mirror of
https://github.com/amadvance/snapraid.git
synced 2026-10-26 02:49:02 -04:00
Adds SSSE3 and AVX parity computation.
This commit is contained in:
Andrea Mazzoleni
parent
1b46d03d70
commit
823bc8a8c0
8 changed files with 871 additions and 325 deletions
4
HISTORY
4
HISTORY
|
|
@ -10,6 +10,8 @@ SnapRAID HISTORY
|
|||
up to 89 disks with Hexa (6) parity is possible, but not yet implemented.
|
||||
Note that more general approaches without these limitations are also
|
||||
possible, but they would affect badly the performance.
|
||||
Parity computation uses SSSE3 and AVX instructions if available for the
|
||||
best performance.
|
||||
* In sync, if a silent error is found, directly mark the block as bad
|
||||
just like in scrub, and continue with the sync process.
|
||||
* Sort files by inode when listing the directory. This improves
|
||||
|
|
@ -20,6 +22,8 @@ SnapRAID HISTORY
|
|||
of more info.
|
||||
* For files with changes only in some blocks, updates the parity
|
||||
only for blocks that really are changed.
|
||||
This improves the performance in sync for edited files.
|
||||
* Faster parity computation for ARM 64 bit CPU.
|
||||
* Reduced memory usage avoiding to allocate unnecessary memory.
|
||||
|
||||
4.2 2013/10 (BETA)
|
||||
|
|
|
|||
18
cpu.h
18
cpu.h
|
|
@ -91,6 +91,15 @@ static inline int cpu_has_sse2(void)
|
|||
return (reg[3] >> 26) & 1;
|
||||
}
|
||||
|
||||
static inline int cpu_has_ssse3(void)
|
||||
{
|
||||
uint32_t reg[4];
|
||||
|
||||
cpuid(1, reg);
|
||||
|
||||
return (reg[2] >> 9) & 1;
|
||||
}
|
||||
|
||||
static inline int cpu_has_sse42(void)
|
||||
{
|
||||
uint32_t reg[4];
|
||||
|
|
@ -100,6 +109,15 @@ static inline int cpu_has_sse42(void)
|
|||
return (reg[2] >> 20) & 1;
|
||||
}
|
||||
|
||||
static inline int cpu_has_avx(void)
|
||||
{
|
||||
uint32_t reg[4];
|
||||
|
||||
cpuid(1, reg);
|
||||
|
||||
return (reg[2] >> 28) & 1;
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if the processor has a slow MULT implementation.
|
||||
* If yes, it's better to use a hash not based on multiplication.
|
||||
|
|
|
|||
47
mktables.c
47
mktables.c
|
|
@ -133,6 +133,53 @@ int main(void)
|
|||
}
|
||||
printf("};\n\n");
|
||||
|
||||
/* x4l */
|
||||
printf("#define GFX4L { ");
|
||||
for(i=0;i<16;++i) {
|
||||
v = gfmul(0x4, i);
|
||||
printf("0x%02x", (unsigned)v);
|
||||
if (i != 15)
|
||||
printf(", ");
|
||||
}
|
||||
printf(" }\n");
|
||||
|
||||
/* x4h */
|
||||
printf("#define GFX4H { ");
|
||||
for(i=0;i<16;++i) {
|
||||
v = gfmul(0x40, i);
|
||||
printf("0x%02x", (unsigned)v);
|
||||
if (i != 15)
|
||||
printf(", ");
|
||||
}
|
||||
printf(" }\n");
|
||||
|
||||
/* x8l */
|
||||
printf("#define GFX8L { ");
|
||||
for(i=0;i<16;++i) {
|
||||
v = gfmul(0x8, i);
|
||||
printf("0x%02x", (unsigned)v);
|
||||
if (i != 15)
|
||||
printf(", ");
|
||||
}
|
||||
printf(" }\n");
|
||||
|
||||
/* x8h */
|
||||
printf("#define GFX8H { ");
|
||||
for(i=0;i<16;++i) {
|
||||
v = gfmul(0x80, i);
|
||||
printf("0x%02x", (unsigned)v);
|
||||
if (i != 15)
|
||||
printf(", ");
|
||||
}
|
||||
printf(" }\n");
|
||||
|
||||
/* poly */
|
||||
printf("#define GFPOLY8 { 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d } \n");
|
||||
printf("#define GFPOLY16 { 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d } \n");
|
||||
printf("#define GFMASK16 { 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f } \n");
|
||||
|
||||
printf("\n");
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
|
|
|||
53
raid.h
53
raid.h
|
|
@ -78,30 +78,43 @@ void raid_recov_3data(int x, int y, int z, int i, int j, int k, unsigned char**
|
|||
*/
|
||||
void raid_recov_4data(int x, int y, int z, int v, int i, int j, int k, int l, unsigned char** vbuf, unsigned data, unsigned char* zero, unsigned size);
|
||||
|
||||
/**
|
||||
* Gets the name of the selected function to compute parity.
