Changeset e319fc5

Timestamp:

Jun 19, 2021, 3:53:18 PM (3 years ago)

Author:

Peter A. Buhr <pabuhr@…>

Branches:

ADT, ast-experimental, enum, forall-pointer-decay, jacob/cs343-translation, master, new-ast-unique-expr, pthread-emulation, qualifiedEnum

Children:

15f769c

Parents:

6992f95 (diff), c7d8696a (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc

Files:

• libcfa/src/concurrency/kernel.cfa (modified) (2 diffs)
• libcfa/src/concurrency/kernel.hfa (modified) (1 diff)
• libcfa/src/concurrency/kernel/startup.cfa (modified) (1 diff)
• libcfa/src/concurrency/ready_queue.cfa (modified) (5 diffs)
• libcfa/src/containers/array.hfa (modified) (3 diffs)
• libcfa/src/device/cpu.cfa (modified) (6 diffs)
• libcfa/src/device/cpu.hfa (modified) (1 diff)
• src/AST/Convert.cpp (modified) (1 diff)
• src/AST/Decl.cpp (modified) (1 diff)
• src/AST/Decl.hpp (modified) (1 diff)
• src/CodeGen/CodeGenerator.cc (modified) (1 diff)
• src/CodeGen/CodeGenerator.h (modified) (1 diff)
• src/Common/PassVisitor.h (modified) (3 diffs)
• src/Common/PassVisitor.impl.h (modified) (4 diffs)
• src/Parser/DeclarationNode.cc (modified) (1 diff)
• src/Parser/ExpressionNode.cc (modified) (1 diff)
• src/Parser/ParseNode.h (modified) (1 diff)
• src/Parser/TypedefTable.cc (modified) (2 diffs)
• src/Parser/parser.yy (modified) (5 diffs)
• src/SymTab/Indexer.cc (modified) (3 diffs)
• src/SymTab/Indexer.h (modified) (2 diffs)
• src/SymTab/Validate.cc (modified) (14 diffs)
• src/SynTree/Declaration.h (modified) (1 diff)
• src/SynTree/Expression.h (modified) (1 diff)
• src/SynTree/Mutator.h (modified) (1 diff)
• src/SynTree/SynTree.h (modified) (1 diff)
• src/SynTree/TypeDecl.cc (modified) (1 diff)
• src/SynTree/TypeExpr.cc (modified) (1 diff)
• src/SynTree/Visitor.h (modified) (1 diff)
• tests/array-container/.expect/array-md-sbscr-cases.x86.txt (added)
• tests/array-container/.expect/language-dim-mismatch.txt (added)
• tests/array-container/array-basic.cfa (modified) (3 diffs)
• tests/array-container/array-md-sbscr-cases.cfa (modified) (2 diffs)
• tests/array-container/language-dim-mismatch.cfa (added)
• tests/device/cpu.cfa (modified) (2 diffs)

libcfa/src/concurrency/kernel.cfa

@@ -280 +280 @@
 
                                 // Spin a little on I/O, just in case
-                                        for(5) {
+                                for(5) {
                                         __maybe_io_drain( this );
                                         readyThread = pop_fast( this->cltr );
@@ -287 +287 @@
 
                                 // no luck, try stealing a few times
-                                        for(5) {
+                                for(5) {
                                         if( __maybe_io_drain( this ) ) {
                                                 readyThread = pop_fast( this->cltr );

libcfa/src/concurrency/kernel.hfa

@@ -66 +66 @@
                 unsigned id;
                 unsigned target;
+                unsigned last;
                 unsigned long long int cutoff;
         } rdq;

libcfa/src/concurrency/kernel/startup.cfa

@@ -541 +541 @@
         this.rdq.id  = -1u;
         this.rdq.target = -1u;
+        this.rdq.last = -1u;
         this.rdq.cutoff = 0ull;
         do_terminate = false;

libcfa/src/concurrency/ready_queue.cfa

@@ -24 +24 @@
 
 #include "bits/defs.hfa"
+#include "device/cpu.hfa"
 #include "kernel_private.hfa"
 
@@ -47 +48 @@
 #endif
 
-#if   defined(USE_RELAXED_FIFO)
+#if   defined(USE_CPU_WORK_STEALING)
+        #define READYQ_SHARD_FACTOR 2
+#elif defined(USE_RELAXED_FIFO)
         #define BIAS 4
         #define READYQ_SHARD_FACTOR 4
@@ -215 +218 @@
 //=======================================================================
 void ?{}(__ready_queue_t & this) with (this) {
-        lanes.data  = 0p;
-        lanes.tscs  = 0p;
-        lanes.count = 0;
+        #if defined(USE_CPU_WORK_STEALING)
+                lanes.count = cpu_info.hthrd_count * READYQ_SHARD_FACTOR;
+                lanes.data = alloc( lanes.count );
+                lanes.tscs = alloc( lanes.count );
+
+                for( idx; (size_t)lanes.count ) {
+                        (lanes.data[idx]){};
+                        lanes.tscs[idx].tv = rdtscl();
+                }
+        #else
+                lanes.data  = 0p;
+                lanes.tscs  = 0p;
+                lanes.count = 0;
+        #endif
 }
 
 void ^?{}(__ready_queue_t & this) with (this) {
-        verify( SEQUENTIAL_SHARD == lanes.count );
+        #if !defined(USE_CPU_WORK_STEALING)
+                verify( SEQUENTIAL_SHARD == lanes.count );
+        #endif
+
         free(lanes.data);
         free(lanes.tscs);
@@ -227 +244 @@
 
