Use aligned allocation via C11 or Windows API where possible

4 files changed, 63 insertions, 12 deletions

diff --git a/src/byte_sink.h b/src/byte_sink.h
index f39bc2ba..876add2b 100644
--- a/src/byte_sink.h
+++ b/src/byte_sink.h
@@ -42,7 +42,7 @@ serd_byte_sink_new(SerdSink sink, void* stream, size_t block_size)
 	bsink.size       = 0;
 	bsink.block_size = block_size;
 	bsink.buf        = ((block_size > 1)
-	                    ? (uint8_t*)serd_bufalloc(block_size)
+	                    ? (uint8_t*)serd_allocate_buffer(block_size)
 	                    : NULL);
 	return bsink;
 }
@@ -60,7 +60,7 @@ static inline void
 serd_byte_sink_free(SerdByteSink* bsink)
 {
 	serd_byte_sink_flush(bsink);
-	free(bsink->buf);
+	serd_free_aligned(bsink->buf);
 	bsink->buf = NULL;
 }
 
diff --git a/src/byte_source.c b/src/byte_source.c
index d783959c..5c0404a7 100644
--- a/src/byte_source.c
+++ b/src/byte_source.c
@@ -22,7 +22,6 @@
 
 #include <stdbool.h>
 #include <stdint.h>
-#include <stdlib.h>
 #include <string.h>
 
 SerdStatus
@@ -63,7 +62,7 @@ serd_byte_source_open_source(SerdByteSource*     source,
 	source->read_func   = read_func;
 
 	if (page_size > 1) {
-		source->file_buf = (uint8_t*)serd_bufalloc(page_size);
+		source->file_buf = (uint8_t*)serd_allocate_buffer(page_size);
 		source->read_buf = source->file_buf;
 		memset(source->file_buf, '\0', page_size);
 	} else {
@@ -102,7 +101,7 @@ SerdStatus
 serd_byte_source_close(SerdByteSource* source)
 {
 	if (source->page_size > 1) {
-		free(source->file_buf);
+		serd_free_aligned(source->file_buf);
 	}
 	memset(source, '\0', sizeof(*source));
 	return SERD_SUCCESS;
diff --git a/src/system.c b/src/system.c
index 0a6a5561..fbff8806 100644
--- a/src/system.c
+++ b/src/system.c
@@ -25,11 +25,19 @@
 #   include <fcntl.h>
 #endif
 
+#ifdef _WIN32
+#	include <malloc.h>
+#endif
+
 #include <errno.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 
+#define SERD_HAVE_ALIGNED_ALLOCATION                  \
+	defined(_WIN32) || defined(HAVE_ALIGNED_ALLOC) || \
+	    defined(HAVE_POSIX_MEMALIGN)
+
 FILE*
 serd_fopen(const char* path, const char* mode)
 {
@@ -46,14 +54,49 @@ serd_fopen(const char* path, const char* mode)
 }
 
 void*
-serd_bufalloc(size_t size)
+serd_malloc_aligned(const size_t alignment, const size_t size)
 {
-#ifdef HAVE_POSIX_MEMALIGN
+#if defined(_WIN32)
+	return _aligned_malloc(size, alignment);
+#elif defined(HAVE_ALIGNED_ALLOC)
+	return aligned_alloc(alignment, size);
+#elif defined(HAVE_POSIX_MEMALIGN)
 	void*     ptr = NULL;
-	const int ret = posix_memalign(&ptr, SERD_PAGE_SIZE, size);
+	const int ret = posix_memalign(&ptr, alignment, size);
 	return ret ? NULL : ptr;
 #else
+	(void)alignment;
 	return malloc(size);
 #endif
 }
 
+void*
+serd_calloc_aligned(const size_t alignment, const size_t size)
+{
+#ifdef SERD_HAVE_ALIGNED_ALLOCATION
+	void* const ptr = serd_malloc_aligned(alignment, size);
+	if (ptr) {
+		memset(ptr, 0, size);
+	}
+	return ptr;
+#else
+	(void)alignment;
+	return calloc(1, size);
+#endif
+}
+
+void*
+serd_allocate_buffer(const size_t size)
+{
+	return serd_malloc_aligned(SERD_PAGE_SIZE, size);
+}
+
+void
+serd_free_aligned(void* const ptr)
+{
+#ifdef _WIN32
+	_aligned_free(ptr);
+#else
+	free(ptr);
+#endif
+}
diff --git a/src/system.h b/src/system.h
index 0f6a0e0b..60095a5b 100644
--- a/src/system.h
+++ b/src/system.h
@@ -21,10 +21,19 @@
 
 #include <stdio.h>
 
-FILE*
-serd_fopen(const char* path, const char* mode);
+/// Open a file configured for fast sequential reading
+FILE* serd_fopen(const char* path, const char* mode);
 
-SERD_MALLOC_FUNC void*
-serd_bufalloc(size_t size);
+/// Allocate a buffer aligned to `alignment` bytes
+SERD_MALLOC_FUNC void* serd_malloc_aligned(size_t alignment, size_t size);
+
+/// Allocate a zeroed buffer aligned to `alignment` bytes
+SERD_MALLOC_FUNC void* serd_calloc_aligned(size_t alignment, size_t size);
+
+/// Allocate an aligned buffer for I/O
+SERD_MALLOC_FUNC void* serd_allocate_buffer(size_t size);
+
+/// Free a buffer allocated with an aligned allocation function
+void serd_free_aligned(void* ptr);
 
 #endif // SERD_SYSTEM_H