void tpanel::impl_draw_foreground(
surface& frame_buffer
, int x_offset
, int y_offset)
{
std::vector<int> anims;
canvas(1).blit(
frame_buffer
, calculate_blitting_rectangle(x_offset, y_offset), get_dirty(), anims, anims);
}
inline
Uint32 * DynArr256::get(Uint32 pos) const
{
#if defined VM_TRACE || defined ERROR_INSERT
// In debug this function will abort if used
// with pos not already mapped by call to set.
// Use get_dirty if return NULL is wanted instead.
require((m_head.m_sz > 0) && (pos <= m_head.m_high_pos));
#endif
return get_dirty(pos);
}
static GLuint
get_state_size(struct i915_hw_state *state)
{
GLuint dirty = get_dirty(state);
GLuint i;
GLuint sz = 0;
if (dirty & I915_UPLOAD_INVARIENT)
sz += 30 * 4;
if (dirty & I915_UPLOAD_RASTER_RULES)
sz += sizeof(state->RasterRules);
if (dirty & I915_UPLOAD_CTX)
sz += sizeof(state->Ctx);
if (dirty & I915_UPLOAD_BLEND)
sz += sizeof(state->Blend);
if (dirty & I915_UPLOAD_BUFFERS)
sz += sizeof(state->Buffer);
if (dirty & I915_UPLOAD_STIPPLE)
sz += sizeof(state->Stipple);
if (dirty & I915_UPLOAD_TEX_ALL) {
int nr = 0;
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i))
nr++;
sz += (2 + nr * 3) * sizeof(GLuint) * 2;
}
if (dirty & I915_UPLOAD_CONSTANTS)
sz += state->ConstantSize * sizeof(GLuint);
if (dirty & I915_UPLOAD_PROGRAM)
sz += state->ProgramSize * sizeof(GLuint);
return sz;
}
static GLuint
get_state_size(struct i915_hw_state *state)
{
GLuint dirty = get_dirty(state);
GLuint i;
GLuint sz = 0;
if (dirty & I915_UPLOAD_CTX)
sz += sizeof(state->Ctx);
if (dirty & I915_UPLOAD_BUFFERS)
sz += sizeof(state->Buffer);
if (dirty & I915_UPLOAD_STIPPLE)
sz += sizeof(state->Stipple);
if (dirty & I915_UPLOAD_FOG)
sz += sizeof(state->Fog);
if (dirty & I915_UPLOAD_TEX_ALL) {
int nr = 0;
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i))
nr++;
sz += (2 + nr * 3) * sizeof(GLuint) * 2;
}
if (dirty & I915_UPLOAD_CONSTANTS)
sz += state->ConstantSize * sizeof(GLuint);
if (dirty & I915_UPLOAD_PROGRAM)
sz += state->ProgramSize * sizeof(GLuint);
return sz;
}
static int aio_event_thread(void *data)
{
struct aio_threadinfo *tinfo = data;
struct aio_output *output = tinfo->output;
struct aio_threadinfo *other = &output->tinfo[2];
int err = -ENOMEM;
MARS_DBG("event thread has started.\n");
//set_user_nice(current, -20);
use_fake_mm();
if (!current->mm)
goto err;
err = aio_start_thread(output, &output->tinfo[2], aio_sync_thread, 'y');
if (unlikely(err < 0))
goto err;
while (!brick_thread_should_stop() || atomic_read(&tinfo->queued_sum) > 0) {
mm_segment_t oldfs;
int count;
int i;
struct timespec timeout = {
.tv_sec = 1,
};
struct io_event events[MARS_MAX_AIO_READ];
oldfs = get_fs();
set_fs(get_ds());
/* TODO: don't timeout upon termination.
* Probably we should submit a dummy request.
