Merge pull request #142 from devnexen/arm_aal_proposal

AAL, basic arm implementation proposal.

CMakeLists.txt

@@ -94,7 +94,15 @@ if (WIN32)
 endif()
 
 if (CMAKE_CXX_COMPILER_ID MATCHES "Clang")
-  target_compile_options(snmalloc_lib INTERFACE -mcx16)
+  if(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "x86_64")
+    target_compile_options(snmalloc_lib INTERFACE -mcx16)
+  elseif(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "amd64")
+   target_compile_options(snmalloc_lib INTERFACE -mcx16)
+  elseif(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "AMD64")
+   target_compile_options(snmalloc_lib INTERFACE -mcx16)
+  elseif(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "x86")
+   target_compile_options(snmalloc_lib INTERFACE -mcx16)
+  endif()
 endif ()
 
 # Have to set this globally, as can't be set on an interface target.
@@ -136,7 +144,8 @@ if(NOT DEFINED SNMALLOC_ONLY_HEADER_LIBRARY)
      add_compile_options(-march=native)
     elseif(${CMAKE_SYSTEM_PROCESSOR} STREQUAL "x86")
      add_compile_options(-march=native)
-    # XXX elseif ARM?
+    elseif(${CMAKE_SYSTEM_PROCESSOR} MATCHES "arm")
+     add_compile_options(-march=native)
     endif()
   endif()
 

src/aal/aal.h

@@ -1,6 +1,7 @@
 #pragma once
 #include "../ds/defines.h"
 
+#include <chrono>
 #include <cstdint>
 
 #if defined(__i386__) || defined(_M_IX86) || defined(_X86_) || \
@@ -9,6 +10,10 @@
 #  define PLATFORM_IS_X86
 #endif
 
+#if defined(__arm__) || defined(__aarch64__)
+#  define PLATFORM_IS_ARM
+#endif
+
 namespace snmalloc
 {
   /**
@@ -22,6 +27,10 @@ namespace snmalloc
      * and so may use bit operations on pointer values.
      */
     IntegerPointers = (1 << 0),
+    /**
+     * This architecture cannot access cpu cycles counters.
+     */
+    NoCpuCycleCounters = (1 << 1),
   };
 
   /**
@@ -59,6 +68,15 @@ namespace snmalloc
      */
     static inline uint64_t tick()
     {
+      if constexpr (
+        (Arch::aal_features & NoCpuCycleCounters) == NoCpuCycleCounters)
+      {
+        auto tick = std::chrono::high_resolution_clock::now();
+        return static_cast<uint64_t>(
+          std::chrono::duration_cast<std::chrono::nanoseconds>(
+            tick.time_since_epoch())
+            .count());
+      }
 #if __has_builtin(__builtin_readcyclecounter) && \
   !defined(SNMALLOC_NO_AAL_BUILTINS)
       return __builtin_readcyclecounter();
@@ -70,8 +88,10 @@ namespace snmalloc
 
 } // namespace snmalloc
 
-#ifdef PLATFORM_IS_X86
+#if defined(PLATFORM_IS_X86)
 #  include "aal_x86.h"
+#elif defined(PLATFORM_IS_ARM)
+#  include "aal_arm.h"
 #endif
 
 namespace snmalloc

src/aal/aal_arm.h

@@ -0,0 +1,34 @@
+#pragma once
+
+#if defined(__arch64__)
+#  define SNMALLOC_VA_BITS_64
+#else
+#  define SNMALLOC_VA_BITS_32
+#endif
+
+#include <iostream>
+namespace snmalloc
+{
+  /**
+   * ARM-specific architecture abstraction layer.
+   */
+  class AAL_arm
+  {
+  public:
+    /**
+     * Bitmap of AalFeature flags
+     */
+    static constexpr uint64_t aal_features =
+      IntegerPointers | NoCpuCycleCounters;
+    /**
+     * On pipelined processors, notify the core that we are in a spin loop and
+     * that speculative execution past this point may not be a performance gain.
+     */
+    static inline void pause()
+    {
+      __asm__ volatile("yield");
+    }
+  };
+
+  using AAL_Arch = AAL_arm;
+} // namespace snmalloc

src/ds/bits.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <cstddef>
 #include <limits>
 
 // #define USE_LZCNT

src/pal/pal_posix.h

@@ -3,6 +3,7 @@
 #include "../ds/address.h"
 #include "../mem/allocconfig.h"
 
+#include <stdlib.h>
 #include <string.h>
 #include <sys/mman.h>