First step at speeding up io_u rand refill

[fio.git] / io_u.c

diff --git a/io_u.c b/io_u.c
index 232f6a6f6929ff9514e035336b6ba9626c0a45fd..4d3116b7fd4fc0901f4419e8690270c2c18a2d43 100644 (file)
--- a/io_u.c
+++ b/io_u.c
@@ -8,6 +8,7 @@
 #include "fio.h"
 #include "hash.h"
 #include "verify.h"
+#include "lib/rand.h"
 
 struct io_completion_data {
        int nr;                         /* input */
@@ -187,7 +188,7 @@ static int get_next_rand_offset(struct thread_data *td, struct fio_file *f,
  * until we find a free one. For sequential io, just return the end of
  * the last io issued.
  */
-static int get_next_offset(struct thread_data *td, struct io_u *io_u)
+static int __get_next_offset(struct thread_data *td, struct io_u *io_u)
 {
        struct fio_file *f = io_u->file;
        unsigned long long b;
@@ -231,7 +232,17 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u)
        return 0;
 }
 
-static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
+static int get_next_offset(struct thread_data *td, struct io_u *io_u)
+{
+       struct prof_io_ops *ops = &td->prof_io_ops;
+
+       if (ops->fill_io_u_off)
+               return ops->fill_io_u_off(td, io_u);
+
+       return __get_next_offset(td, io_u);
+}
+
+static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u)
 {
        const int ddir = io_u->ddir;
        unsigned int uninitialized_var(buflen);
@@ -276,6 +287,16 @@ static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
        return buflen;
 }
 
+static unsigned int get_next_buflen(struct thread_data *td, struct io_u *io_u)
+{
+       struct prof_io_ops *ops = &td->prof_io_ops;
+
+       if (ops->fill_io_u_size)
+               return ops->fill_io_u_size(td, io_u);
+
+       return __get_next_buflen(td, io_u);
+}
+
 static void set_rwmix_bytes(struct thread_data *td)
 {
        unsigned int diff;
@@ -374,6 +395,14 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
             td->io_issues[DDIR_WRITE] && should_fsync(td))
                return DDIR_DATASYNC;
 
+       /*
+        * see if it's time to sync_file_range
+        */
+       if (td->sync_file_range_nr &&
+          !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) &&
+            td->io_issues[DDIR_WRITE] && should_fsync(td))
+               return DDIR_SYNC_FILE_RANGE;
+
        if (td_rw(td)) {
                /*
                 * Check if it's time to seed a new data direction.
@@ -413,7 +442,6 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 {
        td_io_u_lock(td);
 
-       assert((io_u->flags & IO_U_F_FREE) == 0);
        io_u->flags |= IO_U_F_FREE;
        io_u->flags &= ~IO_U_F_FREE_DEF;
 
@@ -770,7 +798,7 @@ static struct fio_file *get_next_file_rr(struct thread_data *td, int goodf,
        return f;
 }
 
-static struct fio_file *get_next_file(struct thread_data *td)
+static struct fio_file *__get_next_file(struct thread_data *td)
 {
        struct fio_file *f;
 
@@ -805,6 +833,16 @@ out:
        return f;
 }
 
+static struct fio_file *get_next_file(struct thread_data *td)
+{
+       struct prof_io_ops *ops = &td->prof_io_ops;
+
+       if (ops->get_next_file)
+               return ops->get_next_file(td);
+
+       return __get_next_file(td);
+}
+
 static int set_io_u_file(struct thread_data *td, struct io_u *io_u)
 {
        struct fio_file *f;
@@ -851,24 +889,22 @@ again:
                io_u->end_io = NULL;
        }
 
