361 lines
10 KiB
ArmAsm
Executable File
361 lines
10 KiB
ArmAsm
Executable File
/*
|
||
* (C) Copyright 2007-2013
|
||
* Allwinner Technology Co., Ltd. <www.allwinnertech.com>
|
||
* Jerry Wang <wangflord@allwinnertech.com>
|
||
*
|
||
* See file CREDITS for list of people who contributed to this
|
||
* project.
|
||
*
|
||
* This program is free software; you can redistribute it and/or
|
||
* modify it under the terms of the GNU General Public License as
|
||
* published by the Free Software Foundation; either version 2 of
|
||
* the License, or (at your option) any later version.
|
||
*
|
||
* This program is distributed in the hope that it will be useful,
|
||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||
* GNU General Public License for more details.
|
||
*
|
||
* You should have received a copy of the GNU General Public License
|
||
* along with this program; if not, write to the Free Software
|
||
* Foundation, Inc., 59 Temple Place, Suite 330, Boston,
|
||
* MA 02111-1307 USA
|
||
*/
|
||
|
||
#include <asm/assembler.h>
|
||
|
||
|
||
/*
|
||
************************************************************************************************************
|
||
*
|
||
* arm_neon_init
|
||
*
|
||
* name :
|
||
*
|
||
* parmeters :
|
||
*
|
||
* return :
|
||
*
|
||
* note :
|
||
*
|
||
*
|
||
************************************************************************************************************
|
||
*/
|
||
.balign 4
|
||
.global arm_neon_init @r0存放起始地址，r1存放数据字节数
|
||
|
||
arm_neon_init:
|
||
|
||
stmfd sp!, {lr}
|
||
|
||
ldr r0, =(0xf<<20)
|
||
mcr p15, 0, r0, c1, c0, 2 @(enable CP10/CP11, and disable ASEDIS/D32DIS)
|
||
|
||
mov r0, #0
|
||
mcr p15, 0, r0, c7, c5, 4 @(CP15ISB)
|
||
|
||
mov r0, #0x40000000
|
||
fmxr fpexc, r0 @(enable NEON)
|
||
|
||
isb @(wait all code are executed including pipeline)
|
||
dsb @(wait all register access are finished)
|
||
|
||
mov r0, #0
|
||
|
||
ldmfd sp!, {pc}
|
||
|
||
|
||
/*
|
||
************************************************************************************************************
|
||
*
|
||
* add_sum_neon
|
||
*
|
||
* name :
|
||
*
|
||
* parmeters : r0, 起始地址 r1, 数据字节数
|
||
*
|
||
* return :
|
||
*
|
||
* note :
|
||
*
|
||
*
|
||
************************************************************************************************************
|
||
*/
|
||
.balign 4
|
||
.global add_sum_neon @r0存放起始地址，r1存放数据字节数
|
||
.arm
|
||
|
||
add_sum_neon:
|
||
stmfd sp!, {r2-r6, lr}
|
||
|
||
mov r2, r0
|
||
mov r3, r1
|
||
mov r5, #0 @初始化r5=0，保存返回值
|
||
vbic.I32 q0, #0x000000ff
|
||
vbic.I32 q0, #0x0000ff00
|
||
vbic.I32 q0, #0x00ff0000
|
||
vbic.I32 q0, #0xff000000
|
||
|
||
cmp r3, #31
|
||
bls __data_deal_32byte_unalign_0
|
||
|
||
__data_read_loop:
|
||
vld1.I32 {d2, d3, d4, d5}, [r2]! @把r2开始的数据，读取8(64/8)*4个字节到d0-d3当中，即Q0到Q1，完成后，r2自动变化到下一个读取地址
|
||
vadd.I32 q3, q1, q2
|
||
vadd.I32 q0, q0, q3 @结果保存到Q0中
|
||
sub r3, r3, #32 @r3(字节数)减去32字节
