sravan/system76-edk2
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

BaseMemoryLibSse2: Take advantage of write combining buffers

Browse Source 
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>
This commit is contained in:
Compostella, Jeremy
2020-10-10 04:42:34 +08:00
committed by mergify[bot]
parent 19c87b7d44
commit d25fd8710d
10 changed files with 85 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
MdePkg/Library/BaseMemoryLibSse2/X64/SetMem.nasm
View File

@@ -42,8 +42,8 @@ ASM_PFX(InternalMemSetMem):
rep stosb
.0:
mov rcx, rdx
and rdx, 15
shr rcx, 4
and rdx, 63
shr rcx, 6
jz @SetBytes
mov ah, al ; ax <- Value repeats twice
movdqa [rsp + 0x10], xmm0 ; save xmm0
@@ -52,7 +52,10 @@ ASM_PFX(InternalMemSetMem):
movlhps xmm0, xmm0 ; xmm0 <- Value repeats 16 times
.1:
movntdq [rdi], xmm0 ; rdi should be 16-byte aligned
add rdi, 16
movntdq [rdi + 16], xmm0
movntdq [rdi + 32], xmm0
movntdq [rdi + 48], xmm0
add rdi, 64
loop .1
mfence
movdqa xmm0, [rsp + 0x10] ; restore xmm0

11
MdePkg/Library/BaseMemoryLibSse2/X64/SetMem16.nasm
View File

@@ -33,7 +33,7 @@ ASM_PFX(InternalMemSetMem16):
mov r9, rdi
xor rcx, rcx
sub rcx, rdi
and rcx, 15
and rcx, 63
mov rax, r8
jz .0
shr rcx, 1
@@ -43,15 +43,18 @@ ASM_PFX(InternalMemSetMem16):
rep stosw
.0:
mov rcx, rdx
and edx, 7
shr rcx, 3
and edx, 31
shr rcx, 5
jz @SetWords
movd xmm0, eax
pshuflw xmm0, xmm0, 0
movlhps xmm0, xmm0
.1:
movntdq [rdi], xmm0
add rdi, 16
movntdq [rdi + 16], xmm0
movntdq [rdi + 32], xmm0
movntdq [rdi + 48], xmm0
add rdi, 64
loop .1
mfence
@SetWords:

9
MdePkg/Library/BaseMemoryLibSse2/X64/SetMem32.nasm
View File

@@ -43,14 +43,17 @@ ASM_PFX(InternalMemSetMem32):
rep stosd
.0:
mov rcx, rdx
and edx, 3
shr rcx, 2
and edx, 15
shr rcx, 4
jz @SetDwords
movd xmm0, eax
pshufd xmm0, xmm0, 0
.1:
movntdq [rdi], xmm0
add rdi, 16
movntdq [rdi + 16], xmm0
movntdq [rdi + 32], xmm0
movntdq [rdi + 48], xmm0
add rdi, 64
loop .1
mfence
@SetDwords:

19
MdePkg/Library/BaseMemoryLibSse2/X64/SetMem64.nasm
View File

@@ -37,17 +37,28 @@ ASM_PFX(InternalMemSetMem64):
add rdx, 8
dec rcx
.0:
shr rcx, 1
push rbx
mov rbx, rcx
and rbx, 7
shr rcx, 3
jz @SetQwords
movlhps xmm0, xmm0
.1:
movntdq [rdx], xmm0
lea rdx, [rdx + 16]
movntdq [rdx + 16], xmm0
movntdq [rdx + 32], xmm0
movntdq [rdx + 48], xmm0
lea rdx, [rdx + 64]
loop .1
mfence
@SetQwords:
jnc .2
mov [rdx], r8
push rdi
mov rcx, rbx
mov rax, r8
mov rdi, rdx
rep stosq
pop rdi
.2:
pop rbx
ret

13
MdePkg/Library/BaseMemoryLibSse2/X64/ZeroMem.nasm
View File

@@ -32,7 +32,7 @@ ASM_PFX(InternalMemZeroMem):
xor rcx, rcx
xor eax, eax
sub rcx, rdi
and rcx, 15
and rcx, 63
mov r8, rdi
jz .0
cmp rcx, rdx
@@ -41,13 +41,16 @@ ASM_PFX(InternalMemZeroMem):
rep stosb
.0:
mov rcx, rdx
and edx, 15
shr rcx, 4
and edx, 63
shr rcx, 6
jz @ZeroBytes
pxor xmm0, xmm0
.1:
movntdq [rdi], xmm0 ; rdi should be 16-byte aligned
add rdi, 16
movntdq [rdi], xmm0
movntdq [rdi + 16], xmm0
movntdq [rdi + 32], xmm0
movntdq [rdi + 48], xmm0
add rdi, 64
loop .1
mfence
@ZeroBytes: