fix: CMake arch auto-detection SIGILL, typos, and improved error messages

README.md

@@ -1,4 +1,4 @@
-<div align=" center">
+<div align="center">
   <picture>
     <source media="(prefers-color-scheme: dark)" srcset="https://zvec.oss-cn-hongkong.aliyuncs.com/logo/github_log_2.svg" />
     <img src="https://zvec.oss-cn-hongkong.aliyuncs.com/logo/github_logo_1.svg" width="400" alt="zvec logo" />

cmake/option.cmake

@@ -89,23 +89,20 @@ function(_detect_armv8_best)
 endfunction()
 
 function(_detect_x86_best)
-  set(_x86_flags
-    "graniterapids" "emeraldrapids" "sapphirerapids"
-    "skylake-avx512" "skylake"
-    "broadwell" "haswell" "sandybridge" "nehalem"
-    "znver3" "znver2" "znver1"
-  )
-  foreach(_arch IN LISTS _x86_flags)
-    check_c_compiler_flag("-march=${_arch}" _COMP_SUPP_${_arch})
-    if(_COMP_SUPP_${_arch})
-      _AppendFlags(CMAKE_C_FLAGS "-march=${_arch}")
-      _AppendFlags(CMAKE_CXX_FLAGS "-march=${_arch}")
-      set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" PARENT_SCOPE)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" PARENT_SCOPE)
-      return()
-    endif()
-  endforeach()
-  message(WARNING "No known x86 microarchitecture flag supported; falling back to generic.")
+  # Use -march=native to target the actual build machine's CPU.
+  # The previous approach iterated from highest to lowest arch and checked
+  # whether the *compiler* accepted the flag — but the compiler will accept
+  # e.g. -march=skylake-avx512 on any machine, which then emits AVX-512
+  # instructions that crash on CPUs without those extensions (see #128, #92).
+  check_c_compiler_flag("-march=native" _COMP_SUPP_NATIVE)
+  if(_COMP_SUPP_NATIVE)
+    _AppendFlags(CMAKE_C_FLAGS "-march=native")
+    _AppendFlags(CMAKE_CXX_FLAGS "-march=native")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" PARENT_SCOPE)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" PARENT_SCOPE)
+    return()
+  endif()
+  message(WARNING "Compiler does not support -march=native; falling back to generic x86-64.")
 endfunction()
 
 if(MSVC)

src/ailego/internal/cpu_features.cc

@@ -327,7 +327,7 @@ bool CpuFeatures::VMX(void) {
   return !!(flags_.L1_ECX & (1u << 5));
 }
 
-// ！Running on a hypervisor
+//! Running on a hypervisor
 bool CpuFeatures::HYPERVISOR(void) {
   return !!(flags_.L1_ECX & (1u << 31));
 }

src/ailego/internal/cpu_features.h

@@ -179,7 +179,7 @@ class CpuFeatures {
   //! Hardware virtualization
   static bool VMX(void);
 
-  // ！Running on a hypervisor
+  //! Running on a hypervisor
   static bool HYPERVISOR(void);
 
   //! Intrinsics of compiling
@@ -359,7 +359,7 @@ class CpuFeatures {
     //! Hardware virtualization
     bool VMX = CpuFeatures::VMX();
 
-    // ！Running on a hypervisor
+    //! Running on a hypervisor
     bool HYPERVISOR = CpuFeatures::HYPERVISOR();
   };
   static StaticFlags static_flags_;

src/core/algorithm/hnsw/hnsw_builder_entity.h

@@ -123,7 +123,7 @@ class HnswBuilderEntity : public HnswEntity {
   std::string neighbors_buffer_{};        // level 0 neighbors buffer
   std::string upper_neighbors_buffer_{};  // upper layer neighbors buffer
 
-  std::string sparse_data_buffer_{};  // aligned spase data buffer
+  std::string sparse_data_buffer_{};  // aligned sparse data buffer
   size_t sparse_data_offset_{0};      //
 
   // upper layer offset + level in upper_neighbors_buffer_

src/core/algorithm/hnsw/hnsw_context.cc

@@ -77,7 +77,7 @@ int HnswContext::init(ContextType type) {
       update_heap_.limit(entity_->l0_neighbor_cnt() + 1);
       candidates_.limit(max_scan_num_);
 
-      check_need_adjuct_ctx();
+      check_need_adjust_ctx();
       break;
 
     default:

src/core/algorithm/hnsw/hnsw_context.h

@@ -361,8 +361,8 @@ class HnswContext : public IndexContext {
     return level_topks_[level];
   }
 
