 4, 2, .LCPI1_4@toc@ha
addi 4, 4, .LCPI1_4@toc@l
vadduwm 11, 6, 11
xxswapd 0, 38
vadduwm 1, 1, 13
xxsldwi 50, 45, 45, 1
xxlxor 32, 43, 32
xxsldwi 43, 43, 43, 3
xxsldwi 33, 33, 33, 1
vperm 12, 8, 9, 12
vrlw 0, 0, 5
vadduwm 1, 0, 1
xxlxor 38, 33, 0
vadduwm 1, 1, 12
vperm 6, 6, 6, 2
vadduwm 15, 6, 11
lvx 11, 0, 4
addis 4, 2, .LCPI1_6@toc@ha
addi 4, 4, .LCPI1_6@toc@l
xxlxor 32, 47, 32
lvx 17, 0, 4
addi 4, 30, .LCPI1_5@toc@l
vperm 14, 10, 7, 11
vrlw 0, 0, 3
vadduwm 1, 0, 1
xxlxor 38, 33, 38
vrlw 6, 6, 4
vadduwm 8, 6, 15
xxswapd 0, 38
lvx 6, 0, 4
addis 4, 2, .LCPI1_9@toc@ha
addi 4, 4, .LCPI1_9@toc@l
xxlxor 32, 40, 32
xxsldwi 40, 40, 40, 1
vperm 13, 12, 18, 6
vrlw 9, 0, 5
vadduwm 0, 1, 14
lvx 1, 0, 12
xxsldwi 46, 46, 46, 3
xxsldwi 32, 32, 32, 3
vperm 7, 7, 7, 1
vadduwm 15, 9, 0
xxlxor 32, 47, 0
vperm 16, 0, 0, 2
lvx 0, 0, 11
addis 11, 2, .LCPI1_10@toc@ha
vcmpequh 0, 0, 17
vadduwm 19, 16, 8
xxlxor 40, 51, 41
xxsel 45, 39, 45, 32
vrlw 31, 8, 3
lvx 8, 0, 4
addis 4, 2, .LCPI1_11@toc@ha
addi 4, 4, .LCPI1_11@toc@l
vcmpequh 7, 8, 17
vadduwm 8, 15, 13
vadduwm 15, 31, 8
lvx 8, 0, 4
addi 4, 11, .LCPI1_10@toc@l
lvx 17, 0, 4
addis 4, 2, .LCPI1_12@toc@ha
xxlxor 41, 47, 48
xxsldwi 47, 47, 47, 1
addi 4, 4, .LCPI1_12@toc@l
xxlnor 48, 39, 39
vrlw 29, 9, 4
vperm 9, 16, 16, 8
xxland 48, 50, 39
vperm 17, 30, 12, 17
vperm 16, 16, 16, 8
vmrghw 12, 12, 10
lvx 10, 0, 4
addis 4, 2, .LCPI1_13@toc@ha
vadduwm 19, 29, 19
addi 4, 4, .LCPI1_13@toc@l
xxlxor 63, 51, 63
xxsldwi 51, 51, 51, 3
xxland 0, 49, 41
vrlw 17, 31, 5
xxlor 48, 0, 48
xxswapd 0, 61
vperm 18, 12, 18, 10
vadduwm 15, 15, 16
xxland 60, 48, 39
vadduwm 15, 17, 15
vperm 28, 28, 28, 8
xxlxor 63, 47, 0
vadduwm 15, 15, 18
vperm 31, 31, 31, 2
vperm 30, 18, 16, 6
vadduwm 19, 31, 19
xxlxor 44, 51, 49
vrlw 12, 12, 3
vadduwm 15, 12, 15
xxlxor 49, 47, 63
vperm 31, 13, 14, 11
vrlw 17, 17, 4
vperm 14, 14, 14, 1
vadduwm 15, 15, 31
vadduwm 19, 17, 19
xxswapd 0, 49
xxsldwi 47, 47, 47, 3
xxsel 46, 46, 62, 32
xxlxor 44, 51, 44
xxsldwi 51, 51, 51, 1
vrlw 12, 12, 5
vadduwm 15, 12, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 19, 17, 19
xxlxor 44, 51, 44
vrlw 29, 12, 3
vadduwm 12, 15, 14
vadduwm 15, 29, 12
lvx 12, 0, 4
addis 4, 2, .LCPI1_14@toc@ha
addi 4, 4, .LCPI1_14@toc@l
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
vperm 30, 13, 18, 12
vrlw 17, 17, 4
vmrghw 13, 18, 13
xxland 0, 62, 41
vadduwm 19, 17, 19
vperm 16, 13, 16, 10
xxlxor 61, 51, 61
xxsldwi 50, 51, 51, 3
xxsldwi 51, 63, 63, 3
vrlw 30, 29, 5
xxlor 61, 60, 0
xxswapd 0, 49
vperm 31, 14, 19, 11
vadduwm 15, 15, 29
vperm 19, 19, 19, 1
vadduwm 15, 30, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 16
vperm 17, 17, 17, 2
vadduwm 18, 17, 18
xxlxor 45, 50, 62
vperm 30, 16, 29, 6
vrlw 13, 13, 3
vadduwm 15, 13, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 31
xxsldwi 63, 63, 63, 3
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 18, 17, 18
xxswapd 0, 49
xxlxor 45, 50, 45
xxsldwi 50, 50, 50, 1
vrlw 13, 13, 5
vadduwm 15, 13, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 18, 17, 18
xxlxor 45, 50, 45
vrlw 28, 13, 3
xxsel 45, 51, 62, 32
xxland 51, 61, 39
vperm 30, 14, 16, 12
vadduwm 15, 15, 13
vperm 19, 19, 19, 8
vmrghw 14, 16, 14
vadduwm 15, 28, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 62, 41
vrlw 17, 17, 4
xxlor 51, 51, 0
vadduwm 15, 15, 19
