fix aligned-new error in jitkernel (#15626)

* fix aligned-new error in jitkernel

test=develop

* override genbase new to fix mis-align

test=develop

paddle/fluid/operators/jit/gen_base.cc

@@ -17,7 +17,13 @@
 #include <iostream>
 #include <sstream>
 #include <vector>
+#include "paddle/fluid/memory/allocation/cpu_allocator.h"  // for posix_memalign
 #include "paddle/fluid/platform/cpu_info.h"
+#include "paddle/fluid/platform/enforce.h"
+
+#ifndef _WIN32
+#define posix_memalign_free free
+#endif
 
 DEFINE_bool(dump_jitcode, false, "Whether to dump the jitcode to file");
 
@@ -40,6 +46,17 @@ void GenBase::dumpCode(const unsigned char* code) const {
   }
 }
 
+void* GenBase::operator new(size_t size) {
+  void* ptr;
+  constexpr size_t alignment = 32ul;
+  PADDLE_ENFORCE_EQ(posix_memalign(&ptr, alignment, size), 0,
+                    "GenBase Alloc %ld error!", size);
+  PADDLE_ENFORCE(ptr, "Fail to allocate GenBase CPU memory: size = %d .", size);
+  return ptr;
+}
+
+void GenBase::operator delete(void* ptr) { posix_memalign_free(ptr); }
+
 std::vector<int> packed_groups(int n, int k, int* block_out, int* rest_out) {
   int block;
   int max_num_regs;

paddle/fluid/operators/jit/gen_base.h

@@ -42,6 +42,11 @@ class GenBase : public Kernel {
     return reinterpret_cast<Func>(const_cast<unsigned char*>(code));
   }
 
+  void* operator new(size_t size);
+  void operator delete(void* ptr);
+  void* operator new[](size_t size) { return operator new(size); }
+  void operator delete[](void* ptr) { operator delete(ptr); }
+
  protected:
   void dumpCode(const unsigned char* code) const;
 };