Support printing dual socket. Fix bug where cache sizes were not displayed correctly (they were truncated)

This commit is contained in:
Dr-Noob
2020-06-29 17:13:37 +02:00
parent 941bf35d03
commit d7b7e2b62d
4 changed files with 168 additions and 114 deletions

View File

@@ -445,19 +445,19 @@ struct cache* get_cache_info(struct cpuInfo* cpu) {
// Sanity checks. If we read values greater than this, they can't be valid ones // Sanity checks. If we read values greater than this, they can't be valid ones
// The values were chosen by me // The values were chosen by me
if(cach->L1i > 64 * 1024) { if(cach->L1i > 64 * 1024) {
printBug("Invalid L1i size: %dKB\n", cach->L1i/1024); printBug("Invalid L1i size: %dKB", cach->L1i/1024);
return NULL; return NULL;
} }
if(cach->L1d > 64 * 1024) { if(cach->L1d > 64 * 1024) {
printBug("Invalid L1d size: %dKB\n", cach->L1d/1024); printBug("Invalid L1d size: %dKB", cach->L1d/1024);
return NULL; return NULL;
} }
if(cach->L2 != UNKNOWN && cach->L2 > 2 * 1048576) { if(cach->L2 != UNKNOWN && cach->L2 > 2 * 1048576) {
printBug("Invalid L2 size: %dMB\n", cach->L2/(1048576)); printBug("Invalid L2 size: %dMB", cach->L2/(1048576));
return NULL; return NULL;
} }
if(cach->L3 != UNKNOWN && cach->L3 > 100 * 1048576) { if(cach->L3 != UNKNOWN && cach->L3 > 100 * 1048576) {
printBug("Invalid L3 size: %dMB\n", cach->L3/(1048576)); printBug("Invalid L3 size: %dMB", cach->L3/(1048576));
return NULL; return NULL;
} }
@@ -493,6 +493,10 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
return freq; return freq;
} }
uint32_t get_nsockets(struct topology* topo) {
return topo->sockets;
}
int64_t get_freq(struct frequency* freq) { int64_t get_freq(struct frequency* freq) {
return freq->max; return freq->max;
} }
@@ -590,22 +594,38 @@ char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64
return string; return string;
} }
char* get_str_topology(struct topology* topo) { char* get_str_topology(struct topology* topo, bool dual_socket) {
char* string; char* string;
if(topo->smt > 1) { if(topo->smt > 1) {
//3 for digits, 8 for ' cores (', 3 for digits, 9 for ' threads)' //3 for digits, 8 for ' cores (', 3 for digits, 9 for ' threads)'
uint32_t size = 3+8+3+9+1; uint32_t size = 3+8+3+9+1;
string = malloc(sizeof(char)*size); string = malloc(sizeof(char)*size);
if(dual_socket)
snprintf(string, size, "%d cores (%d threads)",topo->physical_cores * topo->sockets, topo->logical_cores * topo->sockets);
else
snprintf(string, size, "%d cores (%d threads)",topo->physical_cores,topo->logical_cores); snprintf(string, size, "%d cores (%d threads)",topo->physical_cores,topo->logical_cores);
} }
else { else {
uint32_t size = 3+7+1; uint32_t size = 3+7+1;
string = malloc(sizeof(char)*size); string = malloc(sizeof(char)*size);
if(dual_socket)
snprintf(string, size, "%d cores",topo->physical_cores * topo->sockets);
else
snprintf(string, size, "%d cores",topo->physical_cores); snprintf(string, size, "%d cores",topo->physical_cores);
} }
return string; return string;
} }
char* get_str_sockets(struct topology* topo) {