vadduwm 18, 17, 18
xxswapd 0, 49
xxlxor 60, 50, 60
xxsldwi 48, 50, 50, 3
vperm 18, 14, 29, 10
vrlw 30, 28, 5
vperm 29, 18, 19, 6
vadduwm 15, 30, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 18
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 46, 48, 62
vperm 30, 13, 31, 11
vrlw 14, 14, 3
vperm 31, 31, 31, 1
vadduwm 15, 14, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 30
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 16, 17, 16
xxswapd 0, 49
xxlxor 46, 48, 46
xxsldwi 48, 48, 48, 1
vrlw 14, 14, 5
vadduwm 15, 14, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 46, 48, 46
vrlw 28, 14, 3
xxsel 46, 63, 61, 32
xxland 63, 51, 39
vperm 29, 13, 18, 12
vadduwm 15, 15, 14
vperm 31, 31, 31, 8
vmrghw 13, 18, 13
vadduwm 15, 28, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 61, 41
vrlw 17, 17, 4
xxlor 63, 63, 0
vperm 13, 13, 19, 10
xxsldwi 51, 62, 62, 3
vadduwm 15, 15, 31
vperm 30, 14, 19, 11
vadduwm 16, 17, 16
xxswapd 0, 49
xxlxor 60, 48, 60
xxsldwi 48, 48, 48, 3
vrlw 29, 28, 5
vadduwm 15, 29, 15
xxlxor 49, 47, 0
vadduwm 15, 15, 13
vperm 17, 17, 17, 2
vadduwm 16, 17, 16
xxlxor 50, 48, 61
vrlw 18, 18, 3
vadduwm 15, 18, 15
xxlxor 49, 47, 49
vadduwm 15, 15, 30
vrlw 17, 17, 4
xxsldwi 47, 47, 47, 3
vadduwm 11, 17, 16
xxswapd 0, 49
xxlxor 48, 43, 50
xxsldwi 43, 43, 43, 1
vperm 18, 19, 19, 1
vrlw 16, 16, 5
vperm 19, 13, 31, 6
vadduwm 15, 16, 15
xxlxor 49, 47, 0
vperm 17, 17, 17, 2
vadduwm 29, 17, 11
xxlxor 43, 61, 48
vrlw 16, 11, 3
xxsel 43, 50, 51, 32
xxland 50, 63, 39
vperm 19, 14, 13, 12
vadduwm 15, 15, 11
vperm 18, 18, 18, 8
vmrghw 13, 13, 14
vadduwm 15, 16, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 1
xxland 0, 51, 41
lvx 19, 0, 4
vrlw 17, 17, 4
xxlor 50, 50, 0
vperm 13, 13, 31, 10
xxsldwi 63, 62, 62, 3
vadduwm 15, 15, 18
vperm 19, 11, 31, 19
vadduwm 29, 17, 29
xxswapd 0, 49
vperm 1, 31, 31, 1
xxlxor 48, 61, 48
xxsldwi 46, 61, 61, 3
vperm 6, 13, 18, 6
vrlw 16, 16, 5
xxsel 32, 33, 38, 32
xxland 38, 50, 39
vadduwm 15, 16, 15
vperm 7, 11, 13, 12
xxlxor 49, 47, 0
vadduwm 15, 15, 13
vperm 17, 17, 17, 2
vperm 6, 6, 6, 8
vadduwm 14, 17, 14
xxlxor 48, 46, 48
vrlw 16, 16, 3
vadduwm 15, 16, 15
xxlxor 49, 47, 49
xxsldwi 47, 47, 47, 3
vrlw 17, 17, 4
vadduwm 15, 15, 19
vadduwm 14, 17, 14
xxswapd 0, 49
xxlxor 48, 46, 48
xxsldwi 46, 46, 46, 1
vrlw 16, 16, 5
vadduwm 15, 16, 15
xxlxor 49, 47, 0
vadduwm 0, 15, 0
vperm 17, 17, 17, 2
xxland 0, 39, 41
xxlor 38, 38, 0
vadduwm 14, 17, 14
xxlxor 48, 46, 48
vrlw 16, 16, 3
vadduwm 0, 16, 0
xxlxor 33, 32, 49
xxsldwi 32, 32, 32, 1
vrlw 1, 1, 4
vadduwm 0, 0, 6
vadduwm 8, 1, 14
xxswapd 0, 33
xxlxor 44, 40, 48
xxsldwi 38, 40, 40, 3
vrlw 7, 12, 5
vadduwm 0, 7, 0
xxlxor 33, 32, 0
vperm 2, 1, 1, 2
vmrghw 1, 13, 11
vadduwm 6, 2, 6
vperm 1, 1, 18, 10
xxlxor 39, 38, 39
vrlw 3, 7, 3
vadduwm 0, 0, 1
vadduwm 0, 3, 0
xxlxor 34, 32, 34
xxsldwi 0, 32, 32, 3
vrlw 2, 2, 4
vadduwm 4, 2, 6
xxswapd 2, 34
xxlxor 35, 36, 35
xxsldwi 1, 36, 36, 1
vrlw 3, 3, 5
xxlxor 0, 1, 0
xxswapd 0, 0
xxlxor 3, 35, 2
stxvd2x 0, 0, 8
xxswapd 3, 3
stxvd2x 3, 8, 5
lfdx 0, 0, 3
lfd 3, 8(3)
xxmrghd 34, 3, 0
xxlxor 0, 1, 34
xxswapd 0, 0
stxvd2x 0, 8, 10
lfd 0, 16(3)
lfd 1, 24(3)
li 3, -32
xxmrghd 34, 1, 0
xxlxor 0, 2, 34
xxswapd 0, 0
stxvd2x 0, 8, 6
lxvd2x 63, 1, 3
li 3, -48
ld 30, -16(1)
lxvd2x 62, 1, 3
li 3, -64
lxvd2x 61, 1, 3
li 3, -80