-  inline void check_need_adjuct_ctx(void) {
-    check_need_adjuct_ctx(entity_->doc_cnt());
+  inline void check_need_adjust_ctx(void) {
+    check_need_adjust_ctx(entity_->doc_cnt());
   }
 
   inline size_t compute_reserve_cnt(uint32_t cur_doc) const {
@@ -375,7 +375,7 @@ class HnswContext : public IndexContext {
   }
 
   //! candidates heap and visitfilter need to resize as doc cnt growing up
-  inline void check_need_adjuct_ctx(uint32_t doc_cnt) {
+  inline void check_need_adjust_ctx(uint32_t doc_cnt) {
     if (ailego_unlikely(doc_cnt + kTriggerReserveCnt > reserve_max_doc_cnt_)) {
       while (doc_cnt + kTriggerReserveCnt > reserve_max_doc_cnt_) {
         reserve_max_doc_cnt_ =

src/core/algorithm/hnsw/hnsw_streamer.cc

@@ -391,7 +391,7 @@ IndexStreamer::Context::Pointer HnswStreamer::create_context(void) const {
     LOG_DEBUG("HnswStreamer doc_count[%zu] estimate[%zu]",
               (size_t)entity_.doc_cnt(), (size_t)estimate_doc_count);
   }
-  ctx->check_need_adjuct_ctx(std::max(entity_.doc_cnt(), estimate_doc_count));
+  ctx->check_need_adjust_ctx(std::max(entity_.doc_cnt(), estimate_doc_count));
 
   return Context::Pointer(ctx);
 }
@@ -466,7 +466,7 @@ int HnswStreamer::add_with_id_impl(uint32_t id, const void *query,
   ctx->clear();
   ctx->update_dist_caculator_distance(add_distance_, add_batch_distance_);
   ctx->reset_query(query);
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
 
   if (metric_->support_train()) {
     const std::lock_guard<std::mutex> lk(mutex_);
@@ -546,7 +546,7 @@ int HnswStreamer::add_impl(uint64_t pkey, const void *query,
   ctx->clear();
   ctx->update_dist_caculator_distance(add_distance_, add_batch_distance_);
   ctx->reset_query(query);
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
 
   if (metric_->support_train()) {
     const std::lock_guard<std::mutex> lk(mutex_);
@@ -618,7 +618,7 @@ int HnswStreamer::search_impl(const void *query, const IndexQueryMeta &qmeta,
   ctx->clear();
   ctx->update_dist_caculator_distance(search_distance_, search_batch_distance_);
   ctx->resize_results(count);
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
   for (size_t q = 0; q < count; ++q) {
     ctx->reset_query(query);
     ret = alg_->search(ctx);

src/core/algorithm/hnsw_sparse/hnsw_sparse_builder_entity.h

@@ -146,7 +146,7 @@ class HnswSparseBuilderEntity : public HnswSparseEntity {
   std::string neighbors_buffer_{};        // level 0 neighbors buffer
   std::string upper_neighbors_buffer_{};  // upper layer neighbors buffer
 
-  std::string sparse_data_buffer_{};  // aligned spase data buffer
+  std::string sparse_data_buffer_{};  // aligned sparse data buffer
   size_t sparse_data_offset_{0};      //
 
   // upper layer offset + level in upper_neighbors_buffer_

src/core/algorithm/hnsw_sparse/hnsw_sparse_context.cc

@@ -71,7 +71,7 @@ int HnswSparseContext::init(ContextType type) {
       update_heap_.limit(entity_->l0_neighbor_cnt() + 1);
       candidates_.limit(max_scan_num_);
 
-      check_need_adjuct_ctx();
+      check_need_adjust_ctx();
       break;
 
     default:

src/core/algorithm/hnsw_sparse/hnsw_sparse_context.h

@@ -357,8 +357,8 @@ class HnswSparseContext : public IndexContext {
     return level_topks_[level];
   }
 
-  inline void check_need_adjuct_ctx(void) {
-    check_need_adjuct_ctx(entity_->doc_cnt());
+  inline void check_need_adjust_ctx(void) {
+    check_need_adjust_ctx(entity_->doc_cnt());
   }
 
   inline size_t compute_reserve_cnt(uint32_t cur_doc) const {
@@ -371,7 +371,7 @@ class HnswSparseContext : public IndexContext {
   }
 
   //! candidates heap and visitfilter need to resize as doc cnt growing up
-  inline void check_need_adjuct_ctx(uint32_t doc_cnt) {
+  inline void check_need_adjust_ctx(uint32_t doc_cnt) {
     if (ailego_unlikely(doc_cnt + kTriggerReserveCnt > reserve_max_doc_cnt_)) {
       while (doc_cnt + kTriggerReserveCnt > reserve_max_doc_cnt_) {
         reserve_max_doc_cnt_ =

src/core/algorithm/hnsw_sparse/hnsw_sparse_streamer.cc

@@ -485,7 +485,7 @@ int HnswSparseStreamer::add_with_id_impl(uint32_t id,
                                    sparse_query_buffer);
 
   ctx->reset_query(sparse_query_buffer.data());
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
 
   level_t level = alg_->get_random_level();
   ret =
@@ -570,7 +570,7 @@ int HnswSparseStreamer::add_impl(uint64_t pkey, const uint32_t sparse_count,
                                    sparse_query_buffer);
 
   ctx->reset_query(sparse_query_buffer.data());
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
 
   level_t level = alg_->get_random_level();
   node_id_t id;
@@ -637,7 +637,7 @@ int HnswSparseStreamer::search_impl(
   ctx->clear();
   ctx->update_dist_caculator_distance(search_distance_);
   ctx->resize_results(count);
-  ctx->check_need_adjuct_ctx(entity_.doc_cnt());
+  ctx->check_need_adjust_ctx(entity_.doc_cnt());
 
   const uint32_t *sparse_indices_tmp = sparse_indices;
   const void *sparse_query_tmp = sparse_query;

src/core/interface/index.cc

@@ -130,7 +130,10 @@ int Index::CreateAndInitConverterReformer(const QuantizerParam &param,
           } else if (index_param.data_type == DataType::DT_FP32) {
             converter_name = "CosineNormalizeConverter";
           } else {
-            LOG_ERROR("Unsupported data type: ");
+            LOG_ERROR(
+                "Cosine metric without quantizer only supports FP32 and FP16 "
+                "data types. For INT8 data, use a quantizer "
+                "(e.g. QuantizerType::kInt8).");
             return core::IndexError_Unsupported;
           }
           break;
@@ -769,7 +772,7 @@ int Index::Merge(const std::vector<Index::Pointer> &indexes,
     LOG_ERROR("Failed to init reducer");
     return core::IndexError_Runtime;
   }
-  if (reducer->set_target_streamer_wiht_info(builder_, streamer_, converter_,
+  if (reducer->set_target_streamer_with_info(builder_, streamer_, converter_,
                                              reformer_,
                                              input_vector_meta_) != 0) {
     LOG_ERROR("Failed to set target streamer");

src/core/mixed_reducer/mixed_streamer_reducer.cc

@@ -55,7 +55,7 @@ int MixedStreamerReducer::cleanup(void) {
   return 0;
 }
 
-int MixedStreamerReducer::set_target_streamer_wiht_info(
+int MixedStreamerReducer::set_target_streamer_with_info(
     const IndexBuilder::Pointer builder, const IndexStreamer::Pointer streamer,
     const IndexConverter::Pointer converter,
     const IndexReformer::Pointer reformer,

src/core/mixed_reducer/mixed_streamer_reducer.h

@@ -45,7 +45,7 @@ class MixedStreamerReducer : public IndexStreamerReducer {
   int dump(const IndexDumper::Pointer &dumper) override;
 
  public:  // StreamerReducer's unique methods
-  int set_target_streamer_wiht_info(
+  int set_target_streamer_with_info(
       const IndexBuilder::Pointer builder,
       const IndexStreamer::Pointer streamer,
       const IndexConverter::Pointer converter,

src/core/quantizer/cosine_converter.cc

@@ -254,7 +254,7 @@ class CosineConverter : public IndexConverter {
     IndexMeta::DataType type = meta_.data_type();
 
     if (type != original_type_) {
-      LOG_ERROR("Orignal Type Not Matched: (%d, %d)", type, original_type_);
+      LOG_ERROR("Original Type Not Matched: (%d, %d)", type, original_type_);
       return IndexError_Mismatch;
     }
 

src/db/index/column/vector_column/vector_column_indexer.h

@@ -123,7 +123,7 @@ class VectorColumnIndexer {
 
   std::string engine_name_ = "proxima";
   bool is_sparse_{false};  // TODO: eliminate the dynamic flag and make it
-                           // static/template/seperate class
+                           // static/template/separate class
 };
 
 

src/include/zvec/core/framework/index_reducer.h

@@ -213,7 +213,7 @@ class IndexStreamerReducer : public IndexReducerBase {
   //! Index Reducer Pointer
   typedef std::shared_ptr<IndexStreamerReducer> Pointer;
 
-  virtual int set_target_streamer_wiht_info(
+  virtual int set_target_streamer_with_info(
       const IndexBuilder::Pointer builder,
       const IndexStreamer::Pointer streamer,
       const IndexConverter::Pointer converter,