char* string = malloc(sizeof(char) * 2);
int32_t sanity_ret = snprintf(string, 2, "%d", topo->sockets);
if(sanity_ret < 0) {
printBug("get_str_sockets: snprintf returned a negative value for input: '%d'", topo->sockets);
return NULL;
}
return string;
}
char* get_str_cpu_name(struct cpuInfo* cpu) { char* get_str_cpu_name(struct cpuInfo* cpu) {
return cpu->cpu_name; return cpu->cpu_name;
} }
@@ -702,12 +722,12 @@ char* get_str_sha(struct cpuInfo* cpu) {
int32_t get_value_as_smallest_unit(char ** str, uint32_t value) { int32_t get_value_as_smallest_unit(char ** str, uint32_t value) {
int32_t sanity_ret; int32_t sanity_ret;
*str = malloc(sizeof(char)* 7); //4 for digits, 2 for units *str = malloc(sizeof(char)* 11); //8 for digits, 2 for units
if(value/1024 >= 1024) if(value/1024 >= 1024)
sanity_ret = snprintf(*str, 6,"%d"STRING_MEGABYTES,value/(1<<20)); sanity_ret = snprintf(*str, 10,"%.4g"STRING_MEGABYTES, (double)value/(1<<20));
else else
sanity_ret = snprintf(*str, 6,"%d"STRING_KILOBYTES,value/(1<<10)); sanity_ret = snprintf(*str, 10,"%.4g"STRING_KILOBYTES, (double)value/(1<<10));
return sanity_ret; return sanity_ret;
} }
@@ -754,7 +774,7 @@ char* get_str_cache_one(int32_t cache_size) {
int32_t tmp_len = get_value_as_smallest_unit(&tmp, cache_size); int32_t tmp_len = get_value_as_smallest_unit(&tmp, cache_size);
if(tmp_len < 0) { if(tmp_len < 0) {
printBug("get_value_as_smallest_unit: snprintf returned a negative value for input: %d\n", cache_size); printBug("get_value_as_smallest_unit: snprintf returned a negative value for input: %d", cache_size);
return NULL; return NULL;
} }
@@ -762,7 +782,7 @@ char* get_str_cache_one(int32_t cache_size) {
sanity_ret = snprintf(string, size, "%s", tmp); sanity_ret = snprintf(string, size, "%s", tmp);
if(sanity_ret < 0) { if(sanity_ret < 0) {
printBug("get_str_cache_one: snprintf returned a negative value for input: '%s'\n", tmp); printBug("get_str_cache_one: snprintf returned a negative value for input: '%s'", tmp);
return NULL; return NULL;
} }
free(tmp); free(tmp);
@@ -770,14 +790,18 @@ char* get_str_cache_one(int32_t cache_size) {
} }
char* get_str_cache(int32_t cache_size, struct topology* topo, bool llc) { char* get_str_cache(int32_t cache_size, struct topology* topo, bool llc) {
if(llc) { if(topo->sockets == 1) {
if(topo->sockets == 1) if(llc)
return get_str_cache_one(cache_size); return get_str_cache_one(cache_size);
else
return get_str_cache_two(cache_size, topo->sockets);
}
else else
return get_str_cache_two(cache_size, topo->physical_cores); return get_str_cache_two(cache_size, topo->physical_cores);
}
else {
if(llc)
return get_str_cache_two(cache_size, topo->sockets);
else
return get_str_cache_two(cache_size, topo->physical_cores * topo->sockets);
}
} }
char* get_str_l1i(struct cache* cach, struct topology* topo) { char* get_str_l1i(struct cache* cach, struct topology* topo) {

View File

@@ -19,6 +19,7 @@ typedef int32_t VENDOR;
struct cpuInfo* get_cpu_info(); struct cpuInfo* get_cpu_info();
VENDOR get_cpu_vendor(struct cpuInfo* cpu); VENDOR get_cpu_vendor(struct cpuInfo* cpu);