|
||
cmp r3, #31
|
||
bhs __data_read_loop
|
||
|
||
vadd.I32 d0, d0, d1
|
||
vmov r5, r6, d0
|
||
adds r5, r5, r6 @r5保存返回值
|
||
|
||
__data_deal_32byte_unalign_0:
|
||
and r4, r3, #3 @r4保存非4字节对齐的字节数
|
||
lsr r3, r3, #2 @r3保存4字节对齐的个数
|
||
__data_deal_32byte_unalign_1:
|
||
cmp r3, #0 @查看是否还有未处理的4字节对齐
|
||
beq __data_deal_4byte_unalign
|
||
ldr r6, [r2], #4 @读出数据
|
||
adds r5, r5, r6 @计算累加
|
||
sub r3, r3, #1 @对齐数据减1
|
||
b __data_deal_32byte_unalign_1
|
||
|
||
__data_deal_4byte_unalign: @处理未对齐的数据
|
||
cmp r4, #0
|
||
beq __data_deal_return
|
||
rsb r4, r4, #4
|
||
lsl r4, r4, #3
|
||
mvn r6, #0
|
||
lsr r6, r6, r4
|
||
ldr r3, [r2], #4 @读出数据
|
||
and r6, r3, r6
|
||
adds r5, r5, r6
|
||
|
||
__data_deal_return:
|
||
mov r0, r5
|
||
|
||
ldmfd sp!, {r2-r6, pc}
|
||
/*
|
||
************************************************************************************************************
|
||
*
|
||
* memcpy_neon
|
||
*
|
||
* name :
|
||
*
|
||
* parmeters :
|
||
*
|
||
* return :
|
||
*
|
||
* note :
|
||
*
|
||
*
|
||
************************************************************************************************************
|
||
*/
|
||
.align 4
|
||
.global memcpy_neon
|
||
.arm
|
||
|
||
memcpy_neon:
|
||
cmp r2, #0
|
||
moveq pc, lr
|
||
|
||
cmp r0, r1
|
||
moveq pc, lr
|
||
|
||
stmfd sp!, {r3-r9, lr}
|
||
|
||
mov r3, r0 @r3保存目的地址
|
||
mov r4, r1 @r4保存源地址
|
||
__memcpy_loop:
|
||
cmp r2, #31 @判断传输字节是否小于32字节
|
||
bls __memcpy_neon_32byte_unalign_0
|
||
vld1.I32 {d2,d3,d4,d5}, [r4]! @从r4开始，读取32个字节对Q1，Q2
|
||
vst1.I32 {d2,d3,d4,d5}, [r3]! @把Q1，Q2的数据存放到r3开始的地址
|
||
sub r2, r2, #32
|
||
b __memcpy_loop
|
||
|
||
__memcpy_neon_32byte_unalign_0:
|
||
PLD(pld [r3, #0])
|
||
PLD(pld [r4, #0])
|
||
and r5, r2, #3 @r5保存非4字节对齐的字节数
|
||
lsr r6, r2, #2 @r6保存4字节对齐的个数
|
||
__memcpy_neon_32byte_unalign_1:
|
||
cmp r6, #0
|
||
beq __memcpy_neon_4byte_unalign
|
||
ldr r7, [r4], #4
|
||
str r7, [r3], #4
|
||
sub r6,r6,#1
|
||
b __memcpy_neon_32byte_unalign_1
|
||
__memcpy_neon_4byte_unalign:
|
||
cmp r5, #0
|
||
beq __memcpy_neon_exit
|
||
ldr r7, [r4], #4 @读出源地址最后一个word到r7，肯定会多读取
|
||
ldr r8, [r4] @读出目的地址最后一个word到r8
|
||
mvn r9, #0
|
||
rsb r5, r5, #4
|
||
lsl r5, r5, #3 @r5=r5*8
|
||
lsl r9, r9, r5
|
||
and r8, r8, r9 @目的地址数据r8清零需要写入的位
|
||
mvn r6, #0
|
||
eor r9, r9, r6
|
||
and r7, r7, r9 @源地址数据r7清零不能写入的位
|
||
orr r7, r7, r8 @合并r7，r8，得到正确数据
|
||
str r7, [r3], #4
|
||
|
||
__memcpy_neon_exit:
|
||
ldmfd sp!, {r3-r9, lr}
|
||
|
||
/*
|
||
************************************************************************************************************
|
||
*
|
||
* bmpdecode_neon
|
||
*
|
||
* name :
|
||
*
|
||
* parmeters : r0: bmp原始数据地址
|
||
*
|
||
* r1: 解析后数据存放地址
|
||
*
|
||
* r2: 结构体的地址
|
||
*
|
||
* {
|
||
* int x; x方向像素点个数
|
||
* int y; y方向像素点个数
|
||
* int bmp_bpix; 每个像素的字节数
|
||
* }
|
||
*
|
||
* return : 无
|
||
*
|
||
* note : 采取的策略是，每次读取32个字节(当一行字节数(x像素点个数*每像素点字节数)超过时)
|
||
