build : fix most gcc and clang warnings (ggerganov#2861)

* fix most gcc and clang warnings

* baby-llama : remove commented opt_params_adam

* fix some MinGW warnings

* fix more MinGW warnings

CMakeLists.txt

@@ -403,6 +403,7 @@ if (LLAMA_ALL_WARNINGS)
             -Wpointer-arith
             -Wmissing-prototypes
             -Werror=implicit-int
+            -Wno-unused-function
         )
         set(cxx_flags
             -Wall
@@ -412,6 +413,10 @@ if (LLAMA_ALL_WARNINGS)
             -Wno-unused-function
             -Wno-multichar
         )
+        if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
+            # g++ only
+            set(cxx_flags ${cxx_flags} -Wno-format-truncation)
+        endif()
     else()
         # todo : msvc
     endif()

Makefile

@@ -91,9 +91,14 @@ endif # LLAMA_DISABLE_LOGS
 
 # warnings
 CFLAGS   += -Wall -Wextra -Wpedantic -Wcast-qual -Wdouble-promotion -Wshadow -Wstrict-prototypes -Wpointer-arith \
-			-Wmissing-prototypes -Werror=implicit-int
+			-Wmissing-prototypes -Werror=implicit-int -Wno-unused-function
 CXXFLAGS += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wno-multichar
 
+ifeq '' '$(findstring clang++,$(CXX))'
+	# g++ only
+	CXXFLAGS += -Wno-format-truncation
+endif
+
 # OS specific
 # TODO: support Windows
 ifeq ($(UNAME_S),Linux)

common/common.cpp

@@ -24,7 +24,9 @@
 
 #if defined(_WIN32)
 #define WIN32_LEAN_AND_MEAN
-#define NOMINMAX
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
 #include <codecvt>
 #include <locale>
 #include <windows.h>
@@ -1027,7 +1029,7 @@ void dump_non_result_info_yaml(FILE * stream, const gpt_params & params, const l
     dump_string_yaml_multiline(stream, "grammar", params.grammar.c_str());
     fprintf(stream, "grammar-file: # never logged, see grammar instead. Can still be specified for input.\n");
     fprintf(stream, "hellaswag: %s # default: false\n", params.hellaswag ? "true" : "false");
-    fprintf(stream, "hellaswag_tasks: %ld # default: 400\n", params.hellaswag_tasks);
+    fprintf(stream, "hellaswag_tasks: %zu # default: 400\n", params.hellaswag_tasks);
 
     const auto logit_bias_eos = params.logit_bias.find(llama_token_eos(lctx));
     const bool ignore_eos = logit_bias_eos != params.logit_bias.end() && logit_bias_eos->second == -INFINITY;

common/console.cpp

@@ -235,6 +235,7 @@ namespace console {
 
     int estimateWidth(char32_t codepoint) {
 #if defined(_WIN32)
+        (void)codepoint;
         return 1;
 #else
         return wcwidth(codepoint);

examples/baby-llama/baby-llama.cpp

@@ -1617,15 +1617,10 @@ int main(int argc, char ** argv) {
 
         float error_before_opt = ggml_get_f32_1d(e, 0);
 
-        struct ggml_opt_params opt_params_adam = ggml_opt_default_params(GGML_OPT_ADAM);
         struct ggml_opt_params opt_params_lbfgs = ggml_opt_default_params(GGML_OPT_LBFGS);
-        opt_params_adam.print_forward_graph = false;
-        opt_params_adam.print_backward_graph = false;
         opt_params_lbfgs.print_forward_graph = false;
         opt_params_lbfgs.print_backward_graph = false;
-        opt_params_adam.adam.n_iter = 16;
         opt_params_lbfgs.lbfgs.n_iter = 16;
-        // ggml_opt(ctx0, opt_params_adam, e);
         ggml_opt(ctx0, opt_params_lbfgs, e);
         //
         ggml_build_forward_expand(&gf, e);

examples/beam-search/beam-search.cpp

@@ -22,7 +22,9 @@
 #include <unistd.h>
 #elif defined (_WIN32)
 #define WIN32_LEAN_AND_MEAN
-#define NOMINMAX
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
 #include <windows.h>
 #include <signal.h>
 #endif
@@ -73,7 +75,7 @@ void beam_search_callback(void * callback_data_ptr, llama_beams_state beams_stat
         assert(0u < beams_state.n_beams);
         const llama_token * tokens = beams_state.beam_views[0].tokens;
         std::copy(tokens, tokens + n, callback_data.response.end() - n);
-        printf("%lu", n);
+        printf("%zu", n);
     }
     fflush(stdout);
 #if 1 // DEBUG: print current beams for this iteration
@@ -145,7 +147,7 @@ int main(int argc, char ** argv)
 
