thelilfrog/cpufetch
1
0
Fork 0
mirror of https://github.com/Dr-Noob/cpufetch.git synced 2026-03-25 16:00:39 +01:00
Code Issues Packages Projects Releases Wiki Activity

[v0.98][Refactoring] Use int for peak performance, which makes code cleaner

Browse Source
This commit is contained in:
Dr-Noob
2021-08-06 11:04:29 +02:00
parent 6953d8dda5
commit c24dd7cbb6
9 changed files with 106 additions and 115 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
src/ppc/ppc.c
View File

@@ -132,6 +132,32 @@ struct frequency* get_frequency_info() {
return freq;
}
int64_t get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq) {
/*
* Not sure about this
* PP(SP) = N_CORES * FREQUENCY * 4(If altivec)
*/
//First check we have consistent data
if(freq == UNKNOWN_FREQ) {
return -1;
}
struct features* feat = cpu->feat;
int64_t flops = topo->physical_cores * topo->sockets * (freq * 1000000);
if(feat->altivec) flops = flops * 4;
// POWER9 has the concept called "slices". Each SMT4 core has two super-slices,
// and each super-slice is capable of doing two FLOPS per cycle. In the case of
// SMT8, it has 4 super-slices, thus four FLOPS per cycle.
if(is_power9(cpu->arch)) {
int threads_per_core = topo->logical_cores / topo->physical_cores;
flops = flops * (threads_per_core / 2);
}
return flops;
}
struct cpuInfo* get_cpu_info() {
struct cpuInfo* cpu = emalloc(sizeof(struct cpuInfo));
struct features* feat = emalloc(sizeof(struct features));
@@ -152,8 +178,8 @@ struct cpuInfo* get_cpu_info() {
cpu->freq = get_frequency_info();
cpu->topo = get_topology_info(cpu->cach);
cpu->cach = get_cache_info(cpu);
feat->altivec = has_altivec(cpu->arch);
cpu->peak_performance = get_peak_performance(cpu, cpu->topo, get_freq(cpu->freq));
if(cpu->cach == NULL || cpu->topo == NULL) {
return NULL;
@@ -170,32 +196,6 @@ char* get_str_altivec(struct cpuInfo* cpu) {
return string;
}
bool get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq, double* flops) {
/*
* Not sure about this
* PP(SP) = N_CORES * FREQUENCY * 4(If altivec)
*/
//First check we have consistent data
if(freq == UNKNOWN_FREQ) {
return false;
}
struct features* feat = cpu->feat;
*flops = topo->physical_cores * topo->sockets * (freq*1000000);
if(feat->altivec) *flops = *flops * 4;
// POWER9 has the concept called "slices". Each SMT4 core has two super-slices,
// and each super-slice is capable of doing two FLOPS per cycle. In the case of
// SMT8, it has 4 super-slices, thus four FLOPS per cycle.
if(is_power9(cpu->arch)) {
int threads_per_core = topo->logical_cores / topo->physical_cores;
*flops = *flops * (threads_per_core / 2);
}
return true;
}
char* get_str_topology(struct topology* topo, bool dual_socket) {
char* string;
if(topo->smt_supported > 1) {

1
src/ppc/ppc.h
View File

@@ -6,7 +6,6 @@
struct cpuInfo* get_cpu_info();
char* get_str_altivec(struct cpuInfo* cpu);
char* get_str_topology(struct topology* topo, bool dual_socket);
bool get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t freq, double* flops);
void print_debug(struct cpuInfo* cpu);
#endif