 //-----------------------------------------------------------------------
+#if defined(USE_CPU_WORK_STEALING)
+        __attribute__((hot)) void push(struct cluster * cltr, struct $thread * thrd, bool push_local) with (cltr->ready_queue) {
+                __cfadbg_print_safe(ready_queue, "Kernel : Pushing %p on cluster %p\n", thrd, cltr);
+
+                processor * const proc = kernelTLS().this_processor;
+                const bool external = !push_local || (!proc) || (cltr != proc->cltr);
+
+                const int cpu = __kernel_getcpu();
+                /* paranoid */ verify(cpu >= 0);
+                /* paranoid */ verify(cpu < cpu_info.hthrd_count);
+                /* paranoid */ verify(cpu * READYQ_SHARD_FACTOR < lanes.count);
+
+                const cpu_map_entry_t & map = cpu_info.llc_map[cpu];
+                /* paranoid */ verify(map.start * READYQ_SHARD_FACTOR < lanes.count);
+                /* paranoid */ verify(map.self * READYQ_SHARD_FACTOR < lanes.count);
+                /* paranoid */ verifyf((map.start + map.count) * READYQ_SHARD_FACTOR <= lanes.count, "have %u lanes but map can go up to %u", lanes.count, (map.start + map.count) * READYQ_SHARD_FACTOR);
+
+                const int start = map.self * READYQ_SHARD_FACTOR;
+                unsigned i;
+                do {
+                        unsigned r;
+                        if(unlikely(external)) { r = __tls_rand(); }
+                        else { r = proc->rdq.its++; }
+                        i = start + (r % READYQ_SHARD_FACTOR);
+                        // If we can't lock it retry
+                } while( !__atomic_try_acquire( &lanes.data[i].lock ) );
+
+                // Actually push it
+                push(lanes.data[i], thrd);
+
+                // Unlock and return
+                __atomic_unlock( &lanes.data[i].lock );
+
+                #if !defined(__CFA_NO_STATISTICS__)
+                        if(unlikely(external)) __atomic_fetch_add(&cltr->stats->ready.push.extrn.success, 1, __ATOMIC_RELAXED);
+                        else __tls_stats()->ready.push.local.success++;
+                #endif
+
+                __cfadbg_print_safe(ready_queue, "Kernel : Pushed %p on cluster %p (idx: %u, mask %llu, first %d)\n", thrd, cltr, i, used.mask[0], lane_first);
+
+        }
+
+        // Pop from the ready queue from a given cluster
+        __attribute__((hot)) $thread * pop_fast(struct cluster * cltr) with (cltr->ready_queue) {
+                /* paranoid */ verify( lanes.count > 0 );
+                /* paranoid */ verify( kernelTLS().this_processor );
+
+                const int cpu = __kernel_getcpu();
+                /* paranoid */ verify(cpu >= 0);
+                /* paranoid */ verify(cpu < cpu_info.hthrd_count);
+                /* paranoid */ verify(cpu * READYQ_SHARD_FACTOR < lanes.count);
+
+                const cpu_map_entry_t & map = cpu_info.llc_map[cpu];
+                /* paranoid */ verify(map.start * READYQ_SHARD_FACTOR < lanes.count);
+                /* paranoid */ verify(map.self * READYQ_SHARD_FACTOR < lanes.count);
+                /* paranoid */ verifyf((map.start + map.count) * READYQ_SHARD_FACTOR <= lanes.count, "have %u lanes but map can go up to %u", lanes.count, (map.start + map.count) * READYQ_SHARD_FACTOR);
+
+                processor * const proc = kernelTLS().this_processor;
+                const int start = map.self * READYQ_SHARD_FACTOR;
+
+                // Did we already have a help target
+                if(proc->rdq.target == -1u) {
+                        // if We don't have a
+                        unsigned long long min = ts(lanes.data[start]);
+                        for(i; READYQ_SHARD_FACTOR) {
+                                unsigned long long tsc = ts(lanes.data[start + i]);
+                                if(tsc < min) min = tsc;
+                        }
+                        proc->rdq.cutoff = min;
+                        proc->rdq.target = (map.start * READYQ_SHARD_FACTOR) + (__tls_rand() % (map.count* READYQ_SHARD_FACTOR));
+                }
+                else {
+                        const unsigned long long bias = 0; //2_500_000_000;
+                        const unsigned long long cutoff = proc->rdq.cutoff > bias ? proc->rdq.cutoff - bias : proc->rdq.cutoff;
+                        {
+                                unsigned target = proc->rdq.target;
+                                proc->rdq.target = -1u;
+                                if(lanes.tscs[target].tv < cutoff && ts(lanes.data[target]) < cutoff) {
+                                        $thread * t = try_pop(cltr, target __STATS(, __tls_stats()->ready.pop.help));
+                                        proc->rdq.last = target;
+                                        if(t) return t;
+                                }
+                        }
+
+                        unsigned last = proc->rdq.last;
+                        if(last != -1u && lanes.tscs[last].tv < cutoff && ts(lanes.data[last]) < cutoff) {
+                                $thread * t = try_pop(cltr, last __STATS(, __tls_stats()->ready.pop.help));
+                                if(t) return t;
+                        }
+                        else {
+                                proc->rdq.last = -1u;
+                        }
+                }
+
+                for(READYQ_SHARD_FACTOR) {
+                        unsigned i = start + (proc->rdq.itr++ % READYQ_SHARD_FACTOR);
+                        if($thread * t = try_pop(cltr, i __STATS(, __tls_stats()->ready.pop.local))) return t;
+                }
+
+                // All lanes where empty return 0p
+                return 0p;
+        }
+
+        __attribute__((hot)) struct $thread * pop_slow(struct cluster * cltr) with (cltr->ready_queue) {
+                processor * const proc = kernelTLS().this_processor;
+                unsigned last = proc->rdq.last;
+
+                unsigned i = __tls_rand() % lanes.count;
+                return try_pop(cltr, i __STATS(, __tls_stats()->ready.pop.steal));
+        }
+        __attribute__((hot)) struct $thread * pop_search(struct cluster * cltr) {
+                return search(cltr);
+        }
+#endif
 #if defined(USE_RELAXED_FIFO)
         //-----------------------------------------------------------------------
@@ -580 +711 @@
 }
 
-// Grow the ready queue
-void ready_queue_grow(struct cluster * cltr) {
-        size_t ncount;
-        int target = cltr->procs.total;
-
-        /* paranoid */ verify( ready_mutate_islocked() );
-        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue\n");
-
-        // Make sure that everything is consistent
-        /* paranoid */ check( cltr->ready_queue );
-
-        // grow the ready queue
-        with( cltr->ready_queue ) {
-                // Find new count
-                // Make sure we always have atleast 1 list
-                if(target >= 2) {
-                        ncount = target * READYQ_SHARD_FACTOR;
-                } else {
-                        ncount = SEQUENTIAL_SHARD;
-                }
-
-                // Allocate new array (uses realloc and memcpies the data)
-                lanes.data = alloc( ncount, lanes.data`realloc );
-
-                // Fix the moved data
-                for( idx; (size_t)lanes.count ) {
-                        fix(lanes.data[idx]);
-                }
-
-                // Construct new data
-                for( idx; (size_t)lanes.count ~ ncount) {
-                        (lanes.data[idx]){};
-                }
-
-                // Update original
-                lanes.count = ncount;
-        }
-
-        fix_times(cltr);
-
-        reassign_cltr_id(cltr);
-
-        // Make sure that everything is consistent
-        /* paranoid */ check( cltr->ready_queue );
-
-        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue done\n");
-
-        /* paranoid */ verify( ready_mutate_islocked() );
-}
-
-// Shrink the ready queue
-void ready_queue_shrink(struct cluster * cltr) {
-        /* paranoid */ verify( ready_mutate_islocked() );
-        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue\n");
-
-        // Make sure that everything is consistent
-        /* paranoid */ check( cltr->ready_queue );
-
-        int target = cltr->procs.total;
-
-        with( cltr->ready_queue ) {
-                // Remember old count
-                size_t ocount = lanes.count;
-
-                // Find new count
-                // Make sure we always have atleast 1 list
-                lanes.count = target >= 2 ? target * READYQ_SHARD_FACTOR: SEQUENTIAL_SHARD;
-                /* paranoid */ verify( ocount >= lanes.count );
-                /* paranoid */ verify( lanes.count == target * READYQ_SHARD_FACTOR || target < 2 );
-
-                // for printing count the number of displaced threads
-                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
-                        __attribute__((unused)) size_t displaced = 0;
-                #endif
-
-                // redistribute old data
-                for( idx; (size_t)lanes.count ~ ocount) {
-                        // Lock is not strictly needed but makes checking invariants much easier
-                        __attribute__((unused)) bool locked = __atomic_try_acquire(&lanes.data[idx].lock);
-                        verify(locked);
-
-                        // As long as we can pop from this lane to push the threads somewhere else in the queue
-                        while(!is_empty(lanes.data[idx])) {
-                                struct $thread * thrd;
-                                unsigned long long _;
-                                [thrd, _] = pop(lanes.data[idx]);
-
-                                push(cltr, thrd, true);
-
-                                // for printing count the number of displaced threads
-                                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
-                                        displaced++;
-                                #endif
-                        }
-
-                        // Unlock the lane
-                        __atomic_unlock(&lanes.data[idx].lock);
-
-                        // TODO print the queue statistics here
-
-                        ^(lanes.data[idx]){};
-                }
-
-                __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue displaced %zu threads\n", displaced);
-
-                // Allocate new array (uses realloc and memcpies the data)
-                lanes.data = alloc( lanes.count, lanes.data`realloc );
-
-                // Fix the moved data
-                for( idx; (size_t)lanes.count ) {
-                        fix(lanes.data[idx]);
-                }
-        }
-
-        fix_times(cltr);
-
-        reassign_cltr_id(cltr);
-
-        // Make sure that everything is consistent
-        /* paranoid */ check( cltr->ready_queue );
-
-        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue done\n");
-        /* paranoid */ verify( ready_mutate_islocked() );
-}
+#if defined(USE_CPU_WORK_STEALING)
+        // ready_queue size is fixed in this case
+        void ready_queue_grow(struct cluster * cltr) {}
+        void ready_queue_shrink(struct cluster * cltr) {}
+#else
+        // Grow the ready queue
+        void ready_queue_grow(struct cluster * cltr) {
+                size_t ncount;
+                int target = cltr->procs.total;
+
+                /* paranoid */ verify( ready_mutate_islocked() );
+                __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue\n");
+
+                // Make sure that everything is consistent
+                /* paranoid */ check( cltr->ready_queue );
+
+                // grow the ready queue
+                with( cltr->ready_queue ) {
+                        // Find new count
+                        // Make sure we always have atleast 1 list
+                        if(target >= 2) {
+                                ncount = target * READYQ_SHARD_FACTOR;
+                        } else {
+                                ncount = SEQUENTIAL_SHARD;
+                        }
+
+                        // Allocate new array (uses realloc and memcpies the data)
+                        lanes.data = alloc( ncount, lanes.data`realloc );
+
+                        // Fix the moved data
+                        for( idx; (size_t)lanes.count ) {
+                                fix(lanes.data[idx]);
+                        }
+
+                        // Construct new data
+                        for( idx; (size_t)lanes.count ~ ncount) {
+                                (lanes.data[idx]){};
+                        }
+
+                        // Update original
+                        lanes.count = ncount;
+                }
+
+                fix_times(cltr);
+
+                reassign_cltr_id(cltr);
+
+                // Make sure that everything is consistent
+                /* paranoid */ check( cltr->ready_queue );
+
+                __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue done\n");
+
+                /* paranoid */ verify( ready_mutate_islocked() );
+        }
+
+        // Shrink the ready queue
+        void ready_queue_shrink(struct cluster * cltr) {
+                /* paranoid */ verify( ready_mutate_islocked() );
+                __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue\n");
+
+                // Make sure that everything is consistent
+                /* paranoid */ check( cltr->ready_queue );
+
+                int target = cltr->procs.total;
+
+                with( cltr->ready_queue ) {
+                        // Remember old count
+                        size_t ocount = lanes.count;
+
+                        // Find new count
+                        // Make sure we always have atleast 1 list
+                        lanes.count = target >= 2 ? target * READYQ_SHARD_FACTOR: SEQUENTIAL_SHARD;
+                        /* paranoid */ verify( ocount >= lanes.count );
+                        /* paranoid */ verify( lanes.count == target * READYQ_SHARD_FACTOR || target < 2 );
+
+                        // for printing count the number of displaced threads
+                        #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
+                                __attribute__((unused)) size_t displaced = 0;
+                        #endif
+
+                        // redistribute old data
+                        for( idx; (size_t)lanes.count ~ ocount) {
+                                // Lock is not strictly needed but makes checking invariants much easier
+                                __attribute__((unused)) bool locked = __atomic_try_acquire(&lanes.data[idx].lock);
+                                verify(locked);
+
+                                // As long as we can pop from this lane to push the threads somewhere else in the queue
+                                while(!is_empty(lanes.data[idx])) {
+                                        struct $thread * thrd;
+                                        unsigned long long _;
+                                        [thrd, _] = pop(lanes.data[idx]);
+
+                                        push(cltr, thrd, true);
+
+                                        // for printing count the number of displaced threads
+                                        #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
+                                                displaced++;
+                                        #endif
+                                }
+
+                                // Unlock the lane
+                                __atomic_unlock(&lanes.data[idx].lock);
+
+                                // TODO print the queue statistics here
+
+                                ^(lanes.data[idx]){};
+                        }
+
+                        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue displaced %zu threads\n", displaced);
+
+                        // Allocate new array (uses realloc and memcpies the data)
+                        lanes.data = alloc( lanes.count, lanes.data`realloc );
+
+                        // Fix the moved data
+                        for( idx; (size_t)lanes.count ) {
+                                fix(lanes.data[idx]);
+                        }
+                }
+
+                fix_times(cltr);
+
+                reassign_cltr_id(cltr);
+
+                // Make sure that everything is consistent
+                /* paranoid */ check( cltr->ready_queue );
+
+                __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue done\n");
+                /* paranoid */ verify( ready_mutate_islocked() );
+        }
+#endif
 
