mirror of
https://github.com/amadvance/snapraid.git
synced 2026-10-25 18:40:19 -04:00
Add portable 128-bit vector parity routines for gen1/gen2/genz to optimize non-x86 CPUs
This commit is contained in:
Andrea Mazzoleni
parent
d03a583523
commit
b78542685f
10 changed files with 336 additions and 6 deletions
|
|
@ -65,16 +65,30 @@
|
|||
#define __noreturn __attribute__((noreturn))
|
||||
#endif
|
||||
|
||||
/**
|
||||
* Enable C-based SIMD-accelerated vector optimizations on:
|
||||
* - x86/x64 (SSE2)
|
||||
* - ARM (NEON)
|
||||
* - RISC-V (Vector Extension)
|
||||
*/
|
||||
#if defined(__SSE2__) || defined(__ARM_NEON) || defined(__riscv_vector)
|
||||
#if defined(__has_attribute)
|
||||
#if __has_attribute(__vector_size__)
|
||||
#define CONFIG_VEC128 1
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/**
|
||||
* Architecture for inline assembly.
|
||||
* Enable Assembly-based SIMD-accelerated vector optimizations on:
|
||||
* - x86_32 (32 bit)
|
||||
* - x86_64 (64 bit)
|
||||
*/
|
||||
#if HAVE_ASSEMBLY
|
||||
#if defined(__i386__)
|
||||
#define CONFIG_X86 1
|
||||
#define CONFIG_X86_32 1
|
||||
#endif
|
||||
|
||||
#if defined(__x86_64__)
|
||||
#define CONFIG_X86 1
|
||||
#define CONFIG_X86_64 1
|
||||
|
|
|
|||
|
|
@ -157,6 +157,11 @@ void speed(int period, int nd, int size)
|
|||
#else
|
||||
printf("128-bit integers are not supported\n");
|
||||
#endif
|
||||
#if CONFIG_VEC128
|
||||
printf("128-bit vectors are supported\n");
|
||||
#else
|
||||
printf("128-bit vectors are not supported\n");
|
||||
#endif
|
||||
|
||||
#if HAVE_FUTIMENS
|
||||
printf("Support nanosecond timestamps with futimens()\n");
|
||||
|
|
@ -274,6 +279,9 @@ void speed(int period, int nd, int size)
|
|||
printf("%8s", "int8");
|
||||
printf("%8s", "int32");
|
||||
printf("%8s", "int64");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "vec128");
|
||||
#endif
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
printf("%8s", "sse2");
|
||||
|
|
@ -325,6 +333,15 @@ void speed(int period, int nd, int size)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_gen1_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -388,6 +405,15 @@ void speed(int period, int nd, int size)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_gen2_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -473,6 +499,15 @@ void speed(int period, int nd, int size)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_genz_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -525,6 +560,9 @@ void speed(int period, int nd, int size)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -605,6 +643,9 @@ void speed(int period, int nd, int size)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -685,6 +726,9 @@ void speed(int period, int nd, int size)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -764,6 +808,9 @@ void speed(int period, int nd, int size)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -857,7 +904,11 @@ void speed(int period, int nd, int size)
|
|||
|
||||
SPEED_START {
|
||||
/* ensure to use same hardware in the delta step */
|
||||
#if CONFIG_VEC128
|
||||
raid_gen_force(1, raid_gen1_vec128);
|
||||
#else
|
||||
raid_gen_force(1, sizeof(void*) == 8 ? raid_gen1_int64 : raid_gen1_int32);
|
||||
#endif
|
||||
/* +1 to avoid GEN1 optimized case */
|
||||
raid_rec1_int8(1, id, ip + 1, nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
|
@ -932,7 +983,11 @@ void speed(int period, int nd, int size)
|
|||
|
||||
SPEED_START {
|
||||
/* ensure to use same hardware in the delta step */
|
||||
#if CONFIG_VEC128
|
||||
raid_gen_force(2, raid_gen2_vec128);
|
||||
#else
|
||||
raid_gen_force(2, sizeof(void*) == 8 ? raid_gen2_int64 : raid_gen2_int32);
|
||||
#endif
|
||||
/* +1 to avoid GEN2 optimized case */
|
||||
raid_rec2_int8(2, id, ip + 1, nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
|
|
|||
39
raid/gf.h
39
raid/gf.h
|
|
@ -70,6 +70,14 @@ static __always_inline uint8_t A(int p, int d)
|
|||
*/
|
||||
#define v_64(p) (*(uint64_t *)&(p))
|
||||
|
||||
#if CONFIG_VEC128
|
||||
/*
|
||||
* Dereference as uvec128_t
|
||||
*/
|
||||
typedef uint64_t uvec128_t __attribute__ ((__vector_size__ (16)));
|
||||
#define v_128(p) (*(uvec128_t *)&(p))
|
||||
#endif
|
||||
|
||||
/*
|
||||
* Polynomial-dependent XOR masks for GF(2^8) multiply and divide by 2.
