rt/snapraid
1
0
Fork 0
mirror of https://github.com/amadvance/snapraid.git synced 2026-10-26 10:51:54 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Doesn't use extended SSE registers in the Bulldozer AMD family.

Browse source
This commit is contained in:
Andrea Mazzoleni 2013-09-29 15:24:15 +02:00
parent 02e11256de
commit 3d689c53de
3 changed files with 61 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

43
cpu.h
View file

@ -119,7 +119,7 @@ static inline int cpu_has_slowmult(void)
* Memory is little-endian 64-bit
* Speed test with 8 disk and 262144 buffer size...
* memset0 1849 [MB/s]
* HASH Murmur3 378 [MB/s]
* HASH Murmur3 378 [MB/s] (extremely slower than Spooky2)
* HASH Spooky2 3413 [MB/s]
* RAID5 int32x2 707 [MB/s]
* RAID5 mmxx2 1264 [MB/s]
@ -138,6 +138,47 @@ static inline int cpu_has_slowmult(void)
return 0;
}
/**
* Check if the processor has a slow extended set of SSE registers.
* If yes, it's better to unroll without using the second part of registers.
*/
static inline int cpu_has_slowextendedreg(void)
{
char vendor[CPU_VENDOR_MAX];
unsigned family;
unsigned model;
cpu_info(vendor, &family, &model);
if (strcmp(vendor, "AuthenticAMD") == 0) {
/* AMD Bulldozer (from user)
* CPU AuthenticAMD, family 21, model 19, flags mmx sse2
* Memory is little-endian 64-bit
* Speed test with 8 disk and 262144 buffer size, for a total of 2560 KiB...
* memset0 5721 [MB/s]
* CRC table 1080 [MB/s]
* CRC intel-crc32 2845 [MB/s]
* HASH Murmur3 2970 [MB/s]
* HASH Spooky2 7503 [MB/s]
* RAID5 int32x2 4595 [MB/s]
* RAID5 mmxx2 5856 [MB/s]
* RAID5 mmxx4 6157 [MB/s]
* RAID5 sse2x2 7151 [MB/s]
* RAID5 sse2x4 8447 [MB/s]
* RAID5 sse2x8 8155 [MB/s] (slower than sse2x4)
* RAID6 int32x2 1892 [MB/s]
* RAID6 mmxx2 3744 [MB/s]
* RAID6 sse2x2 4922 [MB/s]
* RAID6 sse2x4 4464 [MB/s] (slower than sse2x2)
*/
if (family == 21)
return 1;
}
return 0;
}
/* other cases */
/* Intel(R) Core(TM) i7-3740QM CPU @ 2.70GHz

9
raid.c
View file

@ -2505,8 +2505,13 @@ void raid_init(void)
}
if (cpu_has_sse2()) {
#if defined(__x86_64__)
raid5_gen = raid5_sse2r8;
raid6_gen = raid6_sse2r4;
if (cpu_has_slowextendedreg()) {
raid5_gen = raid5_sse2r4;
raid6_gen = raid6_sse2r2;
} else {
raid5_gen = raid5_sse2r8;
raid6_gen = raid6_sse2r4;
}
raidTP_gen = raidTP_sse2r2;
raidQP_gen = raidQP_sse2r2;
#else

18
speed.c
View file

@ -132,13 +132,16 @@ void speed(void)
printf("Expected fastest RAID5 is ");
#if defined(__i386__) || defined(__x86_64__)
if (cpu_has_sse2())
if (cpu_has_sse2()) {
#if defined(__x86_64__)
printf("sse2x8");
if (cpu_has_slowextendedreg())
printf("sse2x4");
else
printf("sse2x8");
#else
printf("sse2x4");
#endif
else if (cpu_has_mmx())
} else if (cpu_has_mmx())
printf("mmxx4");
else
printf("int32x2");
@ -152,13 +155,16 @@ void speed(void)
printf("Expected fastest RAID6 is ");
#if defined(__i386__) || defined(__x86_64__)
if (cpu_has_sse2())
if (cpu_has_sse2()) {
#if defined(__x86_64__)
printf("sse2x4");
if (cpu_has_slowextendedreg())
printf("sse2x2");
else
printf("sse2x4");
#else
printf("sse2x2");
#endif
else if (cpu_has_mmx())
} else if (cpu_has_mmx())
printf("mmxx2");
else
printf("int32x2");