*/
count = sys_io_getevents(output->ctxp, 1, MARS_MAX_AIO_READ, events, &timeout);
set_fs(oldfs);
if (likely(count > 0)) {
atomic_sub(count, &output->submit_count);
}
for (i = 0; i < count; i++) {
struct aio_mref_aspect *mref_a = (void*)events[i].data;
struct mref_object *mref;
int err = events[i].res;
if (!mref_a) {
continue; // this was a dummy request
}
mref = mref_a->object;
MARS_IO("AIO done %p pos = %lld len = %d rw = %d\n", mref, mref->ref_pos, mref->ref_len, mref->ref_rw);
mapfree_set(output->mf, mref->ref_pos, mref->ref_pos + mref->ref_len);
if (output->brick->o_fdsync
&& err >= 0
&& mref->ref_rw != READ
&& !mref->ref_skip_sync
&& !mref_a->resubmit++) {
// workaround for non-implemented AIO FSYNC operation
if (output->mf &&
output->mf->mf_filp &&
output->mf->mf_filp->f_op &&
!output->mf->mf_filp->f_op->aio_fsync) {
mars_trace(mref, "aio_fsync");
_enqueue(other, mref_a, mref->ref_prio, true);
continue;
}
err = aio_submit(output, mref_a, true);
if (likely(err >= 0))
continue;
}
_complete(output, mref_a, err);
}
}
err = 0;
err:
MARS_DBG("event thread has stopped, err = %d\n", err);
aio_stop_thread(output, 2, false);
unuse_fake_mm();
tinfo->terminated = true;
wake_up_interruptible_all(&tinfo->terminate_event);
return err;
}
#if 1
/* This should go to fs/open.c (as long as vfs_submit() is not implemented)
*/
#include <linux/fdtable.h>
void fd_uninstall(unsigned int fd)
{
struct files_struct *files = current->files;
struct fdtable *fdt;
MARS_DBG("fd = %d\n", fd);
if (unlikely(fd < 0)) {
MARS_ERR("bad fd = %d\n", fd);
return;
}
spin_lock(&files->file_lock);
fdt = files_fdtable(files);
rcu_assign_pointer(fdt->fd[fd], NULL);
spin_unlock(&files->file_lock);
}
EXPORT_SYMBOL(fd_uninstall);
#endif
static
atomic_t ioctx_count = ATOMIC_INIT(0);
static
void _destroy_ioctx(struct aio_output *output)
{
if (unlikely(!output))
goto done;
aio_stop_thread(output, 1, true);
use_fake_mm();
if (likely(output->ctxp)) {
mm_segment_t oldfs;
int err;
MARS_DBG("ioctx count = %d destroying %p\n", atomic_read(&ioctx_count), (void*)output->ctxp);
oldfs = get_fs();
set_fs(get_ds());
err = sys_io_destroy(output->ctxp);
set_fs(oldfs);
atomic_dec(&ioctx_count);
MARS_DBG("ioctx count = %d status = %d\n", atomic_read(&ioctx_count), err);
output->ctxp = 0;
}
if (likely(output->fd >= 0)) {
MARS_DBG("destroying fd %d\n", output->fd);
fd_uninstall(output->fd);
put_unused_fd(output->fd);
output->fd = -1;
}
done:
if (likely(current->mm)) {
unuse_fake_mm();
}
}
static
int _create_ioctx(struct aio_output *output)
{
struct file *file;
mm_segment_t oldfs;
int err = -EINVAL;
CHECK_PTR_NULL(output, done);
CHECK_PTR_NULL(output->mf, done);
file = output->mf->mf_filp;
CHECK_PTR_NULL(file, done);
/* TODO: this is provisionary. We only need it for sys_io_submit()
* which uses userspace concepts like file handles.
* This should be accompanied by a future kernelsapce vfs_submit() or
* do_submit() which currently does not exist :(
*/
err = get_unused_fd();
MARS_DBG("file %p '%s' new fd = %d\n", file, output->mf->mf_name, err);
if (unlikely(err < 0)) {
MARS_ERR("cannot get fd, err=%d\n", err);
goto done;
}
output->fd = err;
fd_install(err, file);
use_fake_mm();
err = -ENOMEM;
if (unlikely(!current->mm)) {
MARS_ERR("cannot fake mm\n");
goto done;
}
MARS_DBG("ioctx count = %d old = %p\n", atomic_read(&ioctx_count), (void*)output->ctxp);
output->ctxp = 0;
oldfs = get_fs();
set_fs(get_ds());
err = sys_io_setup(MARS_MAX_AIO, &output->ctxp);
set_fs(oldfs);
if (likely(output->ctxp))
atomic_inc(&ioctx_count);
MARS_DBG("ioctx count = %d new = %p status = %d\n", atomic_read(&ioctx_count), (void*)output->ctxp, err);
if (unlikely(err < 0)) {
MARS_ERR("io_setup failed, err=%d\n", err);
goto done;
}
err = aio_start_thread(output, &output->tinfo[1], aio_event_thread, 'e');
if (unlikely(err < 0)) {
MARS_ERR("could not start event thread\n");
goto done;
}
done:
if (likely(current->mm)) {
unuse_fake_mm();
}
return err;
}
static int aio_submit_thread(void *data)
{
struct aio_threadinfo *tinfo = data;
struct aio_output *output = tinfo->output;
struct file *file;
int err = -EINVAL;
MARS_DBG("submit thread has started.\n");
file = output->mf->mf_filp;
use_fake_mm();
while (!brick_thread_should_stop() || atomic_read(&output->read_count) + atomic_read(&output->write_count) + atomic_read(&tinfo->queued_sum) > 0) {
struct aio_mref_aspect *mref_a;
struct mref_object *mref;
int sleeptime;
int status;
wait_event_interruptible_timeout(
tinfo->event,
atomic_read(&tinfo->queued_sum) > 0,
HZ / 4);
mref_a = _dequeue(tinfo);
if (!mref_a) {
continue;
}
mref = mref_a->object;
status = -EINVAL;
CHECK_PTR(mref, error);
mapfree_set(output->mf, mref->ref_pos, -1);
if (mref->ref_rw) {
insert_dirty(output, mref_a);
}
// check for reads exactly at EOF (special case)
if (mref->ref_pos == mref->ref_total_size &&
!mref->ref_rw &&
mref->ref_timeout > 0) {
loff_t total_size = i_size_read(file->f_mapping->host);
loff_t len = total_size - mref->ref_pos;
if (len > 0) {
mref->ref_total_size = total_size;
mref->ref_len = len;
} else {
if (!mref_a->start_jiffies) {
mref_a->start_jiffies = jiffies;
}
if ((long long)jiffies - mref_a->start_jiffies <= mref->ref_timeout) {
if (atomic_read(&tinfo->queued_sum) <= 0) {
atomic_inc(&output->total_msleep_count);
brick_msleep(1000 * 4 / HZ);
}
_enqueue(tinfo, mref_a, MARS_PRIO_LOW, true);
continue;
}
MARS_DBG("ENODATA %lld\n", len);
_complete(output, mref_a, -ENODATA);
continue;
}
}
sleeptime = 1;
for (;;) {
status = aio_submit(output, mref_a, false);
if (likely(status != -EAGAIN)) {
break;
}
atomic_inc(&output->total_delay_count);
brick_msleep(sleeptime);
if (sleeptime < 100) {
sleeptime++;
}
}
error:
if (unlikely(status < 0)) {
MARS_IO("submit_count = %d status = %d\n", atomic_read(&output->submit_count), status);
_complete_mref(output, mref, status);
}
}
MARS_DBG("submit thread has stopped, status = %d.\n", err);
if (likely(current->mm)) {
unuse_fake_mm();
}
tinfo->terminated = true;
wake_up_interruptible_all(&tinfo->terminate_event);
return err;
}
static int aio_get_info(struct aio_output *output, struct mars_info *info)
{
struct file *file;
loff_t min;
loff_t max;
if (unlikely(!output ||
!output->mf ||
!(file = output->mf->mf_filp) ||
!file->f_mapping ||
!file->f_mapping->host))
return -EINVAL;
info->tf_align = 1;
info->tf_min_size = 1;
/* Workaround for races in the page cache.
*
* It appears that concurrent reads and writes seem to
* result in inconsistent reads in some very rare cases, due to
* races. Sometimes, the inode claims that the file has been already
* appended by a write operation, but the data has not actually hit
* the page cache, such that a concurrent read gets NULL blocks.