lxvd2x 60, 1, 3
blr
.long 0
.quad 0
.Lfunc_end1:
.size zfs_blake3_compress_xof_sse2, .Lfunc_end1-.Lfunc_begin1
.cfi_endproc
.globl zfs_blake3_hash_many_sse2
.p2align 2
.type zfs_blake3_hash_many_sse2,@function
zfs_blake3_hash_many_sse2:
.Lfunc_begin2:
.cfi_startproc
.Lfunc_gep2:
addis 2, 12, .TOC.-.Lfunc_gep2@ha
addi 2, 2, .TOC.-.Lfunc_gep2@l
.Lfunc_lep2:
.localentry zfs_blake3_hash_many_sse2, .Lfunc_lep2-.Lfunc_gep2
mfocrf 12, 32
mflr 0
std 0, 16(1)
stw 12, 8(1)
stdu 1, -256(1)
.cfi_def_cfa_offset 256
.cfi_offset lr, 16
.cfi_offset r17, -120
.cfi_offset r18, -112
.cfi_offset r19, -104
.cfi_offset r20, -96
.cfi_offset r21, -88
.cfi_offset r22, -80
.cfi_offset r23, -72
.cfi_offset r24, -64
.cfi_offset r25, -56
.cfi_offset r26, -48
.cfi_offset r27, -40
.cfi_offset r28, -32
.cfi_offset r29, -24
.cfi_offset r30, -16
.cfi_offset cr2, 8
std 26, 208(1)
mr 26, 4
cmpldi 1, 4, 4
andi. 4, 8, 1
std 18, 144(1)
std 19, 152(1)
crmove 8, 1
ld 19, 360(1)
lwz 18, 352(1)
std 24, 192(1)
std 25, 200(1)
std 27, 216(1)
std 28, 224(1)
mr 24, 10
mr 28, 6
mr 27, 5
mr 25, 3
std 29, 232(1)
std 30, 240(1)
mr 30, 9
mr 29, 7
std 17, 136(1)
std 20, 160(1)
std 21, 168(1)
std 22, 176(1)
std 23, 184(1)
blt 1, .LBB2_3
li 3, 0
li 4, 1
clrldi 23, 30, 32
isel 22, 4, 3, 8
clrldi 21, 24, 32
clrldi 20, 18, 32
.LBB2_2:
mr 3, 25
mr 4, 27
mr 5, 28
mr 6, 29
mr 7, 22
mr 8, 23
mr 9, 21
mr 10, 20
std 19, 32(1)
bl blake3_hash4_sse2
addi 26, 26, -4
addi 3, 29, 4
addi 25, 25, 32
addi 19, 19, 128
cmpldi 26, 3
isel 29, 3, 29, 8
bgt 0, .LBB2_2
.LBB2_3:
cmpldi 26, 0
beq 0, .LBB2_11
li 3, 0
li 4, 1
or 21, 24, 30
li 20, 16
addi 24, 1, 96
isel 22, 4, 3, 8
.LBB2_5:
lxvd2x 0, 28, 20
ld 23, 0(25)
mr 17, 27
mr 3, 21
stxvd2x 0, 24, 20
lxvd2x 0, 0, 28
stxvd2x 0, 0, 24
.LBB2_6:
cmpldi 17, 1
beq 0, .LBB2_8
cmpldi 17, 0
bne 0, .LBB2_9
b .LBB2_10
.LBB2_8:
or 3, 3, 18
.LBB2_9:
clrldi 7, 3, 56
mr 3, 24
mr 4, 23
li 5, 64
mr 6, 29
bl zfs_blake3_compress_in_place_sse2
addi 23, 23, 64
addi 17, 17, -1
mr 3, 30
b .LBB2_6
.LBB2_10:
lxvd2x 0, 24, 20
addi 26, 26, -1
add 29, 29, 22
addi 25, 25, 8
cmpldi 26, 0
stxvd2x 0, 19, 20
lxvd2x 0, 0, 24
stxvd2x 0, 0, 19
addi 19, 19, 32
bne 0, .LBB2_5
.LBB2_11:
ld 30, 240(1)
ld 29, 232(1)
ld 28, 224(1)
ld 27, 216(1)
ld 26, 208(1)
ld 25, 200(1)
ld 24, 192(1)
ld 23, 184(1)
ld 22, 176(1)
ld 21, 168(1)
ld 20, 160(1)
ld 19, 152(1)
ld 18, 144(1)
ld 17, 136(1)
addi 1, 1, 256
ld 0, 16(1)
lwz 12, 8(1)
mtocrf 32, 12
mtlr 0
blr
.long 0
.quad 0
.Lfunc_end2:
.size zfs_blake3_hash_many_sse2, .Lfunc_end2-.Lfunc_begin2
.cfi_endproc
.section .rodata.cst16,"aM",@progbits,16
.p2align 4
.LCPI3_0:
.quad 4294967296
.quad 12884901890
.LCPI3_1:
.byte 29
.byte 28
.byte 31
.byte 30
.byte 25
.byte 24
.byte 27
.byte 26
.byte 21
.byte 20
.byte 23
.byte 22
.byte 17
.byte 16
.byte 19
.byte 18
.LCPI3_2:
.long 1779033703
.long 1779033703
.long 1779033703
.long 1779033703
.LCPI3_3:
.long 3144134277
.long 3144134277
.long 3144134277
.long 3144134277
.LCPI3_4:
.long 1013904242
.long 1013904242
.long 1013904242
.long 1013904242
.LCPI3_5:
.long 2773480762
.long 2773480762
.long 2773480762
.long 2773480762
.text
.p2align 2
.type blake3_hash4_sse2,@function
blake3_hash4_sse2:
.Lfunc_begin3:
.cfi_startproc
.Lfunc_gep3:
addis 2, 12, .TOC.-.Lfunc_gep3@ha
addi 2, 2, .TOC.-.Lfunc_gep3@l
.Lfunc_lep3:
.localentry blake3_hash4_sse2, .Lfunc_lep3-.Lfunc_gep3