|
||||
*/
|
||||
const char* raid5_tag(void);
|
||||
const char* raid6_tag(void);
|
||||
const char* raidTP_tag(void);
|
||||
const char* raidQP_tag(void);
|
||||
|
||||
/**
|
||||
* Internal specialized parity computation.
|
||||
* Only for testing.
|
||||
*/
|
||||
void raid5_int32r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_int64r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_mmxr4(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_sse2r4(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_sse2r8(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_int32r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_int64r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_mmxr2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_sse2r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_sse2r4(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_int32r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_int64r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_mmxr1(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_sse2r1(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_sse2r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_int32r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_int64r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_mmxr1(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_sse2r1(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_sse2r2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_int32(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_int64(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_mmx(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_sse2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid5_sse2ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_int32(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_int64(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_mmx(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_sse2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raid6_sse2ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_int32(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_int64(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_mmx(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_sse2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_sse2ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_ssse3(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidTP_ssse3ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_int32(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_int64(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_mmx(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_sse2(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_sse2ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_ssse3(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_ssse3ext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
void raidQP_avxext(unsigned char** vbuf, unsigned data, unsigned size);
|
||||
|
||||
#endif
|
||||
|
||||
|
|
|
|||
55
selftest.c
55
selftest.c
|
|
@ -335,33 +335,48 @@ static void gentest(unsigned diskmax, unsigned block_size)
|
|||
|
||||
/* load all the available functions */
|
||||
mac = 0;
|
||||
map[mac++] = raid5_int32r2;
|
||||
map[mac++] = raid5_int64r2;
|
||||
map[mac++] = raid6_int32r2;
|
||||
map[mac++] = raid6_int64r2;
|
||||
map[mac++] = raidTP_int32r2;
|
||||
map[mac++] = raidTP_int64r2;
|
||||
map[mac++] = raidQP_int32r2;
|
||||
map[mac++] = raidQP_int64r2;
|
||||
map[mac++] = raid5_int32;
|
||||
map[mac++] = raid5_int64;
|
||||
map[mac++] = raid6_int32;
|
||||
map[mac++] = raid6_int64;
|
||||
map[mac++] = raidTP_int32;
|
||||
map[mac++] = raidTP_int64;
|
||||
map[mac++] = raidQP_int32;
|
||||
map[mac++] = raidQP_int64;
|
||||
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_mmx()) {
|
||||
map[mac++] = raid5_mmxr4;
|
||||
map[mac++] = raid6_mmxr2;
|
||||
map[mac++] = raidQP_mmxr1;
