OSHMEM v1.3: adds fetch and set amos and c11 generic

oshmem/include/pshmemx.h

@@ -50,12 +50,10 @@ OSHMEM_DECLSPEC  void pshmemx_int64_p(int64_t* addr, int64_t value, int pe);
 /*
  * Block data put routines
  */
-OSHMEM_DECLSPEC  void pshmemx_put16(void *target, const void *source, size_t len, int pe);
 
 /*
  * Strided put routines
  */
-OSHMEM_DECLSPEC void pshmemx_iput16(void* target, const void* source, ptrdiff_t tst, ptrdiff_t sst,size_t len, int pe);
 
 /*
  * Elemental get routines
@@ -67,12 +65,10 @@ OSHMEM_DECLSPEC  int64_t pshmemx_int64_g(const int64_t* addr, int pe);
 /*
  * Block data get routines
  */
-OSHMEM_DECLSPEC  void pshmemx_get16(void *target, const void *source, size_t len, int pe);
 
 /*
  * Strided get routines
  */
-OSHMEM_DECLSPEC void pshmemx_iget16(void* target, const void* source, ptrdiff_t tst, ptrdiff_t sst,size_t len, int pe);
 
 /*
  * Atomic operations
@@ -81,6 +77,10 @@ OSHMEM_DECLSPEC void pshmemx_iget16(void* target, const void* source, ptrdiff_t
 OSHMEM_DECLSPEC int32_t pshmemx_int32_swap(int32_t *target, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t pshmemx_int64_swap(int64_t *target, int64_t value, int pe);
 
+/*Atomic set */
+OSHMEM_DECLSPEC void pshmemx_int32_set(int32_t *target, int32_t value, int pe);
+OSHMEM_DECLSPEC void pshmemx_int64_set(int64_t *target, int64_t value, int pe);
+
 /* Atomic conditional swap */
 OSHMEM_DECLSPEC int32_t pshmemx_int32_cswap(int32_t *target, int32_t cond, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t pshmemx_int64_cswap(int64_t *target, int64_t cond, int64_t value, int pe);
@@ -89,6 +89,10 @@ OSHMEM_DECLSPEC int64_t pshmemx_int64_cswap(int64_t *target, int64_t cond, int64
 OSHMEM_DECLSPEC int32_t pshmemx_int32_fadd(int32_t *target, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t pshmemx_int64_fadd(int64_t *target, int64_t value, int pe);
 
+/* Atomic Fetch */
+OSHMEM_DECLSPEC int32_t pshmemx_int32_fetch(const int32_t *target, int pe);
+OSHMEM_DECLSPEC int64_t pshmemx_int64_fetch(const int64_t *target, int pe);
+
 /* Atomic Fetch&Inc */
 OSHMEM_DECLSPEC int32_t pshmemx_int32_finc(int32_t *target, int pe);
 OSHMEM_DECLSPEC int64_t pshmemx_int64_finc(int64_t *target, int pe);
@@ -148,27 +152,31 @@ OSHMEM_DECLSPEC void pshmemx_int64_prod_to_all(int64_t *target, const int64_t *s
 #define pshmem_int32_p               pshmemx_int32_p
 #define pshmem_int64_p               pshmemx_int64_p
 
-#define pshmem_put16                 pshmemx_put16
-#define pshmem_iput16                pshmemx_iput16
-
 #define pshmem_int16_g               pshmemx_int16_g
 #define pshmem_int32_g               pshmemx_int32_g
 #define pshmem_int64_g               pshmemx_int64_g
 
-#define pshmem_get16                 pshmemx_get16
-#define pshmem_iget16                pshmemx_iget16
-
 #define pshmem_int32_swap            pshmemx_int32_swap
 #define pshmem_int64_swap            pshmemx_int64_swap
+
+#define pshmem_int32_set             pshmemx_int32_set
+#define pshmem_int64_set             pshmemx_int64_set
+
 #define pshmem_int32_cswap           pshmemx_int32_cswap
 #define pshmem_int64_cswap           pshmemx_int64_cswap
 
 #define pshmem_int32_fadd            pshmemx_int32_fadd
 #define pshmem_int64_fadd            pshmemx_int64_fadd
+
+#define pshmem_int32_fetch           pshmemx_int32_fetch
+#define pshmem_int64_fetch           pshmemx_int64_fetch
+
 #define pshmem_int32_finc            pshmemx_int32_finc
 #define pshmem_int64_finc            pshmemx_int64_finc
+
 #define pshmem_int32_add             pshmemx_int32_add
 #define pshmem_int64_add             pshmemx_int64_add
+
 #define pshmem_int32_inc             pshmemx_int32_inc
 #define pshmem_int64_inc             pshmemx_int64_inc
 

oshmem/include/shmemx.h

@@ -35,12 +35,10 @@ OSHMEM_DECLSPEC  void shmemx_int64_p(int64_t* addr, int64_t value, int pe);
 /*
  * Block data put routines
  */
-OSHMEM_DECLSPEC  void shmemx_put16(void *target, const void *source, size_t len, int pe);
 
 /*
  * Strided put routines
  */
-OSHMEM_DECLSPEC void shmemx_iput16(void* target, const void* source, ptrdiff_t tst, ptrdiff_t sst,size_t len, int pe);
 
 /*
  * Elemental get routines
@@ -52,12 +50,10 @@ OSHMEM_DECLSPEC  int64_t shmemx_int64_g(const int64_t* addr, int pe);
 /*
  * Block data get routines
  */
-OSHMEM_DECLSPEC  void shmemx_get16(void *target, const void *source, size_t len, int pe);
 
 /*
  * Strided get routines
  */
-OSHMEM_DECLSPEC void shmemx_iget16(void* target, const void* source, ptrdiff_t tst, ptrdiff_t sst,size_t len, int pe);
 
 /*
  * Atomic operations
@@ -66,6 +62,10 @@ OSHMEM_DECLSPEC void shmemx_iget16(void* target, const void* source, ptrdiff_t t
 OSHMEM_DECLSPEC int32_t shmemx_int32_swap(int32_t *target, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t shmemx_int64_swap(int64_t *target, int64_t value, int pe);
 
+/* Atomic set */
+OSHMEM_DECLSPEC void shmemx_int32_set(int32_t *target, int32_t value, int pe);
+OSHMEM_DECLSPEC void shmemx_int64_set(int64_t *target, int64_t value, int pe);
+
 /* Atomic conditional swap */
 OSHMEM_DECLSPEC int32_t shmemx_int32_cswap(int32_t *target, int32_t cond, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t shmemx_int64_cswap(int64_t *target, int64_t cond, int64_t value, int pe);
@@ -74,6 +74,10 @@ OSHMEM_DECLSPEC int64_t shmemx_int64_cswap(int64_t *target, int64_t cond, int64_
 OSHMEM_DECLSPEC int32_t shmemx_int32_fadd(int32_t *target, int32_t value, int pe);
 OSHMEM_DECLSPEC int64_t shmemx_int64_fadd(int64_t *target, int64_t value, int pe);
 
+/* Atomic Fetch */
+OSHMEM_DECLSPEC int32_t shmemx_int32_fetch(const int32_t *target, int pe);
+OSHMEM_DECLSPEC int64_t shmemx_int64_fetch(const int64_t *target, int pe);
+
 /* Atomic Fetch&Inc */
 OSHMEM_DECLSPEC int32_t shmemx_int32_finc(int32_t *target, int pe);
 OSHMEM_DECLSPEC int64_t shmemx_int64_finc(int64_t *target, int pe);
@@ -133,25 +137,28 @@ OSHMEM_DECLSPEC void shmemx_int64_prod_to_all(int64_t *target, const int64_t *so
 #define shmem_int32_p               shmemx_int32_p
 #define shmem_int64_p               shmemx_int64_p
 
-#define shmem_put16                 shmemx_put16
-#define shmem_iput16                shmemx_iput16
-
 #define shmem_int16_g               shmemx_int16_g
 #define shmem_int32_g               shmemx_int32_g
 #define shmem_int64_g               shmemx_int64_g
 
-#define shmem_get16                 shmemx_get16
-#define shmem_iget16                shmemx_iget16
-
 #define shmem_int32_swap            shmemx_int32_swap
 #define shmem_int64_swap            shmemx_int64_swap
+
+#define shmem_int32_set             shmemx_int32_set
+#define shmem_int64_set             shmemx_int64_set
+
 #define shmem_int32_cswap           shmemx_int32_cswap
 #define shmem_int64_cswap           shmemx_int64_cswap
 
 #define shmem_int32_fadd            shmemx_int32_fadd
 #define shmem_int64_fadd            shmemx_int64_fadd
+
+#define shmem_int32_fetch           shmemx_int32_fetch
+#define shmem_int64_fetch           shmemx_int64_fetch
+
 #define shmem_int32_finc            shmemx_int32_finc
 #define shmem_int64_finc            shmemx_int64_finc
+
 #define shmem_int32_add             shmemx_int32_add
 #define shmem_int64_add             shmemx_int64_add
 #define shmem_int32_inc             shmemx_int32_inc

oshmem/shmem/c/Makefile.am

@@ -51,8 +51,10 @@ OSHMEM_API_SOURCES = \
 	shmem_clear_cache_line_inv.c \
 	shmem_reduce.c \
 	shmem_swap.c \
+	shmem_set.c \
 	shmem_cswap.c \
 	shmem_fadd.c \
+	shmem_fetch.c \
 	shmem_finc.c \
 	shmem_add.c \
 	shmem_inc.c \

oshmem/shmem/c/profile/Makefile.am

@@ -63,8 +63,10 @@ OSHMEM_API_SOURCES = \
 	pshmem_clear_cache_line_inv.c \
 	pshmem_reduce.c \
 	pshmem_swap.c \
+	pshmem_set.c \
 	pshmem_cswap.c \
 	pshmem_fadd.c \
+	pshmem_fetch.c \
 	pshmem_finc.c \
 	pshmem_add.c \
 	pshmem_inc.c \

oshmem/shmem/c/profile/defines.h

@@ -86,7 +86,9 @@
 #define shmem_double_put             pshmem_double_put
 #define shmem_longlong_put           pshmem_longlong_put
 #define shmem_longdouble_put         pshmem_longdouble_put
-#define shmemx_put16                 pshmemx_put16
+
+#define shmem_put8                   pshmem_put8
+#define shmem_put16                  pshmem_put16
 #define shmem_put32                  pshmem_put32
 #define shmem_put64                  pshmem_put64
 #define shmem_put128                 pshmem_put128
@@ -95,14 +97,17 @@
 /*
  * Strided put routines
  */
-#define shmem_int_iput               pshmem_int_iput
+#define shmem_char_iput              pshmem_char_iput
 #define shmem_short_iput             pshmem_short_iput
+#define shmem_int_iput               pshmem_int_iput
 #define shmem_float_iput             pshmem_float_iput
 #define shmem_double_iput            pshmem_double_iput
 #define shmem_longlong_iput          pshmem_longlong_iput
 #define shmem_longdouble_iput        pshmem_longdouble_iput
 #define shmem_long_iput              pshmem_long_iput
-#define shmemx_iput16                pshmemx_iput16
+
+#define shmem_iput8                  pshmem_iput8
+#define shmem_iput16                 pshmem_iput16
 #define shmem_iput32                 pshmem_iput32
 #define shmem_iput64                 pshmem_iput64
 #define shmem_iput128                pshmem_iput128
@@ -151,7 +156,9 @@
 #define shmem_double_get             pshmem_double_get
 #define shmem_longlong_get           pshmem_longlong_get
 #define shmem_longdouble_get         pshmem_longdouble_get
-#define shmemx_get16                 pshmemx_get16
+
+#define shmem_get8                   pshmem_get8
+#define shmem_get16                  pshmem_get16
 #define shmem_get32                  pshmem_get32
 #define shmem_get64                  pshmem_get64
 #define shmem_get128                 pshmem_get128
@@ -160,14 +167,17 @@
 /*
  * Strided get routines
  */
-#define shmem_int_iget               pshmem_int_iget
+#define shmem_char_iget              pshmem_char_iget
 #define shmem_short_iget             pshmem_short_iget
+#define shmem_int_iget               pshmem_int_iget
 #define shmem_float_iget             pshmem_float_iget
 #define shmem_double_iget            pshmem_double_iget
 #define shmem_longlong_iget          pshmem_longlong_iget
 #define shmem_longdouble_iget        pshmem_longdouble_iget
 #define shmem_long_iget              pshmem_long_iget
-#define shmemx_iget16                pshmemx_iget16
+
+#define shmem_iget8                  pshmem_iget8
+#define shmem_iget16                 pshmem_iget16
 #define shmem_iget32                 pshmem_iget32
 #define shmem_iget64                 pshmem_iget64
 #define shmem_iget128                pshmem_iget128
@@ -194,7 +204,6 @@
  * Atomic operations
  */
 /* Atomic swap */
-#define shmem_swap                   pshmem_swap
 #define shmem_double_swap            pshmem_double_swap
 #define shmem_float_swap             pshmem_float_swap
 #define shmem_int_swap               pshmem_int_swap
@@ -203,6 +212,14 @@
 #define shmemx_int32_swap            pshmemx_int32_swap
 #define shmemx_int64_swap            pshmemx_int64_swap
 
+/* Atomic set */
+#define shmem_double_set            pshmem_double_set
+#define shmem_float_set             pshmem_float_set
+#define shmem_int_set               pshmem_int_set
+#define shmem_long_set              pshmem_long_set
+#define shmem_longlong_set          pshmem_longlong_set
+#define shmemx_int32_set            pshmemx_int32_set
+#define shmemx_int64_set            pshmemx_int64_set
 
 /* Atomic conditional swap */
 #define shmem_int_cswap              pshmem_int_cswap
@@ -219,6 +236,15 @@
 #define shmemx_int32_fadd            pshmemx_int32_fadd
 #define shmemx_int64_fadd            pshmemx_int64_fadd
 
+/* Atomic Fetch */
+#define shmem_double_fetch            pshmem_double_fetch
+#define shmem_float_fetch             pshmem_float_fetch
+#define shmem_int_fetch               pshmem_int_fetch
+#define shmem_long_fetch              pshmem_long_fetch
+#define shmem_longlong_fetch          pshmem_longlong_fetch
+#define shmemx_int32_fetch            pshmemx_int32_fetch
+#define shmemx_int64_fetch            pshmemx_int64_fetch
+
 /* Atomic Fetch&Inc */
 #define shmem_int_finc               pshmem_int_finc
 #define shmem_long_finc              pshmem_long_finc

oshmem/shmem/c/shmem_fetch.c

@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2016      Mellanox Technologies, Inc.
+ *                         All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+#include "oshmem_config.h"
+
+#include "oshmem/constants.h"
+#include "oshmem/include/shmem.h"
+
+#include "oshmem/runtime/runtime.h"
+
+#include "oshmem/op/op.h"
+#include "oshmem/mca/atomic/atomic.h"
+
+/*
+ * These routines perform an atomic fetch operation.
+ * The fetch routines retrieve the value at address target on PE pe.
+ * The operation must be completed without the possibility of another process 
+ * updating target during the fetch.
+ */
+#define SHMEM_TYPE_FETCH(type_name, type, prefix)    \
+    type prefix##type_name##_fetch(const type *target, int pe) \
+    {                                                               \
+        int rc = OSHMEM_SUCCESS;                                    \
+        size_t size = 0;                                            \
+        type out_value;                                             \
+        type value = 0;                                             \
+        oshmem_op_t* op = oshmem_op_sum##type_name;                 \
+                                                                    \
+        RUNTIME_CHECK_INIT();                                       \
+        RUNTIME_CHECK_PE(pe);                                       \
+        RUNTIME_CHECK_ADDR(target);                                 \
+                                                                    \
+        size = sizeof(out_value);                                   \
+        rc = MCA_ATOMIC_CALL(fadd(                                  \
+            (void*)target,                                          \
+            (void*)&out_value,                                      \
+            (const void*)&value,                                    \
+            size,                                                   \
+            pe,                                                     \
+            op));                                                   \
+        RUNTIME_CHECK_RC(rc);                                       \
+                                                                    \
+        return out_value;                                           \
+    }
+
+#if OSHMEM_PROFILING
+#include "oshmem/include/pshmem.h"
+#pragma weak shmem_int_fetch = pshmem_int_fetch
+#pragma weak shmem_long_fetch = pshmem_long_fetch
+#pragma weak shmem_longlong_fetch = pshmem_longlong_fetch
+#pragma weak shmem_double_fetch = pshmem_double_fetch
+#pragma weak shmem_float_fetch = pshmem_float_fetch
+#pragma weak shmemx_int32_fetch = pshmemx_int32_fetch
+#pragma weak shmemx_int64_fetch = pshmemx_int64_fetch
+#include "oshmem/shmem/c/profile/defines.h"
+#endif
+
+SHMEM_TYPE_FETCH(_int, int, shmem)
+SHMEM_TYPE_FETCH(_long, long, shmem)
+SHMEM_TYPE_FETCH(_longlong, long long, shmem)
+SHMEM_TYPE_FETCH(_double, double, shmem)
+SHMEM_TYPE_FETCH(_float, float, shmem)
+SHMEM_TYPE_FETCH(_int32, int32_t, shmemx)
+SHMEM_TYPE_FETCH(_int64, int64_t, shmemx)
+

oshmem/shmem/c/shmem_get.c

@@ -54,7 +54,8 @@
 #pragma weak shmem_double_get = pshmem_double_get
 #pragma weak shmem_longdouble_get = pshmem_longdouble_get
 #pragma weak shmem_getmem = pshmem_getmem
-#pragma weak shmemx_get16 = pshmemx_get16
+#pragma weak shmem_get8 = pshmem_get8
+#pragma weak shmem_get16 = pshmem_get16
 #pragma weak shmem_get32 = pshmem_get32
 #pragma weak shmem_get64 = pshmem_get64
 #pragma weak shmem_get128 = pshmem_get128
@@ -92,9 +93,9 @@ SHMEM_TYPE_GET(_longdouble, long double)
     }
 
 SHMEM_TYPE_GETMEM(_getmem, 1, shmem)
-SHMEM_TYPE_GETMEM(_get16, 2, shmemx)
+SHMEM_TYPE_GETMEM(_get8, 1, shmem)
+SHMEM_TYPE_GETMEM(_get16, 2, shmem)
 SHMEM_TYPE_GETMEM(_get32, 4, shmem)
 SHMEM_TYPE_GETMEM(_get64, 8, shmem)
 SHMEM_TYPE_GETMEM(_get128, 16, shmem)
 
-SHMEM_TYPE_GETMEM(_get, sizeof(long), shmem)