stdu 1, -400(1)
.cfi_def_cfa_offset 400
.cfi_offset r22, -152
.cfi_offset r23, -144
.cfi_offset r24, -136
.cfi_offset r25, -128
.cfi_offset r26, -120
.cfi_offset r27, -112
.cfi_offset r28, -104
.cfi_offset r29, -96
.cfi_offset r30, -88
.cfi_offset f23, -72
.cfi_offset f24, -64
.cfi_offset f25, -56
.cfi_offset f26, -48
.cfi_offset f27, -40
.cfi_offset f28, -32
.cfi_offset f29, -24
.cfi_offset f30, -16
.cfi_offset f31, -8
.cfi_offset v20, -352
.cfi_offset v21, -336
.cfi_offset v22, -320
.cfi_offset v23, -304
.cfi_offset v24, -288
.cfi_offset v25, -272
.cfi_offset v26, -256
.cfi_offset v27, -240
.cfi_offset v28, -224
.cfi_offset v29, -208
.cfi_offset v30, -192
.cfi_offset v31, -176
li 11, 48
li 0, 8
std 30, 312(1)
li 30, 12
li 12, 4
lfiwzx 0, 0, 5
stxvd2x 52, 1, 11
li 11, 64
lfiwzx 2, 5, 0
li 0, 20
lfiwzx 3, 5, 30
stxvd2x 53, 1, 11
li 11, 80
li 30, 24
lfiwzx 4, 5, 0
li 0, 28
stxvd2x 54, 1, 11
li 11, 96
lfiwzx 1, 5, 12
lfiwzx 6, 5, 30
xxspltw 45, 0, 1
cmpldi 4, 0
std 22, 248(1)
stxvd2x 55, 1, 11
li 11, 112
lfiwzx 7, 5, 0
xxspltw 40, 2, 1
std 23, 256(1)
xxspltw 38, 3, 1
xxspltw 50, 4, 1
std 24, 264(1)
std 25, 272(1)
std 26, 280(1)
xxspltw 54, 7, 1
std 27, 288(1)
std 28, 296(1)
std 29, 304(1)
stxvd2x 56, 1, 11
li 11, 128
stfd 23, 328(1)
stxvd2x 57, 1, 11
li 11, 144
stfd 24, 336(1)
stxvd2x 58, 1, 11
li 11, 160
stfd 25, 344(1)
stxvd2x 59, 1, 11
li 11, 176
xxspltw 59, 1, 1
stxvd2x 60, 1, 11
li 11, 192
stfd 26, 352(1)
stxvd2x 61, 1, 11
li 11, 208
stfd 27, 360(1)
stxvd2x 62, 1, 11
li 11, 224
xxspltw 62, 6, 1
stxvd2x 63, 1, 11
li 11, 16
stfd 28, 368(1)
lfiwzx 5, 5, 11
ld 5, 432(1)
stfd 29, 376(1)
stfd 30, 384(1)
stfd 31, 392(1)
xxspltw 61, 5, 1
beq 0, .LBB3_5
addis 30, 2, .LCPI3_0@toc@ha
neg 7, 7
xxleqv 34, 34, 34
addis 28, 2, .LCPI3_2@toc@ha
addis 27, 2, .LCPI3_3@toc@ha
addis 26, 2, .LCPI3_4@toc@ha
addis 25, 2, .LCPI3_5@toc@ha
ld 29, 24(3)
addi 0, 30, .LCPI3_0@toc@l
mtfprwz 1, 7
addis 7, 2, .LCPI3_1@toc@ha
ld 30, 16(3)
lxvd2x 0, 0, 0
mtfprwz 2, 6
rldicl 6, 6, 32, 32
addi 0, 7, .LCPI3_1@toc@l
ld 7, 8(3)
vslw 2, 2, 2
lvx 5, 0, 0
addi 0, 28, .LCPI3_2@toc@l
addi 28, 27, .LCPI3_3@toc@l
addi 27, 26, .LCPI3_4@toc@l
addi 26, 25, .LCPI3_5@toc@l
or 25, 9, 8
li 9, 0
xxspltw 36, 2, 1
xxswapd 35, 0
xxspltw 0, 1, 1
xxland 35, 0, 35
mtfprwz 0, 6
ld 6, 0(3)
addi 3, 3, -8
vadduwm 4, 3, 4
xxlor 35, 35, 34
xxlxor 34, 36, 34
xxlor 9, 36, 36
vspltisw 4, 4
vcmpgtsw 2, 3, 2
xxspltw 35, 0, 1
xxlor 10, 36, 36
vsubuwm 2, 3, 2
xxlor 11, 34, 34
lvx 2, 0, 0
li 0, 32
xxlor 12, 34, 34
lvx 2, 0, 28
li 28, 48
xxlor 13, 34, 34
lvx 2, 0, 27
li 27, 0
xxlor 31, 34, 34
lvx 2, 0, 26
xxlor 30, 34, 34
.LBB3_2:
mr 26, 27
addi 27, 27, 1
xxlor 28, 40, 40
cmpld 27, 4
sldi 26, 26, 6
xxlor 24, 45, 45
iseleq 24, 10, 9
add 23, 6, 26
add 22, 30, 26
lxvd2x 0, 6, 26
lxvd2x 1, 7, 26
or 25, 24, 25
add 24, 7, 26
lxvd2x 2, 30, 26
lxvd2x 3, 29, 26
xxlor 29, 38, 38
lxvd2x 4, 23, 11
lxvd2x 6, 24, 11
clrlwi 25, 25, 24
lxvd2x 7, 22, 11
lxvd2x 8, 23, 0
mtfprd 5, 25
add 25, 29, 26
xxswapd 34, 0
lxvd2x 0, 25, 11
xxswapd 36, 1
xxswapd 33, 2
lxvd2x 1, 24, 0
lxvd2x 2, 22, 0
xxswapd 39, 3
xxswapd 32, 4
lxvd2x 3, 25, 0
lxvd2x 4, 23, 28
xxswapd 49, 6
xxswapd 51, 7
lxvd2x 6, 24, 28
xxswapd 58, 8
lxvd2x 7, 22, 28
lxvd2x 8, 25, 28
xxswapd 60, 0
mr 25, 3
xxswapd 57, 1
xxswapd 53, 2
xxswapd 52, 3
xxswapd 56, 4