|
||||
map[mac++] = raidTP_mmxr1;
|
||||
map[mac++] = raid5_mmx;
|
||||
map[mac++] = raid6_mmx;
|
||||
map[mac++] = raidQP_mmx;
|
||||
map[mac++] = raidTP_mmx;
|
||||
}
|
||||
|
||||
if (cpu_has_sse2()) {
|
||||
map[mac++] = raid5_sse2r4;
|
||||
map[mac++] = raid6_sse2r2;
|
||||
map[mac++] = raidTP_sse2r1;
|
||||
map[mac++] = raidQP_sse2r1;
|
||||
map[mac++] = raid5_sse2;
|
||||
map[mac++] = raid6_sse2;
|
||||
map[mac++] = raidTP_sse2;
|
||||
map[mac++] = raidQP_sse2;
|
||||
#if defined(__x86_64__)
|
||||
map[mac++] = raid5_sse2r8;
|
||||
map[mac++] = raid6_sse2r4;
|
||||
map[mac++] = raidTP_sse2r2;
|
||||
map[mac++] = raidQP_sse2r2;
|
||||
map[mac++] = raid5_sse2ext;
|
||||
map[mac++] = raid6_sse2ext;
|
||||
map[mac++] = raidTP_sse2ext;
|
||||
map[mac++] = raidQP_sse2ext;
|
||||
#endif
|
||||
}
|
||||
|
||||
if (cpu_has_ssse3()) {
|
||||
map[mac++] = raidTP_ssse3;
|
||||
map[mac++] = raidQP_ssse3;
|
||||
#if defined(__x86_64__)
|
||||
map[mac++] = raidTP_ssse3ext;
|
||||
map[mac++] = raidQP_ssse3ext;
|
||||
#endif
|
||||
}
|
||||
|
||||
if (cpu_has_avx()) {
|
||||
#if defined(__x86_64__)
|
||||
map[mac++] = raidQP_avxext;
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
|
|
|||
220
speed.c
220
speed.c
|
|
@ -89,9 +89,12 @@ void speed(void)
|
|||
|
||||
cpu_info(vendor, &family, &model);
|
||||
|
||||
printf("CPU %s, family %u, model %u, flags%s%s%s\n", vendor, family, model,
|
||||
printf("CPU %s, family %u, model %u, flags%s%s%s%s%s%s\n", vendor, family, model,
|
||||
cpu_has_mmx() ? " mmx" : "",
|
||||
cpu_has_sse2() ? " sse2" : "",
|
||||
cpu_has_ssse3() ? " ssse3" : "",
|
||||
cpu_has_sse42() ? " sse42" : "",
|
||||
cpu_has_avx() ? " avx" : "",
|
||||
cpu_has_slowmult() ? " slowmult" : ""
|
||||
);
|
||||
}
|
||||
|
|
@ -130,93 +133,10 @@ void speed(void)
|
|||
#endif
|
||||
printf("\n");
|
||||
|
||||
printf("Expected fastest RAID5 is ");
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_sse2()) {
|
||||
#if defined(__x86_64__)
|
||||
if (cpu_has_slowextendedreg())
|
||||
printf("sse2x4");
|
||||
else
|
||||
printf("sse2x8");
|
||||
#else
|
||||
printf("sse2x4");
|
||||
#endif
|
||||
} else if (cpu_has_mmx())
|
||||
printf("mmxx4");
|
||||
else
|
||||
printf("int32x2");
|
||||
#else
|
||||
if (sizeof(void*) == 4)
|
||||
printf("int32x2");
|
||||
else
|
||||
printf("int64x2");
|
||||
#endif
|
||||
printf("\n");
|
||||
|
||||
printf("Expected fastest RAID6 is ");
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_sse2()) {
|
||||
#if defined(__x86_64__)
|
||||
if (cpu_has_slowextendedreg())
|
||||
printf("sse2x2");
|
||||
else
|
||||
printf("sse2x4");
|
||||
#else
|
||||
printf("sse2x2");
|
||||
#endif
|
||||
} else if (cpu_has_mmx())
|
||||
printf("mmxx2");
|
||||
else
|
||||
printf("int32x2");
|
||||
#else
|
||||
if (sizeof(void*) == 4)
|
||||
printf("int32x2");
|
||||
else
|
||||
printf("int64x2");
|
||||
#endif
|
||||
printf("\n");
|
||||
|
||||
printf("Expected fastest RAIDTP is ");
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_sse2())
|
||||
#if defined(__x86_64__)
|
||||
printf("sse2x2");
|
||||
#else
|
||||
printf("sse2x1");
|
||||
#endif
|
||||
else if (cpu_has_mmx())
|
||||
printf("mmxx1");
|
||||
else
|
||||
printf("int32x2");
|
||||
#else
|
||||
if (sizeof(void*) == 4)
|
||||
printf("int32x2");
|
||||
else
|
||||
printf("int64x2");
|
||||
#endif
|
||||
printf("\n");
|
||||
|
||||
|
||||
printf("Expected fastest RAIDQP is ");
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_sse2())
|
||||
#if defined(__x86_64__)
|
||||
printf("sse2x2");
|
||||
#else
|
||||
printf("sse2x1");
|
||||
#endif
|
||||
else if (cpu_has_mmx())
|
||||
printf("mmxx1");
|
||||
else
|
||||
printf("int32x2");
|
||||
#else
|
||||
if (sizeof(void*) == 4)
|
||||
printf("int32x2");
|
||||
else
|
||||
printf("int64x2");
|
||||
#endif
|
||||
printf("\n");
|
||||
|
||||
printf("Expected fastest RAID5 is %s\n", raid5_tag());
|
||||