 #if !defined(__CFA_NO_STATISTICS__)

libcfa/src/containers/array.hfa

@@ -1 +1 @@
 
 
-// a type whose size is n
-#define Z(n) char[n]
-
-// the inverse of Z(-)
-#define z(N) sizeof(N)
-
-forall( T & ) struct tag {};
+forall( __CFA_tysys_id_only_X & ) struct tag {};
 #define ttag(T) ((tag(T)){})
-#define ztag(n) ttag(Z(n))
+#define ztag(n) ttag(n)
 
 
@@ -18 +12 @@
 forall( [N], S & | sized(S), Timmed &, Tbase & ) {
     struct arpk {
-        S strides[z(N)];
+        S strides[N];
     };
 
@@ -56 +50 @@
 
     static inline size_t ?`len( arpk(N, S, Timmed, Tbase) & a ) {
-        return z(N);
+        return N;
     }
 
     // workaround #226 (and array relevance thereof demonstrated in mike102/otype-slow-ndims.cfa)
     static inline void ?{}( arpk(N, S, Timmed, Tbase) & this ) {
-        void ?{}( S (&inner)[z(N)] ) {}
+        void ?{}( S (&inner)[N] ) {}
         ?{}(this.strides);
     }
     static inline void ^?{}( arpk(N, S, Timmed, Tbase) & this ) {
-        void ^?{}( S (&inner)[z(N)] ) {}
+        void ^?{}( S (&inner)[N] ) {}
         ^?{}(this.strides);
     }

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;
                 }
 

libcfa/src/device/cpu.hfa

@@ -17 +17 @@
 
 struct cpu_map_entry_t {
+        unsigned self;
         unsigned start;
         unsigned count;

src/AST/Convert.cpp

@@ -2415 +2415 @@
         }
 
+        virtual void visit( const DimensionExpr * old ) override final {
+                // DimensionExpr gets desugared away in Validate.
+                // As long as new-AST passes don't use it, this cheap-cheerful error
+                // detection helps ensure that these occurrences have been compiled
+                // away, as expected.  To move the DimensionExpr boundary downstream
+                // or move the new-AST translation boundary upstream, implement
+                // DimensionExpr in the new AST and implement a conversion.
+                (void) old;
+                assert(false && "DimensionExpr should not be present at new-AST boundary");
+        }
+
         virtual void visit( const AsmExpr * old ) override final {
                 this->node = visitBaseExpr( old,

src/AST/Decl.cpp

@@ -78 +78 @@
 
 const char * TypeDecl::typeString() const {
-        static const char * kindNames[] = { "sized data type", "sized data type", "sized object type", "sized function type", "sized tuple type", "sized array length type" };
+        static const char * kindNames[] = { "sized data type", "sized data type", "sized object type", "sized function type", "sized tuple type", "sized length value" };
         static_assert( sizeof(kindNames) / sizeof(kindNames[0]) == TypeDecl::NUMBER_OF_KINDS, "typeString: kindNames is out of sync." );
         assertf( kind < TypeDecl::NUMBER_OF_KINDS, "TypeDecl kind is out of bounds." );

src/AST/Decl.hpp

@@ -175 +175 @@
 class TypeDecl final : public NamedTypeDecl {
   public:
-        enum Kind { Dtype, DStype, Otype, Ftype, Ttype, ALtype, NUMBER_OF_KINDS };
+        enum Kind { Dtype, DStype, Otype, Ftype, Ttype, Dimension, NUMBER_OF_KINDS };
 
