[PATCH i-g-t] tests/intel/xe_sriov_scheduling: K-in-flight; completion window; --inflight
Kamil Konieczny
kamil.konieczny at linux.intel.com
Tue Aug 19 12:15:34 UTC 2025
Hi Marcin,
On 2025-08-19 at 13:00:55 +0200, Marcin Bernatowicz wrote:
please improve subject or split this into more patches.
For example better subject could be:
[PATCH i-g-t] tests/intel/xe_sriov_scheduling: Refactor to better saturate HW
Regards,
Kamil
> Refactor submission/measurement to better saturate HW and make
> throughput comparisons more robust, especially with short jobs.
> Add the --inflight option.
>
> - Drive a K-in-flight pipeline per VF using per-slot BO/addr/spin and
> binary out-fences; add subm_exec_slot()/subm_wait_slot() (prefill +
> refill).
> - Record complete_ts[] and per-slot submit_ts[]; build the common
> window from completions [max(first), min(last)] and compute
> throughput as count/window.
> - Push durations as submit-to-completion (complete_ts - submit_ts) and
> print "mean submit->signal latency".
> - Add --inflight (0=auto; non-preempt defaults to 1; short jobs pick
> higher K); print chosen K in the banner.
>
> Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz at linux.intel.com>
> Cc: Adam Miszczak <adam.miszczak at linux.intel.com>
> Cc: Jakub Kolakowski <jakub1.kolakowski at intel.com>
> Cc: Lukasz Laguna <lukasz.laguna at intel.com>
> Cc: Satyanarayana K V P <satyanarayana.k.v.p at intel.com>
> ---
> tests/intel/xe_sriov_scheduling.c | 241 +++++++++++++++++++++---------
> 1 file changed, 171 insertions(+), 70 deletions(-)
>
> diff --git a/tests/intel/xe_sriov_scheduling.c b/tests/intel/xe_sriov_scheduling.c
> index d69315690..467eb1e29 100644
> --- a/tests/intel/xe_sriov_scheduling.c
> +++ b/tests/intel/xe_sriov_scheduling.c
> @@ -27,6 +27,8 @@ struct subm_opts {
> uint32_t exec_quantum_ms;
> uint32_t preempt_timeout_us;
> double outlier_treshold;
> + /* --inflight=0 => auto; >=1 => explicit K */
> + unsigned int inflight;
> };
>
> struct subm_work_desc {
> @@ -39,6 +41,7 @@ struct subm_stats {
> igt_stats_t samples;
> uint64_t start_timestamp;
> uint64_t end_timestamp;
> + uint64_t *complete_ts; /* absolute completion timestamps (ns) */
> unsigned int num_early_finish;
> unsigned int concurrent_execs;
> double concurrent_rate;
> @@ -51,13 +54,17 @@ struct subm {
> int vf_num;
> struct subm_work_desc work;
> uint32_t expected_ticks;
> - uint64_t addr;
> uint32_t vm;
> struct drm_xe_engine_class_instance hwe;
> uint32_t exec_queue_id;
> - uint32_t bo;
> + /* K slots (K BOs / addresses / mapped spinners / done fences / submit_ts) */
> + unsigned int slots;
> + uint64_t *submit_ts; /* per-slot submit timestamps (ns) */
> + uint64_t *addr;
> + uint32_t *bo;
> size_t bo_size;
> - struct xe_spin *spin;
> + struct xe_spin **spin;
> + uint32_t *done_fence;
> struct drm_xe_sync sync[1];
> struct drm_xe_exec exec;
> };
> @@ -78,43 +85,62 @@ struct subm_set {
> };
>
> static void subm_init(struct subm *s, int fd, int vf_num, uint64_t addr,
> - struct drm_xe_engine_class_instance hwe)
> + struct drm_xe_engine_class_instance hwe,