|
||||
*
|
||||
|
|
@ -154,4 +162,35 @@ static __always_inline uint64_t d2_64(uint64_t v)
|
|||
return v;
|
||||
}
|
||||
|
||||
#if CONFIG_VEC128
|
||||
#define RAID_POLY_128 ((uvec128_t){ RAID_POLY_64, RAID_POLY_64 })
|
||||
#define RAID_INV2_128 ((uvec128_t){ RAID_INV2_64, RAID_INV2_64 })
|
||||
|
||||
/*
|
||||
* Multiply each byte of a uint128 by 2 in GF(2^8).
|
||||
*/
|
||||
static __always_inline uvec128_t x2_128(uvec128_t v)
|
||||
{
|
||||
uvec128_t mask = v & ((uvec128_t){ 0x8080808080808080ULL, 0x8080808080808080ULL });
|
||||
|
||||
mask = (mask << 1) - (mask >> 7);
|
||||
v = (v << 1) & ((uvec128_t){ 0xfefefefefefefefeULL, 0xfefefefefefefefeULL });
|
||||
v ^= mask & RAID_POLY_128;
|
||||
return v;
|
||||
}
|
||||
|
||||
/*
|
||||
* Divide each byte of a uint128 by 2 in GF(2^8).
|
||||
*/
|
||||
static __always_inline uvec128_t d2_128(uvec128_t v)
|
||||
{
|
||||
uvec128_t mask = v & ((uvec128_t){ 0x0101010101010101ULL, 0x0101010101010101ULL });
|
||||
|
||||
mask = (mask << 8) - mask;
|
||||
v = (v >> 1) & ((uvec128_t){ 0x7f7f7f7f7f7f7f7fULL, 0x7f7f7f7f7f7f7f7fULL });
|
||||
v ^= mask & RAID_INV2_128;
|
||||
return v;
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif
|
||||
|
|
|
|||
79
raid/int.c
79
raid/int.c
|
|
@ -60,6 +60,36 @@ void raid_gen1_int64(int nd, size_t size, void **vv)
|
|||
}
|
||||
}
|
||||
|
||||
#if CONFIG_VEC128
|
||||
/*
|
||||
* GEN1 (RAID5 with xor) 128bit C implementation
|
||||
*/
|
||||
void raid_gen1_vec128(int nd, size_t size, void **vv)
|
||||
{
|
||||
uint8_t **v = (uint8_t **)vv;
|
||||
uint8_t *p;
|
||||
int d, l;
|
||||
size_t i;
|
||||
|
||||
uvec128_t p0;
|
||||
uvec128_t p1;
|
||||
|
||||
l = nd - 1;
|
||||
p = v[nd];
|
||||
|
||||
for (i = 0; i < size; i += 32) {
|
||||
p0 = v_128(v[l][i]);
|
||||
p1 = v_128(v[l][i + 16]);
|
||||
for (d = l - 1; d >= 0; --d) {
|
||||
p0 ^= v_128(v[d][i]);
|
||||
p1 ^= v_128(v[d][i + 16]);
|
||||
}
|
||||
v_128(p[i]) = p0;
|
||||
v_128(p[i + 16]) = p1;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
/*
|
||||
* GEN2 (RAID6 with powers of 2) 32bit C implementation
|
||||
*/
|
||||
|
|
@ -142,6 +172,50 @@ void raid_gen2_int64(int nd, size_t size, void **vv)
|
|||
}
|
||||
}
|
||||
|
||||
#if CONFIG_VEC128
|
||||
/*
|
||||
* GEN2 (RAID6 with powers of 2) 128bit C implementation
|
||||
*/
|
||||
void raid_gen2_vec128(int nd, size_t size, void **vv)
|
||||
{
|
||||
uint8_t **v = (uint8_t **)vv;
|
||||
uint8_t *p;
|
||||
uint8_t *q;
|
||||
int d, l;
|
||||
size_t i;
|
||||
|
||||
uvec128_t d0, d1;
|
||||
uvec128_t p0, p1;
|
||||
uvec128_t q0, q1;
|
||||
|
||||
l = nd - 1;
|
||||
p = v[nd];
|
||||
q = v[nd + 1];
|
||||
|
||||
for (i = 0; i < size; i += 32) {
|
||||
p0 = q0 = v_128(v[l][i]);
|
||||
p1 = q1 = v_128(v[l][i + 16]);
|
||||
for (d = l - 1; d >= 0; --d) {
|
||||
d0 = v_128(v[d][i]);
|
||||
d1 = v_128(v[d][i + 16]);
|
||||
|
||||