printf("Expected fastest RAID6 is %s\n", raid6_tag());
|
||||
printf("Expected fastest RAIDTP is %s\n", raidTP_tag());
|
||||
printf("Expected fastest RAIDQP is %s\n", raidQP_tag());
|
||||
|
||||
printf("If these expectations are false, please report it in the SnapRAID forum\n");
|
||||
|
||||
|
|
@ -268,135 +188,151 @@ void speed(void)
|
|||
printf("HASH Spooky2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
SPEED_START {
|
||||
raid5_int32r2(buffer, diskmax, block_size);
|
||||
raid5_int32(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID5 int32x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID5 int32 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
SPEED_START {
|
||||
raid5_int64r2(buffer, diskmax, block_size);
|
||||
raid5_int64(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID5 int64x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID5 int64 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_mmx()) {
|
||||
SPEED_START {
|
||||
raid5_mmxr4(buffer, diskmax, block_size);
|
||||
raid5_mmx(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID5 mmxx4 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID5 mmx %"PRIu64" [MB/s]\n", ds / dt);
|
||||
}
|
||||
|
||||
if (cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
raid5_sse2r4(buffer, diskmax, block_size);
|
||||
raid5_sse2(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID5 sse2x4 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID5 sse2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raid5_sse2r8(buffer, diskmax, block_size);
|
||||
raid5_sse2ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID5 sse2x8 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID5 sse2ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
SPEED_START {
|
||||
raid6_int32r2(buffer, diskmax, block_size);
|
||||
raid6_int32(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID6 int32x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID6 int32 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
SPEED_START {
|
||||
raid6_int64r2(buffer, diskmax, block_size);
|
||||
raid6_int64(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID6 int64x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID6 int64 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_mmx()) {
|
||||
SPEED_START {
|
||||
raid6_mmxr2(buffer, diskmax, block_size);
|
||||
raid6_mmx(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID6 mmxx2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID6 mmx %"PRIu64" [MB/s]\n", ds / dt);
|
||||
}
|
||||
|
||||
if (cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
raid6_sse2r2(buffer, diskmax, block_size);
|
||||
raid6_sse2(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID6 sse2x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID6 sse2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raid6_sse2r4(buffer, diskmax, block_size);
|
||||
raid6_sse2ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAID6 sse2x4 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAID6 sse2ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
SPEED_START {
|
||||
raidTP_int32r2(buffer, diskmax, block_size);
|
||||
raidTP_int32(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP int32x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP int32 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
SPEED_START {
|
||||
raidTP_int64r2(buffer, diskmax, block_size);
|
||||
raidTP_int64(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP int64x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP int64 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_mmx()) {
|
||||
SPEED_START {
|
||||
raidTP_mmxr1(buffer, diskmax, block_size);
|
||||
raidTP_mmx(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP mmxr1 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP mmx %"PRIu64" [MB/s]\n", ds / dt);
|
||||
}
|
||||
|
||||
if (cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