uint32_t get_nsockets(struct topology* topo);
int64_t get_freq(struct frequency* freq); int64_t get_freq(struct frequency* freq);
struct cache* get_cache_info(struct cpuInfo* cpu); struct cache* get_cache_info(struct cpuInfo* cpu);
struct frequency* get_frequency_info(struct cpuInfo* cpu); struct frequency* get_frequency_info(struct cpuInfo* cpu);
@@ -39,7 +40,8 @@ char* get_str_l3(struct cache* cach, struct topology* topo);
char* get_str_freq(struct frequency* freq); char* get_str_freq(struct frequency* freq);
char* get_str_topology(struct topology* topo); char* get_str_sockets(struct topology* topo);
char* get_str_topology(struct topology* topo, bool dual_socket);
char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq); char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq);

View File

@@ -6,30 +6,7 @@
#include "cpuid.h" #include "cpuid.h"
#include "global.h" #include "global.h"
/*** static const char* VERSION = "0.5";
################
####### #######
#### ####
### ####
### ### Name: Intel(R) Xeon(R) CPU E5-2698 v4 @ 2.20GHz
### ### Sockets: 2
Cores: 20 (40 threads)
Frequency: 3.60 GHz
## ### ######### ###### ###### ### ### AVX: AVX,AVX2
## ### ### ### ### #### #### ### ### SSE: SSE,SSE2,SSE3,SSSE3,SSE4_1,SSE4_2
## ### ### ### ### ### ### ### ### FMA: FMA3
## ## ### ### ##### ######### ## ### L1d Size: 32KB (1MB Total)
## ## L1i Size: 32KB (1MB Total)
### L2 Size: 256KB (4MB Total)
### L3 Size: 50M (100MB Total)
#### #### Peak Perf.: 2.30 TFLOP/s
##### ##########
########## ################
###############################
***/
static const char* VERSION = "0.411";
void print_help(char *argv[]) { void print_help(char *argv[]) {
printf("Usage: %s [--version] [--help] [--style STYLE]\n\ printf("Usage: %s [--version] [--help] [--style STYLE]\n\

View File

@@ -18,50 +18,52 @@
#define COL_AMD_DARK_2 "\x1b[32;1m" #define COL_AMD_DARK_2 "\x1b[32;1m"
#define RESET "\x1b[0m" #define RESET "\x1b[0m"
#define TITLE_NAME "Name: " #define TITLE_NAME "Name:"
#define TITLE_FREQUENCY "Frequency: " #define TITLE_FREQUENCY "Frequency:"
#define TITLE_NCORES "N.Cores: " #define TITLE_SOCKETS "Sockets:"
#define TITLE_AVX "AVX: " #define TITLE_NCORES "Cores:"
#define TITLE_SSE "SSE: " #define TITLE_NCORES_DUAL "Cores (Total):"
#define TITLE_FMA "FMA: " #define TITLE_AVX "AVX:"
#define TITLE_AES "AES: " #define TITLE_SSE "SSE:"
#define TITLE_SHA "SHA: " #define TITLE_FMA "FMA:"
#define TITLE_L1i "L1i Size: " #define TITLE_AES "AES:"
#define TITLE_L1d "L1d Size: " #define TITLE_SHA "SHA:"
#define TITLE_L2 "L2 Size: " #define TITLE_L1i "L1i Size:"
#define TITLE_L3 "L3 Size: " #define TITLE_L1d "L1d Size:"
#define TITLE_PEAK "Peak FLOPS: " #define TITLE_L2 "L2 Size:"
#define TITLE_L3 "L3 Size:"
#define TITLE_PEAK "Peak Perf.:"
/*** CENTER TEXT ***/ #define MAX_ATTRIBUTE_COUNT 15
#define LINES_SPACE_UP 3
#define LINES_SPACE_DOWN 4
#define ATTRIBUTE_COUNT 13
#define ATTRIBUTE_NAME 0 #define ATTRIBUTE_NAME 0
#define ATTRIBUTE_FREQUENCY 1 #define ATTRIBUTE_FREQUENCY 1
#define ATTRIBUTE_NCORES 2 #define ATTRIBUTE_SOCKETS 2
