From 4265304230b4abb966b2b162b522873678527178 Mon Sep 17 00:00:00 2001
From: Brian Barrett <brbarret@open-mpi.org>
Date: Mon, 14 Feb 2005 18:04:23 +0000
Subject: [PATCH] * Clean up some minor build system ick in the atomic
 operations * Make sure the update.sh scripts all use OMPI_WANT_SMP_LOCKS *
 Add sparc (32bit) assembly   - Memory barriers   - spinlocks   - emulate
 add/sub using the spinlock table with hashed lookups, as     suggested by the
 linux kernel folk (better than the other option,     requiring the counters
 only use 24 bits)

This commit was SVN r4429.
---
 configure.ac                        |   1 +
 src/asm/Makefile.am                 |   1 +
 src/asm/asm.c                       |  81 ++++++++++++++++
 src/include/sys/Makefile.am         |   2 +-
 src/include/sys/alpha/update.sh     |   1 +
 src/include/sys/amd64/update.sh     |   1 +
 src/include/sys/atomic.h            |  40 ++++++--
 src/include/sys/atomic_impl.h       |   3 +-
 src/include/sys/ia64/update.sh      |   1 +
 src/include/sys/sparc/Makefile.am   |  26 ++++++
 src/include/sys/sparc/atomic.h      | 140 ++++++++++++++++++++++++++++
 src/include/sys/sparc/update.sh     |  32 +++++++
 src/include/sys/sparc64/Makefile.am |   2 +-
 src/include/sys/sparc64/update.sh   |   1 +
 14 files changed, 322 insertions(+), 10 deletions(-)
 create mode 100644 src/asm/asm.c
 create mode 100644 src/include/sys/sparc/Makefile.am
 create mode 100644 src/include/sys/sparc/atomic.h
 create mode 100644 src/include/sys/sparc/update.sh

diff --git a/configure.ac b/configure.ac
index bcebb47f9b..9a9491e62b 100644
--- a/configure.ac
+++ b/configure.ac
@@ -1492,6 +1492,7 @@ AC_CONFIG_FILES([
     src/include/sys/ia32/Makefile
     src/include/sys/ia64/Makefile
     src/include/sys/powerpc/Makefile
+    src/include/sys/sparc/Makefile
     src/include/sys/sparc64/Makefile
     src/include/sys/win32/Makefile
 
diff --git a/src/asm/Makefile.am b/src/asm/Makefile.am
index f6b1104c0f..ef015f9e28 100644
--- a/src/asm/Makefile.am
+++ b/src/asm/Makefile.am
@@ -47,6 +47,7 @@ endif
 
 libasm_la_DEPENDENCIES = generated/@OMPI_ASM_FILE@
 noinst_LTLIBRARIES = libasm.la
+dist_libasm_la_SOURCES = asm.c
 
 EXTRA_DIST	= \
 	asm-data.txt \
diff --git a/src/asm/asm.c b/src/asm/asm.c
new file mode 100644
index 0000000000..a15c824728
--- /dev/null
+++ b/src/asm/asm.c
@@ -0,0 +1,81 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, 
+ *                         University of Stuttgart.  All rights reserved.
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#include "ompi_config.h"
+
+#include "include/sys/atomic.h"
+#include "include/sys/architecture.h"
+
+#if OMPI_ASSEMBLY_ARCH == OMPI_SPARC32
+
+#if OMPI_WANT_SMP_LOCKS
+
+#define LOCKS_TABLE_SIZE 8
+/* make sure to get into reasonably useful bits (so shift at least 5) */
+#define FIND_LOCK(addr) (&(locks_table[(((unsigned long) addr) >> 8) &  \
+                                       (LOCKS_TABLE_SIZE - 1)]))
+
+/* have to fix if you change LOCKS_TABLE_SIZE */
+static ompi_lock_t locks_table[LOCKS_TABLE_SIZE] = {
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED,
+    OMPI_ATOMIC_UNLOCKED
+};
+
+# else /* OMPI_WANT_SMP_LOCKS */
+
+#define LOCKS_TABLE_SIZE 1
+#define FIND_LOCK(addr) (&(locks_table[0]))
+
+static ompi_lock_t locks_table[1] = { OMPI_ATOMIC_UNLOCKED };
+
+#endif /* OMPI_WANT_SMP_LOCKS */
+
+
+int32_t
+ompi_atomic_sub_32(volatile int32_t *addr, int delta)
+{
+    int32_t ret;
+
+    ompi_atomic_lock(FIND_LOCK(addr));
+
+    ret = (*addr += delta);
+
+    ompi_atomic_unlock(FIND_LOCK(addr));
+
+    return ret;
+}
+
+
+int32_t
+ompi_atomic_sub_32(volatile int32_t *addr, int delta)
+{
+    int32_t ret;
+
+    ompi_atomic_lock(FIND_LOCK(addr));
+
+    ret = (*addr -= delta);
+
+    ompi_atomic_unlock(FIND_LOCK(addr));
+
+    return ret;
+}
+
+
+#endif /* OMPI_ASSEMBLY_ARCH == OMPI_SPARC32 */
diff --git a/src/include/sys/Makefile.am b/src/include/sys/Makefile.am
index d9b1510a2d..1cb9acb836 100644
--- a/src/include/sys/Makefile.am
+++ b/src/include/sys/Makefile.am
@@ -14,7 +14,7 @@
 
 include $(top_srcdir)/config/Makefile.options
 
-SUBDIRS = alpha amd64 ia32 ia64 powerpc sparc64 win32
+SUBDIRS = alpha amd64 ia32 ia64 powerpc sparc sparc64 win32
 
 noinst_HEADERS = atomic.h architecture.h atomic_impl.h cache.h
 
diff --git a/src/include/sys/alpha/update.sh b/src/include/sys/alpha/update.sh
index 9487684052..83737fcbdc 100644
--- a/src/include/sys/alpha/update.sh
+++ b/src/include/sys/alpha/update.sh
@@ -25,6 +25,7 @@ cat > $CFILE<<EOF
 #define static
 #define inline
 #define OMPI_GCC_INLINE_ASSEMBLY 1
+#define OMPI_WANT_SMP_LOCKS 1
 #include "atomic.h"
 EOF
 
diff --git a/src/include/sys/amd64/update.sh b/src/include/sys/amd64/update.sh
index 9487684052..83737fcbdc 100644
--- a/src/include/sys/amd64/update.sh
+++ b/src/include/sys/amd64/update.sh
@@ -25,6 +25,7 @@ cat > $CFILE<<EOF
 #define static
 #define inline
 #define OMPI_GCC_INLINE_ASSEMBLY 1
+#define OMPI_WANT_SMP_LOCKS 1
 #include "atomic.h"
 EOF
 
diff --git a/src/include/sys/atomic.h b/src/include/sys/atomic.h
index c64bc5ba7b..804d1899c9 100644
--- a/src/include/sys/atomic.h
+++ b/src/include/sys/atomic.h
@@ -77,7 +77,7 @@ extern "C" {
 #elif OMPI_ASSEMBLY_ARCH == OMPI_POWERPC64
 #include "include/sys/powerpc/atomic.h"
 #elif OMPI_ASSEMBLY_ARCH == OMPI_SPARC32
-#error "32 bit Sparc support not implemented yet"
+#include "include/sys/sparc/atomic.h"
 #elif OMPI_ASSEMBLY_ARCH == OMPI_SPARC64
 #include "include/sys/sparc64/atomic.h"
 #endif
@@ -153,20 +153,28 @@ void ompi_atomic_wmb(void);
  *********************************************************************/
 /**
  * Volatile lock object (with optional padding).
+ *
+ * \note The internals of the lock are included here, but should be
+ * considered private.  The implementation currently in use may choose
+ * to use an int or unsigned char as the lock value - the user is not
+ * informed either way.
  */
 struct ompi_lock_t {
     union {
         volatile int lock;         /**< The lock address (an integer) */
+        volatile unsigned char sparc_lock; /**< The lock address on sparc */
         char padding[sizeof(int)]; /**< Array for optional padding */
     } u;
 };
 typedef struct ompi_lock_t ompi_lock_t;
 
 #if !defined(OMPI_HAVE_ATOMIC_SPINLOCKS) && !defined(DOXYGEN)
-#define OMPI_HAVE_ATOMIC_SPINLOCKS (OMPI_HAVE_ATOMIC_CMPSET_32 || OMPI_HAVE_ATOMIC_CMPSET_64)
+/* 0 is more like "pending" - we'll fix up at the end after all
+   the static inline functions are declared */
+#define OMPI_HAVE_ATOMIC_SPINLOCKS 0
 #endif
 
-#if defined(DOXYGEN) || OMPI_HAVE_ATOMIC_SPINLOCKS
+#if defined(DOXYGEN) || OMPI_HAVE_ATOMIC_SPINLOCKS || (OMPI_HAVE_ATOMIC_CMPSET_32 || OMPI_HAVE_ATOMIC_CMPSET_64)
 
 /**
  * Enumeration of lock states
@@ -183,7 +191,10 @@ enum {
  * @param lock         Address of the lock
  * @param value        Initial value to set lock to
  */
-static inline void ompi_atomic_init(ompi_lock_t* lock, int value);
+#if OMPI_HAVE_ATOMIC_SPINLOCKS == 0
+static inline 
+#endif
+void ompi_atomic_init(ompi_lock_t* lock, int value);
 
 
 /**
@@ -192,7 +203,10 @@ static inline void ompi_atomic_init(ompi_lock_t* lock, int value);
  * @param lock          Address of the lock.
  * @return              0 if the lock was acquired, 1 otherwise.
  */
-static inline int ompi_atomic_trylock(ompi_lock_t *lock);
+#if OMPI_HAVE_ATOMIC_SPINLOCKS == 0
+static inline
+#endif
+int ompi_atomic_trylock(ompi_lock_t *lock);
 
 
 /**
@@ -200,7 +214,10 @@ static inline int ompi_atomic_trylock(ompi_lock_t *lock);
  *
  * @param lock          Address of the lock.
  */
-static inline void ompi_atomic_lock(ompi_lock_t *lock);
+#if OMPI_HAVE_ATOMIC_SPINLOCKS == 0
+static inline
+#endif
+void ompi_atomic_lock(ompi_lock_t *lock);
 
 
 /**
@@ -208,7 +225,16 @@ static inline void ompi_atomic_lock(ompi_lock_t *lock);
  *
  * @param lock          Address of the lock.
  */
-static inline void ompi_atomic_unlock(ompi_lock_t *lock);
+#if OMPI_HAVE_ATOMIC_SPINLOCKS == 0
+static inline
+#endif
+void ompi_atomic_unlock(ompi_lock_t *lock);
+
+
+#if OMPI_HAVE_ATOMIC_SPINLOCKS == 0
+#define OMPI_HAVE_ATOMIC_SPINLOCKS (OMPI_HAVE_ATOMIC_CMPSET_32 || OMPI_HAVE_ATOMIC_CMPSET_64)
+#define OMPI_NEED_INLINE_ATOMIC_SPINLOCKS
+#endif
 
 #endif /* OMPI_HAVE_ATOMIC_SPINLOCKS */
 
diff --git a/src/include/sys/atomic_impl.h b/src/include/sys/atomic_impl.h
index f6452a8107..ed8437b083 100644
--- a/src/include/sys/atomic_impl.h
+++ b/src/include/sys/atomic_impl.h
@@ -307,7 +307,8 @@ static inline int ompi_atomic_sub_ptr(volatile void* addr,
  * Atomic spinlocks
  *
  *********************************************************************/
-#if OMPI_HAVE_ATOMIC_SPINLOCKS
+#ifdef OMPI_NEED_INLINE_ATOMIC_SPINLOCKS
+
 /* 
  * Lock initialization function. It set the lock to UNLOCKED.
  */
diff --git a/src/include/sys/ia64/update.sh b/src/include/sys/ia64/update.sh
index 9487684052..83737fcbdc 100644
--- a/src/include/sys/ia64/update.sh
+++ b/src/include/sys/ia64/update.sh
@@ -25,6 +25,7 @@ cat > $CFILE<<EOF
 #define static
 #define inline
 #define OMPI_GCC_INLINE_ASSEMBLY 1
+#define OMPI_WANT_SMP_LOCKS 1
 #include "atomic.h"
 EOF
 
diff --git a/src/include/sys/sparc/Makefile.am b/src/include/sys/sparc/Makefile.am
new file mode 100644
index 0000000000..90155cc214
--- /dev/null
+++ b/src/include/sys/sparc/Makefile.am
@@ -0,0 +1,26 @@
+#
+# Copyright (c) 2004-2005 The Trustees of Indiana University.
+#                         All rights reserved.
+# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+#                         All rights reserved.
+# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, 
+#                         University of Stuttgart.  All rights reserved.
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+include $(top_srcdir)/config/Makefile.options
+
+noinst_HEADERS = atomic.h
+
+# Conditionally install the header files
+
+if WANT_INSTALL_HEADERS
+ompidir = $(includedir)/openmpi/include/sys/sparc
+ompi_HEADERS = $(noinst_HEADERS)
+else
+ompidir = $(includedir)
+endif
diff --git a/src/include/sys/sparc/atomic.h b/src/include/sys/sparc/atomic.h
new file mode 100644
index 0000000000..b4a1bfb9ff
--- /dev/null
+++ b/src/include/sys/sparc/atomic.h
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, 
+ *                         University of Stuttgart.  All rights reserved.
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#ifndef OMPI_SYS_ARCH_ATOMIC_H
+#define OMPI_SYS_ARCH_ATOMIC_H 1
+
+
+#if OMPI_WANT_SMP_LOCKS
+#define MB() __asm__  __volatile__ ("" : : : "memory")
+#else
+#define MB()
+#endif
+
+
+/**********************************************************************
+ *
+ * Define constants for UltraSparc 64
+ *
+ *********************************************************************/
+#define OMPI_HAVE_MEM_BARRIER 1
+
+#define OMPI_HAVE_ATOMIC_CMPSET_32 0
+#define OMPI_HAVE_ATOMIC_CMPSET_64 0
+
+#define OMPI_HAVE_ATOMIC_MATH_32 1
+#define OMPI_HAVE_ATOMIC_SUB_32 1
+#define OMPI_HAVE_ATOMIC_ADD_32 1
+
+#define OMPI_HAVE_ATOMIC_SPINLOCKS 1
+
+/**********************************************************************
+ *
+ * Memory Barriers
+ *
+ *********************************************************************/
+#if OMPI_GCC_INLINE_ASSEMBLY
+
+static inline void ompi_atomic_mb(void)
+{
+    MB();
+}
+
+
+static inline void ompi_atomic_rmb(void)
+{
+    MB();
+}
+
+
+static inline void ompi_atomic_wmb(void)
+{
+    MB();
+}
+
+#endif /* OMPI_GCC_INLINE_ASSEMBLY */
+
+
+/**********************************************************************
+ *
+ * Atomic spinlocks
+ *
+ *********************************************************************/
+#if OMPI_GCC_INLINE_ASSEMBLY
+
+/* for these, the lock is held whenever lock.sparc_lock != 0.  We
+   attempt to leave it as OMPI_LOCKED whenever possible */
+
+
+static inline void ompi_atomic_init(ompi_lock_t* lock, int value)
+{
+    lock->sparc_lock = (unsigned char) value;
+}
+
+
+static inline int ompi_atomic_trylock(ompi_lock_t *lock)
+{
+    unsigned char result;
+
+    /* try to load the lock byte (atomically making the memory byte
+       contain all 1s).  If the byte used to be 0, we now have the
+       lock.  Otherwise, someone else has the lock.  Either way, the
+       lock is now held. */
+    __asm__ __volatile__ ("\t"
+                          "ldstub [%1], %0"
+                          : "=r"(result)
+                          : "r"(&(lock->sparc_lock))
+                          : "memory");
+    return (result == 0);
+}
+
+
+static inline void ompi_atomic_lock(ompi_lock_t *lock)
+{
+    /* From page 264 of The SPARC Architecture Manual, Version 8 */
+    __asm__ __volatile__ (
+                          "retry:               \n\t"
+                          "ldstub [%0], %%l0    \n\t"
+                          "tst    %%l0          \n\t"
+                          "bw     out           \n\t"
+                          "nop                  \n"
+                          "loop:                \n\t"
+                          "ldb    [%0], %%l0    \n\t"
+                          "tst    %%l0          \n\t"
+                          "bne    loop          \n\t"
+                          "nop                  \n\t"
+                          "ba,a   retry         \n"
+                          "out:                 \n\t"
+                          "nop"
+                          :
+                          : "r"(&(lock->sparc_lock))
+                          : "%l0", "memory");
+}
+
+
+static inline void ompi_atomic_unlock(ompi_lock_t *lock)
+{
+    /* 0 out that byte in memory */
+    __asm__ __volatile__ ("\t"
+                          "stbar             \n\t"
+                          "stb %%g0, [%0]    \n\t"
+                          :
+                          : "r"(&(lock->sparc_lock))
+                          : "memory");
+}
+
+#endif /* OMPI_GCC_INLINE_ASSEMBLY */
+
+
+#endif /* ! OMPI_SYS_ARCH_ATOMIC_H */
diff --git a/src/include/sys/sparc/update.sh b/src/include/sys/sparc/update.sh
new file mode 100644
index 0000000000..83737fcbdc
--- /dev/null
+++ b/src/include/sys/sparc/update.sh
@@ -0,0 +1,32 @@
+#!/bin/sh
+#
+# Copyright (c) 2004-2005 The Trustees of Indiana University.
+#                         All rights reserved.
+# Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+#                         All rights reserved.
+# Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, 
+#                         University of Stuttgart.  All rights reserved.
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+CFILE=/tmp/ompi_atomic_$$.c
+
+trap "/bin/rm -f $CFILE; exit 0" 0 1 2 15
+
+echo Updating atomic.s from atomic.h using gcc
+
+cat > $CFILE<<EOF
+#include <stdlib.h>
+#include <inttypes.h>
+#define static
+#define inline
+#define OMPI_GCC_INLINE_ASSEMBLY 1
+#define OMPI_WANT_SMP_LOCKS 1
+#include "atomic.h"
+EOF
+
+gcc -I. -S $CFILE -o atomic.s
diff --git a/src/include/sys/sparc64/Makefile.am b/src/include/sys/sparc64/Makefile.am
index 034552fb4f..ff5ceadb3b 100644
--- a/src/include/sys/sparc64/Makefile.am
+++ b/src/include/sys/sparc64/Makefile.am
@@ -14,7 +14,7 @@
 
 include $(top_srcdir)/config/Makefile.options
 
-noinst_HEADERS = atomic.h atomic.s
+noinst_HEADERS = atomic.h
 
 # Conditionally install the header files
 
diff --git a/src/include/sys/sparc64/update.sh b/src/include/sys/sparc64/update.sh
index 9487684052..83737fcbdc 100644
--- a/src/include/sys/sparc64/update.sh
+++ b/src/include/sys/sparc64/update.sh
@@ -25,6 +25,7 @@ cat > $CFILE<<EOF
 #define static
 #define inline
 #define OMPI_GCC_INLINE_ASSEMBLY 1
+#define OMPI_WANT_SMP_LOCKS 1
 #include "atomic.h"
 EOF