     if (tokens_list.size() > max_tokens_list_size)
     {
-        fprintf( stderr , "%s: error: prompt too long (%lu tokens, max %lu)\n" ,
+        fprintf( stderr , "%s: error: prompt too long (%zu tokens, max %zu)\n" ,
              __func__ , tokens_list.size() , max_tokens_list_size );
         return 1;
     }

examples/server/server.cpp

@@ -17,6 +17,8 @@
 #include "completion.js.hpp"
 #include "json-schema-to-grammar.mjs.hpp"
 
+#include <cstddef>
+
 #ifndef SERVER_VERBOSE
 #define SERVER_VERBOSE 1
 #endif
@@ -1038,7 +1040,7 @@ static json format_timings(llama_server_context &llama)
 {
     const auto timings = llama_get_timings(llama.ctx);
 
-    assert(timings.n_eval == llama.num_tokens_predicted);
+    assert(timings.n_eval == ptrdiff_t(llama.num_tokens_predicted));
 
     return json{
         {"prompt_n", timings.n_p_eval},
@@ -1239,7 +1241,7 @@ void beam_search_callback(void * callback_data, llama_beams_state beams_state) {
         const llama_token * tokens = beams_state.beam_views[0].tokens;
         const auto map = [](llama_token tok) { return completion_token_output{{},tok}; };
         std::transform(tokens, tokens + n, llama.generated_token_probs.end() - n, map);
-        printf("%lu", n);
+        printf("%zu", n);
     }
     fflush(stdout);
 #if 0 // DEBUG: print current beams for this iteration
@@ -1548,7 +1550,7 @@ int main(int argc, char **argv)
 
     svr.set_exception_handler([](const Request &, Response &res, std::exception_ptr ep)
                               {
-        const auto * fmt = "500 Internal Server Error\n%s";
+        const char fmt[] = "500 Internal Server Error\n%s";
         char buf[BUFSIZ];
         try {
             std::rethrow_exception(std::move(ep));

k_quants.c

@@ -183,13 +183,9 @@ static float make_qkx1_quants(int n, int nmax, const float * restrict x, uint8_t
         int ntry, float alpha) {
     float min = x[0];
     float max = x[0];
-    float sum_x = 0;
-    float sum_x2 = 0;
     for (int i = 1; i < n; ++i) {
         if (x[i] < min) min = x[i];
         if (x[i] > max) max = x[i];
-        sum_x += x[i];
-        sum_x2 += x[i]*x[i];
     }
     if (max == min) {
         for (int i = 0; i < n; ++i) L[i] = 0;
@@ -2060,7 +2056,7 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
 
     __m256 acc = _mm256_setzero_ps();
 
-    uint32_t *aux;
+    const uint32_t *aux;
 
     for (int i = 0; i < nb; ++i) {
 
@@ -2070,7 +2066,7 @@ void ggml_vec_dot_q3_K_q8_K(const int n, float * restrict s, const void * restri
         const int8_t  * restrict q8 = y[i].qs;
 
         // Set up scales
-        aux = (uint32_t *)x[i].scales;
+        aux = (const uint32_t *)x[i].scales;
         __m128i scales128 = _mm_set_epi32(
                 ((aux[1] >> 4) & kmask2) | (((aux[2] >> 6) & kmask1) << 4),
                 ((aux[0] >> 4) & kmask2) | (((aux[2] >> 4) & kmask1) << 4),

llama.cpp

@@ -3600,7 +3600,7 @@ static void llama_grammar_advance_stack(
         std::vector<std::vector<const llama_grammar_element *>> & new_stacks) {
 
     if (stack.empty()) {
-        new_stacks.push_back(stack);
+        new_stacks.emplace_back(stack);
         return;
     }
 
@@ -3637,7 +3637,7 @@ static void llama_grammar_advance_stack(
         }
         case LLAMA_GRETYPE_CHAR:
         case LLAMA_GRETYPE_CHAR_NOT:
-            new_stacks.push_back(stack);
+            new_stacks.emplace_back(stack);
             break;
         default:
             // end of alternate (LLAMA_GRETYPE_END, LLAMA_GRETYPE_ALT) or middle of char range