#define ATTRIBUTE_AVX 3 #define ATTRIBUTE_NCORES 3
#define ATTRIBUTE_SSE 4 #define ATTRIBUTE_NCORES_DUAL 4
#define ATTRIBUTE_FMA 5 #define ATTRIBUTE_AVX 5
#define ATTRIBUTE_AES 6 #define ATTRIBUTE_SSE 6
#define ATTRIBUTE_SHA 7 #define ATTRIBUTE_FMA 7
#define ATTRIBUTE_L1i 8 #define ATTRIBUTE_AES 8
#define ATTRIBUTE_L1d 9 #define ATTRIBUTE_SHA 9
#define ATTRIBUTE_L2 10 #define ATTRIBUTE_L1i 10
#define ATTRIBUTE_L3 11 #define ATTRIBUTE_L1d 11
#define ATTRIBUTE_PEAK 12 #define ATTRIBUTE_L2 12
#define ATTRIBUTE_L3 13
#define ATTRIBUTE_PEAK 14
static const int STYLES_CODE_LIST [STYLES_COUNT] = {STYLE_DEFAULT, STYLE_DARK}; static const int STYLES_CODE_LIST [STYLES_COUNT] = {STYLE_DEFAULT, STYLE_DARK};
static const char* ATTRIBUTE_FIELDS [ATTRIBUTE_COUNT] = { TITLE_NAME, TITLE_FREQUENCY, static const char* ATTRIBUTE_FIELDS [MAX_ATTRIBUTE_COUNT] = { TITLE_NAME, TITLE_FREQUENCY, TITLE_SOCKETS,
TITLE_NCORES, TITLE_AVX, TITLE_SSE, TITLE_NCORES, TITLE_NCORES_DUAL,
TITLE_AVX, TITLE_SSE,
TITLE_FMA, TITLE_AES, TITLE_SHA, TITLE_FMA, TITLE_AES, TITLE_SHA,
TITLE_L1i, TITLE_L1d, TITLE_L2, TITLE_L3, TITLE_L1i, TITLE_L1d, TITLE_L2, TITLE_L3,
TITLE_PEAK }; TITLE_PEAK
};
static const int ATTRIBUTE_LIST[ATTRIBUTE_COUNT] = { ATTRIBUTE_NAME, ATTRIBUTE_FREQUENCY, static const int ATTRIBUTE_LIST[MAX_ATTRIBUTE_COUNT] = { ATTRIBUTE_NAME, ATTRIBUTE_FREQUENCY, ATTRIBUTE_SOCKETS,
ATTRIBUTE_NCORES, ATTRIBUTE_AVX, ATTRIBUTE_SSE, ATTRIBUTE_NCORES, ATTRIBUTE_NCORES_DUAL, ATTRIBUTE_AVX,
ATTRIBUTE_FMA, ATTRIBUTE_AES, ATTRIBUTE_SHA, ATTRIBUTE_SSE, ATTRIBUTE_FMA, ATTRIBUTE_AES, ATTRIBUTE_SHA,
ATTRIBUTE_L1i, ATTRIBUTE_L1d, ATTRIBUTE_L2, ATTRIBUTE_L3, ATTRIBUTE_L1i, ATTRIBUTE_L1d, ATTRIBUTE_L2, ATTRIBUTE_L3,
ATTRIBUTE_PEAK }; ATTRIBUTE_PEAK };
@@ -70,31 +72,35 @@ struct ascii {
char color1[10]; char color1[10];
char color2[10]; char color2[10];
char reset[10]; char reset[10];
char* atributes[ATTRIBUTE_COUNT]; char* attributes[MAX_ATTRIBUTE_COUNT];
uint32_t n_attributes_set;
VENDOR vendor; VENDOR vendor;
}; };
void setAttribute(struct ascii* art, int type, char* value) { void setAttribute(struct ascii* art, int type, char* value) {
int i = 0; art->attributes[type] = value;
while(i < ATTRIBUTE_COUNT && type != ATTRIBUTE_LIST[i]) art->n_attributes_set++;
i++;
if(i != ATTRIBUTE_COUNT)
art->atributes[i] = value;
else
printBug("Setting attribute failed because it was not found");
} }
struct ascii* set_ascii(VENDOR cpuVendor, STYLE style) { struct ascii* set_ascii(VENDOR cpuVendor, STYLE style) {
/*** Check that number of lines of ascii art matches the number // Sanity checks //
of spaces plus the number of lines filled with text ***/ for(int i=0; i < MAX_ATTRIBUTE_COUNT; i++) {
if(LINES_SPACE_UP+LINES_SPACE_DOWN+ATTRIBUTE_COUNT != NUMBER_OF_LINES) { if(ATTRIBUTE_FIELDS[i] == NULL) {
printBug("Number of lines do not match (%d vs %d)",LINES_SPACE_UP+LINES_SPACE_DOWN+ATTRIBUTE_COUNT,NUMBER_OF_LINES); printBug("Attribute field at position %d is empty", i);
return NULL; return NULL;
} }
if(i > 0 && ATTRIBUTE_LIST[i] == 0) {
printBug("Attribute list at position %d is empty", i);