xxswapd 55, 6
xxswapd 0, 5
xxswapd 40, 7
xxswapd 41, 8
mtctr 12
.LBB3_3:
ldu 24, 8(25)
add 24, 24, 26
addi 24, 24, 256
dcbt 0, 24
bdnz .LBB3_3
vmrgew 3, 4, 2
vspltisw 31, 9
mr 25, 8
vmrglw 10, 4, 2
vspltisw 14, 10
vmrghw 6, 4, 2
xxspltw 0, 0, 3
vmrgew 4, 17, 0
vmrglw 11, 17, 0
vmrghw 16, 17, 0
vmrgew 0, 25, 26
vmrgew 13, 7, 1
vmrglw 2, 7, 1
vmrghw 7, 7, 1
xxlor 25, 36, 36
vmrgew 4, 28, 19
xxlor 26, 32, 32
vmrglw 0, 25, 26
vmrglw 1, 28, 19
xxmrgld 47, 34, 42
xxlor 44, 28, 28
vmrghw 25, 25, 26
xxlor 23, 36, 36
vmrghw 4, 28, 19
vspltisw 19, -16
xxlor 5, 32, 32
vmrgew 0, 20, 21
xxmrgld 34, 33, 43
vmrglw 28, 20, 21
vmrghw 21, 20, 21
vmrglw 20, 23, 24
vmrghw 26, 23, 24
vmrglw 17, 9, 8
xxlor 8, 32, 32
vmrgew 0, 23, 24
xxmrgld 56, 39, 38
vmrgew 23, 9, 8
xxlor 33, 24, 24
xxlor 2, 34, 34
vadduwm 11, 15, 1
xxmrgld 33, 36, 48
xxlor 6, 47, 47
xxlor 27, 32, 32
vmrghw 0, 9, 8
vspltisw 9, 12
vsubuwm 8, 31, 19
xxmrgld 51, 23, 25
vadduwm 31, 2, 12
xxlor 34, 10, 10
vadduwm 10, 14, 14
vslw 15, 2, 2
xxlor 34, 29, 29
vadduwm 14, 24, 27
xxlor 24, 48, 48
vadduwm 16, 1, 2
xxmrgld 34, 45, 35
vadduwm 31, 31, 30
xxmrghd 36, 36, 24
vadduwm 11, 11, 29
vadduwm 14, 14, 18
vadduwm 13, 16, 22
xxlxor 47, 63, 47
xxlor 1, 9, 9
xxlor 1, 11, 11
xxlxor 48, 43, 9
vadduwm 11, 11, 2
xxlor 7, 34, 34
xxmrghd 34, 39, 38
xxlxor 39, 46, 11
xxlor 1, 50, 50
xxlxor 50, 45, 0
vperm 15, 15, 15, 5
vperm 16, 16, 16, 5
vperm 7, 7, 7, 5
vperm 18, 18, 18, 5
xxlor 4, 33, 33
xxlor 33, 31, 31
vadduwm 14, 14, 2
xxlor 3, 34, 34
xxlor 34, 12, 12
xxlor 35, 13, 13
vadduwm 6, 15, 1
xxlor 33, 30, 30
vadduwm 2, 16, 2
vadduwm 3, 7, 3
vadduwm 12, 18, 1
xxlxor 59, 34, 61
xxlxor 61, 35, 1
xxlxor 33, 38, 62
xxlxor 62, 44, 54
vrlw 22, 27, 10
vrlw 29, 29, 10
vrlw 1, 1, 10
vrlw 30, 30, 10
vadduwm 31, 31, 19
vadduwm 13, 13, 4
vadduwm 11, 22, 11
vadduwm 14, 29, 14
vadduwm 31, 1, 31
vadduwm 13, 30, 13
vadduwm 9, 9, 9
xxlor 1, 36, 36
xxlxor 48, 43, 48
xxlxor 36, 46, 39
xxmrgld 39, 60, 5
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 28, 4, 9
xxmrgld 36, 53, 57
vrlw 15, 15, 9
xxmrghd 57, 53, 57
vrlw 18, 18, 9
vadduwm 14, 14, 4
xxlor 0, 36, 36
xxmrgld 36, 49, 52
vadduwm 2, 16, 2
xxmrgld 49, 8, 26
vadduwm 3, 28, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 54, 34, 54
xxlxor 61, 35, 61
xxlxor 33, 38, 33
xxlxor 62, 44, 62
vrlw 29, 29, 8
vrlw 20, 1, 8
xxmrgld 33, 55, 27
vrlw 30, 30, 8
vrlw 22, 22, 8
vadduwm 11, 11, 7
xxlor 5, 39, 39
xxmrgld 39, 32, 58
vadduwm 31, 31, 4
vadduwm 11, 29, 11
vadduwm 13, 13, 7
vadduwm 14, 20, 14
vadduwm 31, 30, 31
vadduwm 13, 22, 13
xxlor 28, 36, 36
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 60
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vadduwm 11, 11, 17
vmr 28, 17
xxmrghd 49, 32, 58
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 21, 4, 2
vadduwm 3, 15, 3
xxlxor 34, 38, 61
xxlxor 61, 44, 52
xxlxor 62, 53, 62
xxlxor 54, 35, 54
vrlw 20, 2, 10
vrlw 29, 29, 10
vrlw 0, 30, 10
vrlw 30, 22, 10
vadduwm 14, 14, 25
vadduwm 31, 31, 1
vadduwm 13, 13, 17
vadduwm 11, 20, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vadduwm 11, 11, 24
xxlor 8, 56, 56
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 21
vadduwm 3, 15, 3
xxlxor 55, 38, 52
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 56, 32
vrlw 30, 30, 8
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
