From 3286bf5ab66c5ea00237af642967a25043581353 Mon Sep 17 00:00:00 2001
From: Ken Steele <ken@tilera.com>
Date: Wed, 20 Mar 2013 12:54:38 -0400
Subject: [PATCH] On the Tile architecture, use atomic instructions for atomic
 ptr and counter.

For atomic_counter and atomic_ptr classes, detect the Tile architecture
using #if defined __tile__ matching ARM and Solaris and then use the
Tile atomic instructions. Without this change, the default Mutex
implementation is used, which is slower.
---
 AUTHORS                |  1 +
 src/atomic_counter.hpp | 10 ++++++++++
 src/atomic_ptr.hpp     |  8 ++++++++
 3 files changed, 19 insertions(+)

diff --git a/AUTHORS b/AUTHORS
index 55d55614..232d77dc 100644
--- a/AUTHORS
+++ b/AUTHORS
@@ -41,6 +41,7 @@ Jacob Rideout <jacob.rideout@returnpath.net>
 Joe Thornber <joe.thornber@gmail.com>
 Jon Dyte <jon@totient.co.uk>
 Kamil Shakirov <kamils80@gmail.com>
+Ken Steele <ken@tilera.com>
 Marc Rossi <mrossi19@gmail.com>
 Martin Hurton <hurtonm@gmail.com>
 Martin Lucina <martin@lucina.net>
diff --git a/src/atomic_counter.hpp b/src/atomic_counter.hpp
index c87642a4..45cfe88d 100644
--- a/src/atomic_counter.hpp
+++ b/src/atomic_counter.hpp
@@ -33,6 +33,8 @@
 #define ZMQ_ATOMIC_COUNTER_WINDOWS
 #elif (defined ZMQ_HAVE_SOLARIS || defined ZMQ_HAVE_NETBSD)
 #define ZMQ_ATOMIC_COUNTER_ATOMIC_H
+#elif defined __tile__
+#define ZMQ_ATOMIC_COUNTER_TILE
 #else
 #define ZMQ_ATOMIC_COUNTER_MUTEX
 #endif
@@ -43,6 +45,8 @@
 #include "windows.hpp"
 #elif defined ZMQ_ATOMIC_COUNTER_ATOMIC_H
 #include <atomic.h>
+#elif defined ZMQ_ATOMIC_COUNTER_TILE
+#include <arch/atomic.h>
 #endif
 
 namespace zmq
@@ -82,6 +86,8 @@ namespace zmq
 #elif defined ZMQ_ATOMIC_COUNTER_ATOMIC_H
             integer_t new_value = atomic_add_32_nv (&value, increment_);
             old_value = new_value - increment_;
+#elif defined ZMQ_ATOMIC_COUNTER_TILE
+	    old_value = arch_atomic_add (&value, increment_);
 #elif defined ZMQ_ATOMIC_COUNTER_X86
             __asm__ volatile (
                 "lock; xadd %0, %1 \n\t"
@@ -123,6 +129,10 @@ namespace zmq
             int32_t delta = - ((int32_t) decrement);
             integer_t nv = atomic_add_32_nv (&value, delta);
             return nv != 0;
+#elif defined ZMQ_ATOMIC_COUNTER_TILE
+            int32_t delta = - ((int32_t) decrement);
+            integer_t nv = arch_atomic_add (&value, delta);
+            return nv != 0;
 #elif defined ZMQ_ATOMIC_COUNTER_X86
             integer_t oldval = -decrement;
             volatile integer_t *val = &value;
diff --git a/src/atomic_ptr.hpp b/src/atomic_ptr.hpp
index 59d78882..066d54be 100644
--- a/src/atomic_ptr.hpp
+++ b/src/atomic_ptr.hpp
@@ -28,6 +28,8 @@
 #define ZMQ_ATOMIC_PTR_X86
 #elif defined __ARM_ARCH_7A__ && defined __GNUC__
 #define ZMQ_ATOMIC_PTR_ARM
+#elif defined __tile__
+#define ZMQ_ATOMIC_PTR_TILE
 #elif defined ZMQ_HAVE_WINDOWS
 #define ZMQ_ATOMIC_PTR_WINDOWS
 #elif (defined ZMQ_HAVE_SOLARIS || defined ZMQ_HAVE_NETBSD)
@@ -42,6 +44,8 @@
 #include "windows.hpp"
 #elif defined ZMQ_ATOMIC_PTR_ATOMIC_H
 #include <atomic.h>
+#elif defined ZMQ_ATOMIC_PTR_TILE
+#include <arch/atomic.h>
 #endif
 
 namespace zmq
@@ -80,6 +84,8 @@ namespace zmq
             return (T*) InterlockedExchangePointer ((PVOID*) &ptr, val_);
 #elif defined ZMQ_ATOMIC_PTR_ATOMIC_H
             return (T*) atomic_swap_ptr (&ptr, val_);
+#elif defined ZMQ_ATOMIC_PTR_TILE
+            return (T*) arch_atomic_exchange (&ptr, val_);
 #elif defined ZMQ_ATOMIC_PTR_X86
             T *old;
             __asm__ volatile (
@@ -123,6 +129,8 @@ namespace zmq
                 (volatile PVOID*) &ptr, val_, cmp_);
 #elif defined ZMQ_ATOMIC_PTR_ATOMIC_H
             return (T*) atomic_cas_ptr (&ptr, cmp_, val_);
+#elif defined ZMQ_ATOMIC_PTR_TILE
+            return (T*) arch_atomic_val_compare_and_exchange (&ptr, cmp_, val_);
 #elif defined ZMQ_ATOMIC_PTR_X86
             T *old;
             __asm__ volatile (