Validation (#71)

Validating of OSACA predictions for IVB, SKX, ZEN1, ZEN2, A64FX and TX2 with different kernels.

build_and_run.py contains the configuration used at RRZE's testcluster and UR's qpace4, Analysis.ipynb contains the analysis script and results. Raw data from measurements (122MB) will be attached to next OSACA release.

For now, find the raw data here: https://hawo.net/~sijuhamm/d/UPIhBOtz/validation-data.tar.gz

The analysis report can be viewed at https://nbviewer.jupyter.org/github/RRZE-HPC/OSACA/blob/validation/validation/Analysis.ipynb

Quite a few changes on OSACA included:

Feature: register change tracking via semantic understanding of operations
Feature: recording LCD latency along path and exposing this to frontend
Feature: support for memory reference aliases
Feature: store throughput scaling (similar to load throughput scaling)
Fix: model importer works with latest uops.info export
Fix: immediate type tracking on ARM now preserves type in internal representaion
Removed unused KerncraftAPI

validation/kernels/striad.c

@@ -0,0 +1,80 @@
+#ifdef MAIN
+#include <stdio.h>
+#include <stdlib.h>
+#include <likwid.h>
+#endif
+
+#define DTYPE double
+
+void dummy(void *);
+
+void kernel(DTYPE* a, DTYPE* b, DTYPE* c, DTYPE* d, const int repeat, const int cur_elements)
+#ifndef MAIN
+{
+    for(int r=0; r < repeat; r++) {
+        for(int i=0; i<cur_elements; i++) {
+            a[i] = b[i] + c[i] * d[i];
+        }
+        dummy((void*)a);
+    }
+}
+#else
+;
+int main(int argc, char *argv[]) {
+    if(argc < 2) {
+        printf("Usage: %s (repeat elements)...\n", argv[0]);
+        return 1;
+    }
+    const int tests = (argc-1) / 2;
+    int repeats[tests];
+    int elements[tests];
+    int maxelements = 0;
+    for(int t=0; t<tests; t++) {
+        repeats[t] = atoi(argv[1+t*2]);
+        elements[t] = atoi(argv[2+t*2]);
+        if(maxelements < elements[t]) {
+            maxelements = elements[t];
+        }
+    }
+    printf("kernel: triad\n");
+    printf("elementsize: %lu\n", sizeof(DTYPE));
+    
+    //SETUP
+    DTYPE* a = malloc(maxelements*sizeof(DTYPE));
+    DTYPE* b = malloc(maxelements*sizeof(DTYPE));
+    DTYPE* c = malloc(maxelements*sizeof(DTYPE));
+    DTYPE* d = malloc(maxelements*sizeof(DTYPE));
+    for(int i=0; i<maxelements; i++) {
+        a[i] = i;
+        b[i] = maxelements-i;
+        c[i] = 12.34;
+        d[i] = 12.34;
+    }
+
+    likwid_markerInit();
+
+    char cur_region_name[128];
+    for(int t=0; t<tests; t++) {
+        const int cur_elements = elements[t];
+        const int cur_repeats = repeats[t];
+        sprintf(cur_region_name, "triad_%i_%i", cur_repeats, cur_elements);
+        likwid_markerRegisterRegion(cur_region_name);
+        printf("%s:iterations: %i\n", cur_region_name, cur_elements);
+        printf("%s:repetitions: %i\n", cur_region_name, cur_repeats);
+
+        for(int warmup = 1; warmup >= 0; --warmup) {
+            int repeat = 2;
+            if(warmup == 0) {
+                repeat = cur_repeats;
+                likwid_markerStartRegion(cur_region_name);
+            }
+
+            kernel(a, b, c, d, repeat, cur_elements);
+        }
+        likwid_markerStopRegion(cur_region_name);
+    }
+    likwid_markerClose();
+    free(a);
+    return 0;
+}
+#endif