p0 ^= d0;
|
||||
p1 ^= d1;
|
||||
|
||||
q0 = x2_128(q0);
|
||||
q1 = x2_128(q1);
|
||||
|
||||
q0 ^= d0;
|
||||
q1 ^= d1;
|
||||
}
|
||||
v_128(p[i]) = p0;
|
||||
v_128(p[i + 16]) = p1;
|
||||
v_128(q[i]) = q0;
|
||||
v_128(q[i + 16]) = q1;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
/*
|
||||
* GEN3 (triple parity with Cauchy matrix) 8bit C implementation
|
||||
*
|
||||
|
|
@ -554,6 +628,11 @@ void raid_register_int(void)
|
|||
raid_gen_register(RAID_ALGO_CAUCHY_PAR2, "int64", raid_gen2_int64);
|
||||
raid_gen_register(RAID_ALGO_VANDERMONDE_PAR3, "int64", raid_genz_int64);
|
||||
}
|
||||
#if CONFIG_VEC128
|
||||
raid_gen_register(RAID_ALGO_CAUCHY_PAR1, "vec128", raid_gen1_vec128);
|
||||
raid_gen_register(RAID_ALGO_CAUCHY_PAR2, "vec128", raid_gen2_vec128);
|
||||
raid_gen_register(RAID_ALGO_VANDERMONDE_PAR3, "vec128", raid_genz_vec128);
|
||||
#endif
|
||||
raid_gen_register(RAID_ALGO_CAUCHY_PAR3, "int8", raid_gen3_int8);
|
||||
raid_gen_register(RAID_ALGO_CAUCHY_PAR4, "int8", raid_gen4_int8);
|
||||
raid_gen_register(RAID_ALGO_CAUCHY_PAR5, "int8", raid_gen5_int8);
|
||||
|
|
|
|||
|
|
@ -12,9 +12,26 @@
|
|||
#define HAVE_ASSEMBLY 1
|
||||
#endif
|
||||
|
||||
/* If the platforms supports assembly */
|
||||
/**
|
||||
* Enable C-base SIMD-accelerated vector optimizations on:
|
||||
* - x86/x64 (SSE2)
|
||||
* - ARM (NEON)
|
||||
* - RISC-V (Vector Extension)
|
||||
*/
|
||||
#if defined(__SSE2__) || defined(__ARM_NEON) || defined(__riscv_vector)
|
||||
#if defined(__has_attribute)
|
||||
#if __has_attribute(__vector_size__)
|
||||
#define CONFIG_VEC128 1
|
||||
#endif
|
||||
#endif
|
||||
#endif
|
||||
|
||||
/**
|
||||
* Enable Assembly-base SIMD-accelerated vector optimizations on:
|
||||
* - x86_32 (32 bit)
|
||||
* - x86_64 (64 bit)
|
||||
*/
|
||||
#if HAVE_ASSEMBLY
|
||||
/* Autodetect from the compiler */
|
||||
#if defined(__i386__)
|
||||
#define CONFIG_X86 1
|
||||
#define CONFIG_X86_32 1
|
||||
|
|
@ -82,11 +99,13 @@ void raid_rec1of1(int *id, int nd, size_t size, void **v);
|
|||
void raid_rec2of2_int8(int *id, int *ip, int nd, size_t size, void **vv);
|
||||
void raid_gen1_int32(int nd, size_t size, void **vv);
|
||||
void raid_gen1_int64(int nd, size_t size, void **vv);
|
||||
void raid_gen1_vec128(int nd, size_t size, void **vv);
|
||||
void raid_gen1_sse2(int nd, size_t size, void **vv);
|
||||
void raid_gen1_avx2(int nd, size_t size, void **vv);
|
||||
void raid_gen1_avx512bw(int nd, size_t size, void **vv);
|
||||
void raid_gen2_int32(int nd, size_t size, void **vv);
|
||||
void raid_gen2_int64(int nd, size_t size, void **vv);
|
||||
void raid_gen2_vec128(int nd, size_t size, void **vv);
|
||||
void raid_gen2_sse2(int nd, size_t size, void **vv);
|
||||
void raid_gen2_avx2(int nd, size_t size, void **vv);
|
||||
void raid_gen2_sse2ext(int nd, size_t size, void **vv);
|
||||
|
|
