sergiotarxz/nuttx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

arch_memcpy: Optimize arch memcpy for armv7-m and armv8-m

Browse Source 
Use ldm and stm instruction to optimize performance when
both src and dst are 32-bit aligned.

Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
This commit is contained in:
zhangyuan21 2023-06-19 20:48:24 +08:00 committed by Petro Karashchenko
parent d563717827
commit 3625385541
2 changed files with 42 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
libs/libc/machine/arm/armv7-m/gnu/arch_memcpy.S
View File

@ -66,12 +66,18 @@
#if __OPT_BIG_BLOCK_SIZE == 16 #if __OPT_BIG_BLOCK_SIZE == 16
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12 .irp offset, 0,4,8,12
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0
#elif __OPT_BIG_BLOCK_SIZE == 32 #elif __OPT_BIG_BLOCK_SIZE == 32
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12,16,20,24,28 .irp offset, 0,4,8,12,16,20,24,28
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0,16
#elif __OPT_BIG_BLOCK_SIZE == 64 #elif __OPT_BIG_BLOCK_SIZE == 64
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60 .irp offset, 0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0,16,32,48
#else #else
#error "Illegal __OPT_BIG_BLOCK_SIZE" #error "Illegal __OPT_BIG_BLOCK_SIZE"
#endif #endif
@ -113,6 +119,21 @@ memcpy:
ands r3, r3, #3 ands r3, r3, #3
bne .Lmisaligned_copy bne .Lmisaligned_copy
.Lbig_aligned:
subs r2, __OPT_BIG_BLOCK_SIZE
blo .Lmid_block
stmfd sp!, {r4-r7}
.Lbig_aligned_loop:
BEGIN_UNROLL_BIG_BLOCK_X4
ldmia r1!, {r4, r5, r6, r7}
stmia r0!, {r4, r5, r6, r7}
END_UNROLL
subs r2, __OPT_BIG_BLOCK_SIZE
bhs .Lbig_aligned_loop
ldmfd sp!, {r4-r7}
b .Lmid_block
.Lbig_block: .Lbig_block:
subs r2, __OPT_BIG_BLOCK_SIZE subs r2, __OPT_BIG_BLOCK_SIZE
blo .Lmid_block blo .Lmid_block

21
libs/libc/machine/arm/armv8-m/gnu/arch_memcpy.S
View File

@ -69,12 +69,18 @@
#if __OPT_BIG_BLOCK_SIZE == 16 #if __OPT_BIG_BLOCK_SIZE == 16
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12 .irp offset, 0,4,8,12
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0
#elif __OPT_BIG_BLOCK_SIZE == 32 #elif __OPT_BIG_BLOCK_SIZE == 32
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12,16,20,24,28 .irp offset, 0,4,8,12,16,20,24,28
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0,16
#elif __OPT_BIG_BLOCK_SIZE == 64 #elif __OPT_BIG_BLOCK_SIZE == 64
#define BEGIN_UNROLL_BIG_BLOCK \ #define BEGIN_UNROLL_BIG_BLOCK \
.irp offset, 0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60 .irp offset, 0,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60
#define BEGIN_UNROLL_BIG_BLOCK_X4 \
.irp offset, 0,16,32,48
#else #else
#error "Illegal __OPT_BIG_BLOCK_SIZE" #error "Illegal __OPT_BIG_BLOCK_SIZE"
#endif #endif
@ -134,6 +140,21 @@ memcpy:
ands r3, r3, #3 ands r3, r3, #3
bne .Lmisaligned_copy bne .Lmisaligned_copy
.Lbig_aligned:
subs r2, __OPT_BIG_BLOCK_SIZE
blo .Lmid_block
stmfd sp!, {r4-r7}
.Lbig_aligned_loop:
BEGIN_UNROLL_BIG_BLOCK_X4
ldmia r1!, {r4, r5, r6, r7}
stmia r0!, {r4, r5, r6, r7}
END_UNROLL
subs r2, __OPT_BIG_BLOCK_SIZE
bhs .Lbig_aligned_loop
ldmfd sp!, {r4-r7}
b .Lmid_block
.Lbig_block: .Lbig_block:
subs r2, __OPT_BIG_BLOCK_SIZE subs r2, __OPT_BIG_BLOCK_SIZE
blo .Lmid_block blo .Lmid_block