*/
min = i_size_read(file->f_mapping->host);
max = 0;
if (!output->brick->is_static_device) {
get_dirty(output, &min, &max);
}
info->current_size = min;
MARS_DBG("determined file size = %lld\n", info->current_size);
return 0;
}
//////////////// informational / statistics ///////////////
static noinline
char *aio_statistics(struct aio_brick *brick, int verbose)
{
struct aio_output *output = brick->outputs[0];
char *res = brick_string_alloc(4096);
char *sync = NULL;
int pos = 0;
if (!res)
return NULL;
pos += report_timing(&timings[0], res + pos, 4096 - pos);
pos += report_timing(&timings[1], res + pos, 4096 - pos);
pos += report_timing(&timings[2], res + pos, 4096 - pos);
snprintf(res + pos, 4096 - pos,
"total "
"reads = %d "
"writes = %d "
"allocs = %d "
"submits = %d "
"again = %d "
"delays = %d "
"msleeps = %d "
"fdsyncs = %d "
"fdsync_waits = %d "
"map_free = %d | "
"flying reads = %d "
"writes = %d "
"allocs = %d "
"submits = %d "
"q0 = %d "
"q1 = %d "
"q2 = %d "
"| total "
"q0 = %d "
"q1 = %d "
"q2 = %d "
"%s\n",
atomic_read(&output->total_read_count),
atomic_read(&output->total_write_count),
atomic_read(&output->total_alloc_count),
atomic_read(&output->total_submit_count),
atomic_read(&output->total_again_count),
atomic_read(&output->total_delay_count),
atomic_read(&output->total_msleep_count),
atomic_read(&output->total_fdsync_count),
atomic_read(&output->total_fdsync_wait_count),
atomic_read(&output->total_mapfree_count),
atomic_read(&output->read_count),
atomic_read(&output->write_count),
atomic_read(&output->alloc_count),
atomic_read(&output->submit_count),
atomic_read(&output->tinfo[0].queued_sum),
atomic_read(&output->tinfo[1].queued_sum),
atomic_read(&output->tinfo[2].queued_sum),
atomic_read(&output->tinfo[0].total_enqueue_count),
atomic_read(&output->tinfo[1].total_enqueue_count),
atomic_read(&output->tinfo[2].total_enqueue_count),
sync ? sync : "");
if (sync)
brick_string_free(sync);
return res;
}
static noinline
void aio_reset_statistics(struct aio_brick *brick)
{
struct aio_output *output = brick->outputs[0];
int i;
atomic_set(&output->total_read_count, 0);
atomic_set(&output->total_write_count, 0);
atomic_set(&output->total_alloc_count, 0);
atomic_set(&output->total_submit_count, 0);
atomic_set(&output->total_again_count, 0);
atomic_set(&output->total_delay_count, 0);
atomic_set(&output->total_msleep_count, 0);
atomic_set(&output->total_fdsync_count, 0);
atomic_set(&output->total_fdsync_wait_count, 0);
atomic_set(&output->total_mapfree_count, 0);
for (i = 0; i < 3; i++) {
struct aio_threadinfo *tinfo = &output->tinfo[i];
atomic_set(&tinfo->total_enqueue_count, 0);
}
}
//////////////// object / aspect constructors / destructors ///////////////
static int aio_mref_aspect_init_fn(struct generic_aspect *_ini)
{
struct aio_mref_aspect *ini = (void*)_ini;
INIT_LIST_HEAD(&ini->io_head);
INIT_LIST_HEAD(&ini->dirty_head);
return 0;
}
static void aio_mref_aspect_exit_fn(struct generic_aspect *_ini)
{
struct aio_mref_aspect *ini = (void*)_ini;
CHECK_HEAD_EMPTY(&ini->dirty_head);
CHECK_HEAD_EMPTY(&ini->io_head);
}
MARS_MAKE_STATICS(aio);
////////////////////// brick constructors / destructors ////////////////////
static int aio_brick_construct(struct aio_brick *brick)
{
return 0;
}
static int aio_switch(struct aio_brick *brick)
{
static int index;
struct aio_output *output = brick->outputs[0];
const char *path = output->brick->brick_path;
int flags = O_RDWR | O_LARGEFILE;
