Reduce possibility of mistakes.

* Unifies CPU/GPU-shared data structure definitions into a single header (nba/framework/datablock_shared.hh) * Adds "explicit" specifier to memory pool constructors
ANLAB-KAIST · Feb 23, 2016 · f48ff79 · f48ff79
1 parent a7eca05
commit f48ff79
Show file tree

Hide file tree

Showing 7 changed files with 14 additions and 46 deletions.
diff --git a/elements/ip/IPlookup_kernel.cu b/elements/ip/IPlookup_kernel.cu
@@ -15,8 +15,7 @@
 
 #define IGNORED_IP 0xFFffFFffu
 
-/* Compatibility definitions. */
-#include <nba/engines/cuda/compat.hh>
+#include <nba/framework/datablock_shared.hh>
 
 extern "C" {
 

diff --git a/elements/ipsec/IPsecAES_kernel.cu b/elements/ipsec/IPsecAES_kernel.cu
@@ -11,8 +11,7 @@
 #include <openssl/aes.h>
 #include <openssl/md5.h>
 
-/* Compatibility definitions. */
-#include <nba/engines/cuda/compat.hh>
+#include <nba/framework/datablock_shared.hh>
 
 /* The index is given by the order in get_used_datablocks(). */
 #define dbid_enc_payloads_d   (0)

diff --git a/elements/ipsec/IPsecAuthHMACSHA1_kernel.cu b/elements/ipsec/IPsecAuthHMACSHA1_kernel.cu
@@ -5,8 +5,7 @@
 
 #include "IPsecAuthHMACSHA1_kernel.hh"
 
-/* Compatibility definitions. */
-#include <nba/engines/cuda/compat.hh>
+#include <nba/framework/datablock_shared.hh>
 
 /* The index is given by the order in get_used_datablocks(). */
 #define dbid_enc_payloads_d (0)

diff --git a/elements/ipv6/LookupIP6Route_kernel.cu b/elements/ipv6/LookupIP6Route_kernel.cu
@@ -8,7 +8,7 @@
 
 #include "util_jhash.h"
 #include "util_hash_table.hh"
-#include <nba/engines/cuda/compat.hh>
+#include <nba/framework/datablock_shared.hh>
 
 using namespace nba;
 

diff --git a/include/nba/engines/cuda/mempool.hh b/include/nba/engines/cuda/mempool.hh
@@ -13,15 +13,15 @@ namespace nba {
 class CUDAMemoryPool : public MemoryPool<dev_mem_t>
 {
 public:
-    CUDAMemoryPool()
+    explicit CUDAMemoryPool()
         : MemoryPool(), base(nullptr)
     { }
 
-    CUDAMemoryPool(size_t max_size)
+    explicit CUDAMemoryPool(size_t max_size)
         : MemoryPool(max_size), base(nullptr)
     { }
 
-    CUDAMemoryPool(size_t max_size, size_t align)
+    explicit CUDAMemoryPool(size_t max_size, size_t align)
         : MemoryPool(max_size, align), base(nullptr)
     { }
 
@@ -63,15 +63,15 @@ private:
 class CPUMemoryPool : public MemoryPool<host_mem_t>
 {
 public:
-    CPUMemoryPool(int cuda_flags = 0)
+    explicit CPUMemoryPool(int cuda_flags)
         : MemoryPool(), base(nullptr), flags(cuda_flags), use_external(false)
     { }
 
-    CPUMemoryPool(size_t max_size, int cuda_flags = 0)
+    explicit CPUMemoryPool(size_t max_size, int cuda_flags)
         : MemoryPool(max_size), base(nullptr), flags(cuda_flags), use_external(false)
     { }
 
-    CPUMemoryPool(size_t max_size, size_t align, int cuda_flags = 0)
+    explicit CPUMemoryPool(size_t max_size, size_t align, int cuda_flags)
         : MemoryPool(max_size, align), base(nullptr), flags(cuda_flags), use_external(false)
     { }
 

diff --git a/include/nba/framework/datablock.hh b/include/nba/framework/datablock.hh
@@ -4,6 +4,7 @@
 #include <nba/framework/config.hh>
 #include <nba/framework/computecontext.hh>
 #include <nba/core/shiftedint.hh> // should come after cuda headers
+#include <nba/framework/datablock_shared.hh>
 #include <vector>
 #include <string>
 #include <tuple>
@@ -113,35 +114,6 @@ struct datablock_tracker {
     dev_mem_t aligned_item_sizes_d;
 };
 
-/**
- * Datablock batch info struct.
- *
- * It contains item offset/size information for variable-length datablocks.
- *
- * NOTE: The alignment of this struct should match with CUDA.
- */
-struct datablock_batch_info {
-    void *buffer_bases_in;
-    void *buffer_bases_out;
-    uint32_t item_count_in;
-    uint32_t item_count_out;
-    uint16_t *item_sizes_in;
-    uint16_t *item_sizes_out;
-    dev_offset_t *item_offsets_in;
-    dev_offset_t *item_offsets_out;
-}; // __cuda_aligned
-
-/**
- * Kernel argument spec for datablocks.
- */
-struct datablock_kernel_arg {
-    uint32_t total_item_count_in;
-    uint32_t total_item_count_out;
-    uint16_t item_size_in;  // for fixed-size cases
-    uint16_t item_size_out; // for fixed-size cases
-    struct datablock_batch_info batches[0];
-}; // __cuda_aligned
-
 
 /** Datablock information class.
  *

diff --git a/include/nba/engines/cuda/compat.hh → include/nba/framework/datablock_shared.hh b/include/nba/engines/cuda/compat.hh → include/nba/framework/datablock_shared.hh
@@ -1,9 +1,8 @@
-#ifndef __NBA_ENGINES_CUDA_COMPAT_HH__
-#define __NBA_ENGINES_CUDA_COMPAT_HH__
+#ifndef __NBA_DATABLOCK_SHARED_HH__
+#define __NBA_DATABLOCK_SHARED_HH__
 
 /*
- * This header is included by .cu sources.
- * We put only relevant data structures here for use in CUDA codes.
+ * This header is included by both .cc/.cu sources.
  * Note that the nvcc should support C++11 (CUDA v6.5 or higher).
  */