Make atomic cswap and swap mandatory

The compare and swap and swap atomic interfaces are de facto
mandatory, as no code actually checks for their existance before
using them (with the notable exception of the 128 bit cswap), so
make those interfaces mandatory and remove all the complex code
to detect implementations of them.

At the same time, reorder the implementation files to all have
consistent ordering of the cswap and swap implementations, to
ease maintenance and move to a implementation includes the
compatibility layer mode instead of the previous overly complex
all in one software implementation model.

Finally, remove the untyped macro wrappers as 1) no one uses
them and 2) they're a pain to maintain.

Signed-off-by: Brian Barrett <[email protected]>

Author: bwbarrett

opal/include/opal/sys/Makefile.am

@@ -16,6 +16,8 @@
 # Copyright (c) 2017      Research Organization for Information Science
 #                         and Technology (RIST). All rights reserved.
 # Copyright (c) 2020-2021 Google, LLC. All rights reserved.
+# Copyright (c) 2022      Amazon.com, Inc. or its affiliates.
+#                         All Rights reserved.
 # $COPYRIGHT$
 #
 # Additional copyrights may follow
@@ -30,6 +32,8 @@ headers += \
 	opal/sys/atomic.h \
 	opal/sys/atomic_stdc.h \
 	opal/sys/atomic_impl.h \
+	opal/sys/atomic_impl_ptr_cswap.h \
+	opal/sys/atomic_impl_ptr_swap.h \
 	opal/sys/timer.h \
         opal/sys/cma.h
 

opal/include/opal/sys/arm64/atomic.h

@@ -30,10 +30,6 @@
 #ifndef OPAL_SYS_ARCH_ATOMIC_H
 #define OPAL_SYS_ARCH_ATOMIC_H 1
 
-#define OPAL_HAVE_ATOMIC_COMPARE_EXCHANGE_32 1
-#define OPAL_HAVE_ATOMIC_SWAP_32             1
-#define OPAL_HAVE_ATOMIC_COMPARE_EXCHANGE_64 1
-#define OPAL_HAVE_ATOMIC_SWAP_64             1
 #define OPAL_HAVE_ATOMIC_ADD_32              1
 #define OPAL_HAVE_ATOMIC_AND_32              1
 #define OPAL_HAVE_ATOMIC_OR_32               1
@@ -72,9 +68,10 @@ static inline void opal_atomic_isync(void)
     __asm__ __volatile__("isb");
 }
 
+
 /**********************************************************************
  *
- * Atomic math operations
+ * Compare and Swap
  *
  *********************************************************************/
 
@@ -99,20 +96,6 @@ static inline bool opal_atomic_compare_exchange_strong_32(opal_atomic_int32_t *a
     return ret;
 }
 
-static inline int32_t opal_atomic_swap_32(opal_atomic_int32_t *addr, int32_t newval)
-{
-    int32_t ret, tmp;
-
-    __asm__ __volatile__("1:  ldaxr   %w0, [%2]       \n"
-                         "    stlxr   %w1, %w3, [%2]  \n"
-                         "    cbnz    %w1, 1b         \n"
-                         : "=&r"(ret), "=&r"(tmp)
-                         : "r"(addr), "r"(newval)
-                         : "cc", "memory");
-
-    return ret;
-}
-
 /* these two functions aren't inlined in the non-gcc case because then
    there would be two function calls (since neither cmpset_32 nor
    atomic_?mb can be inlined).  Instead, we "inline" them by hand in
@@ -182,21 +165,6 @@ static inline bool opal_atomic_compare_exchange_strong_64(opal_atomic_int64_t *a
     return ret;
 }
 
-static inline int64_t opal_atomic_swap_64(opal_atomic_int64_t *addr, int64_t newval)
-{
-    int64_t ret;
-    int tmp;
-
-    __asm__ __volatile__("1:  ldaxr   %0, [%2]        \n"
-                         "    stlxr   %w1, %3, [%2]   \n"
-                         "    cbnz    %w1, 1b         \n"
-                         : "=&r"(ret), "=&r"(tmp)
-                         : "r"(addr), "r"(newval)
-                         : "cc", "memory");
-
-    return ret;
-}
-
 /* these two functions aren't inlined in the non-gcc case because then
    there would be two function calls (since neither cmpset_64 nor
    atomic_?mb can be inlined).  Instead, we "inline" them by hand in
@@ -246,6 +214,53 @@ static inline bool opal_atomic_compare_exchange_strong_rel_64(opal_atomic_int64_
     return ret;
 }
 
+#include "opal/sys/atomic_impl_ptr_cswap.h"
+
+
+/**********************************************************************
+ *
+ * Swap
+ *
+ *********************************************************************/
+
+static inline int32_t opal_atomic_swap_32(opal_atomic_int32_t *addr, int32_t newval)
+{
+    int32_t ret, tmp;
+
+    __asm__ __volatile__("1:  ldaxr   %w0, [%2]       \n"
+                         "    stlxr   %w1, %w3, [%2]  \n"
+                         "    cbnz    %w1, 1b         \n"
+                         : "=&r"(ret), "=&r"(tmp)
+                         : "r"(addr), "r"(newval)
+                         : "cc", "memory");
+
+    return ret;
+}
+
+static inline int64_t opal_atomic_swap_64(opal_atomic_int64_t *addr, int64_t newval)
+{
+    int64_t ret;
+    int tmp;
+
+    __asm__ __volatile__("1:  ldaxr   %0, [%2]        \n"
+                         "    stlxr   %w1, %3, [%2]   \n"
+                         "    cbnz    %w1, 1b         \n"
+                         : "=&r"(ret), "=&r"(tmp)
+                         : "r"(addr), "r"(newval)
+                         : "cc", "memory");
+
+    return ret;
+}
+
+#include "opal/sys/atomic_impl_ptr_swap.h"
+
+
+/**********************************************************************
+ *
+ * Atomic math operations
+ *
+ *********************************************************************/
+
 #define OPAL_ASM_MAKE_ATOMIC(type, bits, name, inst, reg)                          \
     static inline type opal_atomic_fetch_##name##_##bits(opal_atomic_##type *addr, \
                                                          type value)               \