compress_parallel.c: Use more appropriate type for shift

[wimlib] / src / compress_parallel.c

diff --git a/src/compress_parallel.c b/src/compress_parallel.c
index fa7f5dde113750adea8a1b43f12227748cd5109d..cb49a24c75d23e2618217d6149ef7d0a47393c8f 100644 (file)
--- a/src/compress_parallel.c
+++ b/src/compress_parallel.c
@@ -43,6 +43,9 @@
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
+#ifdef HAVE_SYS_SYSCTL_H
+#  include <sys/sysctl.h>
+#endif
 
 struct message_queue {
        struct list_head list;
@@ -64,8 +67,8 @@ struct compressor_thread_data {
 struct message {
        u8 *uncompressed_chunks[MAX_CHUNKS_PER_MSG];
        u8 *compressed_chunks[MAX_CHUNKS_PER_MSG];
-       unsigned uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
-       unsigned compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
+       u32 uncompressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
+       u32 compressed_chunk_sizes[MAX_CHUNKS_PER_MSG];
        size_t num_filled_chunks;
        size_t num_alloc_chunks;
        struct list_head list;
@@ -116,17 +119,24 @@ get_avail_memory(void)
        if (phys_bytes == 0)
                goto default_size;
        return phys_bytes;
-#else
+#elif defined(_SC_PAGESIZE) && defined(_SC_PHYS_PAGES)
        long page_size = sysconf(_SC_PAGESIZE);
        long num_pages = sysconf(_SC_PHYS_PAGES);
        if (page_size <= 0 || num_pages <= 0)
                goto default_size;
        return ((u64)page_size * (u64)num_pages);
+#else
+       int mib[2] = {CTL_HW, HW_MEMSIZE};
+       u64 memsize;
+       size_t len = sizeof(memsize);
+       if (sysctl(mib, ARRAY_LEN(mib), &memsize, &len, NULL, 0) < 0 || len != 8)
+               goto default_size;
+       return memsize;
 #endif
 
 default_size:
        WARNING("Failed to determine available memory; assuming 1 GiB");
-       return 1U << 30;
+       return 1ULL << 30;
 }
 
 static int
@@ -322,7 +332,7 @@ submit_compression_msg(struct parallel_chunk_compressor *ctx)
 
 static bool
 parallel_chunk_compressor_submit_chunk(struct chunk_compressor *_ctx,
-                                      const void *chunk, size_t size)
+                                      const void *chunk, u32 size)
 {
        struct parallel_chunk_compressor *ctx = (struct parallel_chunk_compressor *)_ctx;
        struct message *msg;
@@ -351,8 +361,8 @@ parallel_chunk_compressor_submit_chunk(struct chunk_compressor *_ctx,
 
 static bool
 parallel_chunk_compressor_get_chunk(struct chunk_compressor *_ctx,
-                                   const void **cdata_ret, unsigned *csize_ret,
-                                   unsigned *usize_ret)
+                                   const void **cdata_ret, u32 *csize_ret,
+                                   u32 *usize_ret)
 {
        struct parallel_chunk_compressor *ctx = (struct parallel_chunk_compressor *)_ctx;
        struct message *msg;
@@ -407,7 +417,6 @@ new_parallel_chunk_compressor(int out_ctype, u32 out_chunk_size,
        unsigned desired_num_threads;
 
        wimlib_assert(out_chunk_size > 0);
-       wimlib_assert(out_ctype != WIMLIB_COMPRESSION_TYPE_NONE);
 
        if (num_threads == 0)
                num_threads = get_default_num_threads();
@@ -423,16 +432,26 @@ new_parallel_chunk_compressor(int out_ctype, u32 out_chunk_size,
 
        desired_num_threads = num_threads;
 
-       chunks_per_msg = MAX_CHUNKS_PER_MSG;
-       msgs_per_thread = 2;
+       if (out_chunk_size < ((u32)1 << 23)) {
+               chunks_per_msg = MAX_CHUNKS_PER_MSG;
+               msgs_per_thread = 2;
+       } else {
+               /* Big chunks: Just have one buffer per thread --- more would
+                * just waste memory.  */
+               chunks_per_msg = 1;
+               msgs_per_thread = 1;
+       }
        for (;;) {
                approx_mem_required =
                        (u64)chunks_per_msg *
                        (u64)msgs_per_thread *
                        (u64)num_threads *
                        (u64)out_chunk_size
+                       + out_chunk_size
                        + 1000000
-                       + (out_chunk_size * num_threads * 4);
+                       + num_threads * wimlib_get_compressor_needed_memory(out_ctype,
+                                                                           out_chunk_size,
+                                                                           NULL);
                if (approx_mem_required <= max_memory)
                        break;
 
@@ -511,7 +530,9 @@ new_parallel_chunk_compressor(int out_ctype, u32 out_chunk_size,
                if (ret) {
                        errno = ret;
                        ret = WIMLIB_ERR_NOMEM;
-                       WARNING_WITH_ERRNO("Failed to create compressor thread %u of %u");
+                       WARNING_WITH_ERRNO("Failed to create compressor thread %u of %u",
+                                          ctx->num_started_threads + 1,
+                                          num_threads);
                        goto err;
                }
        }