void
intel_winsys_destroy(struct intel_winsys *winsys)
{
if (winsys->decode)
drm_intel_decode_context_free(winsys->decode);
if (winsys->first_gem_ctx)
drm_intel_gem_context_destroy(winsys->first_gem_ctx);
pipe_mutex_destroy(winsys->mutex);
drm_intel_bufmgr_destroy(winsys->bufmgr);
FREE(winsys);
}
int main(int argc, char **argv)
{
int fd;
int devid;
if (argc != 1) {
fprintf(stderr, "usage: %s\n", argv[0]);
exit(-1);
}
fd = drm_open_any();
devid = intel_get_drm_devid(fd);
if (!HAS_BLT_RING(devid)) {
fprintf(stderr, "inter ring check needs gen6+\n");
return 77;
}
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
if (!bufmgr) {
fprintf(stderr, "failed to init libdrm\n");
exit(-1);
}
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, devid);
if (!batch) {
fprintf(stderr, "failed to create batch buffer\n");
exit(-1);
}
target_buffer = drm_intel_bo_alloc(bufmgr, "target bo", 4096, 4096);
if (!target_buffer) {
fprintf(stderr, "failed to alloc target buffer\n");
exit(-1);
}
store_dword_loop(I915_EXEC_RENDER);
drm_intel_bo_unreference(target_buffer);
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
int main(int argc, char **argv)
{
int fd;
int object_size = OBJECT_WIDTH * OBJECT_HEIGHT * 4;
double start_time, end_time;
drm_intel_bo *dst_bo;
drm_intel_bufmgr *bufmgr;
struct intel_batchbuffer *batch;
int i;
fd = drm_open_any();
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
dst_bo = drm_intel_bo_alloc(bufmgr, "dst", object_size, 4096);
/* Prep loop to get us warmed up. */
for (i = 0; i < 60; i++) {
do_render(bufmgr, batch, dst_bo, OBJECT_WIDTH, OBJECT_HEIGHT);
}
drm_intel_bo_wait_rendering(dst_bo);
/* Do the actual timing. */
start_time = get_time_in_secs();
for (i = 0; i < 200; i++) {
do_render(bufmgr, batch, dst_bo, OBJECT_WIDTH, OBJECT_HEIGHT);
}
drm_intel_bo_wait_rendering(dst_bo);
end_time = get_time_in_secs();
printf("%d iterations in %.03f secs: %.01f MB/sec\n", i,
end_time - start_time,
(double)i * OBJECT_WIDTH * OBJECT_HEIGHT * 4 / 1024.0 / 1024.0 /
(end_time - start_time));
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
int main(int argc, char **argv)
{
int fd;
int devid;
if (argc != 1) {
fprintf(stderr, "usage: %s\n", argv[0]);
igt_fail(-1);
}
fd = drm_open_any();
devid = intel_get_drm_devid(fd);
if (HAS_BSD_RING(devid))
num_rings++;
if (HAS_BLT_RING(devid))
num_rings++;
printf("num rings detected: %i\n", num_rings);
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
if (!bufmgr) {
fprintf(stderr, "failed to init libdrm\n");
igt_fail(-1);
}
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, devid);
if (!batch) {
fprintf(stderr, "failed to create batch buffer\n");
igt_fail(-1);
}
mi_lri_loop();
gem_quiescent_gpu(fd);
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
int main(int argc, char **argv)
{
int fd;
int i;
drm_intel_bo *src_bo, *dst_bo;
fd = drm_open_any();
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
src_bo = drm_intel_bo_alloc(bufmgr, "src bo", size, 4096);
dst_bo = drm_intel_bo_alloc(bufmgr, "src bo", size, 4096);
/* The ring we've been using is 128k, and each rendering op
* will use at least 8 dwords:
*
* BATCH_START
* BATCH_START offset
* MI_FLUSH
* STORE_DATA_INDEX
* STORE_DATA_INDEX offset
* STORE_DATA_INDEX value
* MI_USER_INTERRUPT
* (padding)
*
* So iterate just a little more than that -- if we don't fill the ring
* doing this, we aren't likely to with this test.
