thelilfrog/cpufetch
1
0
Fork 0
mirror of https://github.com/Dr-Noob/cpufetch.git synced 2026-03-25 07:50:40 +01:00
Code Issues Packages Projects Releases Wiki Activity

[v0.98][Refactoring] Unify the use of get_str_peak_performance

Browse Source
This commit is contained in:
Dr-Noob
2021-08-06 10:26:07 +02:00
parent 7e1dde3c71
commit 6953d8dda5
9 changed files with 73 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files

61
src/x86/cpuid.c
View File

@@ -655,65 +655,48 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
return freq;
}
/*** STRING FUNCTIONS ***/
// STRING FUNCTIONS
bool get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq, double* flops) {
/*
* PP = PeakPerformance
* SP = SinglePrecision
*
* PP(SP) =
* N_CORES *
* FREQUENCY *
* 2(Two vector units) *
* 2(If cpu has fma) *
* 16(If AVX512), 8(If AVX), 4(If SSE) *
*/
char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq) {
/***
PP = PeakPerformance
SP = SinglePrecision
PP(SP) =
N_CORES *
FREQUENCY *
2(Two vector units) *
2(If cpu has fma) *
16(If AVX512), 8(If AVX), 4(If SSE) *
***/
//7 for GFLOP/s and 6 for digits,eg 412.14
uint32_t size = 7+6+1+1;
assert(strlen(STRING_UNKNOWN)+1 <= size);
char* string = emalloc(sizeof(char)*size);
//First check we have consistent data
//First, check we have consistent data
if(freq == UNKNOWN_FREQ) {
snprintf(string,strlen(STRING_UNKNOWN)+1,STRING_UNKNOWN);
return string;
return false;
}
struct features* feat = cpu->feat;
double flops = topo->physical_cores * topo->sockets * (freq*1000000);
int vpus = get_number_of_vpus(cpu);
flops = flops * vpus;
*flops = topo->physical_cores * topo->sockets * (freq*1000000) * vpus;
if(feat->FMA3 || feat->FMA4)
flops = flops*2;
*flops = *flops*2;
// Ice Lake has AVX512, but it has 1 VPU for AVX512, while
// it has 2 for AVX2. If this is a Ice Lake CPU, we are computing
// the peak performance supposing AVX2, not AVX512
if(feat->AVX512 && vpus_are_AVX512(cpu))
flops = flops*16;
*flops = *flops*16;
else if(feat->AVX || feat->AVX2)
flops = flops*8;
*flops = *flops*8;
else if(feat->SSE)
flops = flops*4;
*flops = *flops*4;
// See https://sites.utexas.edu/jdm4372/2018/01/22/a-peculiar-
// throughput-limitation-on-intels-xeon-phi-x200-knights-landing/
if(is_knights_landing(cpu))
flops = flops * 6 / 7;
*flops = *flops * 6 / 7;
if(flops >= (double)1000000000000.0)
snprintf(string,size,"%.2f TFLOP/s",flops/1000000000000);
else if(flops >= 1000000000.0)
snprintf(string,size,"%.2f GFLOP/s",flops/1000000000);
else
snprintf(string,size,"%.2f MFLOP/s",flops/1000000);
return string;
return true;
}
// TODO: Refactoring

3
src/x86/cpuid.h
View File

@@ -12,7 +12,8 @@ char* get_str_avx(struct cpuInfo* cpu);
char* get_str_sse(struct cpuInfo* cpu);
char* get_str_fma(struct cpuInfo* cpu);
char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_socket);
char* get_str_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq);
bool get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq, double* flops);
void print_debug(struct cpuInfo* cpu);
void print_raw(struct cpuInfo* cpu);