fix CUDA split buffers

Author: slaren

ggml-backend.c

@@ -1051,8 +1051,9 @@ struct ggml_backend_sched {
     struct ggml_cgraph * graph;
 
     // graph splits
-    struct ggml_backend_sched_split splits[GGML_SCHED_MAX_SPLITS];
+    struct ggml_backend_sched_split * splits;
     int n_splits;
+    int splits_capacity;
 
     // pipeline parallelism support
     int n_copies;
@@ -1443,6 +1444,10 @@ static void ggml_backend_sched_split_graph(ggml_backend_sched_t sched, struct gg
             if (node_backend_id != cur_backend_id || offload) {
                 split->i_end = i;
                 i_split++;
+                if (i_split >= sched->splits_capacity) {
+                    sched->splits_capacity *= 2;
+                    sched->splits = realloc(sched->splits, sched->splits_capacity * sizeof(struct ggml_backend_sched_split));
+                }
                 GGML_ASSERT(i_split < GGML_SCHED_MAX_SPLITS);
                 split = &sched->splits[i_split];
                 split->backend_id = node_backend_id;
@@ -1711,7 +1716,9 @@ ggml_backend_sched_t ggml_backend_sched_new(
 
     sched->n_copies = parallel ? GGML_SCHED_MAX_COPIES : 1;
 
-    GGML_ASSERT(sched->n_copies <= GGML_SCHED_MAX_COPIES);
+    const int initial_splits_capacity = 16;
+    sched->splits = calloc(sizeof(sched->splits[0]), initial_splits_capacity);
+    sched->splits_capacity = initial_splits_capacity;
 
     for (int b = 0; b < n_backends; b++) {
         sched->backends[b] = backends[b];
@@ -1742,6 +1749,7 @@ void ggml_backend_sched_free(ggml_backend_sched_t sched) {
     }
     ggml_gallocr_free(sched->galloc);
     ggml_free(sched->ctx);
+    free(sched->splits);
     free(sched->hash_set.keys);
     free(sched->tensor_backend_id);
     free(sched->tensor_copies);

ggml-cuda.cu

@@ -10755,6 +10755,8 @@ static ggml_backend_buffer_type_i ggml_backend_cuda_buffer_type_interface = {
 };
 
 GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_buffer_type(int device) {
+    ggml_init_cublas();
+
     // FIXME: this is not thread safe
     if (device >= ggml_backend_cuda_get_device_count()) {
         return nullptr;
@@ -11039,6 +11041,8 @@ static ggml_backend_buffer_type_i ggml_backend_cuda_split_buffer_type_interface
 };
 
 GGML_CALL ggml_backend_buffer_type_t ggml_backend_cuda_split_buffer_type(const float * tensor_split) {
+    ggml_init_cublas();
+
     // FIXME: this is not thread safe
     static std::map<std::array<float, GGML_CUDA_MAX_DEVICES>, struct ggml_backend_buffer_type> buft_map;
 
@@ -11389,15 +11393,9 @@ GGML_CALL static bool ggml_backend_cuda_supports_op(ggml_backend_t backend, cons
 }
 
 GGML_CALL static bool ggml_backend_cuda_offload_op(ggml_backend_t backend, const ggml_tensor * op) {
-    const ggml_tensor * dst = op;
-
     const int min_batch_size = 32;
 
-    if (dst->ne[1] > min_batch_size && dst->op != GGML_OP_GET_ROWS) {
-        return true;
-    }
-
-    return false;
+    return op->ne[1] > min_batch_size && op->op != GGML_OP_GET_ROWS;
 }
 
 static ggml_backend_event_t ggml_backend_cuda_event_new(ggml_backend_t backend) {
@@ -11476,7 +11474,7 @@ static ggml_guid_t ggml_backend_cuda_guid() {
 }
 
 GGML_CALL ggml_backend_t ggml_backend_cuda_init(int device) {
-    ggml_init_cublas(); // TODO: remove from ggml.c
+    ggml_init_cublas();
 
     if (device < 0 || device >= ggml_cuda_get_device_count()) {
         fprintf(stderr, "%s: error: invalid device %d\n", __func__, device);

llama.cpp

@@ -5039,7 +5039,11 @@ static bool llm_load_tensors(
             ml.get_mapping_range(&first, &last, ctx);
             buf = ggml_backend_cpu_buffer_from_ptr((char *) ml.mapping->addr + first, last - first);
 #ifdef GGML_USE_CUBLAS
-            ggml_backend_cuda_register_host_buffer((char *) ml.mapping->addr + first, last - first);
+            if (n_layer >= n_gpu_layers) {
+                ggml_backend_cuda_register_host_buffer(
+                        ggml_backend_buffer_get_base(buf),
+                        ggml_backend_buffer_get_size(buf));
+            }
 #endif
         }
 #ifdef GGML_USE_METAL