raidTP_sse2r1(buffer, diskmax, block_size);
|
||||
raidTP_sse2(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP sse2x1 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP sse2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raidTP_sse2r2(buffer, diskmax, block_size);
|
||||
raidTP_sse2ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP sse2x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP sse2ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
SPEED_START {
|
||||
raidQP_int32r2(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
if (cpu_has_ssse3()) {
|
||||
SPEED_START {
|
||||
raidTP_ssse3(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP int32x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDTP ssse3 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raidTP_ssse3ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDTP ssse3ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
|
||||
SPEED_START {
|
||||
raidQP_int64r2(buffer, diskmax, block_size);
|
||||
raidQP_int32(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP int64x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDQP int32 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
SPEED_START {
|
||||
raidQP_int64(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP int64 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__i386__) || defined(__x86_64__)
|
||||
if (cpu_has_mmx()) {
|
||||
SPEED_START {
|
||||
raidQP_mmxr1(buffer, diskmax, block_size);
|
||||
raidQP_mmx(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP mmxr1 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
|
@ -404,21 +340,47 @@ void speed(void)
|
|||
|
||||
if (cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
raidQP_sse2r1(buffer, diskmax, block_size);
|
||||
raidQP_sse2(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP sse2x1 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDQP sse2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raidQP_sse2r2(buffer, diskmax, block_size);
|
||||
raidQP_sse2ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP sse2x2 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
printf("RAIDQP sse2ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
|
||||
if (cpu_has_ssse3()) {
|
||||
SPEED_START {
|
||||
raidQP_ssse3(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP ssse3 %"PRIu64" [MB/s]\n", ds / dt);
|
||||
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raidQP_ssse3ext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP ssse3ext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
|
||||
if (cpu_has_avx()) {
|
||||
#if defined(__x86_64__)
|
||||
SPEED_START {
|
||||
raidQP_avxext(buffer, diskmax, block_size);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("RAIDQP avxext %"PRIu64" [MB/s]\n", ds / dt);
|
||||
#endif
|
||||
}
|
||||
|
||||
free(buffer_alloc);
|
||||
free(buffer);
|
||||
}
|
||||
|
|
|
|||
8
tables.h
8
tables.h
|
|
@ -8868,3 +8868,11 @@ const unsigned char __attribute__((aligned(256))) gfinv[256] =
|
|||
0x42, 0xd4, 0xe8, 0x75, 0x7f, 0xff, 0x7e, 0xfd,
|
||||
};
|
||||
|
||||
#define GFX4L { 0x00, 0x04, 0x08, 0x0c, 0x10, 0x14, 0x18, 0x1c, 0x20, 0x24, 0x28, 0x2c, 0x30, 0x34, 0x38, 0x3c }
|
||||
#define GFX4H { 0x00, 0x40, 0x80, 0xc0, 0x1d, 0x5d, 0x9d, 0xdd, 0x3a, 0x7a, 0xba, 0xfa, 0x27, 0x67, 0xa7, 0xe7 }
|
||||
#define GFX8L { 0x00, 0x08, 0x10, 0x18, 0x20, 0x28, 0x30, 0x38, 0x40, 0x48, 0x50, 0x58, 0x60, 0x68, 0x70, 0x78 }
|
||||
#define GFX8H { 0x00, 0x80, 0x1d, 0x9d, 0x3a, 0xba, 0x27, 0xa7, 0x74, 0xf4, 0x69, 0xe9, 0x4e, 0xce, 0x53, 0xd3 }
|
||||
#define GFPOLY8 { 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d }
|
||||
#define GFPOLY16 { 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d, 0x1d }
|
||||
#define GFMASK16 { 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f, 0x0f }
|
||||
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue