io_u: sequence random buflen generation individually

[fio.git] / fio.h

diff --git a/fio.h b/fio.h
index 806f7ed55bf5a46e9f71bf4c525639436eca5307..39d775c0677dfadbda23c25b68b55014149b44ff 100644 (file)
--- a/fio.h
+++ b/fio.h
@@ -25,7 +25,7 @@
 #include "debug.h"
 #include "file.h"
 #include "io_ddir.h"
-#include "ioengine.h"
+#include "ioengines.h"
 #include "iolog.h"
 #include "helpers.h"
 #include "options.h"
@@ -35,6 +35,7 @@
 #include "oslib/getopt.h"
 #include "lib/rand.h"
 #include "lib/rbtree.h"
+#include "lib/num2str.h"
 #include "client.h"
 #include "server.h"
 #include "stat.h"
@@ -58,6 +59,10 @@
 #define MPOL_LOCAL MPOL_MAX
 #endif
 
+#ifdef CONFIG_CUDA
+#include <cuda.h>
+#endif
+
 /*
  * offset generator types
  */
@@ -86,6 +91,8 @@ enum {
 
 enum {
        FIO_RAND_BS_OFF         = 0,
+       FIO_RAND_BS1_OFF,
+       FIO_RAND_BS2_OFF,
        FIO_RAND_VER_OFF,
        FIO_RAND_MIX_OFF,
        FIO_RAND_FILE_OFF,
@@ -100,6 +107,8 @@ enum {
        FIO_DEDUPE_OFF,
        FIO_RAND_POISSON_OFF,
        FIO_RAND_ZONE_OFF,
+       FIO_RAND_POISSON2_OFF,
+       FIO_RAND_POISSON3_OFF,
        FIO_RAND_NR_OFFS,
 };
 
@@ -142,7 +151,7 @@ struct thread_data {
        unsigned int thread_number;
        unsigned int subjob_number;
        unsigned int groupid;
-       struct thread_stat ts;
+       struct thread_stat ts __attribute__ ((aligned(8)));
 
        int client_type;
 
@@ -158,10 +167,10 @@ struct thread_data {
        struct thread_data *parent;
 
        uint64_t stat_io_bytes[DDIR_RWDIR_CNT];
-       struct timeval bw_sample_time;
+       struct timespec bw_sample_time;
 
        uint64_t stat_io_blocks[DDIR_RWDIR_CNT];
-       struct timeval iops_sample_time;
+       struct timespec iops_sample_time;
 
        volatile int update_rusage;
        struct fio_mutex *rusage_sem;
@@ -207,7 +216,7 @@ struct thread_data {
 
        unsigned long rand_seeds[FIO_RAND_NR_OFFS];
 
-       struct frand_state bsrange_state;
+       struct frand_state bsrange_state[DDIR_RWDIR_CNT];
        struct frand_state verify_state;
        struct frand_state trim_state;
        struct frand_state delay_state;
@@ -280,9 +289,9 @@ struct thread_data {
        unsigned long rate_bytes[DDIR_RWDIR_CNT];
        unsigned long rate_blocks[DDIR_RWDIR_CNT];
        unsigned long long rate_io_issue_bytes[DDIR_RWDIR_CNT];
-       struct timeval lastrate[DDIR_RWDIR_CNT];
-       int64_t last_usec;
-       struct frand_state poisson_state;
+       struct timespec lastrate[DDIR_RWDIR_CNT];
+       int64_t last_usec[DDIR_RWDIR_CNT];
+       struct frand_state poisson_state[DDIR_RWDIR_CNT];
 
        /*
         * Enforced rate submission/completion workqueue
@@ -316,21 +325,21 @@ struct thread_data {
         */
        struct frand_state random_state;
 
-       struct timeval start;   /* start of this loop */
-       struct timeval epoch;   /* time job was started */
+       struct timespec start;  /* start of this loop */
+       struct timespec epoch;  /* time job was started */
        unsigned long long unix_epoch; /* Time job was started, unix epoch based. */
-       struct timeval last_issue;
+       struct timespec last_issue;
        long time_offset;
-       struct timeval tv_cache;
-       struct timeval terminate_time;
-       unsigned int tv_cache_nr;
-       unsigned int tv_cache_mask;
+       struct timespec ts_cache;
+       struct timespec terminate_time;
+       unsigned int ts_cache_nr;
+       unsigned int ts_cache_mask;
        unsigned int ramp_time_over;
 
        /*
         * Time since last latency_window was started
         */
-       struct timeval latency_ts;
+       struct timespec latency_ts;
        unsigned int latency_qd;
        unsigned int latency_qd_high;
        unsigned int latency_qd_low;
@@ -405,6 +414,18 @@ struct thread_data {
        struct steadystate_data ss;
 
        char verror[FIO_VERROR_SIZE];
+
+#ifdef CONFIG_CUDA
+       /*
+        * for GPU memory management
+        */
+       int gpu_dev_cnt;
+       int gpu_dev_id;
+       CUdevice  cu_dev;
+       CUcontext cu_ctx;
+       CUdeviceptr dev_mem_ptr;
+#endif 
+
 };
 
 /*
@@ -520,7 +541,6 @@ extern void fio_options_mem_dupe(struct thread_data *);
 extern void td_fill_rand_seeds(struct thread_data *);
 extern void td_fill_verify_state_seed(struct thread_data *);
 extern void add_job_opts(const char **, int);
-extern char *num2str(uint64_t, int, int, int, int);
 extern int ioengine_load(struct thread_data *);
 extern bool parse_dryrun(void);
 extern int fio_running_or_pending_io_threads(void);
@@ -533,13 +553,6 @@ extern uintptr_t page_size;
 extern int initialize_fio(char *envp[]);
 extern void deinitialize_fio(void);
 
-#define N2S_NONE       0
-#define N2S_BITPERSEC  1       /* match unit_base for bit rates */
-#define N2S_PERSEC     2
-#define N2S_BIT                3
-#define N2S_BYTE       4
-#define N2S_BYTEPERSEC         8       /* match unit_base for byte rates */
-
 #define FIO_GETOPT_JOB         0x89000000
 #define FIO_GETOPT_IOENGINE    0x98000000
 #define FIO_NR_OPTIONS         (FIO_MAX_OPTS + 128)
@@ -585,7 +598,8 @@ static inline enum fio_ioengine_flags td_ioengine_flags(struct thread_data *td)
 
 static inline void td_set_ioengine_flags(struct thread_data *td)
 {
-       td->flags |= (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
+       td->flags = (~(TD_ENG_FLAG_MASK << TD_ENG_FLAG_SHIFT) & td->flags) |
+                   (td->io_ops->flags << TD_ENG_FLAG_SHIFT);
 }
 
 static inline bool td_ioengine_flagged(struct thread_data *td,
@@ -628,17 +642,9 @@ extern void free_threads_shm(void);
  */
 extern void reset_all_stats(struct thread_data *);
 
-/*
- * blktrace support
- */
-#ifdef FIO_HAVE_BLKTRACE
-extern int is_blktrace(const char *, int *);
-extern int load_blktrace(struct thread_data *, const char *, int);
-#endif
-
 extern int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
                   enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
-                  struct timeval *comp_time);
+                  struct timespec *comp_time);
 
 /*
  * Latency target helpers