int status = 0;
MARS_DBG("power.button = %d\n", brick->power.button);
if (!brick->power.button)
goto cleanup;
if (brick->power.led_on || output->mf)
goto done;
mars_power_led_off((void*)brick, false);
if (brick->o_creat) {
flags |= O_CREAT;
MARS_DBG("using O_CREAT on %s\n", path);
}
if (brick->o_direct) {
flags |= O_DIRECT;
MARS_DBG("using O_DIRECT on %s\n", path);
}
output->mf = mapfree_get(path, flags);
if (unlikely(!output->mf)) {
MARS_ERR("could not open file = '%s' flags = %d\n", path, flags);
status = -ENOENT;
goto err;
}
output->index = ++index;
status = _create_ioctx(output);
if (unlikely(status < 0)) {
MARS_ERR("could not create ioctx, status = %d\n", status);
goto err;
}
status = aio_start_thread(output, &output->tinfo[0], aio_submit_thread, 's');
if (unlikely(status < 0)) {
MARS_ERR("could not start theads, status = %d\n", status);
goto err;
}
MARS_DBG("opened file '%s'\n", path);
mars_power_led_on((void*)brick, true);
done:
return 0;
err:
MARS_ERR("status = %d\n", status);
cleanup:
if (brick->power.led_off) {
goto done;
}
mars_power_led_on((void*)brick, false);
aio_stop_thread(output, 0, false);
_destroy_ioctx(output);
mars_power_led_off((void*)brick,
(output->tinfo[0].thread == NULL &&
output->tinfo[1].thread == NULL &&
output->tinfo[2].thread == NULL));
MARS_DBG("switch off led_off = %d status = %d\n", brick->power.led_off, status);
if (brick->power.led_off) {
if (output->mf) {
MARS_DBG("closing file = '%s'\n", output->mf->mf_name);
mapfree_put(output->mf);
output->mf = NULL;
}
}
return status;
}
static int aio_output_construct(struct aio_output *output)
{
INIT_LIST_HEAD(&output->dirty_anchor);
spin_lock_init(&output->dirty_lock);
init_waitqueue_head(&output->fdsync_event);
output->fd = -1;
return 0;
}
/* Push the state into the sarea and/or texture memory.
*/
static void
i915_emit_state(struct intel_context *intel)
{
struct i915_context *i915 = i915_context(&intel->ctx);
struct i915_hw_state *state = &i915->state;
int i, count, aper_count;
GLuint dirty;
drm_intel_bo *aper_array[3 + I915_TEX_UNITS];
GET_CURRENT_CONTEXT(ctx);
BATCH_LOCALS;
/* We don't hold the lock at this point, so want to make sure that
* there won't be a buffer wrap between the state emits and the primitive
* emit header.
*
* It might be better to talk about explicit places where
* scheduling is allowed, rather than assume that it is whenever a
* batchbuffer fills up.
*/
intel_batchbuffer_require_space(intel,
get_state_size(state) +
INTEL_PRIM_EMIT_SIZE);
count = 0;
again:
if (intel->batch.bo == NULL) {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "i915 emit state");
assert(0);
}
aper_count = 0;
dirty = get_dirty(state);
aper_array[aper_count++] = intel->batch.bo;
if (dirty & I915_UPLOAD_BUFFERS) {
if (state->draw_region)
aper_array[aper_count++] = state->draw_region->bo;
if (state->depth_region)
aper_array[aper_count++] = state->depth_region->bo;
}
if (dirty & I915_UPLOAD_TEX_ALL) {
for (i = 0; i < I915_TEX_UNITS; i++) {
if (dirty & I915_UPLOAD_TEX(i)) {
if (state->tex_buffer[i]) {
aper_array[aper_count++] = state->tex_buffer[i];
}
}
}
}
if (dri_bufmgr_check_aperture_space(aper_array, aper_count)) {
if (count == 0) {
count++;
intel_batchbuffer_flush(intel);
goto again;
} else {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "i915 emit state");
assert(0);
}
}
/* work out list of buffers to emit */
/* Do this here as we may have flushed the batchbuffer above,
* causing more state to be dirty!
