enderice2/Fennix
1
0
Fork 0
mirror of https://github.com/EnderIce2/Fennix.git synced 2025-07-05 20:39:16 +00:00
Code Issues Actions Releases Wiki Activity

TODO: implement a faster mem*

Browse Source
This commit is contained in:
Alex
2023-01-04 06:45:22 +02:00
parent fd477325a3
commit f91503d704
3 changed files with 28 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

118
Library/MemorySetSIMD.cpp
View File

@ -5,110 +5,40 @@
#include <debug.h>
#include <cpu.hpp>
// TODO: Implement these functions properly
/*
TODO: Replace these functions with even more optimized versions.
The current versions are fast but not as fast as they could be and also we need implementation for avx, not only sse.
*/
// TODO: Implement these functions
EXTERNC void *memset_sse(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
char *d = (char *)dest;
if (((uintptr_t)d & 0xF) == 0)
{
size_t num_vectors = n / 16;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movaps (%0), %%xmm0\n"
"movaps %%xmm0, (%1)\n"
:
: "r"(c), "r"(d)
: "xmm0");
d += 16;
}
n -= num_vectors * 16;
}
memset_unsafe(d, c, n);
return dest;
}
EXTERNC void *memset_sse2(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
char *d = (char *)dest;
if (((uintptr_t)d & 0xF) == 0)
{
size_t num_vectors = n / 16;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movdqa (%0), %%xmm0\n"
"movdqa %%xmm0, (%1)\n"
:
: "r"(c), "r"(d)
: "xmm0");
d += 16;
}
n -= num_vectors * 16;
}
memset_unsafe(d, c, n);
return dest;
}
EXTERNC void *memset_sse3(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
char *d = (char *)dest;
if (((uintptr_t)d & 0x7) == 0)
{
size_t num_vectors = n / 8;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movq (%0), %%xmm0\n"
"movddup %%xmm0, %%xmm1\n"
"movq %%xmm1, (%1)\n"
:
: "r"(c), "r"(d)
: "xmm0", "xmm1");
d += 16;
}
n -= num_vectors * 16;
}
memset_unsafe(d, c, n);
return dest;
}
EXTERNC void *memset_ssse3(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
char *d = (char *)dest;
if (((uintptr_t)d & 0xF) == 0)
{
size_t num_vectors = n / 16;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movdqa (%0), %%xmm0\n"
"movdqa 16(%0), %%xmm1\n"
"palignr $8, %%xmm0, %%xmm1\n"
"movdqa %%xmm1, (%1)\n"
:
: "r"(c), "r"(d)
: "xmm0", "xmm1");
d += 16;
}
n -= num_vectors * 16;
}
memset_unsafe(d, c, n);
return dest;
}
EXTERNC void *memset_sse4_1(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
}
EXTERNC void *memset_sse4_2(void *dest, int c, size_t n)
{
char *d = (char *)dest;
if (((uintptr_t)d & 0xF) == 0)
@ -116,7 +46,8 @@ EXTERNC void *memset_sse4_1(void *dest, int c, size_t n)
size_t num_vectors = n / 16;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movdqa (%0), %%xmm0\n"
asmv("movd %0, %%xmm0\n"
"pshufd $0, %%xmm0, %%xmm0\n"
"movdqa %%xmm0, (%1)\n"
:
: "r"(c), "r"(d)
@ -129,28 +60,3 @@ EXTERNC void *memset_sse4_1(void *dest, int c, size_t n)
memset_unsafe(d, c, n);
return dest;
}
EXTERNC void *memset_sse4_2(void *dest, int c, size_t n)
{
return memset_unsafe(dest, c, n);
char *d = (char *)dest;
if (((uintptr_t)d & 0xF) == 0)
{
size_t num_vectors = n / 16;
for (size_t i = 0; i < num_vectors; i++)
{
asmv("movdqa (%0), %%xmm0\n"
"pcmpistri $0, (%0), %%xmm0\n"
"movdqa %%xmm0, (%1)\n"
:
: "r"(d), "r"(c)
: "xmm0");
d += 16;
}
n -= num_vectors * 16;
}
memset_unsafe(d, c, n);
return dest;
}