tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-15 12:42:40 -04:00
Code Issues Packages Projects Releases Wiki Activity

ggml-rpc: chunk send()/recv() to avoid EINVAL for very large tensors over RPC (macOS & others) (#15188)

Browse Source 
* ggml-rpc: chunk send()/recv() to avoid EINVAL for very large tensors over RPC (macOS & others). Fixes #15055

* ggml-rpc: rename RPC_IO_CHUNK->MAX_CHUNK_SIZE, use std::min() for cap, switch to GGML_LOG_ERROR, handle 0-length send/recv

* rpc: drop n==0 special case in send_data(); retry in loop per review

* rpc: remove trailing whitespace in send_data()

---------

Co-authored-by: Shinnosuke Takagi <nosuke@nosukenoMacBook-Pro.local>
This commit is contained in:
Tak-RS
2025-08-13 14:54:30 +09:00
committed by GitHub
parent b0493156fa
commit e71d48e326
1 changed files with 18 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
ggml/src/ggml-rpc/ggml-rpc.cpp
View File

@@ -29,9 +29,12 @@
#include <cstring> #include <cstring>
#include <fstream> #include <fstream>
#include <filesystem> #include <filesystem>
#include <algorithm>
namespace fs = std::filesystem; namespace fs = std::filesystem;
static constexpr size_t MAX_CHUNK_SIZE = 1024ull * 1024ull * 1024ull; // 1 GiB
#ifdef _WIN32 #ifdef _WIN32
typedef SOCKET sockfd_t; typedef SOCKET sockfd_t;
using ssize_t = __int64; using ssize_t = __int64;
@@ -323,11 +326,14 @@ static std::shared_ptr<socket_t> create_server_socket(const char * host, int por
static bool send_data(sockfd_t sockfd, const void * data, size_t size) { static bool send_data(sockfd_t sockfd, const void * data, size_t size) {
size_t bytes_sent = 0; size_t bytes_sent = 0;
while (bytes_sent < size) { while (bytes_sent < size) {
ssize_t n = send(sockfd, (const char *)data + bytes_sent, size - bytes_sent, 0); size_t size_to_send = std::min(size - bytes_sent, MAX_CHUNK_SIZE);
ssize_t n = send(sockfd, (const char *)data + bytes_sent, size_to_send, 0);
if (n < 0) { if (n < 0) {
GGML_LOG_ERROR("send failed (bytes_sent=%zu, size_to_send=%zu)\n",
bytes_sent, size_to_send);
return false; return false;
} }
bytes_sent += n; bytes_sent += (size_t)n;
} }
return true; return true;
} }
@@ -335,11 +341,18 @@ static bool send_data(sockfd_t sockfd, const void * data, size_t size) {
static bool recv_data(sockfd_t sockfd, void * data, size_t size) { static bool recv_data(sockfd_t sockfd, void * data, size_t size) {
size_t bytes_recv = 0; size_t bytes_recv = 0;
while (bytes_recv < size) { while (bytes_recv < size) {
ssize_t n = recv(sockfd, (char *)data + bytes_recv, size - bytes_recv, 0); size_t size_to_recv = std::min(size - bytes_recv, MAX_CHUNK_SIZE);
if (n <= 0) { ssize_t n = recv(sockfd, (char *)data + bytes_recv, size_to_recv, 0);
if (n < 0) {
GGML_LOG_ERROR("recv failed (bytes_recv=%zu, size_to_recv=%zu)\n",
bytes_recv, size_to_recv);
return false; return false;
} }
bytes_recv += n; if (n == 0) {
GGML_LOG_ERROR("recv returned 0 (peer closed?)\n");
return false;
}
bytes_recv += (size_t)n;
} }
return true; return true;
} }