@ -95,6 +114,7 @@ void raid_gen2_avx2gfni(int nd, size_t size, void **vv);
|
|||
void raid_gen2_avx512gfni(int nd, size_t size, void **vv);
|
||||
void raid_genz_int32(int nd, size_t size, void **vv);
|
||||
void raid_genz_int64(int nd, size_t size, void **vv);
|
||||
void raid_genz_vec128(int nd, size_t size, void **vv);
|
||||
void raid_genz_sse2(int nd, size_t size, void **vv);
|
||||
void raid_genz_sse2ext(int nd, size_t size, void **vv);
|
||||
void raid_genz_avx2ext(int nd, size_t size, void **vv);
|
||||
|
|
|
|||
58
raid/intz.c
58
raid/intz.c
|
|
@ -106,3 +106,61 @@ void raid_genz_int64(int nd, size_t size, void **vv)
|
|||
}
|
||||
}
|
||||
|
||||
#if CONFIG_VEC128
|
||||
/*
|
||||
* GENz (triple parity with powers of 2^-1) 128bit C implementation
|
||||
*/
|
||||
void raid_genz_vec128(int nd, size_t size, void **vv)
|
||||
{
|
||||
uint8_t **v = (uint8_t **)vv;
|
||||
uint8_t *p;
|
||||
uint8_t *q;
|
||||
uint8_t *r;
|
||||
int d, l;
|
||||
size_t i;
|
||||
|
||||
uvec128_t d0, d1;
|
||||
uvec128_t p0, p1;
|
||||
uvec128_t q0, q1;
|
||||
uvec128_t r0, r1;
|
||||
|
||||
l = nd - 1;
|
||||
p = v[nd];
|
||||
q = v[nd + 1];
|
||||
r = v[nd + 2];
|
||||
|
||||
for (i = 0; i < size; i += 32) {
|
||||
r0 = q0 = p0 = v_128(v[l][i]);
|
||||
r1 = q1 = p1 = v_128(v[l][i + 16]);
|
||||
for (d = l - 1; d >= 0; --d) {
|
||||
d0 = v_128(v[d][i]);
|
||||
d1 = v_128(v[d][i + 16]);
|
||||
|
||||
p0 ^= d0;
|
||||
p1 ^= d1;
|
||||
|
||||
q0 = x2_128(q0);
|
||||
q1 = x2_128(q1);
|
||||
|
||||
q0 ^= d0;
|
||||
q1 ^= d1;
|
||||
|
||||
r0 = d2_128(r0);
|
||||
r1 = d2_128(r1);
|
||||
|
||||
r0 ^= d0;
|
||||
r1 ^= d1;
|
||||
}
|
||||
v_128(p[i]) = p0;
|
||||
v_128(p[i + 16]) = p1;
|
||||
|
||||
v_128(q[i]) = q0;
|
||||
v_128(q[i + 16]) = q1;
|
||||
|
||||
v_128(r[i]) = r0;
|
||||
v_128(r[i + 16]) = r1;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -367,8 +367,14 @@ int raid_test_par(int mode, int nd, size_t size)
|
|||
/* setup all the available functions */
|
||||
test_setup(1) = raid_gen1_int32;
|
||||
test_setup(1) = raid_gen1_int64;
|
||||
#if CONFIG_VEC128
|
||||
test_setup(1) = raid_gen1_vec128;
|
||||
#endif
|
||||
test_setup(2) = raid_gen2_int32;
|
||||
test_setup(2) = raid_gen2_int64;
|
||||
#if CONFIG_VEC128
|
||||
test_setup(2) = raid_gen2_vec128;
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -449,6 +455,9 @@ int raid_test_par(int mode, int nd, size_t size)
|
|||
} else {
|
||||
test_setup(3) = raid_genz_int32;
|
||||
test_setup(3) = raid_genz_int64;
|
||||
#if CONFIG_VEC128
|
||||
test_setup(3) = raid_genz_vec128;
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
|
|||
|
|
@ -30,9 +30,9 @@ SDE = sde64
|
|||
endif
|
||||
CC = gcc
|
||||
LD = ld
|
||||
CFLAGS = -I.. -Wall -Wextra -g
|
||||
CFLAGS = -I.. -Wall -Wextra -g -DUSE_RAID_AES
|
||||
ifeq ($(COVERAGE),)
|
||||
CFLAGS += -O2 -DUSE_RAID_AES
|
||||
CFLAGS += -O2
|
||||
else
|
||||
