[fix, build] Remove -fno-threadsafe-statics option to guarantee hw initialization thread-safety (#636)

* reverted to using thread-safe statics to make sure Singleton works as expected (thread-safe)
* removed limitation notice
* added note w.r.t c++17 standard and c++ runtime

Author: mzhukova

cmake/CompileOptions.cmake

@@ -6,7 +6,7 @@
 
 set(QPL_LINUX_TOOLCHAIN_REQUIRED_FLAGS "-Wall;-Wextra;-fPIC;-fstack-protector;-fstack-protector-strong;-fstack-clash-protection;--param=ssp-buffer-size=4")
 set(QPL_LINUX_TOOLCHAIN_DYNAMIC_LIBRARY_FLAGS "-Wl,-z,relro,-z,now;-Wl,-z,noexecstack")
-set(QPL_LINUX_TOOLCHAIN_CPP_EMBEDDED_FLAGS "-fno-exceptions;-fno-rtti;-fno-threadsafe-statics")
+set(QPL_LINUX_TOOLCHAIN_CPP_EMBEDDED_FLAGS "-fno-exceptions;-fno-rtti")
 
 set(QPL_WINDOWS_TOOLCHAIN_REQUIRED_FLAGS "/W3;/WX;/GS")
 set(QPL_WINDOWS_TOOLCHAIN_DYNAMIC_LIBRARY_FLAGS "")

doc/source/documentation/get_started_docs/installation.rst

@@ -98,7 +98,8 @@ Before building Intel® QPL, install and set up the following tools:
 
 - nasm 2.15.0 or higher (e.g., can be obtained from https://www.nasm.us)
 
-- GCC 8.2 or higher (Clang 12.0.1 or higher for building fuzz tests)
+- Intel QPL requires C++ compiler with C++17 standard support.
+  For instance, GCC 8.2+ (or Clang 12.0.1+ for building fuzz tests).
 
 - Universally Unique ID library ``uuid-dev`` version 2.35.2 or higher
 

doc/source/documentation/get_started_docs/quick_start.rst

@@ -63,6 +63,11 @@ Alternatively, in order to build ``compression_example.cpp`` individually using
 
     g++ -I/<install_dir>/include -o compression_example compression_example.cpp /<install_dir>/lib64/libqpl.a -ldl
 
+.. attention::
+
+   Intel QPL could be also used from C applications.
+   This would still require C++ runtime library installed on the system, and adding `-lstdc++`.
+
 To run the example on the Hardware Path, use:
 
 .. code-block:: shell

include/qpl/c_api/job.h

@@ -164,19 +164,8 @@ QPL_API(qpl_status, qpl_get_job_size, (qpl_path_t qpl_path, uint32_t * job_size_
  *                              @ref qpl_path_hardware or @ref qpl_path_software
  * @param[in,out]  qpl_job_ptr  a pointer to the @ref qpl_job structure
  *
- * @warning If multiple processes are created to submit jobs to the accelerator in the same application,
- * forking child process(es) must be done before the first @ref qpl_init_job function with @ref qpl_path_hardware.
- * Otherwise, the child process(es) may crash.
- *
  * @warning Memory for qpl_job structure must be allocated at the application side. Size (in bytes)
- * must be obtained with the @ref qpl_get_job_size function and then type-casted to @ref qpl_job :
- *
- * @code
- *  uint32_t  size;
- *  qpl_job *qpl_job_ptr;
- *  status = qpl_get_job_size(&size);
- *  qpl_job_ptr = (qpl_job*)malloc(size);
- * @endcode
+ * must be obtained with the @ref qpl_get_job_size function.
  *
  * @note qpl_job is an alias to the @ref qpl_job structure - must contain additional internal memory buffers for
  * SW path of compression/decompression/etc.

sources/core-iaa/include/hw_iaa_flags.h

@@ -17,7 +17,7 @@ typedef uint64_t hw_operation_flags_t; /**< Represents operation's flags */
 
 /* ################# COMPRESSION FLAGS ################# */
 /**
- * @name Operation codes
+ * @name Operation codes for Intel® In-Memory Analytics Accelerator (Intel® IAA)
  * @anchor HW_OPCODES
  * @todo Opcode values
  * @{

sources/core-iaa/sources/descriptors/hw_crc64_descriptor.c

@@ -13,7 +13,8 @@
 #include "qplc_memop.h"
 
 /**
- * @brief Defines a type of the Intel IAA crc64 descriptor
+ * @brief Defines a type of the Intel® In-Memory Analytics Accelerator (Intel® IAA)
+ * crc64 descriptor
  */
 HW_PATH_BYTE_PACKED_STRUCTURE_BEGIN {
     uint32_t trusted_fields;         /**< 19:0 PASID - process address space ID; 30:20 - reserved; 31 - User/Supervisor */

sources/middle-layer/dispatcher/hw_dispatcher.cpp

@@ -20,21 +20,9 @@
 
 namespace qpl::ml::dispatcher {
 
-
-static int is_initialized = 0;
-static std::mutex hw_init_lock;
-
 hw_dispatcher::hw_dispatcher() noexcept {
-    if (!is_initialized) {
-        hw_init_lock.lock();
-        if (!is_initialized) {
-            hw_init_status_ = hw_dispatcher::initialize_hw();
-            hw_support_     = hw_init_status_ == HW_ACCELERATOR_STATUS_OK;
-
-            is_initialized = 1;
-        }
-        hw_init_lock.unlock();
-    }
+    hw_init_status_ = hw_dispatcher::initialize_hw();
+    hw_support_     = hw_init_status_ == HW_ACCELERATOR_STATUS_OK;
 }
 
 auto hw_dispatcher::initialize_hw() noexcept -> hw_accelerator_status {
@@ -100,9 +88,12 @@ hw_dispatcher::~hw_dispatcher() noexcept {
 #endif
 }
 
+// starting from C++11 standard,
+// it is guarantued that the following would be thread-safe
+// and created only once
+// (case: static variables with block scope)
 auto hw_dispatcher::get_instance() noexcept -> hw_dispatcher & {
     static hw_dispatcher instance{};
-
     return instance;
 }