Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

rpc : fix load/store misaligned addresses

Browse Source
This commit is contained in:
Georgi Gerganov 2024-06-15 14:39:20 +03:00
parent 7b2f4a7d19
commit 34bdbed481
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 8 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
ggml-rpc.cpp
View File

@ -73,9 +73,13 @@ struct rpc_tensor {
uint64_t view_offs; uint64_t view_offs;
uint64_t data; uint64_t data;
char name[GGML_MAX_NAME]; char name[GGML_MAX_NAME];
char padding[4];
}; };
#pragma pack(pop) #pragma pack(pop)
static_assert(sizeof(rpc_tensor) % 8 == 0, "rpc_tensor size must be multiple of 8");
// RPC commands // RPC commands
enum rpc_cmd { enum rpc_cmd {
ALLOC_BUFFER = 0, ALLOC_BUFFER = 0,
@ -599,9 +603,8 @@ static void serialize_graph(const ggml_cgraph * cgraph, std::vector<uint8_t> & o
int output_size = sizeof(uint32_t) + n_nodes * sizeof(uint64_t) + sizeof(uint32_t) + n_tensors * sizeof(rpc_tensor); int output_size = sizeof(uint32_t) + n_nodes * sizeof(uint64_t) + sizeof(uint32_t) + n_tensors * sizeof(rpc_tensor);
output.resize(output_size, 0); output.resize(output_size, 0);
memcpy(output.data(), &n_nodes, sizeof(n_nodes)); memcpy(output.data(), &n_nodes, sizeof(n_nodes));
uint64_t * out_nodes = (uint64_t *)(output.data() + sizeof(n_nodes));
for (uint32_t i = 0; i < n_nodes; i++) { for (uint32_t i = 0; i < n_nodes; i++) {
out_nodes[i] = reinterpret_cast<uint64_t>(cgraph->nodes[i]); memcpy(output.data() + sizeof(n_nodes) + i * sizeof(uint64_t), &cgraph->nodes[i], sizeof(uint64_t));
} }
uint32_t * out_ntensors = (uint32_t *)(output.data() + sizeof(n_nodes) + n_nodes * sizeof(uint64_t)); uint32_t * out_ntensors = (uint32_t *)(output.data() + sizeof(n_nodes) + n_nodes * sizeof(uint64_t));
*out_ntensors = n_tensors; *out_ntensors = n_tensors;
@ -1036,7 +1039,9 @@ bool rpc_server::graph_compute(const std::vector<uint8_t> & input, std::vector<u
} }
std::unordered_map<uint64_t, ggml_tensor*> tensor_map; std::unordered_map<uint64_t, ggml_tensor*> tensor_map;
for (uint32_t i = 0; i < n_nodes; i++) { for (uint32_t i = 0; i < n_nodes; i++) {
graph->nodes[i] = create_node(nodes[i], ctx, tensor_ptrs, tensor_map); int64_t id;
memcpy(&id, &nodes[i], sizeof(id));
graph->nodes[i] = create_node(id, ctx, tensor_ptrs, tensor_map);
} }
ggml_status status = ggml_backend_graph_compute(backend, graph); ggml_status status = ggml_backend_graph_compute(backend, graph);
// output serialization format: | status (1 byte) | // output serialization format: | status (1 byte) |