xxlor 25, 51, 51
vmr 26, 17
xxlor 49, 3, 3
xxlor 52, 1, 1
xxlor 51, 2, 2
vadduwm 14, 14, 17
vadduwm 31, 31, 20
vadduwm 13, 13, 19
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
xxlor 29, 39, 39
xxlor 59, 4, 4
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 30, 30, 10
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
xxlor 53, 0, 0
xxlor 39, 6, 6
vadduwm 11, 11, 27
vadduwm 14, 14, 21
vadduwm 31, 31, 7
vadduwm 13, 13, 1
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
xxlor 34, 7, 7
vadduwm 31, 31, 28
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vrlw 30, 30, 8
vadduwm 11, 11, 2
xxlor 34, 28, 28
vadduwm 13, 13, 26
vadduwm 14, 14, 2
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
xxlor 2, 58, 58
xxlor 39, 25, 25
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 32, 56, 32
xxlxor 62, 35, 62
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vrlw 30, 30, 10
xxlor 54, 29, 29
xxlor 58, 5, 5
vadduwm 11, 11, 25
vadduwm 14, 14, 7
vadduwm 31, 31, 22
vadduwm 13, 13, 26
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vadduwm 11, 11, 17
vadduwm 14, 14, 21
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 56, 32
vrlw 30, 30, 8
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vadduwm 31, 31, 1
vadduwm 13, 13, 20
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
xxlor 0, 33, 33
xxlor 33, 8, 8
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 30, 30, 10
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vadduwm 11, 11, 19
vadduwm 14, 14, 2
vadduwm 31, 31, 1
vadduwm 13, 13, 22
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
vadduwm 11, 11, 27
vadduwm 14, 14, 28
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vrlw 30, 30, 8
vadduwm 31, 31, 25
vadduwm 13, 13, 26
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
xxlor 3, 7, 7
vadduwm 11, 11, 7
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 32, 56, 32
xxlxor 62, 35, 62
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vrlw 30, 30, 10
xxlor 33, 6, 6
xxlor 58, 2, 2
xxlor 39, 3, 3
vadduwm 14, 14, 1
vadduwm 31, 31, 26
vadduwm 13, 13, 7
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
xxlor 52, 0, 0
vadduwm 11, 11, 21
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 56, 32
vrlw 30, 30, 8
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vadduwm 14, 14, 2
vadduwm 31, 31, 22
vadduwm 13, 13, 20
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
xxlor 7, 49, 49
vmr 17, 2
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 30, 30, 10
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
xxlor 54, 1, 1
xxlor 34, 7, 7
vadduwm 11, 11, 22
vadduwm 14, 14, 28
vadduwm 31, 31, 2
vadduwm 13, 13, 26
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
xxlor 59, 25, 25
vadduwm 11, 11, 19
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vrlw 30, 30, 8
vadduwm 14, 14, 25
vadduwm 31, 31, 27
vadduwm 13, 13, 7
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vmr 2, 19
xxlor 0, 7, 7
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 32, 56, 32
xxlxor 62, 35, 62
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vrlw 30, 30, 10
xxlor 1, 51, 51
xxlor 7, 39, 39
xxlor 51, 8, 8
xxlor 39, 5, 5
xxlor 34, 4, 4
vadduwm 11, 11, 1
vadduwm 14, 14, 19
vadduwm 31, 31, 7
vadduwm 13, 13, 2
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
