rpc : set SO_REUSEADDR for the server socket (#7320)

ref: #7293

examples/llama.android/app/src/main/cpp/CMakeLists.txt

@@ -12,17 +12,15 @@ cmake_minimum_required(VERSION 3.22.1)
 # build script scope).
 project("llama-android")
 
-#include(FetchContent)
-#FetchContent_Declare(
-#        llama
-#        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
-#        GIT_TAG        ci-android
-#)
-#
-## Also provides "common"
-#FetchContent_MakeAvailable(llama)
+include(FetchContent)
+FetchContent_Declare(
+        llama
+        GIT_REPOSITORY https://github.com/ggerganov/llama.cpp
+        GIT_TAG        master
+)
 
-add_subdirectory(../../../../../../ please-work)
+# Also provides "common"
+FetchContent_MakeAvailable(llama)
 
 # Creates and names a library, sets it as either STATIC
 # or SHARED, and provides the relative paths to its source code.

examples/rpc/rpc-server.cpp

@@ -56,6 +56,10 @@ static bool rpc_server_params_parse(int argc, char ** argv, rpc_server_params &
         } else if (arg == "-h" || arg == "--help") {
             print_usage(argc, argv, params);
             exit(0);
+        } else {
+            fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
+            print_usage(argc, argv, params);
+            exit(0);
         }
     }
     return true;

ggml-rpc.cpp

@@ -134,7 +134,13 @@ static bool set_no_delay(sockfd_t sockfd) {
     int flag = 1;
     // set TCP_NODELAY to disable Nagle's algorithm
     int ret = setsockopt(sockfd, IPPROTO_TCP, TCP_NODELAY, (char *)&flag, sizeof(int));
-    return ret >= 0;
+    return ret == 0;
+}
+
+static bool set_reuse_addr(sockfd_t sockfd) {
+    int flag = 1;
+    int ret = setsockopt(sockfd, SOL_SOCKET, SO_REUSEADDR, (char *)&flag, sizeof(int));
+    return ret == 0;
 }
 
 static std::shared_ptr<socket_t> socket_connect(const char * host, int port) {
@@ -181,7 +187,10 @@ static std::shared_ptr<socket_t> create_server_socket(const char * host, int por
     if (sock == nullptr) {
         return nullptr;
     }
-
+    if (!set_reuse_addr(sockfd)) {
+        fprintf(stderr, "Failed to set SO_REUSEADDR\n");
+        return nullptr;
+    }
     struct sockaddr_in serv_addr;
     serv_addr.sin_family = AF_INET;
     serv_addr.sin_addr.s_addr = inet_addr(host);

ggml.c

@@ -2076,7 +2076,7 @@ inline static float ggml_silu_f32(float x) {
     return x/(1.0f + expf(-x));
 }
 
-#if defined(__ARM_NEON) && defined(__aarch64__)
+#if defined(__ARM_NEON)
 
 // adapted from arm limited optimized routine
 // the maximum error is 1.45358 plus 0.5 ulps
@@ -2288,7 +2288,7 @@ static void ggml_vec_silu_f32(const int n, float * y, const float * x) {
     for (; i + 3 < n; i += 4) {
         _mm_storeu_ps(y + i, ggml_v_silu(_mm_loadu_ps(x + i)));
     }
-#elif defined(__ARM_NEON) && defined(__aarch64__)
+#elif defined(__ARM_NEON)
     for (; i + 3 < n; i += 4) {
         vst1q_f32(y + i, ggml_v_silu(vld1q_f32(x + i)));
     }
@@ -2335,7 +2335,7 @@ static ggml_float ggml_vec_soft_max_f32(const int n, float * y, const float * x,
 #endif
         sum += (ggml_float)_mm_cvtss_f32(val);
     }
-#elif defined(__ARM_NEON) && defined(__aarch64__)
+#elif defined(__ARM_NEON)
     for (; i + 3 < n; i += 4) {
         float32x4_t val = ggml_v_expf(vsubq_f32(vld1q_f32(x + i),
                                                 vdupq_n_f32(max)));