> + unsigned int inflight)
> {
> + uint64_t base, stride;
> +
> memset(s, 0, sizeof(*s));
> s->fd = fd;
> s->vf_num = vf_num;
> s->hwe = hwe;
> snprintf(s->id, sizeof(s->id), "VF%d %d:%d:%d", vf_num,
> hwe.engine_class, hwe.engine_instance, hwe.gt_id);
> - s->addr = addr ? addr : 0x1a0000;
> + s->slots = inflight ? inflight : 1;
> s->vm = xe_vm_create(s->fd, 0, 0);
> s->exec_queue_id = xe_exec_queue_create(s->fd, s->vm, &s->hwe, 0);
> s->bo_size = ALIGN(sizeof(struct xe_spin) + xe_cs_prefetch_size(s->fd),
> xe_get_default_alignment(s->fd));
> - s->bo = xe_bo_create(s->fd, s->vm, s->bo_size,
> - vram_if_possible(fd, s->hwe.gt_id),
> - DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
> - s->spin = xe_bo_map(s->fd, s->bo, s->bo_size);
> - xe_vm_bind_sync(s->fd, s->vm, s->bo, 0, s->addr, s->bo_size);
> - /* out fence */
> - s->sync[0].type = DRM_XE_SYNC_TYPE_SYNCOBJ;
> - s->sync[0].flags = DRM_XE_SYNC_FLAG_SIGNAL;
> - s->sync[0].handle = syncobj_create(s->fd, 0);
> - s->exec.num_syncs = 1;
> - s->exec.syncs = to_user_pointer(&s->sync[0]);
> + s->addr = calloc(s->slots, sizeof(*s->addr));
> + s->bo = calloc(s->slots, sizeof(*s->bo));
> + s->spin = calloc(s->slots, sizeof(*s->spin));
> + s->done_fence = calloc(s->slots, sizeof(*s->done_fence));
> + s->submit_ts = calloc(s->slots, sizeof(*s->submit_ts));
> + igt_assert(s->addr && s->bo && s->spin && s->done_fence && s->submit_ts);
> +
> + base = addr ? addr : 0x1a0000;
> + stride = ALIGN(s->bo_size, 0x10000);
> + for (unsigned int i = 0; i < s->slots; i++) {
> + s->addr[i] = base + i * stride;
> + s->bo[i] = xe_bo_create(s->fd, s->vm, s->bo_size,
> + vram_if_possible(fd, s->hwe.gt_id),
> + DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM);
> + s->spin[i] = xe_bo_map(s->fd, s->bo[i], s->bo_size);
> + xe_vm_bind_sync(s->fd, s->vm, s->bo[i], 0, s->addr[i], s->bo_size);
> + s->done_fence[i] = syncobj_create(s->fd, 0);
> + }
> +
> s->exec.num_batch_buffer = 1;
> s->exec.exec_queue_id = s->exec_queue_id;
> - s->exec.address = s->addr;
> + /* s->exec.address set per submission */
> }
>
> static void subm_fini(struct subm *s)
> {
> - xe_vm_unbind_sync(s->fd, s->vm, 0, s->addr, s->bo_size);
> - gem_munmap(s->spin, s->bo_size);
> - gem_close(s->fd, s->bo);
> + for (unsigned int i = 0; i < s->slots; i++) {
> + xe_vm_unbind_sync(s->fd, s->vm, 0, s->addr[i], s->bo_size);
> + gem_munmap(s->spin[i], s->bo_size);
> + gem_close(s->fd, s->bo[i]);
> + if (s->done_fence[i])
> + syncobj_destroy(s->fd, s->done_fence[i]);
> + }
> xe_exec_queue_destroy(s->fd, s->exec_queue_id);
> xe_vm_destroy(s->fd, s->vm);
> - syncobj_destroy(s->fd, s->sync[0].handle);
> + free(s->addr);
> + free(s->bo);
> + free(s->spin);
> + free(s->done_fence);
> + free(s->submit_ts);
> }
>
> static void subm_workload_init(struct subm *s, struct subm_work_desc *work)
> @@ -122,25 +148,41 @@ static void subm_workload_init(struct subm *s, struct subm_work_desc *work)
> s->work = *work;
> s->expected_ticks = xe_spin_nsec_to_ticks(s->fd, s->hwe.gt_id,
> s->work.duration_ms * 1000000);
> - xe_spin_init_opts(s->spin, .addr = s->addr, .preempt = s->work.preempt,