return NULL;
}
}
char *COL_DEFAULT_1, *COL_DEFAULT_2, *COL_DARK_1, *COL_DARK_2; char *COL_DEFAULT_1, *COL_DEFAULT_2, *COL_DARK_1, *COL_DARK_2;
struct ascii* art = malloc(sizeof(struct ascii)); struct ascii* art = malloc(sizeof(struct ascii));
art->n_attributes_set = 0;
art->vendor = cpuVendor; art->vendor = cpuVendor;
for(int i=0; i < MAX_ATTRIBUTE_COUNT; i++)
art->attributes[i] = NULL;
strcpy(art->reset,RESET); strcpy(art->reset,RESET);
if(cpuVendor == VENDOR_INTEL) { if(cpuVendor == VENDOR_INTEL) {
@@ -144,12 +150,21 @@ struct ascii* set_ascii(VENDOR cpuVendor, STYLE style) {
return art; return art;
} }
void print_ascii_intel(struct ascii* art) { uint32_t get_next_attribute(struct ascii* art, uint32_t last_attr) {
last_attr++;
while(art->attributes[last_attr] == NULL) last_attr++;
return last_attr;
}
void print_ascii_intel(struct ascii* art, uint32_t la) {
bool flag = false; bool flag = false;
int attr_to_print = -1;
uint32_t space_right;
uint32_t space_up = (NUMBER_OF_LINES - art->n_attributes_set)/2;
uint32_t space_down = NUMBER_OF_LINES - art->n_attributes_set - space_up;
for(int n=0;n<NUMBER_OF_LINES;n++) { for(uint32_t n=0;n<NUMBER_OF_LINES;n++) {
/*** PRINT ASCII-ART ***/
for(int i=0;i<LINE_SIZE;i++) { for(int i=0;i<LINE_SIZE;i++) {
if(flag) { if(flag) {
if(art->art[n][i] == ' ') { if(art->art[n][i] == ' ') {
@@ -169,17 +184,22 @@ void print_ascii_intel(struct ascii* art) {
} }
} }
/*** PRINT ATTRIBUTE ***/
if(n>LINES_SPACE_UP-1 && n<NUMBER_OF_LINES-LINES_SPACE_DOWN) if(n > space_up-1 && n < NUMBER_OF_LINES-space_down) {
printf("%s%s%s%s%s\n",art->color1,ATTRIBUTE_FIELDS[n-LINES_SPACE_UP],art->color2,art->atributes[n-LINES_SPACE_UP],art->reset); attr_to_print = get_next_attribute(art, attr_to_print);
space_right = 1 + (la - strlen(ATTRIBUTE_FIELDS[attr_to_print]));
printf("%s%s%*s%s%s%s\n",art->color1, ATTRIBUTE_FIELDS[attr_to_print], space_right, "", art->color2, art->attributes[attr_to_print], art->reset);
}
else printf("\n"); else printf("\n");
} }
} }
void print_ascii_amd(struct ascii* art) { void print_ascii_amd(struct ascii* art, uint32_t la) {
int attr_to_print = -1;
uint32_t space_up = (NUMBER_OF_LINES - art->n_attributes_set)/2;
uint32_t space_down = NUMBER_OF_LINES - art->n_attributes_set - space_up;
for(int n=0;n<NUMBER_OF_LINES;n++) { for(uint32_t n=0;n<NUMBER_OF_LINES;n++) {
/*** PRINT ASCII-ART ***/
for(int i=0;i<LINE_SIZE;i++) { for(int i=0;i<LINE_SIZE;i++) {
if(art->art[n][i] == '@') if(art->art[n][i] == '@')
printf("%s%c%s", art->color1, art->art[n][i], art->reset); printf("%s%c%s", art->color1, art->art[n][i], art->reset);
@@ -189,29 +209,47 @@ void print_ascii_amd(struct ascii* art) {
printf("%c",art->art[n][i]); printf("%c",art->art[n][i]);
} }
/*** PRINT ATTRIBUTE ***/ if(n > space_up-1 && n < NUMBER_OF_LINES-space_down) {
if(n>LINES_SPACE_UP-1 && n<NUMBER_OF_LINES-LINES_SPACE_DOWN) attr_to_print = get_next_attribute(art, attr_to_print);
printf("%s%s%s%s%s\n",art->color1,ATTRIBUTE_FIELDS[n-LINES_SPACE_UP],art->color2,art->atributes[n-LINES_SPACE_UP], art->reset); printf("%s%s%s%s%s\n",art->color1, ATTRIBUTE_FIELDS[attr_to_print], art->color2, art->attributes[attr_to_print], art->reset);
