t/io_uring: pre-calculate per-file depth

Rather than do this math every time, just calculate it at setup time
and use that.

Note that there's something wonky with how the depths are distributed,
it's quite common to see the first file eat the majority of the queue
depth, causing multiple file runs to be both slower than they should be,
and with very unevenly distributed depths. To be investigated.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/t/io_uring.c b/t/io_uring.c
index 8fd9a4cca3b7030a32795d151a05068bfcb1d088..46b153dcbefc32ee778f75023d0cb69cd680e216 100644 (file)
--- a/t/io_uring.c
+++ b/t/io_uring.c
@@ -110,6 +110,7 @@ struct submitter {
 #endif
 
        int numa_node;
+       int per_file_depth;
        const char *filename;
 
        struct file files[MAX_FDS];
@@ -491,11 +492,6 @@ static int io_uring_enter(struct submitter *s, unsigned int to_submit,
 #endif
 }
 
-static unsigned file_depth(struct submitter *s)
-{
-       return (depth + s->nr_files - 1) / s->nr_files;
-}
-
 static unsigned long long get_offset(struct submitter *s, struct file *f)
 {
        unsigned long long offset;
@@ -517,7 +513,7 @@ static unsigned long long get_offset(struct submitter *s, struct file *f)
        return offset;
 }
 
-static struct file *init_new_io(struct submitter *s)
+static struct file *get_next_file(struct submitter *s)
 {
        struct file *f;
 
@@ -525,7 +521,7 @@ static struct file *init_new_io(struct submitter *s)
                f = &s->files[0];
        } else {
                f = &s->files[s->cur_file];
-               if (f->pending_ios >= file_depth(s)) {
+               if (f->pending_ios >= s->per_file_depth) {
                        s->cur_file++;
                        if (s->cur_file == s->nr_files)
                                s->cur_file = 0;
@@ -547,7 +543,7 @@ static void init_io(struct submitter *s, unsigned index)
                return;
        }
 
-       f = init_new_io(s);
+       f = get_next_file(s);
 
        if (register_files) {
                sqe->flags = IOSQE_FIXED_FILE;
@@ -588,7 +584,7 @@ static void init_io_pt(struct submitter *s, unsigned index)
        unsigned long long slba;
        unsigned long long nlb;
 
-       f = init_new_io(s);
+       f = get_next_file(s);
 
        offset = get_offset(s, f);
 
@@ -872,6 +868,7 @@ static int setup_aio(struct submitter *s)
                fixedbufs = register_files = 0;
        }
 
+       s->per_file_depth = (depth + s->nr_files - 1) / s->nr_files;
        return io_queue_init(roundup_pow2(depth), &s->aio_ctx);
 #else
        fprintf(stderr, "Legacy AIO not available on this system/build\n");
@@ -976,6 +973,7 @@ static int setup_ring(struct submitter *s)
        for (i = 0; i < p.sq_entries; i++)
                sring->array[i] = i;
 
+       s->per_file_depth = (depth + s->nr_files - 1) / s->nr_files;
        return 0;
 }
 
@@ -1002,8 +1000,8 @@ static int submitter_init(struct submitter *s)
        static int init_printed;
        char buf[80];
        s->tid = gettid();
-       printf("submitter=%d, tid=%d, file=%s, node=%d\n", s->index, s->tid,
-                                                       s->filename, s->numa_node);
+       printf("submitter=%d, tid=%d, file=%s, nfiles=%d, node=%d\n", s->index, s->tid,
+                                                       s->filename, s->nr_files, s->numa_node);
 
        set_affinity(s);
 
@@ -1082,7 +1080,7 @@ static int prep_more_ios_aio(struct submitter *s, int max_ios, struct iocb *iocb
        while (index < max_ios) {
                struct iocb *iocb = &iocbs[index];
 
-               f = init_new_io(s);
+               f = get_next_file(s);
 
                io_prep_pread(iocb, f->real_fd, s->iovecs[index].iov_base,
                                s->iovecs[index].iov_len, get_offset(s, f));
@@ -1388,7 +1386,7 @@ static void *submitter_sync_fn(void *data)
                uint64_t offset;
                struct file *f;
 
-               f = init_new_io(s);
+               f = get_next_file(s);
 
 #ifdef ARCH_HAVE_CPU_CLOCK
                if (stats)