> - .ctx_ticks = s->expected_ticks);
> + for (unsigned int i = 0; i < s->slots; i++)
> + xe_spin_init_opts(s->spin[i], .addr = s->addr[i],
> + .preempt = s->work.preempt,
> + .ctx_ticks = s->expected_ticks);
> }
>
> -static void subm_wait(struct subm *s, uint64_t abs_timeout_nsec)
> +static void subm_wait_slot(struct subm *s, unsigned int slot, uint64_t abs_timeout_nsec)
> {
> - igt_assert(syncobj_wait(s->fd, &s->sync[0].handle, 1, abs_timeout_nsec,
> - 0, NULL));
> + igt_assert(syncobj_wait(s->fd, &s->done_fence[slot], 1,
> + abs_timeout_nsec, 0, NULL));
> }
>
> -static void subm_exec(struct subm *s)
> +static void subm_exec_slot(struct subm *s, unsigned int slot)
> {
> - syncobj_reset(s->fd, &s->sync[0].handle, 1);
> + struct timespec tv;
> + int nsync = 0;
> +
> + syncobj_reset(s->fd, &s->done_fence[slot], 1);
> + memset(&s->sync[0], 0, sizeof(s->sync));
> + s->sync[nsync].type = DRM_XE_SYNC_TYPE_SYNCOBJ;
> + s->sync[nsync].flags = DRM_XE_SYNC_FLAG_SIGNAL;
> + s->sync[nsync].handle = s->done_fence[slot];
> + nsync++;
> + s->exec.num_syncs = nsync;
> + s->exec.syncs = to_user_pointer(&s->sync[0]);
> + s->exec.address = s->addr[slot];
> +
> + igt_gettime(&tv);
> + s->submit_ts[slot] = (uint64_t)tv.tv_sec * (uint64_t)NSEC_PER_SEC + (uint64_t)tv.tv_nsec;
> xe_exec(s->fd, &s->exec);
> }
>
> -static bool subm_is_work_complete(struct subm *s)
> +static bool subm_is_work_complete(struct subm *s, unsigned int slot)
> {
> - return s->expected_ticks <= ~s->spin->ticks_delta;
> + return s->expected_ticks <= ~s->spin[slot]->ticks_delta;
> }
>
> static bool subm_is_exec_queue_banned(struct subm *s)
> @@ -157,6 +199,8 @@ static bool subm_is_exec_queue_banned(struct subm *s)
> static void subm_exec_loop(struct subm *s, struct subm_stats *stats,
> const struct subm_opts *opts)
> {
> + const unsigned int inflight = s->slots;
> + unsigned int submitted = 0;
> struct timespec tv;
> unsigned int i;
>
> @@ -165,16 +209,27 @@ static void subm_exec_loop(struct subm *s, struct subm_stats *stats,
> tv.tv_sec * (uint64_t)NSEC_PER_SEC + tv.tv_nsec;
> igt_debug("[%s] start_timestamp: %f\n", s->id, stats->start_timestamp * 1e-9);
>
> - for (i = 0; i < s->work.repeats; ++i) {
> - igt_gettime(&tv);
> + /* Prefill */
> + if (s->work.repeats) {
> + unsigned int can_prefill = min(inflight, s->work.repeats);
>
> - subm_exec(s);
> + for (i = 0; i < can_prefill; i++)
> + subm_exec_slot(s, i % inflight);
> + submitted = can_prefill;
> + }
>
> - subm_wait(s, INT64_MAX);
> + /* Process completions in order: sample i -> slot (i % inflight) */
> + for (i = 0; i < s->work.repeats; ++i) {
> + unsigned int slot = i % inflight;
> +
> + subm_wait_slot(s, slot, INT64_MAX);
>
> - igt_stats_push(&stats->samples, igt_nsec_elapsed(&tv));
> + igt_gettime(&tv);
> + stats->complete_ts[i] = (uint64_t)tv.tv_sec * (uint64_t)NSEC_PER_SEC +
> + (uint64_t)tv.tv_nsec;
> + igt_stats_push(&stats->samples, stats->complete_ts[i] - s->submit_ts[slot]);
>
> - if (!subm_is_work_complete(s)) {
> + if (!subm_is_work_complete(s, slot)) {
> stats->num_early_finish++;
>
> igt_debug("[%s] subm #%d early_finish=%u\n",
> @@ -183,6 +238,14 @@ static void subm_exec_loop(struct subm *s, struct subm_stats *stats,
> if (subm_is_exec_queue_banned(s))
> break;
> }
> +
> + /* Keep the pipeline full */
> + if (submitted < s->work.repeats) {
> + unsigned int next_slot = submitted % inflight;
> +
> + subm_exec_slot(s, next_slot);
> + submitted++;
> + }
> }
>
> igt_gettime(&tv);
> @@ -272,8 +335,10 @@ static void subm_set_fini(struct subm_set *set)
>
> subm_set_close_handles(set);
>
> - for (i = 0; i < set->ndata; ++i)
> + for (i = 0; i < set->ndata; ++i) {
> igt_stats_fini(&set->data[i].stats.samples);
> + free(set->data[i].stats.complete_ts);
> + }
>
> subm_set_free_data(set);
> }
> @@ -334,16 +399,22 @@ static void compute_common_time_frame_stats(struct subm_set *set)
> struct subm_stats *stats;
> uint64_t common_start = 0;
> uint64_t common_end = UINT64_MAX;
> + uint64_t first_ts, last_ts;
>
> - /* Find the common time frame */
> + /* Find common window from completion timestamps */
> for (i = 0; i < ndata; i++) {
> stats = &data[i].stats;
>
> - if (stats->start_timestamp > common_start)
> - common_start = stats->start_timestamp;
> + if (!stats->samples.n_values)
> + continue;
>
> - if (stats->end_timestamp < common_end)
> - common_end = stats->end_timestamp;
> + first_ts = stats->complete_ts[0];
> + last_ts = stats->complete_ts[stats->samples.n_values - 1];
> +
> + if (first_ts > common_start)
> + common_start = first_ts;
> + if (last_ts < common_end)
> + common_end = last_ts;
> }
>
> igt_info("common time frame: [%" PRIu64 ";%" PRIu64 "] %.2fms\n",
> @@ -354,8 +425,7 @@ static void compute_common_time_frame_stats(struct subm_set *set)
>
> /* Compute concurrent_rate for each sample set within the common time frame */
> for (i = 0; i < ndata; i++) {
> - uint64_t total_samples_duration = 0;
> - uint64_t samples_duration_in_common_frame = 0;
> + const double window_s = (common_end - common_start) * 1e-9;
>
> stats = &data[i].stats;
> stats->concurrent_execs = 0;
> @@ -363,29 +433,21 @@ static void compute_common_time_frame_stats(struct subm_set *set)
> stats->concurrent_mean = 0.0;
>
> for (j = 0; j < stats->samples.n_values; j++) {
> - uint64_t sample_start = stats->start_timestamp + total_samples_duration;
> - uint64_t sample_end = sample_start + stats->samples.values_u64[j];
> + uint64_t cts = stats->complete_ts[j];
>
> - if (sample_start >= common_start &&
> - sample_end <= common_end) {
> + if (cts >= common_start && cts <= common_end) {
> stats->concurrent_execs++;
> - samples_duration_in_common_frame +=
> - stats->samples.values_u64[j];
> + stats->concurrent_mean += stats->samples.values_u64[j];
> }
> -
> - total_samples_duration += stats->samples.values_u64[j];
> }
>
> - stats->concurrent_rate = samples_duration_in_common_frame ?
> - (double)stats->concurrent_execs /
> - (samples_duration_in_common_frame *
> - 1e-9) :
> - 0.0;
> + stats->concurrent_rate = (window_s > 0.0) ?
> + ((double)stats->concurrent_execs / window_s) : 0.0;
> +
> stats->concurrent_mean = stats->concurrent_execs ?
> - (double)samples_duration_in_common_frame /
> - stats->concurrent_execs :
> - 0.0;
> - igt_info("[%s] Throughput = %.4f execs/s mean duration=%.4fms nsamples=%d\n",
> + (double)stats->concurrent_mean /
> + stats->concurrent_execs : 0.0;
> + igt_info("[%s] Throughput = %.4f execs/s mean submit->signal latency=%.4fms nsamples=%d\n",
> data[i].subm.id, stats->concurrent_rate, stats->concurrent_mean * 1e-6,
> stats->concurrent_execs);
> }
> @@ -439,9 +501,9 @@ static void log_sample_values(char *id, struct subm_stats *stats,
> }
>
> #define MIN_NUM_REPEATS 25
> -#define MIN_EXEC_QUANTUM_MS 8
> +#define MIN_EXEC_QUANTUM_MS 1
> #define MAX_EXEC_QUANTUM_MS 32
> -#define MIN_JOB_DURATION_MS 16
> +#define MIN_JOB_DURATION_MS 2
> #define MAX_TOTAL_DURATION_MS 15000
> #define PREFERRED_TOTAL_DURATION_MS 10000
> #define MAX_PREFERRED_REPEATS 100
> @@ -546,6 +608,25 @@ static struct vf_sched_params prepare_vf_sched_params(int num_threads,
> return params;
> }
>
> +/* inflight K selection:
> + * user_k == 0 => auto
> + * user_k >= 1 => explicit K
> + */
> +static unsigned int select_inflight_k(unsigned int duration_ms,
> + unsigned int user_k,
> + bool nonpreempt)
> +{
> + if (user_k)
> + return user_k >= 1 ? user_k : 1;
> + if (nonpreempt)
> + return 1;
> + if (duration_ms <= 12)
> + return 4;
> + if (duration_ms <= 20)
> + return 3;
> + return 2;
> +}
> +
> static struct job_sched_params
> prepare_job_sched_params(int num_threads, int job_timeout_ms, const struct subm_opts *opts)
> {
> @@ -573,12 +654,14 @@ static void throughput_ratio(int pf_fd, int num_vfs, const struct subm_opts *opt
> struct job_sched_params job_sched_params = prepare_job_sched_params(num_vfs + 1,
> job_timeout_ms,
> opts);
> + const unsigned int k = select_inflight_k(job_sched_params.duration_ms,
> + opts->inflight, false);
>
> - igt_info("eq=%ums pt=%uus duration=%ums repeats=%d num_vfs=%d job_timeout=%ums\n",
> + igt_info("eq=%ums pt=%uus duration=%ums repeats=%d inflight=%u num_vfs=%d job_timeout=%ums\n",
> job_sched_params.sched_params.exec_quantum_ms,
> job_sched_params.sched_params.preempt_timeout_us,
> job_sched_params.duration_ms, job_sched_params.num_repeats,
> - num_vfs + 1, job_timeout_ms);
> + k, num_vfs + 1, job_timeout_ms);
>
> init_vf_ids(vf_ids, ARRAY_SIZE(vf_ids),
> &(struct init_vf_ids_opts){ .shuffle = true,
> @@ -607,7 +690,7 @@ static void throughput_ratio(int pf_fd, int num_vfs, const struct subm_opts *opt
> igt_assert_fd(vf_fd);
> set->data[n].opts = opts;
> subm_init(&set->data[n].subm, vf_fd, vf_ids[n], 0,
> - xe_engine(vf_fd, 0)->instance);
> + xe_engine(vf_fd, 0)->instance, k);
> subm_workload_init(&set->data[n].subm,
> &(struct subm_work_desc){
> .duration_ms = job_sched_params.duration_ms,
> @@ -615,6 +698,8 @@ static void throughput_ratio(int pf_fd, int num_vfs, const struct subm_opts *opt
> .repeats = job_sched_params.num_repeats });
> igt_stats_init_with_size(&set->data[n].stats.samples,
> set->data[n].subm.work.repeats);
> + set->data[n].stats.complete_ts = calloc(set->data[n].subm.work.repeats,
> + sizeof(uint64_t));
> if (set->sync_method == SYNC_BARRIER)