-       /*
-        * We ran out, wait for async verify threads to finish and return one
-        */
-       if (!io_u && td->o.verify_async) {
-               pthread_cond_wait(&td->free_cond, &td->io_u_lock);
-               goto again;
-       }
-
        if (io_u) {
                assert(io_u->flags & IO_U_F_FREE);
-               io_u->flags &= ~IO_U_F_FREE;
-               io_u->flags &= ~IO_U_F_FREE_DEF;
+               io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF);
 
                io_u->error = 0;
                flist_del(&io_u->list);
                flist_add(&io_u->list, &td->io_u_busylist);
                td->cur_depth++;
                io_u->flags |= IO_U_F_IN_CUR_DEPTH;
+       } else if (td->o.verify_async) {
+               /*
+                * We ran out, wait for async verify threads to finish and
+                * return one
+                */
+               pthread_cond_wait(&td->free_cond, &td->io_u_lock);
+               goto again;
        }
 
        td_io_u_unlock(td);
@@ -968,14 +1004,22 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
         * initialized, silence that warning.
         */
        unsigned long uninitialized_var(usec);
+       struct fio_file *f;
 
        dprint_io_u(io_u, "io complete");
 
+       td_io_u_lock(td);
        assert(io_u->flags & IO_U_F_FLIGHT);
        io_u->flags &= ~IO_U_F_FLIGHT;
+       td_io_u_unlock(td);
 
        if (ddir_sync(io_u->ddir)) {
                td->last_was_sync = 1;
+               f = io_u->file;
+               if (f) {
+                       f->first_write = -1ULL;
+                       f->last_write = -1ULL;
+               }
                return;
        }
 
@@ -984,23 +1028,31 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
        if (!io_u->error) {
                unsigned int bytes = io_u->buflen - io_u->resid;
                const enum fio_ddir idx = io_u->ddir;
+               const enum fio_ddir odx = io_u->ddir ^ 1;
                int ret;
 
                td->io_blocks[idx]++;
                td->io_bytes[idx] += bytes;
                td->this_io_bytes[idx] += bytes;
 
+               if (idx == DDIR_WRITE) {
+                       f = io_u->file;
+                       if (f) {
+                               if (f->first_write == -1ULL ||
+                                   io_u->offset < f->first_write)
+                                       f->first_write = io_u->offset;
+                               if (f->last_write == -1ULL ||
+                                   ((io_u->offset + bytes) > f->last_write))
+                                       f->last_write = io_u->offset + bytes;
+                       }
+               }
+
                if (ramp_time_over(td)) {
                        unsigned long uninitialized_var(lusec);
-                       unsigned long uninitialized_var(rusec);
 
                        if (!td->o.disable_clat || !td->o.disable_bw)
                                lusec = utime_since(&io_u->issue_time,
                                                        &icd->time);
-                       if (__should_check_rate(td, idx) ||
-                           __should_check_rate(td, idx ^ 1))
-                               rusec = utime_since(&io_u->start_time,
-                                                       &icd->time);
 
                        if (!td->o.disable_clat) {
                                add_clat_sample(td, idx, lusec, bytes);
@@ -1009,11 +1061,16 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
                        if (!td->o.disable_bw)
                                add_bw_sample(td, idx, bytes, &icd->time);
                        if (__should_check_rate(td, idx)) {
-                               td->rate_pending_usleep[idx] +=
-                                       (long) td->rate_usec_cycle[idx] - rusec;
+                               td->rate_pending_usleep[idx] =
+                                       ((td->this_io_bytes[idx] *
+                                         td->rate_nsec_cycle[idx]) / 1000 -
+                                        utime_since_now(&td->start));
                        }
                        if (__should_check_rate(td, idx ^ 1))
-                               td->rate_pending_usleep[idx ^ 1] -= rusec;
+                               td->rate_pending_usleep[odx] =
+                                       ((td->this_io_bytes[odx] *
+                                         td->rate_nsec_cycle[odx]) / 1000 -
+                                        utime_since_now(&td->start));
                }
 
                if (td_write(td) && idx == DDIR_WRITE &&
@@ -1161,7 +1218,7 @@ void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
 
        if (!td->o.zero_buffers) {
                while ((void *) ptr - io_u->buf < max_bs) {
-                       *ptr = rand() * GOLDEN_RATIO_PRIME;
+                       *ptr = __rand(&__fio_rand_state);
                        ptr++;
                }
        } else