         Kind kind;

src/CodeGen/CodeGenerator.cc

@@ -589 +589 @@
                         output << nameExpr->get_name();
                 } // if
+        }
+
+        void CodeGenerator::postvisit( DimensionExpr * dimensionExpr ) {
+                extension( dimensionExpr );
+                output << "/*non-type*/" << dimensionExpr->get_name();
         }
 

src/CodeGen/CodeGenerator.h

@@ -92 +92 @@
                 void postvisit( TupleIndexExpr * tupleExpr );
                 void postvisit( TypeExpr *typeExpr );
+                void postvisit( DimensionExpr *dimensionExpr );
                 void postvisit( AsmExpr * );
                 void postvisit( StmtExpr * );

src/Common/PassVisitor.h

@@ -167 +167 @@
         virtual void visit( TypeExpr * typeExpr ) override final;
         virtual void visit( const TypeExpr * typeExpr ) override final;
+        virtual void visit( DimensionExpr * dimensionExpr ) override final;
+        virtual void visit( const DimensionExpr * dimensionExpr ) override final;
         virtual void visit( AsmExpr * asmExpr ) override final;
         virtual void visit( const AsmExpr * asmExpr ) override final;
@@ -309 +311 @@
         virtual Expression * mutate( CommaExpr * commaExpr ) override final;
         virtual Expression * mutate( TypeExpr * typeExpr ) override final;
+        virtual Expression * mutate( DimensionExpr * dimensionExpr ) override final;
         virtual Expression * mutate( AsmExpr * asmExpr ) override final;
         virtual Expression * mutate( ImplicitCopyCtorExpr * impCpCtorExpr ) override final;
@@ -542 +545 @@
 class WithIndexer {
 protected:
-        WithIndexer() {}
+        WithIndexer( bool trackIdentifiers = true ) : indexer(trackIdentifiers) {}
         ~WithIndexer() {}
 

src/Common/PassVisitor.impl.h

@@ -2519 +2519 @@
 
 //--------------------------------------------------------------------------
+// DimensionExpr
+template< typename pass_type >
+void PassVisitor< pass_type >::visit( DimensionExpr * node ) {
+        VISIT_START( node );
+
+        indexerScopedAccept( node->result, *this );
+
+        VISIT_END( node );
+}
+
+template< typename pass_type >
+void PassVisitor< pass_type >::visit( const DimensionExpr * node ) {
+        VISIT_START( node );
+
+        indexerScopedAccept( node->result, *this );
+
+        VISIT_END( node );
+}
+
+template< typename pass_type >
+Expression * PassVisitor< pass_type >::mutate( DimensionExpr * node ) {
+        MUTATE_START( node );
+
+        indexerScopedMutate( node->env   , *this );
+        indexerScopedMutate( node->result, *this );
+
+        MUTATE_END( Expression, node );
+}
+
+//--------------------------------------------------------------------------
 // AsmExpr
 template< typename pass_type >
@@ -3157 +3187 @@
 
         maybeAccept_impl( node->forall, *this );
-        // xxx - should PointerType visit/mutate dimension?
+        maybeAccept_impl( node->dimension, *this );
         maybeAccept_impl( node->base, *this );
 
@@ -3168 +3198 @@
 
         maybeAccept_impl( node->forall, *this );
-        // xxx - should PointerType visit/mutate dimension?
+        maybeAccept_impl( node->dimension, *this );
         maybeAccept_impl( node->base, *this );
 
@@ -3179 +3209 @@
 
         maybeMutate_impl( node->forall, *this );
-        // xxx - should PointerType visit/mutate dimension?
+        maybeMutate_impl( node->dimension, *this );
         maybeMutate_impl( node->base, *this );
 

src/Parser/DeclarationNode.cc

@@ -1076 +1076 @@
         if ( variable.tyClass != TypeDecl::NUMBER_OF_KINDS ) {
                 // otype is internally converted to dtype + otype parameters
-                static const TypeDecl::Kind kindMap[] = { TypeDecl::Dtype, TypeDecl::DStype, TypeDecl::Dtype, TypeDecl::Ftype, TypeDecl::Ttype, TypeDecl::Dtype };
+                static const TypeDecl::Kind kindMap[] = { TypeDecl::Dtype, TypeDecl::DStype, TypeDecl::Dtype, TypeDecl::Ftype, TypeDecl::Ttype, TypeDecl::Dimension };
                 static_assert( sizeof(kindMap) / sizeof(kindMap[0]) == TypeDecl::NUMBER_OF_KINDS, "DeclarationNode::build: kindMap is out of sync." );
                 assertf( variable.tyClass < sizeof(kindMap)/sizeof(kindMap[0]), "Variable's tyClass is out of bounds." );
-                TypeDecl * ret = new TypeDecl( *name, Type::StorageClasses(), nullptr, kindMap[ variable.tyClass ], variable.tyClass == TypeDecl::Otype || variable.tyClass == TypeDecl::ALtype, variable.initializer ? variable.initializer->buildType() : nullptr );
+                TypeDecl * ret = new TypeDecl( *name, Type::StorageClasses(), nullptr, kindMap[ variable.tyClass ], variable.tyClass == TypeDecl::Otype, variable.initializer ? variable.initializer->buildType() : nullptr );
                 buildList( variable.assertions, ret->get_assertions() );
                 return ret;

src/Parser/ExpressionNode.cc

@@ -509 +509 @@
 } // build_varref
 
+DimensionExpr * build_dimensionref( const string * name ) {
+        DimensionExpr * expr = new DimensionExpr( *name );
+        delete name;
+        return expr;
+} // build_varref
 // TODO: get rid of this and OperKinds and reuse code from OperatorTable
 static const char * OperName[] = {                                              // must harmonize with OperKinds

src/Parser/ParseNode.h

@@ -183 +183 @@
 
 NameExpr * build_varref( const std::string * name );
+DimensionExpr * build_dimensionref( const std::string * name );
 
 Expression * build_cast( DeclarationNode * decl_node, ExpressionNode * expr_node );

src/Parser/TypedefTable.cc

@@ -10 +10 @@
 // Created On       : Sat May 16 15:20:13 2015
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Mon Mar 15 20:56:47 2021
-// Update Count     : 260
+// Last Modified On : Wed May 19 08:30:14 2021
+// Update Count     : 262
 //
 
@@ -31 +31 @@
         switch ( kind ) {
           case IDENTIFIER: return "identifier";
+          case TYPEDIMname: return "typedim";
           case TYPEDEFname: return "typedef";
           case TYPEGENname: return "typegen";

src/Parser/parser.yy

@@ -10 +10 @@
 // Created On       : Sat Sep  1 20:22:55 2001
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Mon Apr 26 18:41:54 2021
-// Update Count     : 4990
+// Last Modified On : Wed May 19 14:20:36 2021
+// Update Count     : 5022
 //
 
@@ -287 +287 @@
 
 // names and constants: lexer differentiates between identifier and typedef names
-%token<tok> IDENTIFIER          QUOTED_IDENTIFIER       TYPEDEFname             TYPEGENname
+%token<tok> IDENTIFIER          QUOTED_IDENTIFIER       TYPEDIMname             TYPEDEFname             TYPEGENname
 %token<tok> TIMEOUT                     WOR                                     CATCH                   RECOVER                 CATCHRESUME             FIXUP           FINALLY         // CFA
 %token<tok> INTEGERconstant     CHARACTERconstant       STRINGliteral
@@ -586 +586 @@
         | quasi_keyword
                 { $$ = new ExpressionNode( build_varref( $1 ) ); }
+        | TYPEDIMname                                                                           // CFA, generic length argument
+                // { $$ = new ExpressionNode( new TypeExpr( maybeMoveBuildType( DeclarationNode::newFromTypedef( $1 ) ) ) ); }
+                // { $$ = new ExpressionNode( build_varref( $1 ) ); }
+                { $$ = new ExpressionNode( build_dimensionref( $1 ) ); }
         | tuple
         | '(' comma_expression ')'
@@ -2535 +2539 @@
         | '[' identifier_or_type_name ']'
                 {
-                        typedefTable.addToScope( *$2, TYPEDEFname, "9" );
-                        $$ = DeclarationNode::newTypeParam( TypeDecl::ALtype, $2 );
+                        typedefTable.addToScope( *$2, TYPEDIMname, "9" );
+                        $$ = DeclarationNode::newTypeParam( TypeDecl::Dimension, $2 );
                 }
         // | type_specifier identifier_parameter_declarator
@@ -2590 +2594 @@
                 { $$ = new ExpressionNode( new TypeExpr( maybeMoveBuildType( $1 ) ) ); }
         | assignment_expression
-                { SemanticError( yylloc, toString("Expression generic parameters are currently unimplemented: ", $1->build()) ); $$ = nullptr; }
         | type_list ',' type
                 { $$ = (ExpressionNode *)($1->set_last( new ExpressionNode( new TypeExpr( maybeMoveBuildType( $3 ) ) ) )); }
         | type_list ',' assignment_expression
-                { SemanticError( yylloc, toString("Expression generic parameters are currently unimplemented: ", $3->build()) ); $$ = nullptr; }
-                // { $$ = (ExpressionNode *)( $1->set_last( $3 )); }
+                { $$ = (ExpressionNode *)( $1->set_last( $3 )); }
         ;
 

src/SymTab/Indexer.cc

@@ -74 +74 @@
         }
 
-        Indexer::Indexer()
+        Indexer::Indexer( bool trackIdentifiers )
         : idTable(), typeTable(), structTable(), enumTable(), unionTable(), traitTable(),
-          prevScope(), scope( 0 ), repScope( 0 ) { ++* stats().count; }
+          prevScope(), scope( 0 ), repScope( 0 ), trackIdentifiers( trackIdentifiers ) { ++* stats().count; }
 