xxlor 2, 53, 53
vmr 21, 28
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 56, 32
vrlw 30, 30, 8
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
xxlor 53, 29, 29
vadduwm 11, 11, 17
vadduwm 14, 14, 28
vadduwm 31, 31, 26
vadduwm 13, 13, 21
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
vadduwm 11, 11, 20
xxlor 5, 52, 52
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 30, 30, 10
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
xxlor 52, 2, 2
vadduwm 14, 14, 25
vadduwm 31, 31, 20
vadduwm 13, 13, 7
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
vadduwm 11, 11, 22
vadduwm 14, 14, 27
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vrlw 30, 30, 8
vadduwm 31, 31, 1
vadduwm 13, 13, 2
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
xxlor 3, 29, 29
xxlor 4, 49, 49
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 32, 56, 32
xxlxor 62, 35, 62
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vrlw 30, 30, 10
vmr 17, 28
xxlor 2, 54, 54
xxlor 3, 34, 34
xxlor 34, 8, 8
xxlor 51, 0, 0
xxlor 60, 7, 7
xxlor 54, 1, 1
vadduwm 11, 11, 2
vadduwm 14, 14, 19
vadduwm 31, 31, 28
vadduwm 13, 13, 22
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vadduwm 11, 11, 17
vadduwm 14, 14, 25
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 56, 32
vrlw 30, 30, 8
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vadduwm 31, 31, 7
vadduwm 13, 13, 26
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
xxlor 6, 39, 39
xxlor 39, 4, 4
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 30, 30, 10
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vadduwm 11, 11, 21
vadduwm 14, 14, 27
vadduwm 31, 31, 7
vadduwm 13, 13, 28
vadduwm 11, 30, 11
vadduwm 14, 23, 14
vadduwm 31, 29, 31
vadduwm 13, 0, 13
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 63, 47
xxlxor 50, 45, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
xxlor 0, 49, 49
xxlor 49, 5, 5
vadduwm 24, 16, 24
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 62, 56, 62
xxlxor 55, 35, 55
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 23, 23, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
vrlw 30, 30, 8
vadduwm 11, 11, 17
vadduwm 14, 14, 1
vadduwm 31, 31, 2
vadduwm 13, 13, 22
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
xxlor 34, 3, 3
xxlor 49, 2, 2
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 24, 4, 24
vadduwm 3, 15, 3
xxlxor 55, 38, 55
xxlxor 61, 44, 61
xxlxor 32, 56, 32
xxlxor 62, 35, 62
vrlw 23, 23, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
vrlw 30, 30, 10
vadduwm 11, 11, 19
vadduwm 14, 14, 20
vadduwm 31, 31, 2
vadduwm 13, 13, 17
vadduwm 11, 23, 11
vadduwm 14, 29, 14
vadduwm 31, 0, 31
vadduwm 13, 30, 13
xxlxor 50, 43, 50
xxlxor 48, 46, 48
xxlxor 36, 63, 36
xxlxor 47, 45, 47
vrlw 18, 18, 9
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vadduwm 14, 14, 27
vadduwm 11, 11, 25
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 27, 4, 24
vadduwm 3, 15, 3
xxlxor 57, 38, 55
xxlxor 61, 44, 61
xxlxor 62, 35, 62
xxlxor 32, 59, 32
xxlor 39, 7, 7
vrlw 30, 30, 8
vrlw 25, 25, 8
vrlw 29, 29, 8
vrlw 0, 0, 8
xxlor 1, 58, 58
vmr 26, 19
vadduwm 19, 31, 7
xxlor 39, 6, 6
vadduwm 11, 30, 11
vadduwm 7, 13, 7
vadduwm 13, 25, 14
vadduwm 14, 29, 19
vadduwm 7, 0, 7
xxlxor 48, 43, 48
xxlxor 36, 45, 36
xxlxor 47, 46, 47
xxlxor 50, 39, 50
