Count instructions where timing cannot tell 3% apart (110)

016's second stage passed its criterion: six interleaved pairs against
ee21aec gave TPC-C (RAM, W=1) -0.6% at the median and the bank's
queries 0.0% to +2.2%, no run interfered. The ranks query was slower
all six times, and chasing it showed that on this Pi timing cannot
separate 1-3%: two servers asked in turns disagree with themselves
(ee21aec to e02bfc9 +1.6%, e02bfc9 to 481642f -1.2%, ee21aec to 481642f
+2.8%), and a query whose instructions grew 0.6% ran 3.5% faster.

Instructions do not move so: pcount.c counts the server's user-mode
instructions with perf_event_open (no perf here), and psample.c samples
where they go. ranks had grown 0.4% in instructions (the rest was where
code and data landed), per_branch 3.6% and tables 20 instructions a row,
which 07047c3 halved. The six pairs of 07047c3 were interfered (another
session's build, load 4-12) and are kept apart, unused.

1 files changed, 58 insertions, 0 deletions

diff --git a/research/experiments/025-i64-cost/pcount.c b/research/experiments/025-i64-cost/pcount.c
new file mode 100644
index 0000000..77519b2
--- /dev/null
+++ b/research/experiments/025-i64-cost/pcount.c
@@ -0,0 +1,58 @@
+/* pcount PID: count user-mode instructions and cycles of every thread of
+ * PID from "ready" (on stderr) until stdin closes; prints them as JSON.
+ * perf_event_open alone: a perf_event_paranoid of 2 allows it for one's own
+ * processes, and no perf tool is needed. */
+#define _GNU_SOURCE
+#include <dirent.h>
+#include <linux/perf_event.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/syscall.h>
+#include <unistd.h>
+
+static int open_counter(pid_t tid, uint64_t config) {
+    struct perf_event_attr attr;
+    memset(&attr, 0, sizeof(attr));
+    attr.size = sizeof(attr);
+    attr.type = PERF_TYPE_HARDWARE;
+    attr.config = config;
+    attr.disabled = 1;
+    attr.exclude_kernel = 1;
+    attr.exclude_hv = 1;
+    return (int)syscall(SYS_perf_event_open, &attr, tid, -1, -1, 0);
+}
+
+int main(int argc, char **argv) {
+    if (argc != 2) return 2;
+    char path[64];
+    snprintf(path, sizeof(path), "/proc/%s/task", argv[1]);
+    DIR *dir = opendir(path);
+    if (!dir) { perror("opendir"); return 1; }
+    int fds[512][2], n = 0;
+    struct dirent *e;
+    while ((e = readdir(dir)) && n < 512) {
+        if (e->d_name[0] == '.') continue;
+        pid_t tid = atoi(e->d_name);
+        fds[n][0] = open_counter(tid, PERF_COUNT_HW_INSTRUCTIONS);
+        fds[n][1] = open_counter(tid, PERF_COUNT_HW_CPU_CYCLES);
+        if (fds[n][0] < 0 || fds[n][1] < 0) { perror("perf_event_open"); return 1; }
+        ++n;
+    }
+    closedir(dir);
+    for (int i = 0; i < n; ++i) for (int k = 0; k < 2; ++k) ioctl(fds[i][k], PERF_EVENT_IOC_RESET, 0), ioctl(fds[i][k], PERF_EVENT_IOC_ENABLE, 0);
+    fprintf(stderr, "ready\n");
+    char buffer[64];
+    while (fread(buffer, 1, sizeof(buffer), stdin) > 0) {}
+    uint64_t total[2] = {0, 0};
+    for (int i = 0; i < n; ++i)
+        for (int k = 0; k < 2; ++k) {
+            uint64_t v = 0;
+            ioctl(fds[i][k], PERF_EVENT_IOC_DISABLE, 0);
+            if (read(fds[i][k], &v, sizeof(v)) == sizeof(v)) total[k] += v;
+        }
+    printf("{\"threads\": %d, \"instructions\": %llu, \"cycles\": %llu}\n", n, (unsigned long long)total[0], (unsigned long long)total[1]);
+    return 0;
+}