         Indexer::~Indexer() {
@@ -110 +110 @@
 
         void Indexer::lookupId( const std::string & id, std::list< IdData > &out ) const {
+                assert( trackIdentifiers );
+
                 ++* stats().lookup_calls;
                 if ( ! idTable ) return;
@@ -434 +436 @@
                         const Declaration * deleteStmt ) {
                 ++* stats().add_calls;
+                if ( ! trackIdentifiers ) return;
                 const std::string &name = decl->name;
                 if ( name == "" ) return;

src/SymTab/Indexer.h

@@ -31 +31 @@
         class Indexer : public std::enable_shared_from_this<SymTab::Indexer> {
         public:
-                explicit Indexer();
+                explicit Indexer( bool trackIdentifiers = true );
                 virtual ~Indexer();
 
@@ -180 +180 @@
                 /// returns true if there exists a declaration with C linkage and the given name with a different mangled name
                 bool hasIncompatibleCDecl( const std::string & id, const std::string & mangleName ) const;
+
+            bool trackIdentifiers;
         };
 } // namespace SymTab

src/SymTab/Validate.cc

@@ -105 +105 @@
 
         struct FixQualifiedTypes final : public WithIndexer {
+                FixQualifiedTypes() : WithIndexer(false) {}
                 Type * postmutate( QualifiedType * );
         };
@@ -174 +175 @@
         };
 
+        /// Does early resolution on the expressions that give enumeration constants their values
+        struct ResolveEnumInitializers final : public WithIndexer, public WithGuards, public WithVisitorRef<ResolveEnumInitializers>, public WithShortCircuiting {
+                ResolveEnumInitializers( const Indexer * indexer );
+                void postvisit( EnumDecl * enumDecl );
+
+          private:
+                const Indexer * local_indexer;
+
+        };
+
         /// Replaces array and function types in forall lists by appropriate pointer type and assigns each Object and Function declaration a unique ID.
         struct ForallPointerDecay_old final {
@@ -260 +271 @@
                 void previsit( StructInstType * inst );
                 void previsit( UnionInstType * inst );
+        };
+
+        /// desugar declarations and uses of dimension paramaters like [N],
+        /// from type-system managed values, to tunnneling via ordinary types,
+        /// as char[-] in and sizeof(-) out
+        struct TranslateDimensionGenericParameters : public WithIndexer, public WithGuards {
+                static void translateDimensions( std::list< Declaration * > &translationUnit );
+                TranslateDimensionGenericParameters();
+
+                bool nextVisitedNodeIsChildOfSUIT = false; // SUIT = Struct or Union -Inst Type
+                bool visitingChildOfSUIT = false;
+                void changeState_ChildOfSUIT( bool newVal );
+                void premutate( StructInstType * sit );
+                void premutate( UnionInstType * uit );
+                void premutate( BaseSyntaxNode * node );
+
+                TypeDecl * postmutate( TypeDecl * td );
+                Expression * postmutate( DimensionExpr * de );
+                Expression * postmutate( Expression * e );
         };
 
@@ -307 +337 @@
                 PassVisitor<EnumAndPointerDecay_old> epc;
                 PassVisitor<LinkReferenceToTypes_old> lrt( nullptr );
+                PassVisitor<ResolveEnumInitializers> rei( nullptr );
                 PassVisitor<ForallPointerDecay_old> fpd;
                 PassVisitor<CompoundLiteral> compoundliteral;
@@ -326 +357 @@
                         Stats::Heap::newPass("validate-B");
                         Stats::Time::BlockGuard guard("validate-B");
-                        Stats::Time::TimeBlock("Link Reference To Types", [&]() {
-                                acceptAll( translationUnit, lrt ); // must happen before autogen, because sized flag needs to propagate to generated functions
-                        });
-                        Stats::Time::TimeBlock("Fix Qualified Types", [&]() {
-                                mutateAll( translationUnit, fixQual ); // must happen after LinkReferenceToTypes_old, because aggregate members are accessed
-                        });
-                        Stats::Time::TimeBlock("Hoist Structs", [&]() {
-                                HoistStruct::hoistStruct( translationUnit ); // must happen after EliminateTypedef, so that aggregate typedefs occur in the correct order
-                        });
-                        Stats::Time::TimeBlock("Eliminate Typedefs", [&]() {
-                                EliminateTypedef::eliminateTypedef( translationUnit ); //
-                        });
+                        acceptAll( translationUnit, lrt ); // must happen before autogen, because sized flag needs to propagate to generated functions
+                        mutateAll( translationUnit, fixQual ); // must happen after LinkReferenceToTypes_old, because aggregate members are accessed
+                        HoistStruct::hoistStruct( translationUnit );
+                        EliminateTypedef::eliminateTypedef( translationUnit );
                 }
                 {
                         Stats::Heap::newPass("validate-C");
                         Stats::Time::BlockGuard guard("validate-C");
-                        acceptAll( translationUnit, genericParams );  // check as early as possible - can't happen before LinkReferenceToTypes_old
-                        ReturnChecker::checkFunctionReturns( translationUnit );
-                        InitTweak::fixReturnStatements( translationUnit ); // must happen before autogen
+                        Stats::Time::TimeBlock("Validate Generic Parameters", [&]() {
+                                acceptAll( translationUnit, genericParams );  // check as early as possible - can't happen before LinkReferenceToTypes_old; observed failing when attempted before eliminateTypedef
+                        });
+                        Stats::Time::TimeBlock("Translate Dimensions", [&]() {
+                                TranslateDimensionGenericParameters::translateDimensions( translationUnit );
+                        });
+                        Stats::Time::TimeBlock("Resolve Enum Initializers", [&]() {
+                                acceptAll( translationUnit, rei ); // must happen after translateDimensions because rei needs identifier lookup, which needs name mangling
+                        });
+                        Stats::Time::TimeBlock("Check Function Returns", [&]() {
+                                ReturnChecker::checkFunctionReturns( translationUnit );
+                        });
+                        Stats::Time::TimeBlock("Fix Return Statements", [&]() {
+                                InitTweak::fixReturnStatements( translationUnit ); // must happen before autogen
+                        });
                 }
                 {
@@ -644 +679 @@
         }
 
-        LinkReferenceToTypes_old::LinkReferenceToTypes_old( const Indexer * other_indexer ) {
+        LinkReferenceToTypes_old::LinkReferenceToTypes_old( const Indexer * other_indexer ) : WithIndexer( false ) {
                 if ( other_indexer ) {
                         local_indexer = other_indexer;
@@ -664 +699 @@
         }
 
-        void checkGenericParameters( ReferenceToType * inst ) {
-                for ( Expression * param : inst->parameters ) {
-                        if ( ! dynamic_cast< TypeExpr * >( param ) ) {
-                                SemanticError( inst, "Expression parameters for generic types are currently unsupported: " );
-                        }
-                }
-        }
-
         void LinkReferenceToTypes_old::postvisit( StructInstType * structInst ) {
                 const StructDecl * st = local_indexer->lookupStruct( structInst->name );
@@ -682 +709 @@
                         forwardStructs[ structInst->name ].push_back( structInst );
                 } // if
-                checkGenericParameters( structInst );
         }
 
@@ -695 +721 @@
                         forwardUnions[ unionInst->name ].push_back( unionInst );
                 } // if
-                checkGenericParameters( unionInst );
         }
 