*/
for (i = 0; i < 128 * 1024 / (8 * 4) * 1.25; i++) {
intel_copy_bo(batch, dst_bo, src_bo, width, height);
intel_batchbuffer_flush(batch);
}
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
struct intel_winsys *
intel_winsys_create_for_fd(int fd)
{
struct intel_winsys *winsys;
winsys = CALLOC_STRUCT(intel_winsys);
if (!winsys)
return NULL;
winsys->fd = fd;
winsys->bufmgr = drm_intel_bufmgr_gem_init(winsys->fd, BATCH_SZ);
if (!winsys->bufmgr) {
debug_error("failed to create GEM buffer manager");
FREE(winsys);
return NULL;
}
pipe_mutex_init(winsys->mutex);
if (!probe_winsys(winsys)) {
drm_intel_bufmgr_destroy(winsys->bufmgr);
FREE(winsys);
return NULL;
}
/*
* No need to implicitly set up a fence register for each non-linear reloc
* entry. When a fence register is needed for a reloc entry,
* drm_intel_bo_emit_reloc_fence() will be called explicitly.
*
* intel_bo_add_reloc() currently lacks "bool fenced" for this to work.
* But we never need a fence register on GEN4+ so we do not need to worry
* about it yet.
*/
drm_intel_bufmgr_gem_enable_fenced_relocs(winsys->bufmgr);
drm_intel_bufmgr_gem_enable_reuse(winsys->bufmgr);
return winsys;
}
struct intel_winsys *
intel_winsys_create_for_fd(int fd)
{
/* so that we can have enough (up to 4094) relocs per bo */
const int batch_size = sizeof(uint32_t) * 8192;
struct intel_winsys *winsys;
winsys = CALLOC_STRUCT(intel_winsys);
if (!winsys)
return NULL;
winsys->fd = fd;
winsys->bufmgr = drm_intel_bufmgr_gem_init(winsys->fd, batch_size);
if (!winsys->bufmgr) {
debug_error("failed to create GEM buffer manager");
FREE(winsys);
return NULL;
}
pipe_mutex_init(winsys->mutex);
if (!probe_winsys(winsys)) {
pipe_mutex_destroy(winsys->mutex);
drm_intel_bufmgr_destroy(winsys->bufmgr);
FREE(winsys);
return NULL;
}
/*
* No need to implicitly set up a fence register for each non-linear reloc
* entry. INTEL_RELOC_FENCE will be set on reloc entries that need them.
*/
drm_intel_bufmgr_gem_enable_fenced_relocs(winsys->bufmgr);
drm_intel_bufmgr_gem_enable_reuse(winsys->bufmgr);
return winsys;
}
int main(int argc, char **argv)
{
drm_intel_bo *src;
int fd;
fd = drm_open_any();
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
src = drm_intel_bo_alloc(bufmgr, "src", 128 * 128, 4096);
bad_blit(src, batch->devid);
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
static void
drm_shutdown(struct buffer *my_buf)
{
drm_intel_bufmgr_destroy(my_buf->bufmgr);
close(my_buf->drm_fd);
}
static void run_test(int fd, int num_fences, int expected_errno,
unsigned flags)
{
struct drm_i915_gem_execbuffer2 execbuf[2];
struct drm_i915_gem_exec_object2 exec[2][2*MAX_FENCES+3];
struct drm_i915_gem_relocation_entry reloc[2*MAX_FENCES+2];
int i, n;
int loop = 1000;
if (flags & BUSY_LOAD) {
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
batch = intel_batchbuffer_alloc(bufmgr, devid);
/* Takes forever otherwise. */
loop = 50;
}
if (flags & INTERRUPTIBLE)
igt_fork_signal_helper();
memset(execbuf, 0, sizeof(execbuf));
memset(exec, 0, sizeof(exec));
memset(reloc, 0, sizeof(reloc));
for (n = 0; n < 2*num_fences; n++) {
uint32_t handle = tiled_bo_create(fd);
exec[1][2*num_fences - n-1].handle = exec[0][n].handle = handle;
fill_reloc(&reloc[n], handle);
}
for (i = 0; i < 2; i++) {
for (n = 0; n < num_fences; n++)
exec[i][n].flags = EXEC_OBJECT_NEEDS_FENCE;
exec[i][2*num_fences].handle = batch_create(fd);
exec[i][2*num_fences].relocs_ptr = (uintptr_t)reloc;
exec[i][2*num_fences].relocation_count = 2*num_fences;
execbuf[i].buffers_ptr = (uintptr_t)exec[i];
execbuf[i].buffer_count = 2*num_fences+1;
execbuf[i].batch_len = 2*sizeof(uint32_t);
}
do {
if (flags & BUSY_LOAD)
emit_dummy_load();
igt_assert_eq(__gem_execbuf(fd, &execbuf[0]), expected_errno);
igt_assert_eq(__gem_execbuf(fd, &execbuf[1]), expected_errno);
} while (--loop);
if (flags & INTERRUPTIBLE)
igt_stop_signal_helper();
/* Cleanup */
for (n = 0; n < 2*num_fences; n++)
gem_close(fd, exec[0][n].handle);
for (i = 0; i < 2; i++)
gem_close(fd, exec[i][2*num_fences].handle);
if (flags & BUSY_LOAD) {
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
}
}
int main(int argc, char **argv)
{
drm_intel_bo *bo[4096];
uint32_t bo_start_val[4096];
uint32_t start = 0;
int fd, i, count;
igt_skip_on_simulation();
fd = drm_open_any();
count = 3 * gem_aperture_size(fd) / (1024*1024) / 2;
if (count > intel_get_total_ram_mb() * 9 / 10) {
count = intel_get_total_ram_mb() * 9 / 10;
printf("not enough RAM to run test, reducing buffer count\n");
}
count |= 1;
printf("Using %d 1MiB buffers\n", count);
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
for (i = 0; i < count; i++) {
bo[i] = create_bo(fd, start);
bo_start_val[i] = start;
/*
printf("Creating bo %d\n", i);
check_bo(bo[i], bo_start_val[i]);
*/
start += 1024 * 1024 / 4;
}
for (i = 0; i < count; i++) {
int src = count - i - 1;
intel_copy_bo(batch, bo[i], bo[src], width, height);
bo_start_val[i] = bo_start_val[src];
}
for (i = 0; i < count * 4; i++) {
int src = random() % count;
int dst = random() % count;
if (src == dst)
continue;
intel_copy_bo(batch, bo[dst], bo[src], width, height);
bo_start_val[dst] = bo_start_val[src];
/*
check_bo(bo[dst], bo_start_val[dst]);
printf("%d: copy bo %d to %d\n", i, src, dst);
*/
}
for (i = 0; i < count; i++) {
/*
printf("check %d\n", i);
*/
check_bo(fd, bo[i], bo_start_val[i]);
drm_intel_bo_unreference(bo[i]);
bo[i] = NULL;
}
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(fd);
return 0;
}
VOID
media_drv_bufmgr_destroy (MEDIA_DRV_CONTEXT * drv_ctx)
{
drm_intel_bufmgr_destroy (drv_ctx->drv_data.bufmgr);
}
static void render_timeout(int fd)
{
drm_intel_bufmgr *bufmgr;
struct intel_batchbuffer *batch;
int64_t timeout = ENOUGH_WORK_IN_SECONDS * NSEC_PER_SEC;
int64_t negative_timeout = -1;
int ret;
const bool do_signals = true; /* signals will seem to make the operation
* use less process CPU time */
bool done = false;
int i, iter = 1;
igt_skip_on_simulation();
bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
drm_intel_bufmgr_gem_enable_reuse(bufmgr);
batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
dst = drm_intel_bo_alloc(bufmgr, "dst", BUF_SIZE, 4096);
dst2 = drm_intel_bo_alloc(bufmgr, "dst2", BUF_SIZE, 4096);
igt_skip_on_f(gem_bo_wait_timeout(fd, dst->handle, &timeout) == -EINVAL,
"kernel doesn't support wait_timeout, skipping test\n");
timeout = ENOUGH_WORK_IN_SECONDS * NSEC_PER_SEC;
/* Figure out a rough number of fills required to consume 1 second of
* GPU work.
