diff options
author | Compostella, Jeremy <jeremy.compostella@intel.com> | 2020-10-10 04:42:34 +0800 |
---|---|---|
committer | mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> | 2020-10-16 01:12:05 +0000 |
commit | d25fd8710d6c8fc11582210fb1f8480c0d98416b (patch) | |
tree | c8da622ea074b171f13b3244125cd3d05d11c7e7 /MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm | |
parent | 19c87b7d446c3273e84b238cb02cd1c0ae69c43e (diff) | |
download | edk2-d25fd8710d6c8fc11582210fb1f8480c0d98416b.tar.gz edk2-d25fd8710d6c8fc11582210fb1f8480c0d98416b.tar.bz2 edk2-d25fd8710d6c8fc11582210fb1f8480c0d98416b.zip |
BaseMemoryLibSse2: Take advantage of write combining buffers
The current SSE2 implementation of the ZeroMem(), SetMem(),
SetMem16(), SetMem32 and SetMem64 functions is writing 16 bytes per 16
bytes. It hurts the performances so bad that this is even slower than
a simple 'rep stos' (4% slower) in regular DRAM.
To take full advantages of the 'movntdq' instruction it is better to
"queue" a total of 64 bytes in the write combining buffers. This
patch implement such a change. Below is a table where I measured
(with 'rdtsc') the time to write an entire 100MB RAM buffer. These
functions operate almost two times faster.
| Function | Arch | Untouched | 64 bytes | Result |
|----------+------+-----------+----------+--------|
| ZeroMem | Ia32 | 17765947 | 9136062 | 1.945x |
| ZeroMem | X64 | 17525170 | 9233391 | 1.898x |
| SetMem | Ia32 | 17522291 | 9137272 | 1.918x |
| SetMem | X64 | 17949261 | 9176978 | 1.956x |
| SetMem16 | Ia32 | 18219673 | 9372062 | 1.944x |
| SetMem16 | X64 | 17523331 | 9275184 | 1.889x |
| SetMem32 | Ia32 | 18495036 | 9273053 | 1.994x |
| SetMem32 | X64 | 17368864 | 9285885 | 1.870x |
| SetMem64 | Ia32 | 18564473 | 9241362 | 2.009x |
| SetMem64 | X64 | 17506951 | 9280148 | 1.886x |
Signed-off-by: Jeremy Compostella <jeremy.compostella@intel.com>
Reviewed-by: Liming Gao <gaoliming@byosoft.com.cn>
Diffstat (limited to 'MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm')
-rw-r--r-- | MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm | 11 |
1 files changed, 7 insertions, 4 deletions
diff --git a/MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm b/MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm index 6e308b5594..d461ee086c 100644 --- a/MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm +++ b/MdePkg/Library/BaseMemoryLibSse2/Ia32/SetMem16.nasm @@ -33,7 +33,7 @@ ASM_PFX(InternalMemSetMem16): mov edi, [esp + 8]
xor ecx, ecx
sub ecx, edi
- and ecx, 15 ; ecx + edi aligns on 16-byte boundary
+ and ecx, 63 ; ecx + edi aligns on 16-byte boundary
mov eax, [esp + 16]
jz .0
shr ecx, 1
@@ -43,15 +43,18 @@ ASM_PFX(InternalMemSetMem16): rep stosw
.0:
mov ecx, edx
- and edx, 7
- shr ecx, 3
+ and edx, 31
+ shr ecx, 5
jz @SetWords
movd xmm0, eax
pshuflw xmm0, xmm0, 0
movlhps xmm0, xmm0
.1:
movntdq [edi], xmm0 ; edi should be 16-byte aligned
- add edi, 16
+ movntdq [edi + 16], xmm0
+ movntdq [edi + 32], xmm0
+ movntdq [edi + 48], xmm0
+ add edi, 64
loop .1
mfence
@SetWords:
|