@@ -807 +832 @@
                                 forwardEnums.erase( fwds );
                         } // if
-
-                        for ( Declaration * member : enumDecl->members ) {
-                                ObjectDecl * field = strict_dynamic_cast<ObjectDecl *>( member );
-                                if ( field->init ) {
-                                        // need to resolve enumerator initializers early so that other passes that determine if an expression is constexpr have the appropriate information.
-                                        SingleInit * init = strict_dynamic_cast<SingleInit *>( field->init );
-                                        ResolvExpr::findSingleExpression( init->value, new BasicType( Type::Qualifiers(), BasicType::SignedInt ), indexer );
-                                }
-                        }
                 } // if
         }
@@ -878 +894 @@
                                 typeInst->set_isFtype( typeDecl->kind == TypeDecl::Ftype );
                         } // if
+                } // if
+        }
+
+        ResolveEnumInitializers::ResolveEnumInitializers( const Indexer * other_indexer ) : WithIndexer( true ) {
+                if ( other_indexer ) {
+                        local_indexer = other_indexer;
+                } else {
+                        local_indexer = &indexer;
+                } // if
+        }
+
+        void ResolveEnumInitializers::postvisit( EnumDecl * enumDecl ) {
+                if ( enumDecl->body ) {
+                        for ( Declaration * member : enumDecl->members ) {
+                                ObjectDecl * field = strict_dynamic_cast<ObjectDecl *>( member );
+                                if ( field->init ) {
+                                        // need to resolve enumerator initializers early so that other passes that determine if an expression is constexpr have the appropriate information.
+                                        SingleInit * init = strict_dynamic_cast<SingleInit *>( field->init );
+                                        ResolvExpr::findSingleExpression( init->value, new BasicType( Type::Qualifiers(), BasicType::SignedInt ), indexer );
+                                }
+                        }
                 } // if
         }
@@ -1223 +1260 @@
         }
 
+        // Test for special name on a generic parameter.  Special treatment for the
+        // special name is a bootstrapping hack.  In most cases, the worlds of T's
+        // and of N's don't overlap (normal treamtemt).  The foundations in
+        // array.hfa use tagging for both types and dimensions.  Tagging treats
+        // its subject parameter even more opaquely than T&, which assumes it is
+        // possible to have a pointer/reference to such an object.  Tagging only
+        // seeks to identify the type-system resident at compile time.  Both N's
+        // and T's can make tags.  The tag definition uses the special name, which
+        // is treated as "an N or a T."  This feature is not inteded to be used
+        // outside of the definition and immediate uses of a tag.
+        static inline bool isReservedTysysIdOnlyName( const std::string & name ) {
+                // name's prefix was __CFA_tysys_id_only, before it got wrapped in __..._generic
+                int foundAt = name.find("__CFA_tysys_id_only");
+                if (foundAt == 0) return true;
+                if (foundAt == 2 && name[0] == '_' && name[1] == '_') return true;
+                return false;
+        }
+
         template< typename Aggr >
         void validateGeneric( Aggr * inst ) {
@@ -1239 +1294 @@
                         TypeSubstitution sub;
                         auto paramIter = params->begin();
-                        for ( size_t i = 0; paramIter != params->end(); ++paramIter, ++i ) {
-                                if ( i < args.size() ) {
-                                        TypeExpr * expr = strict_dynamic_cast< TypeExpr * >( * std::next( args.begin(), i ) );
-                                        sub.add( (* paramIter)->get_name(), expr->get_type()->clone() );
-                                } else if ( i == args.size() ) {
+                        auto argIter = args.begin();
+                        for ( ; paramIter != params->end(); ++paramIter, ++argIter ) {
+                                if ( argIter != args.end() ) {
+                                        TypeExpr * expr = dynamic_cast< TypeExpr * >( * argIter );
+                                        if ( expr ) {
+                                                sub.add( (* paramIter)->get_name(), expr->get_type()->clone() );
+                                        }
+                                } else {
                                         Type * defaultType = (* paramIter)->get_init();
                                         if ( defaultType ) {
                                                 args.push_back( new TypeExpr( defaultType->clone() ) );
                                                 sub.add( (* paramIter)->get_name(), defaultType->clone() );
+                                                argIter = std::prev(args.end());
+                                        } else {
+                                                SemanticError( inst, "Too few type arguments in generic type " );
                                         }
                                 }
+                                assert( argIter != args.end() );
+                                bool typeParamDeclared = (*paramIter)->kind != TypeDecl::Kind::Dimension;
+                                bool typeArgGiven;
+                                if ( isReservedTysysIdOnlyName( (*paramIter)->name ) ) {
+                                        // coerce a match when declaration is reserved name, which means "either"
+                                        typeArgGiven = typeParamDeclared;
+                                } else {
+                                        typeArgGiven = dynamic_cast< TypeExpr * >( * argIter );
+                                }
+                                if ( ! typeParamDeclared &&   typeArgGiven ) SemanticError( inst, "Type argument given for value parameter: " );
+                                if (   typeParamDeclared && ! typeArgGiven ) SemanticError( inst, "Expression argument given for type parameter: " );
                         }
 
                         sub.apply( inst );
-                        if ( args.size() < params->size() ) SemanticError( inst, "Too few type arguments in generic type " );
                         if ( args.size() > params->size() ) SemanticError( inst, "Too many type arguments in generic type " );
                 }
@@ -1264 +1335 @@
         void ValidateGenericParameters::previsit( UnionInstType * inst ) {
                 validateGeneric( inst );
+        }
+
+        void TranslateDimensionGenericParameters::translateDimensions( std::list< Declaration * > &translationUnit ) {
+                PassVisitor<TranslateDimensionGenericParameters> translator;
+                mutateAll( translationUnit, translator );
+        }
+
+        TranslateDimensionGenericParameters::TranslateDimensionGenericParameters() : WithIndexer( false ) {}
+
+        // Declaration of type variable:           forall( [N] )          ->  forall( N & | sized( N ) )
+        TypeDecl * TranslateDimensionGenericParameters::postmutate( TypeDecl * td ) {
+                if ( td->kind == TypeDecl::Dimension ) {
+                        td->kind = TypeDecl::Dtype;
+                        if ( ! isReservedTysysIdOnlyName( td->name ) ) {
+                                td->sized = true;
+                        }
+                }
+                return td;
+        }
+
+        // Situational awareness:
+        // array( float, [[currentExpr]]     )  has  visitingChildOfSUIT == true
+        // array( float, [[currentExpr]] - 1 )  has  visitingChildOfSUIT == false
+        // size_t x =    [[currentExpr]]        has  visitingChildOfSUIT == false
+        void TranslateDimensionGenericParameters::changeState_ChildOfSUIT( bool newVal ) {
+                GuardValue( nextVisitedNodeIsChildOfSUIT );
+                GuardValue( visitingChildOfSUIT );
+                visitingChildOfSUIT = nextVisitedNodeIsChildOfSUIT;
+                nextVisitedNodeIsChildOfSUIT = newVal;
+        }
+        void TranslateDimensionGenericParameters::premutate( StructInstType * sit ) {
+                (void) sit;
+                changeState_ChildOfSUIT(true);
+        }
+        void TranslateDimensionGenericParameters::premutate( UnionInstType * uit ) {
+                (void) uit;
+                changeState_ChildOfSUIT(true);
+        }
+        void TranslateDimensionGenericParameters::premutate( BaseSyntaxNode * node ) {
+                (void) node;
+                changeState_ChildOfSUIT(false);
+        }
+
+        // Passing values as dimension arguments:  array( float,     7 )  -> array( float, char[             7 ] )
+        // Consuming dimension parameters:         size_t x =    N - 1 ;  -> size_t x =          sizeof(N) - 1   ;
+        // Intertwined reality:                    array( float, N     )  -> array( float,              N        )
+        //                                         array( float, N - 1 )  -> array( float, char[ sizeof(N) - 1 ] )
+        // Intertwined case 1 is not just an optimization.
+        // Avoiding char[sizeof(-)] is necessary to enable the call of f to bind the value of N, in:
+        //   forall([N]) void f( array(float, N) & );
+        //   array(float, 7) a;
+        //   f(a);
+
+        Expression * TranslateDimensionGenericParameters::postmutate( DimensionExpr * de ) {
+                // Expression de is an occurrence of N in LHS of above examples.
+                // Look up the name that de references.
+                // If we are in a struct body, then this reference can be to an entry of the stuct's forall list.
+                // Whether or not we are in a struct body, this reference can be to an entry of a containing function's forall list.
+                // If we are in a struct body, then the stuct's forall declarations are innermost (functions don't occur in structs).
+                // Thus, a potential struct's declaration is highest priority.
+                // A struct's forall declarations are already renamed with _generic_ suffix.  Try that name variant first.
+
+                std::string useName = "__" + de->name + "_generic_";
+                TypeDecl * namedParamDecl = const_cast<TypeDecl *>( strict_dynamic_cast<const TypeDecl *, nullptr >( indexer.lookupType( useName ) ) );
+
+                if ( ! namedParamDecl ) {
+                        useName = de->name;
+                        namedParamDecl = const_cast<TypeDecl *>( strict_dynamic_cast<const TypeDecl *, nullptr >( indexer.lookupType( useName ) ) );
+                }
+
+                // Expect to find it always.  A misspelled name would have been parsed as an identifier.
+                assert( namedParamDecl && "Type-system-managed value name not found in symbol table" );
+
+                delete de;
+
+                TypeInstType * refToDecl = new TypeInstType( 0, useName, namedParamDecl );
+
+                if ( visitingChildOfSUIT ) {
+                        // As in postmutate( Expression * ), topmost expression needs a TypeExpr wrapper
+                        // But avoid ArrayType-Sizeof
+                        return new TypeExpr( refToDecl );
+                } else {
+                        // the N occurrence is being used directly as a runtime value,
+                        // if we are in a type instantiation, then the N is within a bigger value computation
+                        return new SizeofExpr( refToDecl );
+                }
+        }
+
+        Expression * TranslateDimensionGenericParameters::postmutate( Expression * e ) {
+                if ( visitingChildOfSUIT ) {
+                        // e is an expression used as an argument to instantiate a type
+                        if (! dynamic_cast< TypeExpr * >( e ) ) {
+                                // e is a value expression
+                                // but not a DimensionExpr, which has a distinct postmutate
+                                Type * typeExprContent = new ArrayType( 0, new BasicType( 0, BasicType::Char ), e, true, false );
+                                TypeExpr * result = new TypeExpr( typeExprContent );
+                                return result;
+                        }
+                }
+                return e;
         }
 

