ARMv7-a的多寄存器Load/Store

多寄存器的加载与存储

ARMv7-a架构的芯片提供了高效的寄存器数据加载与存储指令，一条指令可以从内存中加载多个寄存器，也可以将多个寄存器的数据存储至内存中。读写的内存必须是连续的；虽然数据量不大，但基于这些指令可以实现高效的数据拷贝、（软）中断的上下文保存、恢复，以及内核的任务切换、应用层的协程实现。Arm Architecture Reference Manual对这些指令做了详尽的说明：

编写多寄存器的加载和存储示例

ARMv7芯片提供了以上多种指令，原因是这些指令在加载、存储数据时，具体功能实现有细微的差别。如加载或存储是递增的(increment)，还是递减的(decrement)？在加载或存储的内存操作在增减之前(before)还是之后(after)？这些为了探究以上指令的具体功能实现及差别，笔者编写了简单的应用，其中汇编代码为：

.macro load_store_entry, ls_func
	.global \ls_func
	.type \ls_func, %function
\ls_func:
	push {r4, r5, r6, lr}
.endm

.macro load_store_exit, ls_func
	pop {r4, r5, r6, pc}
	.size \ls_func, . - \ls_func
	.align 2
.endm

	.arch armv7-a
	.text
	.arm

	load_store_entry ls_stmda
	stmda r0, {r1, r2, r3}
	load_store_exit ls_stmda

	load_store_entry ls_stmda1
	stmda r0!, {r1, r2, r3}
	load_store_exit ls_stmda1

	load_store_entry ls_stmia
	stmia r0, {r1, r2, r3}
	load_store_exit ls_stmia

	load_store_entry ls_stmia1
	stmia r0!, {r1, r2, r3}
	load_store_exit ls_stmia1

	load_store_entry ls_stmdb
	stmdb r0, {r1, r2, r3}
	load_store_exit ls_stmdb

	load_store_entry ls_stmdb1
	stmdb r0!, {r1, r2, r3}
	load_store_exit ls_stmdb1

	load_store_entry ls_stmib
	stmib r0, {r1, r2, r3}
	load_store_exit ls_stmib

	load_store_entry ls_stmib1
	stmib r0!, {r1, r2, r3}
	load_store_exit ls_stmib1

	load_store_entry ls_ldmda
	ldmda r0, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmda

	load_store_entry ls_ldmda1
	ldmda r0!, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmda1

	load_store_entry ls_ldmia
	ldmia r0, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmia

	load_store_entry ls_ldmia1
	ldmia r0!, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmia1

	load_store_entry ls_ldmdb
	ldmdb r0, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmdb

	load_store_entry ls_ldmdb1
	ldmdb r0!, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmdb1

	load_store_entry ls_ldmib
	ldmib r0, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmib

	load_store_entry ls_ldmib1
	ldmib r0!, {r1, r2, r3}
	bl ls_dump_result
	load_store_exit ls_ldmib1

其中ls_dump_result函数定义于C代码文件中，用于将r0, r1, r2, r3四个寄存器的数据输出；为节省篇幅，相关的C代码省略。在设备上测试，最后的结果为：

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmda r0, {r1, r2, r3}, r0: 0x129a120 ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00002019
	[0x129a11c]: 00002020
	[0x129a120]: 00002021
	[0x129a124]: 00000000
	[0x129a128]: 00000000
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmda r0!, {r1, r2, r3}, r0: 0x129a114 ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00002019
	[0x129a11c]: 00002020
	[0x129a120]: 00002021
	[0x129a124]: 00000000
	[0x129a128]: 00000000
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmia r0, {r1, r2, r3}, r0: 0x129a120 ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00000000
	[0x129a11c]: 00000000
	[0x129a120]: 00002019
	[0x129a124]: 00002020
	[0x129a128]: 00002021
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmia r0!, {r1, r2, r3}, r0: 0x129a12c ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00000000
	[0x129a11c]: 00000000
	[0x129a120]: 00002019
	[0x129a124]: 00002020
	[0x129a128]: 00002021
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmdb r0, {r1, r2, r3}, r0: 0x129a120 ->
	[0x129a110]: 00000000
	[0x129a114]: 00002019
	[0x129a118]: 00002020
	[0x129a11c]: 00002021
	[0x129a120]: 00000000
	[0x129a124]: 00000000
	[0x129a128]: 00000000
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmdb r0!, {r1, r2, r3}, r0: 0x129a114 ->
	[0x129a110]: 00000000
	[0x129a114]: 00002019
	[0x129a118]: 00002020
	[0x129a11c]: 00002021
	[0x129a120]: 00000000
	[0x129a124]: 00000000
	[0x129a128]: 00000000
	[0x129a12c]: 00000000
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmib r0, {r1, r2, r3}, r0: 0x129a120 ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00000000
	[0x129a11c]: 00000000
	[0x129a120]: 00000000
	[0x129a124]: 00002019
	[0x129a128]: 00002020
	[0x129a12c]: 00002021
	[0x129a130]: 00000000

=====================================
r0: 0x129a120; r1 - r3: 0x2019, 0x2020, 0x2021
After stmib r0!, {r1, r2, r3}, r0: 0x129a12c ->
	[0x129a110]: 00000000
	[0x129a114]: 00000000
	[0x129a118]: 00000000
	[0x129a11c]: 00000000
	[0x129a120]: 00000000
	[0x129a124]: 00002019
	[0x129a128]: 00002020
	[0x129a12c]: 00002021
	[0x129a130]: 00000000

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmda r0, {r1, r2, r3} ->
	r0: 0x129a110
	r1: 03030303
	r2: 04040404
	r3: 05050505

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmda r0!, {r1, r2, r3} ->
	r0: 0x129a104
	r1: 03030303
	r2: 04040404
	r3: 05050505

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmia r0, {r1, r2, r3} ->
	r0: 0x129a110
	r1: 05050505
	r2: 06060606
	r3: 07070707

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmia r0!, {r1, r2, r3} ->
	r0: 0x129a11c
	r1: 05050505
	r2: 06060606
	r3: 07070707

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmdb r0, {r1, r2, r3} ->
	r0: 0x129a110
	r1: 02020202
	r2: 03030303
	r3: 04040404

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmdb r0!, {r1, r2, r3} ->
	r0: 0x129a104
	r1: 02020202
	r2: 03030303
	r3: 04040404

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmib r0, {r1, r2, r3} ->
	r0: 0x129a110
	r1: 06060606
	r2: 07070707
	r3: 08080808

=====================================
	[0x129a100]: 01010101
	[0x129a104]: 02020202
	[0x129a108]: 03030303
	[0x129a10c]: 04040404
	[0x129a110]: 05050505
	[0x129a114]: 06060606
	[0x129a118]: 07070707
	[0x129a11c]: 08080808
r0: 0x129a110; After ldmib r0!, {r1, r2, r3} ->
	r0: 0x129a11c
	r1: 06060606
	r2: 07070707
	r3: 08080808

函数入口和返回时的寄存器操作

笔者编写的汇编代码中，函数在入口和返回时有push和pop两条指令，分别用于寄存器的保存和恢复；相关的内存为栈空间：

push {r4, r5, r6, lr}
pop {r4, r5, r5, pc}

经调试，这两条指令实际上是：

stmdb sp!, {r4, r5, r6, lr}
ldmia sp!, {r4, r5, r6, pc}