* 当超过16字节不到32字节，一次读取16字节
|
||
* 当超过8字节且不到16时，一次读取8字节
|
||
* 当不到8字节时且不为0时，一次读取8字节
|
||
*
|
||
* 如果是ARGB格式，如果每行为奇数个像素点，一定多读4字节，然后源地址和目的地址需要修正，减掉4字节
|
||
* 如果每行为偶数个像素点，则刚刚合适，不用修正
|
||
*
|
||
* 如果是RGB格式，每行有效字节数为3x个，补齐字节数为x&3个(4字节对齐)，
|
||
* 由于最少读取8字节，因此需要知道差多少字节可以补齐8字节对齐，然后最后减掉多读取的
|
||
* 算法：3x=有效字节数
|
||
* 3x&7=没有8字节对齐的部分
|
||
* 假设x=8m+n，则n取值为从1到7，3x=3(8m+n)=24m+3n,考虑n的取值(0值直接不考虑)
|
||
* 可能值为3,6,9,12,15,18,21
|
||
* 再和7相与，可能值为3,6,1,4,7,2,5
|
||
* 这也是没有8字节对齐的部分
|
||
* 用8来减掉这些值，得到的是需要补充的字节数，来保证读取能达到一次8字节
|
||
* 多读的值可能是5,2,7,4,1,6,3
|
||
* 考虑对齐因素，需要源地址修正的值应该是(结果&4)
|
||
*
|
||
* 寄存器使用 r0 源地址
|
||
*
|
||
* r1 目的地址
|
||
*
|
||
* r2 结构体地址(读取后就无效)
|
||
* 后面保存x方向像素点个数
|
||
*
|
||
* r3 y方向像素点个数
|
||
*
|
||
* r4 每个像素的字节数
|
||
*
|
||
* r5 目的地址多写入的字节数
|
||
*
|
||
* r6 源地址多读取的字节数
|
||
*
|
||
* r7 一行的有效字节数
|
||
*
|
||
*
|
||
*
|
||
************************************************************************************************************
|
||
*/
|
||
.align 4
|
||
.global bmpdecode_neon
|
||
.arm
|
||
|
||
bmpdecode_neon:
|
||
cmp r2, #0 @检查结构体地址是否合法
|
||
moveq pc, lr
|
||
|
||
cmp r0, r1 @检查原始地址是否等于目的地址
|
||
moveq pc, lr
|
||
|
||
stmfd sp!, {r0-r8, lr}
|
||
|
||
ldr r3, [r2, #4] @r3保存y方向像素点个数
|
||
ldr r4, [r2, #8] @r4保存每个像素的字节数
|
||
ldr r2, [r2] @r2保存x方向像素点个数
|
||
|
||
cmp r4, #0
|
||
beq __bmpdecode_filling_exit
|
||
|
||
cmp r4, #4 @判断一个像素点的字节数
|
||
beq __bmpdecode_x_4pixles
|
||
|
||
@处理一个像素点3个字节的情况
|
||
lsl r7, r4, #2
|
||
sub r7, r7, r4 @r7保存了一行有效的字节数 r7=3*r4
|
||
and r5, r7, #7
|
||
rsb r5, r5, #8 @r5保存了没有8字节对齐的字节部分, 多写入的
|
||
and r6, r5, #4 @r6保存了需要补齐的字节部分，即会多读取的
|
||
|
||
b __bmpdecode_y_loop
|
||
|
||
__bmpdecode_x_4pixles: @处理一个像素点4个字节的情况
|
||
and r5, r2, #1
|
||
cmp r5, #1 @判断是否是奇数个像素点
|
||
lsleq r5, r5, #2 @如果是奇数个，则最终，源地址和目的地址都需要减去4，即多读多写的
|
||
|
||
mov r6, r5
|
||
lsl r7, r4, #2
|
||
|
||
__bmpdecode_y_loop:
|
||
mov r8, r7
|
||
__bmpdecode_x_loop:
|
||
cmp r8, #31 @一次处理32个字节数据
|
||
bls __bmpdecode_filling_32bytes_less
|
||
|
||
vld1.I32 {d2,d3,d4,d5}, [r0]! @一次读取32个字节
|
||
vst1.I32 {d2,d3,d4,d5}, [r1]! @一次存储32个字节
|
||
sub r8, r8, #32
|
||
|
||
b __bmpdecode_x_loop
|
||
|
||
__bmpdecode_filling_32bytes_less:
|
||
cmp r8, #15
|
||
bls __bmpdecode_filling_16bytes_less
|
||
|
||
vld1.I32 {d2, d3}, [r0]! @一次读取16个字节
|
||
vst1.I32 {d2, d3}, [r1]! @一次存储16个字节
|
||
|
||
sub r8, r8, #16
|
||
__bmpdecode_filling_16bytes_less:
|
||
cmp r8, #7
|
||
beq __bmpdecode_filling_8bytes_less
|
||
|
||
vld1.I32 {d2}, [r0]! @一次读取8个字节
|
||
vst1.I32 {d2}, [r1]! @一次存储8个字节
|
||
|
||
sub r8, r8, #8
|
||
__bmpdecode_filling_8bytes_less:
|
||
cmp r8, #0
|
||
beq __bmpdecode_filling_x_finish
|
||
|
||
vld1.I32 {d2}, [r0]! @一次读取8个字节
|
||
vst1.I32 {d2}, [r1]! @一次存储8个字节
|
||
|
||
sub r1, r1, r5 @修正目的地址
|
||
sub r0, r0, r6 @修正源地址
|
||
|
||
__bmpdecode_filling_x_finish:
|
||
|
||
sub r4, r4, #1 @行数减1
|
||
cmp r4, #0
|
||
|
||
bne __bmpdecode_y_loop
|
||
|
||
__bmpdecode_filling_exit:
|
||
|
||
stmfd sp!, {r0-r8, pc}
|
||
|