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
vperm 18, 18, 18, 5
xxlor 51, 1, 1
vadduwm 13, 13, 1
vadduwm 11, 11, 19
vadduwm 19, 16, 27
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 63, 51, 62
xxlxor 62, 35, 57
xxlxor 61, 38, 61
xxlxor 32, 44, 32
vrlw 31, 31, 10
vrlw 30, 30, 10
vrlw 29, 29, 10
vrlw 0, 0, 10
xxlor 33, 0, 0
vadduwm 7, 7, 2
vadduwm 14, 14, 1
vadduwm 11, 31, 11
vadduwm 13, 30, 13
vadduwm 14, 29, 14
vadduwm 7, 0, 7
xxlxor 48, 43, 48
xxlxor 36, 45, 36
xxlxor 47, 46, 47
xxlxor 50, 39, 50
vrlw 16, 16, 9
vrlw 4, 4, 9
vrlw 15, 15, 9
vrlw 18, 18, 9
xxlor 60, 8, 8
vadduwm 1, 11, 21
vadduwm 11, 13, 28
vadduwm 13, 16, 19
vadduwm 3, 4, 3
vadduwm 6, 15, 6
vadduwm 12, 18, 12
xxlxor 51, 45, 63
xxlxor 63, 35, 62
xxlxor 62, 38, 61
xxlxor 32, 44, 32
vrlw 31, 31, 8
vrlw 30, 30, 8
vrlw 0, 0, 8
vrlw 19, 19, 8
vadduwm 14, 14, 26
vadduwm 7, 7, 17
vadduwm 1, 31, 1
vadduwm 11, 30, 11
vadduwm 14, 0, 14
vadduwm 7, 19, 7
xxlxor 50, 33, 50
xxlxor 48, 43, 48
xxlxor 36, 46, 36
xxlxor 47, 39, 47
vperm 18, 18, 18, 5
vperm 16, 16, 16, 5
vperm 4, 4, 4, 5
vperm 15, 15, 15, 5
xxlor 34, 4, 4
vadduwm 14, 14, 22
vadduwm 6, 18, 6
vadduwm 12, 16, 12
vadduwm 13, 4, 13
vadduwm 3, 15, 3
xxlxor 49, 38, 63
xxlxor 63, 44, 62
xxlxor 32, 45, 32
xxlxor 51, 35, 51
vrlw 17, 17, 10
vrlw 31, 31, 10
vrlw 0, 0, 10
vrlw 10, 19, 10
vadduwm 11, 11, 2
xxlor 34, 5, 5
vadduwm 1, 1, 20
vadduwm 2, 7, 2
vadduwm 7, 31, 11
vadduwm 11, 0, 14
vadduwm 2, 10, 2
vadduwm 1, 17, 1
xxlxor 36, 43, 36
xxlxor 46, 34, 47
vrlw 4, 4, 9
vrlw 14, 14, 9
xxlxor 47, 33, 50
xxlxor 48, 39, 48
vrlw 15, 15, 9
vrlw 9, 16, 9
vadduwm 13, 4, 13
vadduwm 3, 14, 3
xxlxor 32, 45, 32
xxlxor 45, 45, 33
xxlxor 33, 35, 42
xxlxor 59, 35, 39
vadduwm 3, 15, 6
vadduwm 6, 9, 12
xxlxor 39, 35, 49
xxlxor 42, 38, 63
vrlw 1, 1, 8
vrlw 7, 7, 8
vrlw 10, 10, 8
vrlw 0, 0, 8
xxlxor 40, 35, 43
xxlxor 38, 38, 34
xxlxor 61, 33, 41
xxlxor 50, 39, 36
xxlxor 62, 42, 46
xxlxor 54, 32, 47
bne 0, .LBB3_2
.LBB3_5:
vmrglw 2, 27, 13
li 3, 32
li 4, 48
vmrglw 4, 6, 8
vmrglw 0, 18, 29
vmrglw 1, 22, 30
vmrghw 3, 27, 13
vmrghw 5, 6, 8
vmrghw 6, 18, 29
vmrghw 7, 22, 30
xxmrgld 40, 36, 34
xxmrghd 34, 36, 34
xxmrgld 41, 33, 32
xxswapd 0, 40
xxmrgld 36, 37, 35
xxmrghd 35, 37, 35
xxmrghd 37, 33, 32
xxswapd 1, 41
xxmrgld 32, 39, 38
xxmrghd 33, 39, 38
xxswapd 2, 34
xxswapd 4, 36
xxswapd 3, 37
stxvd2x 0, 0, 5
xxswapd 5, 32
stxvd2x 1, 5, 11
xxswapd 0, 35
xxswapd 1, 33
stxvd2x 2, 5, 3
li 3, 64
stxvd2x 3, 5, 4
li 4, 80
stxvd2x 4, 5, 3
li 3, 96
stxvd2x 5, 5, 4
li 4, 112
stxvd2x 0, 5, 3
stxvd2x 1, 5, 4
li 3, 224
lxvd2x 63, 1, 3
li 3, 208
lfd 31, 392(1)
ld 30, 312(1)
ld 29, 304(1)
lxvd2x 62, 1, 3
li 3, 192
lfd 30, 384(1)
ld 28, 296(1)
ld 27, 288(1)
lxvd2x 61, 1, 3
li 3, 176
lfd 29, 376(1)
ld 26, 280(1)
ld 25, 272(1)
lxvd2x 60, 1, 3
li 3, 160
lfd 28, 368(1)
ld 24, 264(1)
ld 23, 256(1)
lxvd2x 59, 1, 3
li 3, 144
lfd 27, 360(1)
ld 22, 248(1)
lxvd2x 58, 1, 3
li 3, 128
lfd 26, 352(1)
lxvd2x 57, 1, 3
li 3, 112
lfd 25, 344(1)
lxvd2x 56, 1, 3
li 3, 96
lfd 24, 336(1)
lxvd2x 55, 1, 3
li 3, 80
lfd 23, 328(1)
lxvd2x 54, 1, 3
li 3, 64
lxvd2x 53, 1, 3
li 3, 48
lxvd2x 52, 1, 3
addi 1, 1, 400
blr
.long 0
.quad 0
.Lfunc_end3:
.size blake3_hash4_sse2, .Lfunc_end3-.Lfunc_begin3
.cfi_endproc
.section ".note.GNU-stack","",@progbits
#endif
Reference in New Issue View Git Blame Copy Permalink