src/SynTree/Declaration.h

@@ -201 +201 @@
         typedef NamedTypeDecl Parent;
   public:
-        enum Kind { Dtype, DStype, Otype, Ftype, Ttype, ALtype, NUMBER_OF_KINDS };
+        enum Kind { Dtype, DStype, Otype, Ftype, Ttype, Dimension, NUMBER_OF_KINDS };
 
         Kind kind;

src/SynTree/Expression.h

@@ -587 +587 @@
 };
 
+/// DimensionExpr represents a type-system provided value used in an expression ( forrall([N]) ... N + 1 )
+class DimensionExpr : public Expression {
+  public:
+        std::string name;
+
+        DimensionExpr( std::string name );
+        DimensionExpr( const DimensionExpr & other );
+        virtual ~DimensionExpr();
+
+        const std::string & get_name() const { return name; }
+        void set_name( std::string newValue ) { name = newValue; }
+
+        virtual DimensionExpr * clone() const override { return new DimensionExpr( * this ); }
+        virtual void accept( Visitor & v ) override { v.visit( this ); }
+        virtual void accept( Visitor & v ) const override { v.visit( this ); }
+        virtual Expression * acceptMutator( Mutator & m ) override { return m.mutate( this ); }
+        virtual void print( std::ostream & os, Indenter indent = {} ) const override;
+};
+
 /// AsmExpr represents a GCC 'asm constraint operand' used in an asm statement: [output] "=f" (result)
 class AsmExpr : public Expression {

src/SynTree/Mutator.h

@@ -80 +80 @@
         virtual Expression * mutate( CommaExpr * commaExpr ) = 0;
         virtual Expression * mutate( TypeExpr * typeExpr ) = 0;
+        virtual Expression * mutate( DimensionExpr * dimensionExpr ) = 0;
         virtual Expression * mutate( AsmExpr * asmExpr ) = 0;
         virtual Expression * mutate( ImplicitCopyCtorExpr * impCpCtorExpr ) = 0;

src/SynTree/SynTree.h

@@ -85 +85 @@
 class CommaExpr;
 class TypeExpr;
+class DimensionExpr;
 class AsmExpr;
 class ImplicitCopyCtorExpr;

src/SynTree/TypeDecl.cc

@@ -33 +33 @@
 
 const char * TypeDecl::typeString() const {
-        static const char * kindNames[] = { "sized data type", "sized data type", "sized object type", "sized function type", "sized tuple type", "sized array length type" };
+        static const char * kindNames[] = { "sized data type", "sized data type", "sized object type", "sized function type", "sized tuple type", "sized length value" };
         static_assert( sizeof(kindNames) / sizeof(kindNames[0]) == TypeDecl::NUMBER_OF_KINDS, "typeString: kindNames is out of sync." );
         assertf( kind < TypeDecl::NUMBER_OF_KINDS, "TypeDecl kind is out of bounds." );

src/SynTree/TypeExpr.cc

@@ -35 +35 @@
 }
 
+DimensionExpr::DimensionExpr( std::string name ) : Expression(), name(name) {
+        assertf(name != "0", "Zero is not a valid name");
+        assertf(name != "1", "One is not a valid name");
+}
+
+DimensionExpr::DimensionExpr( const DimensionExpr & other ) : Expression( other ), name( other.name ) {
+}
+
+DimensionExpr::~DimensionExpr() {}
+
+void DimensionExpr::print( std::ostream & os, Indenter indent ) const {
+        os << "Type-Sys Value: " << get_name();
+        Expression::print( os, indent );
+}
 // Local Variables: //
 // tab-width: 4 //

src/SynTree/Visitor.h

@@ -135 +135 @@
         virtual void visit( TypeExpr * node ) { visit( const_cast<const TypeExpr *>(node) ); }
         virtual void visit( const TypeExpr * typeExpr ) = 0;
+        virtual void visit( DimensionExpr * node ) { visit( const_cast<const DimensionExpr *>(node) ); }
+        virtual void visit( const DimensionExpr * typeExpr ) = 0;
         virtual void visit( AsmExpr * node ) { visit( const_cast<const AsmExpr *>(node) ); }
         virtual void visit( const AsmExpr * asmExpr ) = 0;

tests/array-container/array-basic.cfa

@@ -61 +61 @@
 forall( [Nw], [Nx], [Ny], [Nz] )
 void fillHelloData( array( float, Nw, Nx, Ny, Nz ) & wxyz ) {
-    for (w; z(Nw))
-    for (x; z(Nx))
-    for (y; z(Ny))
-    for (z; z(Nz))
+    for (w; Nw)
+    for (x; Nx)
+    for (y; Ny)
+    for (z; Nz)
         wxyz[w][x][y][z] = getMagicNumber(w, x, y, z);
 }
 
-forall( [Zn]
+forall( [N]
       , S & | sized(S)
       )
-float total1d_low( arpk(Zn, S, float, float ) & a ) {
+float total1d_low( arpk(N, S, float, float ) & a ) {
     float total = 0.0f;
-    for (i; z(Zn))
+    for (i; N)
         total += a[i];
     return total;
@@ -98 +98 @@
 
     expect = 0;
-    for (i; z(Nw))
+    for (i; Nw)
         expect += getMagicNumber( i, slice_ix, slice_ix, slice_ix );
     printf("expect Ws             = %f\n", expect);
@@ -117 +117 @@
 
     expect = 0;
-    for (i; z(Nx))
+    for (i; Nx)
         expect += getMagicNumber( slice_ix, i, slice_ix, slice_ix );
     printf("expect Xs             = %f\n", expect);

tests/array-container/array-md-sbscr-cases.cfa

@@ -20 +20 @@
 forall( [Nw], [Nx], [Ny], [Nz] )
 void fillHelloData( array( float, Nw, Nx, Ny, Nz ) & wxyz ) {
-    for (w; z(Nw))
-    for (x; z(Nx))
-    for (y; z(Ny))
-    for (z; z(Nz))
+    for (w; Nw)
+    for (x; Nx)
+    for (y; Ny)
+    for (z; Nz)
         wxyz[w][x][y][z] = getMagicNumber(w, x, y, z);
 }
@@ -246 +246 @@
     assert(( wxyz[[2,  3,  4,  5]] == valExpected ));
 
