Updated some performance evaluation code.

Makefile

@@ -15,7 +15,7 @@ EXECUTABLES=parse
 
 all: $(EXECUTABLES)
 
-parse: main.cpp common_defs.h
+parse: main.cpp common_defs.h linux-perf-events.h
 	$(CXX) $(CXXFLAGS) -o parse main.cpp
 
 

linux-perf-events.h

@@ -11,58 +11,75 @@
 #include <cstring>              // for memset
 #include <stdexcept>
 
+#include <vector>
 
 template <int TYPE = PERF_TYPE_HARDWARE>
 class LinuxEvents {
-
     int fd;
     perf_event_attr attribs;
-
+    int num_events;
+    std::vector<uint64_t> temp_result_vec; 
+    std::vector<uint64_t> ids;
 public:
-    LinuxEvents(int config) : fd(0) {
+    LinuxEvents(std::vector<int> config_vec) : fd(0) {
         memset(&attribs, 0, sizeof(attribs));
         attribs.type        = TYPE;
         attribs.size        = sizeof(attribs);
-        attribs.config      = config;
         attribs.disabled        = 1;
         attribs.exclude_kernel  = 1;
         attribs.exclude_hv      = 1;
 
+        attribs.sample_period = 0;
+        attribs.read_format = PERF_FORMAT_GROUP | PERF_FORMAT_ID;
         const int pid = 0;    // the current process
         const int cpu = -1;   // all CPUs
-        const int group = -1; // no group
         const unsigned long flags = 0;
-        fd = syscall(__NR_perf_event_open, &attribs, pid, cpu, group, flags);
-        if (fd == -1) {
-            report_error("perf_event_open");
+
+        int group = -1; // no group
+        num_events = config_vec.size();
+        u32 i = 0;
+        for (auto config: config_vec) {
+            attribs.config = config;
+            fd = syscall(__NR_perf_event_open, &attribs, pid, cpu, group, flags);
+            if (fd == -1) {
+                report_error("perf_event_open");
+            }
+            ioctl(fd, PERF_EVENT_IOC_ID, &ids[i++]);
+            if (group == -1) {
+                group = fd;
+            }
         }
+
+        temp_result_vec.resize(num_events*2 + 1);
     }
 
     ~LinuxEvents() {
         close(fd);
     }
 
-    void start() {
-        if (ioctl(fd, PERF_EVENT_IOC_RESET, 0) == -1) {
+    really_inline void start() {
+        if (ioctl(fd, PERF_EVENT_IOC_RESET, PERF_IOC_FLAG_GROUP) == -1) {
             report_error("ioctl(PERF_EVENT_IOC_RESET)");
         }
 
-        if (ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) == -1) {
+        if (ioctl(fd, PERF_EVENT_IOC_ENABLE, PERF_IOC_FLAG_GROUP) == -1) {
             report_error("ioctl(PERF_EVENT_IOC_ENABLE)");
         }
     }
 
-    unsigned long end() {
-        if (ioctl(fd, PERF_EVENT_IOC_DISABLE, 0) == -1) {
+    really_inline void end(std::vector<unsigned long long> & results) {
+        if (ioctl(fd, PERF_EVENT_IOC_DISABLE, PERF_IOC_FLAG_GROUP) == -1) {
             report_error("ioctl(PERF_EVENT_IOC_DISABLE)");
         }
 
-        unsigned long result;
-        if (read(fd, &result, sizeof(result)) == -1) {
+        if (read(fd, &temp_result_vec[0], temp_result_vec.size() * 8) == -1) {
             report_error("read");
         }
-
-        return result;
+        // our actual results are in slots 1,3,5, ... of this structure
+        // we really should be checking our ids obtained earlier to be safe
+        for (u32 i = 1; i < temp_result_vec.size(); i+=2) {
+            results[i/2] = temp_result_vec[i];
+        }
     }
 
 private:
@@ -72,3 +89,5 @@ private:
 
 };
 #endif 
+
+

main.cpp

@@ -1,4 +1,3 @@
-#include "linux-perf-events.h"
 #include <iostream>
 #include <iomanip>
 #include <chrono>
@@ -13,6 +12,7 @@
 #include <x86intrin.h>
 #include <assert.h>
 #include "common_defs.h"
+#include "linux-perf-events.h"
 
 using namespace std;
 
@@ -1200,34 +1200,42 @@ int main(int argc, char * argv[]) {
 #endif
 
 #ifndef SQUASH_COUNTERS
-    LinuxEvents<PERF_TYPE_HARDWARE> cycles(PERF_COUNT_HW_CPU_CYCLES);
-    LinuxEvents<PERF_TYPE_HARDWARE> instructions(PERF_COUNT_HW_INSTRUCTIONS);
+    vector<int> evts;
+    evts.push_back(PERF_COUNT_HW_CPU_CYCLES);
+    evts.push_back(PERF_COUNT_HW_INSTRUCTIONS);
+    LinuxEvents<PERF_TYPE_HARDWARE> unified(evts);
+    vector<u64> results;
+    results.resize(evts.size());
     unsigned long cy1 = 0, cy2 = 0, cy3 = 0, cy4 = 0;
     unsigned long cl1 = 0, cl2 = 0, cl3 = 0, cl4 = 0;
 #endif
     for (u32 i = 0; i < iterations; i++) {
         auto start = std::chrono::steady_clock::now();
 #ifndef SQUASH_COUNTERS
-        cycles.start(); instructions.start();
+        unified.start();
 #endif
         find_structural_bits(p.first, p.second, pj);
 #ifndef SQUASH_COUNTERS
-        cl1 += instructions.end(); cy1 += cycles.end();
-        cycles.start(); instructions.start();
+        unified.end(results);
+        cy1 += results[0]; cl1 += results[1];
+        unified.start();
 #endif
         flatten_indexes(p.second, pj);
 #ifndef SQUASH_COUNTERS
-        cl2 += instructions.end(); cy2 += cycles.end();
-        cycles.start(); instructions.start();
+        unified.end(results);
+        cy2 += results[0]; cl2 += results[1];
+        unified.start();
 #endif
         ape_machine(p.first, p.second, pj);
 #ifndef SQUASH_COUNTERS
-        cl3 += instructions.end(); cy3 += cycles.end();
-        cycles.start(); instructions.start();
+        unified.end(results);
+        cy3 += results[0]; cl3 += results[1];
+        unified.start();
 #endif
         shovel_machine(p.first, p.second, pj);
 #ifndef SQUASH_COUNTERS
-        cl4 += instructions.end(); cy4 += cycles.end();
+        unified.end(results);
+        cy4 += results[0]; cl4 += results[1];
 #endif
         auto end = std::chrono::steady_clock::now();
         std::chrono::duration<double> secs = end - start;