*/
do {
struct timespec start, end;
long diff;
#ifndef CLOCK_MONOTONIC_RAW
#define CLOCK_MONOTONIC_RAW CLOCK_MONOTONIC
#endif
igt_assert(clock_gettime(CLOCK_MONOTONIC_RAW, &start) == 0);
for (i = 0; i < iter; i++)
blt_color_fill(batch, dst, BUF_PAGES);
intel_batchbuffer_flush(batch);
drm_intel_bo_wait_rendering(dst);
igt_assert(clock_gettime(CLOCK_MONOTONIC_RAW, &end) == 0);
diff = do_time_diff(&end, &start);
igt_assert(diff >= 0);
if ((diff / MSEC_PER_SEC) > ENOUGH_WORK_IN_SECONDS)
done = true;
else
iter <<= 1;
} while (!done && iter < 1000000);
igt_assert_lt(iter, 1000000);
igt_info("%d iters is enough work\n", iter);
gem_quiescent_gpu(fd);
if (do_signals)
igt_fork_signal_helper();
/* We should be able to do half as much work in the same amount of time,
* but because we might schedule almost twice as much as required, we
* might accidentally time out. Hence add some fudge. */
for (i = 0; i < iter/3; i++)
blt_color_fill(batch, dst2, BUF_PAGES);
intel_batchbuffer_flush(batch);
igt_assert(gem_bo_busy(fd, dst2->handle) == true);
igt_assert_eq(gem_bo_wait_timeout(fd, dst2->handle, &timeout), 0);
igt_assert(gem_bo_busy(fd, dst2->handle) == false);
igt_assert_neq(timeout, 0);
if (timeout == (ENOUGH_WORK_IN_SECONDS * NSEC_PER_SEC))
igt_info("Buffer was already done!\n");
else {
igt_info("Finished with %" PRIu64 " time remaining\n", timeout);
}
/* check that polling with timeout=0 works. */
timeout = 0;
igt_assert_eq(gem_bo_wait_timeout(fd, dst2->handle, &timeout), 0);
igt_assert_eq(timeout, 0);
/* Now check that we correctly time out, twice the auto-tune load should
* be good enough. */
timeout = ENOUGH_WORK_IN_SECONDS * NSEC_PER_SEC;
for (i = 0; i < iter*2; i++)
blt_color_fill(batch, dst2, BUF_PAGES);
intel_batchbuffer_flush(batch);
ret = gem_bo_wait_timeout(fd, dst2->handle, &timeout);
igt_assert_eq(ret, -ETIME);
igt_assert_eq(timeout, 0);
igt_assert(gem_bo_busy(fd, dst2->handle) == true);
/* check that polling with timeout=0 works. */
timeout = 0;
igt_assert_eq(gem_bo_wait_timeout(fd, dst2->handle, &timeout), -ETIME);
igt_assert_eq(timeout, 0);
/* Now check that we can pass negative (infinite) timeouts. */
negative_timeout = -1;
for (i = 0; i < iter; i++)
blt_color_fill(batch, dst2, BUF_PAGES);
intel_batchbuffer_flush(batch);
igt_assert_eq(gem_bo_wait_timeout(fd, dst2->handle, &negative_timeout), 0);
igt_assert_eq(negative_timeout, -1); /* infinity always remains */
igt_assert(gem_bo_busy(fd, dst2->handle) == false);
if (do_signals)
igt_stop_signal_helper();
drm_intel_bo_unreference(dst2);
drm_intel_bo_unreference(dst);
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
}
int main(int argc, char **argv)
{
int i, j;
unsigned *current_permutation, *tmp_permutation;
pid_t signal_helper = -1;
drm_fd = drm_open_any();
devid = intel_get_drm_devid(drm_fd);
parse_options(argc, argv);
/* start our little helper early before too may allocations occur */
signal(SIGUSR1, SIG_IGN);
if (options.use_signal_helper)
signal_helper = fork_signal_helper();
init();
check_render_copyfunc();
tile_permutation = malloc(num_total_tiles*sizeof(uint32_t));
current_permutation = malloc(num_total_tiles*sizeof(uint32_t));
tmp_permutation = malloc(num_total_tiles*sizeof(uint32_t));
assert(tile_permutation);
assert(current_permutation);
assert(tmp_permutation);
fan_out();
for (i = 0; i < options.total_rounds; i++) {
printf("round %i\n", i);
if (i % 64 == 63) {
fan_in_and_check();
printf("everything correct after %i rounds\n", i + 1);
}
target_set = (current_set + 1) & 1;
init_set(target_set);
for (j = 0; j < num_total_tiles; j++)
current_permutation[j] = j;
permute_array(current_permutation, num_total_tiles, exchange_uint);
copy_tiles(current_permutation);
memcpy(tmp_permutation, tile_permutation, sizeof(unsigned)*num_total_tiles);
/* accumulate the permutations */
for (j = 0; j < num_total_tiles; j++)
tile_permutation[j] = current_permutation[tmp_permutation[j]];
current_set = target_set;
}
fan_in_and_check();
fprintf(stderr, "num failed tiles %u, max incoherent bytes %lu\n",
stats.num_failed, stats.max_failed_reads*sizeof(uint32_t));
intel_batchbuffer_free(batch);
drm_intel_bufmgr_destroy(bufmgr);
close(drm_fd);
if (signal_helper != -1)
kill(signal_helper, SIGQUIT);
return 0;
}