-    for ( i; z(Nw) ) {
+    for ( i; Nw ) {
         assert(( wxyz[[ i, 3, 4, 5 ]] == getMagicNumber(i, 3, 4, 5) ));
     }
 
-    for ( i; z(Nx) ) {
+    for ( i; Nx ) {
         assert(( wxyz[[ 2, i, 4, 5 ]] == getMagicNumber(2, i, 4, 5) ));
     }
 
-    for ( i; z(Ny) ) {
+    for ( i; Ny ) {
         assert(( wxyz[[ 2, 3, i, 5 ]] == getMagicNumber(2, 3, i, 5) ));
     }
 
-    for ( i; z(Nz) ) {
+    for ( i; Nz ) {
         assert(( wxyz[[ 2, 3, 4, i ]] == getMagicNumber(2, 3, 4, i) ));
     }
 
-    for ( i; z(Nw) ) {
+    for ( i; Nw ) {
         assert(( wxyz[[ i, all, 4, 5 ]][3] == getMagicNumber(i, 3, 4, 5) ));
     }
 
-    for ( i; z(Nw) ) {
+    for ( i; Nw ) {
         assert(( wxyz[[ all, 3, 4, 5 ]][i] == getMagicNumber(i, 3, 4, 5) ));
     }

tests/device/cpu.cfa

@@ -17 +17 @@
 #include <fstream.hfa>
 #include <device/cpu.hfa>
+#include <stdlib.hfa>
+
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
 extern "C" {
+        #include <dirent.h>
+        #include <sys/types.h>
+        #include <sys/stat.h>
         #include <sys/sysinfo.h>
+        #include <fcntl.h>
+}
+
+// go through a directory calling fn on each file
+static int iterate_dir( const char * path, void (*fn)(struct dirent * ent) ) {
+        // open the directory
+        DIR *dir = opendir(path);
+        if(dir == 0p) { return ENOTDIR; }
+
+        // call fn for each
+        struct dirent * ent;
+        while ((ent = readdir(dir)) != 0p) {
+                fn( ent );
+        }
+
+        // no longer need this
+        closedir(dir);
+        return 0;
+}
+
+// count the number of directories with the specified prefix
+// the directories counted have the form '[prefix]N' where prefix is the parameter
+// and N is an base 10 integer.
+static int count_prefix_dirs(const char * path, const char * prefix) {
+        // read the directory and find the cpu count
+        // and make sure everything is as expected
+        int max = -1;
+        int count = 0;
+        void lambda(struct dirent * ent) {
+                // were are looking for prefixX, where X is a number
+                // check that it starts with 'cpu
+                char * s = strstr(ent->d_name, prefix);
+                if(s == 0p) { return; }
+                if(s != ent->d_name) { return; }
+
+                // check that the next part is a number
+                s += strlen(prefix);
+                char * end;
+                long int val = strtol(s, &end, 10);
+                if(*end != '\0' || val < 0) { return; }
+
+                // check that it's a directory
+                if(ent->d_type != DT_DIR) { return; }
+
+                // it's a match!
+                max = max(val, max);
+                count++;
+        }
+        iterate_dir(path, lambda);
+
+        /* paranoid */ verifyf(count == max + 1, "Inconsistent %s count, counted %d, but max %s was %d", prefix, count, prefix, (int)max);
+
+        return count;
+}
+
+// Count number of cache *indexes* in the system
+// cache indexes are distinct from cache level as Data or Instruction cache
+// can share a level but not an index
+// PITFALL: assumes all cpus have the same indexes as cpu0
+static int count_cache_indexes(void) {
+        return count_prefix_dirs("/sys/devices/system/cpu/cpu0/cache", "index");
+}
+
+// read information about a spcficic cache index/cpu file into the output buffer
+static size_t read_cpuidxinfo_into(unsigned cpu, unsigned idx, const char * file, char * out, size_t out_len) {
+        // Pick the file we want and read it
+        char buf[128];
+        /* paranoid */ __attribute__((unused)) int len =
+        snprintf(buf, 128, "/sys/devices/system/cpu/cpu%u/cache/index%u/%s", cpu, idx, file);
+        /* paranoid */ verifyf(len > 0, "Could not generate '%s' filename for cpu %u, index %u", file, cpu, idx);
+
+        int fd = open(buf, 0, O_RDONLY);
+        /* paranoid */ verifyf(fd > 0, "Could not open file '%s'", buf);
+
+        ssize_t r = read(fd, out, out_len);
+        /* paranoid */ verifyf(r > 0, "Could not read file '%s'", buf);
+
+        /* paranoid */ __attribute__((unused)) int ret =
+        close(fd);
+        /* paranoid */ verifyf(ret == 0, "Could not close file '%s'", buf);
+
+        out[r-1] = '\0';
+        return r-1;
+}
+
+unsigned find_idx() {
+        int idxs = count_cache_indexes();
+
+        unsigned found_level = 0;
+        unsigned found = -1u;
+        for(i; idxs) {
+                unsigned idx = idxs - 1 - i;
+                char buf[32];
+
+                // Level is the cache level: higher means bigger and slower
+                read_cpuidxinfo_into(0, idx, "level", buf, 32);
+                char * end;
+                unsigned long level = strtoul(buf, &end, 10);
+                /* paranoid */ verifyf(level <= 250, "Cpu %u has more than 250 levels of cache, that doesn't sound right", 0);
+                /* paranoid */ verify(*end == '\0');
+
+                if(found_level < level) {
+                        found_level = level;
+                        found = idx;
+                }
+        }
+
+        /* paranoid */ verify(found != -1u);
+        return found;
 }
 
 int main() {
+        //-----------------------------------------------------------------------
         int ret1 = get_nprocs();
         int ret2 = cpu_info.hthrd_count;
@@ -31 +151 @@
         }
 
+        //-----------------------------------------------------------------------
+        // Make sure no one has the same self
+        for(ime; cpu_info.hthrd_count) {
+                unsigned me = cpu_info.llc_map[ime].self;
+                {
+                        unsigned s = cpu_info.llc_map[ime].start;
+                        unsigned e = s + cpu_info.llc_map[ime].count;
+                        if(me < s || me >= e) {
+                                sout | "CPU" | ime | "outside of it's own map: " | s | "<=" | me | "<" | e;
+                        }
+                }
+
+
+                for(ithem; cpu_info.hthrd_count) {
+                        if(ime == ithem) continue;
+
+                        unsigned them = cpu_info.llc_map[ithem].self;
+                        if(me == them) {
+                                sout | "CPU" | ime | "has conflicting self id with" | ithem | "(" | me | ")";
+                        }
+                }
+        }
+
+
+        //-----------------------------------------------------------------------
+        unsigned idx = find_idx();
+        // For all procs check mapping is consistent
+        for(cpu_me; cpu_info.hthrd_count) {
+                char buf_me[32];
+                size_t len_me = read_cpuidxinfo_into(cpu_me, idx, "shared_cpu_list", buf_me, 32);
+                for(cpu_them; cpu_info.hthrd_count) {
+                        if(cpu_me == cpu_them) continue;
+                        char buf_them[32];
+                        size_t len_them = read_cpuidxinfo_into(cpu_them, idx, "shared_cpu_list", buf_them, 32);
+
+                        bool match_file = len_them == len_me && 0 == strncmp(buf_them, buf_me, len_me);
+                        bool match_info = cpu_info.llc_map[cpu_me].start == cpu_info.llc_map[cpu_them].start && cpu_info.llc_map[cpu_me].count == cpu_info.llc_map[cpu_them].count;
+
+                        if(match_file != match_info) {
+                                sout | "CPU" | cpu_me | "and" | cpu_them | "have inconsitent file and cpu_info";
+                                sout | cpu_me | ": <" | cpu_info.llc_map[cpu_me  ].start | "," | cpu_info.llc_map[cpu_me  ].count | "> '" | buf_me   | "'";
+                                sout | cpu_me | ": <" | cpu_info.llc_map[cpu_them].start | "," | cpu_info.llc_map[cpu_them].count | "> '" | buf_them | "'";
+                        }
+                }
+        }
 }