CFLAGS += -O0 --coverage -DCOVERAGE=1 -DNDEBUG=1
|
||||
endif
|
||||
|
|
|
|||
|
|
@ -30,6 +30,9 @@ int main(void)
|
|||
|
||||
raid_init();
|
||||
|
||||
#ifdef CONFIG_VEC128
|
||||
printf("Including 128-bit vectors\n");
|
||||
#endif
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2())
|
||||
printf("Including x86 SSE2\n");
|
||||
|
|
|
|||
|
|
@ -125,6 +125,9 @@ void speed(void)
|
|||
printf("%8s", "int8");
|
||||
printf("%8s", "int32");
|
||||
printf("%8s", "int64");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "vec128");
|
||||
#endif
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
printf("%8s", "sse2");
|
||||
|
|
@ -179,6 +182,15 @@ void speed(void)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_gen1_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -242,6 +254,15 @@ void speed(void)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_gen2_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -327,6 +348,15 @@ void speed(void)
|
|||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
|
||||
#if CONFIG_VEC128
|
||||
SPEED_START {
|
||||
raid_genz_vec128(nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
||||
printf("%8" PRIu64, ds / dt);
|
||||
fflush(stdout);
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
SPEED_START {
|
||||
|
|
@ -380,6 +410,9 @@ void speed(void)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -458,6 +491,9 @@ void speed(void)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -536,6 +572,9 @@ void speed(void)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -614,6 +653,9 @@ void speed(void)
|
|||
|
||||
printf("%8s", "");
|
||||
printf("%8s", "");
|
||||
#if CONFIG_VEC128
|
||||
printf("%8s", "");
|
||||
#endif
|
||||
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2()) {
|
||||
|
|
@ -705,7 +747,11 @@ void speed(void)
|
|||
fflush(stdout);
|
||||
|
||||
SPEED_START {
|
||||
#if CONFIG_VEC128
|
||||
raid_gen_force(1, raid_gen1_vec128);
|
||||
#else
|
||||
raid_gen_force(1, sizeof(void *) == 8 ? raid_gen1_int64 : raid_gen1_int32);
|
||||
#endif
|
||||
/* +1 to avoid GEN1 optimized case */
|
||||
raid_rec1_int8(1, id, ip + 1, nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
|
@ -774,7 +820,11 @@ void speed(void)
|
|||
fflush(stdout);
|
||||
|
||||
SPEED_START {
|
||||
#if CONFIG_VEC128
|
||||
raid_gen_force(2, raid_gen2_vec128);
|
||||
#else
|
||||
raid_gen_force(2, sizeof(void *) == 8 ? raid_gen2_int64 : raid_gen2_int32);
|
||||
#endif
|
||||
/* +1 to avoid GEN2 optimized case */
|
||||
raid_rec2_int8(2, id, ip + 1, nd, size, v);
|
||||
} SPEED_STOP
|
||||
|
|
@ -1133,6 +1183,9 @@ int main(void)
|
|||
|
||||
raid_init();
|
||||
|
||||
#ifdef CONFIG_VEC128
|
||||
printf("Including 128-bit vectors\n");
|
||||
#endif
|
||||
#ifdef CONFIG_X86
|
||||
if (raid_cpu_has_sse2())
|
||||
printf("Including x86 SSE2\n");
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue