[refactor] move dma implem to separate headers

The overall strategy for now is to split the implementations into
distinct headers, but keep the generic APIs inside the main aml.h

Related to #27

benchmarks/dgemm_noprefetch.c

@@ -61,7 +61,6 @@ int main(int argc, char* argv[])
 {
 	AML_ARENA_JEMALLOC_DECL(arns);
 	AML_ARENA_JEMALLOC_DECL(arnf);
-	AML_DMA_LINUX_SEQ_DECL(dma);
 	struct bitmask *slowb, *fastb;
 	aml_init(&argc, &argv);
 	assert(argc == 5);

benchmarks/dgemm_prefetch.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
 #include <assert.h>
 #include <errno.h>
 #include <mkl.h>

benchmarks/stream_add_omp_mt.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-par.h"
 #include <assert.h>
 #include <errno.h>
 #include <omp.h>

benchmarks/stream_add_omp_st.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
 #include <assert.h>
 #include <errno.h>
 #include <omp.h>

benchmarks/stream_add_pth_st.c

@@ -15,6 +15,7 @@
 #include <stdlib.h>
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
 #include "utils.h"
 
 #define ITER 10

include/Makefile.am

 include_HEADERS=aml.h
 
+include_aml_dmadir=$(includedir)/aml/dma
+include_aml_dma_HEADERS = \
+			      aml/dma/linux-seq.h \
+			      aml/dma/linux-par.h
+
+
 include_amlutilsdir=$(includedir)/aml/utils
 include_amlutils_HEADERS = \
 			   aml/utils/vector.h \

include/aml.h

@@ -1186,176 +1186,6 @@ int aml_dma_wait(struct aml_dma *dma, struct aml_dma_request *req);
  */
 int aml_dma_cancel(struct aml_dma *dma, struct aml_dma_request *req);
 
-/*******************************************************************************
- * Linux Sequential DMA API:
- * DMA logic implemented based on general linux API, with the caller thread
- * used as the only execution thread.
- ******************************************************************************/
-
-extern struct aml_dma_ops aml_dma_linux_seq_ops;
-
-struct aml_dma_request_linux_seq {
-	int type;
-	void *dest;
-	void *src;
-	size_t size;
-	int count;
-	void **pages;
-	int *nodes;
-};
-
-struct aml_dma_linux_seq_data {
-	struct aml_vector requests;
-	pthread_mutex_t lock;
-};
-
-struct aml_dma_linux_seq_ops {
-	int (*do_copy)(struct aml_dma_linux_seq_data *dma,
-		       struct aml_dma_request_linux_seq *req);
-	int (*do_move)(struct aml_dma_linux_seq_data *dma,
-		       struct aml_dma_request_linux_seq *req);
-};
-
-struct aml_dma_linux_seq {
-	struct aml_dma_linux_seq_ops ops;
-	struct aml_dma_linux_seq_data data;
-};
-
-#define AML_DMA_LINUX_SEQ_DECL(name) \
-	struct aml_dma_linux_seq __ ##name## _inner_data; \
-	struct aml_dma name = { \
-		&aml_dma_linux_seq_ops, \
-		(struct aml_dma_data *)&__ ## name ## _inner_data, \
-	};
-
-#define AML_DMA_LINUX_SEQ_ALLOCSIZE \
-	(sizeof(struct aml_dma_linux_seq) + \
-	 sizeof(struct aml_dma))
-
-/*
- * Allocates and initializes a new sequential DMA.
- * "dma": an address where the pointer to the newly allocated DMA structure
- *        will be stored.
- * Variadic arguments:
- * - "nbreqs": an argument of type size_t; the initial number of slots for
- *             asynchronous request that are in-flight (will be increased
- *             automatically if necessary).
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_seq_create(struct aml_dma **dma, ...);
-/*
- * Initializes a new sequential DMA.  This is a varargs-variant of the
- * aml_dma_linux_seq_vinit() routine.
- * "dma": an allocated DMA structure.
- * Variadic arguments: see aml_dma_linux_seq_create().
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_seq_init(struct aml_dma *dma, ...);
-/*
- * Initializes a new sequential DMA.
- * "dma": an allocated DMA structure.
- * "args": see the variadic arguments of aml_dma_linux_seq_create().
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_seq_vinit(struct aml_dma *dma, va_list args);
-/*
- * Tears down an initialized sequential DMA.
- * "dma": an initialized DMA structure.
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_seq_destroy(struct aml_dma *dma);
-
-/*******************************************************************************
- * Linux Parallel DMA API:
- * DMA logic implemented based on general linux API, with the caller thread
- * used as the only execution thread.
- ******************************************************************************/
-
-extern struct aml_dma_ops aml_dma_linux_par_ops;
-
-struct aml_dma_linux_par_thread_data {
-	int tid;
-	pthread_t thread;
-	struct aml_dma_linux_par *dma;
-	struct aml_dma_request_linux_par *req;
-};
-
-struct aml_dma_request_linux_par {
-	int type;
-	void *dest;
-	void *src;
-	size_t size;
-	int count;
-	void **pages;
-	int *nodes;
-	struct aml_dma_linux_par_thread_data *thread_data;
-};
-
-struct aml_dma_linux_par_data {
-	size_t nbthreads;
-	struct aml_vector requests;
-	pthread_mutex_t lock;
-};
-
-struct aml_dma_linux_par_ops {
-	void *(*do_thread)(void *);
-	int (*do_copy)(struct aml_dma_linux_par_data *,
-		       struct aml_dma_request_linux_par *, int tid);
-	int (*do_move)(struct aml_dma_linux_par_data *,
-		       struct aml_dma_request_linux_par *, int tid);
-};
-
-struct aml_dma_linux_par {
-	struct aml_dma_linux_par_ops ops;
-	struct aml_dma_linux_par_data data;
-};
-
-#define AML_DMA_LINUX_PAR_DECL(name) \
-	struct aml_dma_linux_par __ ##name## _inner_data; \
-	struct aml_dma name = { \
-		&aml_dma_linux_par_ops, \
-		(struct aml_dma_data *)&__ ## name ## _inner_data, \
-	};
-
-#define AML_DMA_LINUX_PAR_ALLOCSIZE \
-	(sizeof(struct aml_dma_linux_par) + \
-	 sizeof(struct aml_dma))
-
-/*
- * Allocates and initializes a new parallel DMA.
- * "dma": an address where the pointer to the newly allocated DMA structure
- *        will be stored.
- * Variadic arguments:
- * - "nbreqs": an argument of type size_t; the initial number of slots for
- *             asynchronous request that are in-flight (will be increased
- *             automatically if necessary).
- * - "nbthreads": an argument of type size_t; the number of threads to launch
- *                for each request.
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_par_create(struct aml_dma **, ...);
-/*
- * Initializes a new parallel DMA.  This is a varargs-variant of the
- * aml_dma_linux_par_vinit() routine.
- * "dma": an allocated DMA structure.
- * Variadic arguments: see aml_dma_linux_par_create().
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_par_init(struct aml_dma *, ...);
-/*
- * Initializes a new parallel DMA.
- * "dma": an allocated DMA structure.
- * "args": see the variadic arguments of aml_dma_linux_par_create().
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_par_vinit(struct aml_dma *, va_list);
-/*
- * Tears down an initialized parallel DMA.
- * "dma": an initialized DMA structure.
- * Returns 0 if successful; an error code otherwise.
- */
-int aml_dma_linux_par_destroy(struct aml_dma *);
-
 /*******************************************************************************
  * Scratchpad:
  * Use an area to stage data from another area in and out.

include/aml/dma/linux-par.h

+/*******************************************************************************
+ * Copyright 2019 UChicago Argonne, LLC.
+ * (c.f. AUTHORS, LICENSE)
+ *
+ * This file is part of the AML project.
+ * For more info, see https://xgitlab.cels.anl.gov/argo/aml
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+*******************************************************************************/
+
+#ifndef AML_DMA_LINUX_PAR_H
+#define AML_DMA_LINUX_PAR_H 1
+
+/*******************************************************************************
+ * Linux Parallel DMA API:
+ * DMA logic implemented based on general linux API, with the caller thread
+ * used as the only execution thread.
+ ******************************************************************************/
+
+extern struct aml_dma_ops aml_dma_linux_par_ops;
+
+struct aml_dma_linux_par_thread_data {
+	int tid;
+	pthread_t thread;
+	struct aml_dma_linux_par *dma;
+	struct aml_dma_request_linux_par *req;
+};
+
+struct aml_dma_request_linux_par {
+	int type;
+	void *dest;
+	void *src;
+	size_t size;
+	int count;
+	void **pages;
+	int *nodes;
+	struct aml_dma_linux_par_thread_data *thread_data;
+};
+
+struct aml_dma_linux_par_data {
+	size_t nbthreads;
+	struct aml_vector requests;
+	pthread_mutex_t lock;
+};
+
+struct aml_dma_linux_par_ops {
+	void *(*do_thread)(void *);
+	int (*do_copy)(struct aml_dma_linux_par_data *,
+		       struct aml_dma_request_linux_par *, int tid);
+	int (*do_move)(struct aml_dma_linux_par_data *,
+		       struct aml_dma_request_linux_par *, int tid);
+};
+
+struct aml_dma_linux_par {
+	struct aml_dma_linux_par_ops ops;
+	struct aml_dma_linux_par_data data;
+};
+
+#define AML_DMA_LINUX_PAR_DECL(name) \
+	struct aml_dma_linux_par __ ##name## _inner_data; \
+	struct aml_dma name = { \
+		&aml_dma_linux_par_ops, \
+		(struct aml_dma_data *)&__ ## name ## _inner_data, \
+	};
+
+#define AML_DMA_LINUX_PAR_ALLOCSIZE \
+	(sizeof(struct aml_dma_linux_par) + \
+	 sizeof(struct aml_dma))
+
+/*
+ * Allocates and initializes a new parallel DMA.
+ * "dma": an address where the pointer to the newly allocated DMA structure
+ *        will be stored.
+ * Variadic arguments:
+ * - "nbreqs": an argument of type size_t; the initial number of slots for
+ *             asynchronous request that are in-flight (will be increased
+ *             automatically if necessary).
+ * - "nbthreads": an argument of type size_t; the number of threads to launch
+ *                for each request.
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_par_create(struct aml_dma **, ...);
+/*
+ * Initializes a new parallel DMA.  This is a varargs-variant of the
+ * aml_dma_linux_par_vinit() routine.
+ * "dma": an allocated DMA structure.
+ * Variadic arguments: see aml_dma_linux_par_create().
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_par_init(struct aml_dma *, ...);
+/*
+ * Initializes a new parallel DMA.
+ * "dma": an allocated DMA structure.
+ * "args": see the variadic arguments of aml_dma_linux_par_create().
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_par_vinit(struct aml_dma *, va_list);
+/*
+ * Tears down an initialized parallel DMA.
+ * "dma": an initialized DMA structure.
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_par_destroy(struct aml_dma *);
+
+#endif // AML_LINUX_DMA_LINUX_PAR_H

include/aml/dma/linux-seq.h

+/*******************************************************************************
+ * Copyright 2019 UChicago Argonne, LLC.
+ * (c.f. AUTHORS, LICENSE)
+ *
+ * This file is part of the AML project.
+ * For more info, see https://xgitlab.cels.anl.gov/argo/aml
+ *
+ * SPDX-License-Identifier: BSD-3-Clause
+*******************************************************************************/
+
+#ifndef AML_DMA_LINUX_SEQ_H
+#define AML_DMA_LINUX_SEQ_H 1
+
+/*******************************************************************************
+ * Linux Sequential DMA API:
+ * DMA logic implemented based on general linux API, with the caller thread
+ * used as the only execution thread.
+ ******************************************************************************/
+
+extern struct aml_dma_ops aml_dma_linux_seq_ops;
+
+struct aml_dma_request_linux_seq {
+	int type;
+	void *dest;
+	void *src;
+	size_t size;
+	int count;
+	void **pages;
+	int *nodes;
+};
+
+struct aml_dma_linux_seq_data {
+	struct aml_vector requests;
+	pthread_mutex_t lock;
+};
+
+struct aml_dma_linux_seq_ops {
+	int (*do_copy)(struct aml_dma_linux_seq_data *dma,
+		       struct aml_dma_request_linux_seq *req);
+	int (*do_move)(struct aml_dma_linux_seq_data *dma,
+		       struct aml_dma_request_linux_seq *req);
+};
+
+struct aml_dma_linux_seq {
+	struct aml_dma_linux_seq_ops ops;
+	struct aml_dma_linux_seq_data data;
+};
+
+#define AML_DMA_LINUX_SEQ_DECL(name) \
+	struct aml_dma_linux_seq __ ##name## _inner_data; \
+	struct aml_dma name = { \
+		&aml_dma_linux_seq_ops, \
+		(struct aml_dma_data *)&__ ## name ## _inner_data, \
+	};
+
+#define AML_DMA_LINUX_SEQ_ALLOCSIZE \
+	(sizeof(struct aml_dma_linux_seq) + \
+	 sizeof(struct aml_dma))
+
+/*
+ * Allocates and initializes a new sequential DMA.
+ * "dma": an address where the pointer to the newly allocated DMA structure
+ *        will be stored.
+ * Variadic arguments:
+ * - "nbreqs": an argument of type size_t; the initial number of slots for
+ *             asynchronous request that are in-flight (will be increased
+ *             automatically if necessary).
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_create(struct aml_dma **dma, ...);
+/*
+ * Initializes a new sequential DMA.  This is a varargs-variant of the
+ * aml_dma_linux_seq_vinit() routine.
+ * "dma": an allocated DMA structure.
+ * Variadic arguments: see aml_dma_linux_seq_create().
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_init(struct aml_dma *dma, ...);
+/*
+ * Initializes a new sequential DMA.
+ * "dma": an allocated DMA structure.
+ * "args": see the variadic arguments of aml_dma_linux_seq_create().
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_vinit(struct aml_dma *dma, va_list args);
+/*
+ * Tears down an initialized sequential DMA.
+ * "dma": an initialized DMA structure.
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_destroy(struct aml_dma *dma);
+
+/* Performs a copy request.
+ * "dma" the dma_linux_seq_data associated with a linux_seq dma.
+ * "req" a valid linux_seq request.
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_do_copy(struct aml_dma_linux_seq_data *dma,
+			      struct aml_dma_request_linux_seq *req);
+
+/* Performs a move request.
+ * "dma" the dma_linux_seq_data associated with a linux_seq dma.
+ * "req" a valid linux_seq request.
+ * Returns 0 if successful; an error code otherwise.
+ */
+int aml_dma_linux_seq_do_move(struct aml_dma_linux_seq_data *dma,
+			      struct aml_dma_request_linux_seq *req);
+
+#endif // AML_DMA_LINUX_SEQ_H

src/dma/dma_linux_par.c

@@ -9,6 +9,8 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-par.h"
+
 #include <assert.h>
 #include <errno.h>
 #include <sys/mman.h>

src/dma/dma_linux_seq.c

@@ -9,6 +9,8 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
+
 #include <assert.h>
 #include <errno.h>
 #include <sys/mman.h>

tests/dma/test_dma_linux_par.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-par.h"
 #include <assert.h>
 
 #define TILESIZE (2)

tests/dma/test_dma_linux_seq.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
 #include <assert.h>
 
 #define TILESIZE (2)

tests/scratch/test_scratch_par.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-seq.h"
 #include <assert.h>
 
 #define TILESIZE (2)

tests/scratch/test_scratch_seq.c

@@ -9,6 +9,7 @@
 *******************************************************************************/
 
 #include "aml.h"
+#include "aml/dma/linux-par.h"
 #include <assert.h>
 
 #define TILESIZE (2)