split: improve performance

The default buffer size doesn't seem to work very well on Windows.
Increasing the buffer to 16K speeds up the splitting of a 1GB file
into 100MB chunks by a factor of two.

GitHub issue #224.

1 files changed, 7 insertions, 1 deletions

diff --git a/coreutils/split.c b/coreutils/split.c
index 3fcfd95f2..9393423a1 100644
--- a/coreutils/split.c
+++ b/coreutils/split.c
@@ -78,8 +78,10 @@ static char *next_file(char *old, unsigned suffix_len)
         return old;
 }
 
+#if !ENABLE_PLATFORM_MINGW32
 #define read_buffer bb_common_bufsiz1
 enum { READ_BUFFER_SIZE = COMMON_BUFSIZE - 1 };
+#endif
 
 #define SPLIT_OPT_l (1<<0)
 #define SPLIT_OPT_b (1<<1)
@@ -97,8 +99,12 @@ int split_main(int argc UNUSED_PARAM, char **argv)
         unsigned opt;
         ssize_t bytes_read, to_write;
         char *src;
-
+#if ENABLE_PLATFORM_MINGW32
+        size_t READ_BUFFER_SIZE = 16*1024;
+        char *read_buffer = xmalloc(16*1024);
+#else
         setup_common_bufsiz();
+#endif
 
         opt = getopt32(argv, "^"
                         "l:b:a:+" /* -a N */