}
else printf("\n"); else printf("\n");
} }
} }
uint32_t longest_attribute_length(struct ascii* art) {
uint32_t max = 0;
uint64_t len = 0;
for(int i=0; i < MAX_ATTRIBUTE_COUNT; i++) {
if(art->attributes[i] != NULL) {
len = strlen(ATTRIBUTE_FIELDS[i]);
if(len > max) max = len;
}
}
return max;
}
void print_ascii(struct ascii* art) { void print_ascii(struct ascii* art) {
uint32_t longest_attribute = longest_attribute_length(art);
if(art->vendor == VENDOR_INTEL) if(art->vendor == VENDOR_INTEL)
print_ascii_intel(art); print_ascii_intel(art, longest_attribute);
else else
print_ascii_amd(art); print_ascii_amd(art, longest_attribute);
} }
bool print_cpufetch(struct cpuInfo* cpu, struct cache* cach, struct frequency* freq, struct topology* topo, STYLE s) { bool print_cpufetch(struct cpuInfo* cpu, struct cache* cach, struct frequency* freq, struct topology* topo, STYLE s) {
struct ascii* art = set_ascii(get_cpu_vendor(cpu),s); struct ascii* art = set_ascii(get_cpu_vendor(cpu), s);
if(art == NULL) if(art == NULL)
return false; return false;
char* cpu_name = get_str_cpu_name(cpu); char* cpu_name = get_str_cpu_name(cpu);
char* sockets = get_str_sockets(topo);
char* max_frequency = get_str_freq(freq); char* max_frequency = get_str_freq(freq);
char* nCores = get_str_topology(topo); char* n_cores = get_str_topology(topo, false);
char* n_cores_dual = get_str_topology(topo, true);
char* avx = get_str_avx(cpu); char* avx = get_str_avx(cpu);
char* sse = get_str_sse(cpu); char* sse = get_str_sse(cpu);
char* fma = get_str_fma(cpu); char* fma = get_str_fma(cpu);
@@ -225,7 +263,7 @@ bool print_cpufetch(struct cpuInfo* cpu, struct cache* cach, struct frequency* f
setAttribute(art,ATTRIBUTE_NAME,cpu_name); setAttribute(art,ATTRIBUTE_NAME,cpu_name);
setAttribute(art,ATTRIBUTE_FREQUENCY,max_frequency); setAttribute(art,ATTRIBUTE_FREQUENCY,max_frequency);
setAttribute(art,ATTRIBUTE_NCORES,nCores); setAttribute(art,ATTRIBUTE_NCORES,n_cores);
setAttribute(art,ATTRIBUTE_AVX,avx); setAttribute(art,ATTRIBUTE_AVX,avx);
setAttribute(art,ATTRIBUTE_SSE,sse); setAttribute(art,ATTRIBUTE_SSE,sse);
setAttribute(art,ATTRIBUTE_FMA,fma); setAttribute(art,ATTRIBUTE_FMA,fma);
@@ -237,11 +275,24 @@ bool print_cpufetch(struct cpuInfo* cpu, struct cache* cach, struct frequency* f
setAttribute(art,ATTRIBUTE_L3,l3); setAttribute(art,ATTRIBUTE_L3,l3);
setAttribute(art,ATTRIBUTE_PEAK,pp); setAttribute(art,ATTRIBUTE_PEAK,pp);
uint32_t socket_num = get_nsockets(topo);
if (socket_num > 1) {
setAttribute(art, ATTRIBUTE_SOCKETS, sockets);
setAttribute(art, ATTRIBUTE_NCORES_DUAL, n_cores_dual);
}
if(art->n_attributes_set > NUMBER_OF_LINES) {
printBug("The number of attributes set is bigger than the max that can be displayed");
return false;
}
print_ascii(art); print_ascii(art);
free(cpu_name); free(cpu_name);
free(max_frequency); free(max_frequency);
free(nCores); free(sockets);
free(n_cores);
free(n_cores_dual);
free(avx); free(avx);
free(sse); free(sse);
free(fma); free(fma);