Changes in libcfa/src/device/cpu.cfa [45fde9f:cf85f96]

File:

• libcfa/src/device/cpu.cfa (modified) (6 diffs)

libcfa/src/device/cpu.cfa

@@ -256 +256 @@
 }
 
+struct llc_map_t {
+        raw_cache_instance * raw;
+        unsigned count;
+        unsigned start;
+};
+
 // returns an allocate list of all the different distinct last level caches
-static [*idx_range_t, size_t cnt] distinct_llcs(unsigned cpus, unsigned llc_idx, raw_cache_instance ** raw) {
+static [*llc_map_t, size_t cnt] distinct_llcs(unsigned cpus, unsigned llc_idx, raw_cache_instance ** raw) {
         // Allocate at least one element
-        idx_range_t * ranges = alloc();
+        llc_map_t* ranges = alloc();
         size_t range_cnt = 1;
 
         // Initialize with element 0
-        *ranges = raw[0][llc_idx].range;
+        ranges->raw = &raw[0][llc_idx];
+        ranges->count = 0;
+        ranges->start = -1u;
 
         // Go over all other cpus
         CPU_LOOP: for(i; 1~cpus) {
                 // Check if the range is already there
-                idx_range_t candidate = raw[i][llc_idx].range;
+                raw_cache_instance * candidate = &raw[i][llc_idx];
                 for(j; range_cnt) {
-                        idx_range_t exist = ranges[j];
+                        llc_map_t & exist = ranges[j];
                         // If the range is already there just jump to the next cpu
-                        if(0 == strcmp(candidate, exist)) continue CPU_LOOP;
+                        if(0 == strcmp(candidate->range, exist.raw->range)) continue CPU_LOOP;
                 }
 
                 // The range wasn't there, added to the list
                 ranges = alloc(range_cnt + 1, ranges`realloc);
-                ranges[range_cnt] = candidate;
+                ranges[range_cnt].raw = candidate;
+                ranges[range_cnt].count = 0;
+                ranges[range_cnt].start = -1u;
                 range_cnt++;
         }
@@ -287 +297 @@
 struct cpu_pairing_t {
         unsigned cpu;
-        unsigned llc_id;
+        unsigned id;
 };
 
 int ?<?( cpu_pairing_t lhs, cpu_pairing_t rhs ) {
-        return lhs.llc_id < rhs.llc_id;
-}
-
-static [[]cpu_pairing_t] get_cpu_pairings(unsigned cpus, raw_cache_instance ** raw, idx_range_t * maps, size_t map_cnt) {
+        return lhs.id < rhs.id;
+}
+
+static [[]cpu_pairing_t] get_cpu_pairings(unsigned cpus, raw_cache_instance ** raw, llc_map_t * maps, size_t map_cnt) {
         cpu_pairing_t * pairings = alloc(cpus);
 
@@ -301 +311 @@
                 idx_range_t want = raw[i][0].range;
                 MAP_LOOP: for(j; map_cnt) {
-                        if(0 != strcmp(want, maps[j])) continue MAP_LOOP;
-
-                        pairings[i].llc_id = j;
+                        if(0 != strcmp(want, maps[j].raw->range)) continue MAP_LOOP;
+
+                        pairings[i].id = j;
                         continue CPU_LOOP;
                 }
@@ -312 +322 @@
         return pairings;
 }
+
+#include <fstream.hfa>
 
 extern "C" {
@@ -336 +348 @@
 
                 // Find number of distinct cache instances
-                idx_range_t * maps;
+                llc_map_t * maps;
                 size_t map_cnt;
                 [maps, map_cnt] =  distinct_llcs(cpus, cache_levels - llc, raw);
 
                 #if defined(__CFA_WITH_VERIFY__)
+                // Verify that the caches cover the all the cpus
                 {
-                        unsigned width = 0;
+                        unsigned width1 = 0;
+                        unsigned width2 = 0;
                         for(i; map_cnt) {
                                 const char * _;
-                                width += read_width(maps[i], strlen(maps[i]), &_);
+                                width1 += read_width(maps[i].raw->range, strlen(maps[i].raw->range), &_);
+                                width2 += maps[i].raw->width;
                         }
-                        verify(width == cpus);
+                        verify(width1 == cpus);
+                        verify(width2 == cpus);
                 }
                 #endif
@@ -357 +373 @@
                 qsort(pairings, cpus);
 
-                unsigned llc_width = raw[0][cache_levels - llc].width;
-
-                // From the mappins build the actual cpu map we want
+                {
+                        unsigned it = 0;
+                        for(i; cpus) {
+                                unsigned llc_id = pairings[i].id;
+                                if(maps[llc_id].start == -1u) {
+                                        maps[llc_id].start = it;
+                                        it += maps[llc_id].raw->width;
+                                        /* paranoid */ verify(maps[llc_id].start < it);
+                                        /* paranoid */ verify(it != -1u);
+                                }
+                        }
+                        /* paranoid */ verify(it == cpus);
+                }
+
+                // From the mappings build the actual cpu map we want
                 struct cpu_map_entry_t * entries = alloc(cpus);
                 for(i; cpus) { entries[i].count = 0; }
                 for(i; cpus) {
+                        /* paranoid */ verify(pairings[i].id < map_cnt);
                         unsigned c = pairings[i].cpu;
-                        entries[c].start = pairings[i].llc_id * llc_width;
-                        entries[c].count = llc_width;
+                        unsigned llc_id = pairings[i].id;
+                        unsigned width = maps[llc_id].raw->width;
+                        unsigned start = maps[llc_id].start;
+                        unsigned self  = start + (maps[llc_id].count++);
+                        entries[c].count = width;
+                        entries[c].start = start;
+                        entries[c].self  = self;
                 }