> set->data[n].barrier = &set->barrier;
> }
> @@ -670,10 +755,11 @@ static void nonpreempt_engine_resets(int pf_fd, int num_vfs,
> vf_sched_params.preempt_timeout_us / USEC_PER_MSEC;
> int preemptible_end = 1;
> uint8_t vf_ids[num_vfs + 1 /*PF*/];
> + const unsigned int k = select_inflight_k(duration_ms, opts->inflight, true);
>
> - igt_info("eq=%ums pt=%uus duration=%" PRIu64 "ms num_vfs=%d job_timeout=%ums\n",
> + igt_info("eq=%ums pt=%uus duration=%" PRIu64 "ms inflight=%u num_vfs=%d job_timeout=%ums\n",
> vf_sched_params.exec_quantum_ms, vf_sched_params.preempt_timeout_us,
> - duration_ms, num_vfs, job_timeout_ms);
> + duration_ms, k, num_vfs, job_timeout_ms);
>
> init_vf_ids(vf_ids, ARRAY_SIZE(vf_ids),
> &(struct init_vf_ids_opts){ .shuffle = true,
> @@ -702,7 +788,7 @@ static void nonpreempt_engine_resets(int pf_fd, int num_vfs,
> igt_assert_fd(vf_fd);
> set->data[n].opts = opts;
> subm_init(&set->data[n].subm, vf_fd, vf_ids[n], 0,
> - xe_engine(vf_fd, 0)->instance);
> + xe_engine(vf_fd, 0)->instance, k);
> subm_workload_init(&set->data[n].subm,
> &(struct subm_work_desc){
> .duration_ms = duration_ms,
> @@ -710,6 +796,8 @@ static void nonpreempt_engine_resets(int pf_fd, int num_vfs,
> .repeats = MIN_NUM_REPEATS });
> igt_stats_init_with_size(&set->data[n].stats.samples,
> set->data[n].subm.work.repeats);
> + set->data[n].stats.complete_ts = calloc(set->data[n].subm.work.repeats,
> + sizeof(uint64_t));
> if (set->sync_method == SYNC_BARRIER)
> set->data[n].barrier = &set->barrier;
> }
> @@ -738,6 +826,7 @@ static void nonpreempt_engine_resets(int pf_fd, int num_vfs,
> static struct subm_opts subm_opts = {
> .sync_method = SYNC_BARRIER,
> .outlier_treshold = 0.1,
> + .inflight = 0,
> };
>
> static bool extended_scope;
> @@ -764,6 +853,16 @@ static int subm_opts_handler(int opt, int opt_index, void *data)
> subm_opts.outlier_treshold = atoi(optarg) / 100.0;
> igt_info("Outlier threshold: %.2f\n", subm_opts.outlier_treshold);
> break;
> + case 'i': {
> + int val = atoi(optarg);
> +
> + subm_opts.inflight = val > 0 ? val : 0;
> + if (subm_opts.inflight)
> + igt_info("In-flight submissions: %u\n", subm_opts.inflight);
> + else
> + igt_info("In-flight submissions: auto (0)\n");
> + break;
> + }
> default:
> return IGT_OPT_HANDLER_ERROR;
> }
> @@ -777,6 +876,7 @@ static const struct option long_opts[] = {
> { .name = "threshold", .has_arg = true, .val = 't', },
> { .name = "eq_ms", .has_arg = true, .val = 'q', },
> { .name = "pt_us", .has_arg = true, .val = 'p', },
> + { .name = "inflight", .has_arg = true, .val = 'i', },
> {}
> };
>
> @@ -785,7 +885,8 @@ static const char help_str[] =
> " --sync\tThreads synchronization method: 0 - none 1 - barrier (Default 1)\n"
> " --threshold\tSample outlier threshold (Default 0.1)\n"
> " --eq_ms\texec_quantum_ms\n"
> - " --pt_us\tpreempt_timeout_us\n";
> + " --pt_us\tpreempt_timeout_us\n"
> + " --inflight\tNumber of submissions kept in flight per VF (0=auto)\n";
>
> igt_main_args("", long_opts, help_str, subm_opts_handler, NULL)
> {
> --
> 2.31.1
>
More information about the igt-dev
mailing list