*/
dirty = get_dirty(state);
state->emitted |= dirty;
assert(get_dirty(state) == 0);
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
if (dirty & I915_UPLOAD_INVARIENT) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_INVARIENT:\n");
i915_emit_invarient_state(intel);
}
if (dirty & I915_UPLOAD_RASTER_RULES) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_RASTER_RULES:\n");
emit(intel, state->RasterRules, sizeof(state->RasterRules));
}
if (dirty & I915_UPLOAD_CTX) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_CTX:\n");
emit(intel, state->Ctx, sizeof(state->Ctx));
}
if (dirty & I915_UPLOAD_BLEND) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_BLEND:\n");
emit(intel, state->Blend, sizeof(state->Blend));
}
if (dirty & I915_UPLOAD_BUFFERS) {
GLuint count;
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_BUFFERS:\n");
count = 17;
if (state->Buffer[I915_DESTREG_DRAWRECT0] != MI_NOOP)
count++;
BEGIN_BATCH(count);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
if (state->draw_region) {
OUT_RELOC(state->draw_region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
} else {
OUT_BATCH(0);
}
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
if (state->depth_region) {
OUT_RELOC(state->depth_region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
} else {
OUT_BATCH(0);
}
OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
if (state->Buffer[I915_DESTREG_DRAWRECT0] != MI_NOOP)
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT1]);
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT2]);
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT3]);
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT4]);
OUT_BATCH(state->Buffer[I915_DESTREG_DRAWRECT5]);
ADVANCE_BATCH();
}
if (dirty & I915_UPLOAD_STIPPLE) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_STIPPLE:\n");
emit(intel, state->Stipple, sizeof(state->Stipple));
}
/* Combine all the dirty texture state into a single command to
* avoid lockups on I915 hardware.
*/
if (dirty & I915_UPLOAD_TEX_ALL) {
int nr = 0;
GLuint unwind;
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i))
nr++;
BEGIN_BATCH(2 + nr * 3);
OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
OUT_RELOC(state->tex_buffer[i],
I915_GEM_DOMAIN_SAMPLER, 0,
state->tex_offset[i]);
OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
}
ADVANCE_BATCH();
unwind = intel->batch.used;
BEGIN_BATCH(2 + nr * 3);
OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
}
ADVANCE_BATCH();
if (i915->last_sampler &&
memcmp(intel->batch.map + i915->last_sampler,
intel->batch.map + unwind,
(2 + nr*3)*sizeof(int)) == 0)
intel->batch.used = unwind;
else
i915->last_sampler = unwind;
}
/* Push the state into the sarea and/or texture memory.
*/
static void
i915_emit_state(struct intel_context *intel)
{
struct i915_context *i915 = i915_context(&intel->ctx);
struct i915_hw_state *state = i915->current;
int i;
GLuint dirty;
BATCH_LOCALS;
/* We don't hold the lock at this point, so want to make sure that
* there won't be a buffer wrap.
*
* It might be better to talk about explicit places where
* scheduling is allowed, rather than assume that it is whenever a
* batchbuffer fills up.
*/
intel_batchbuffer_require_space(intel->batch, get_state_size(state), 0);
/* Do this here as we may have flushed the batchbuffer above,
* causing more state to be dirty!
*/
dirty = get_dirty(state);
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
if (dirty & I915_UPLOAD_INVARIENT) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_INVARIENT:\n");
i915_emit_invarient_state(intel);
}
if (dirty & I915_UPLOAD_CTX) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_CTX:\n");
emit(intel, state->Ctx, sizeof(state->Ctx));
}
if (dirty & I915_UPLOAD_BUFFERS) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_BUFFERS:\n");
BEGIN_BATCH(I915_DEST_SETUP_SIZE + 2, 0);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
OUT_RELOC(state->draw_region->buffer,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
state->draw_region->draw_offset);
if (state->depth_region) {
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
OUT_RELOC(state->depth_region->buffer,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
state->depth_region->draw_offset);
}
OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
ADVANCE_BATCH();
}
if (dirty & I915_UPLOAD_STIPPLE) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_STIPPLE:\n");
emit(intel, state->Stipple, sizeof(state->Stipple));
}
if (dirty & I915_UPLOAD_FOG) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_FOG:\n");
emit(intel, state->Fog, sizeof(state->Fog));
}
/* Combine all the dirty texture state into a single command to
* avoid lockups on I915 hardware.
*/
if (dirty & I915_UPLOAD_TEX_ALL) {
int nr = 0;
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i))
nr++;
BEGIN_BATCH(2 + nr * 3, 0);
OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
if (state->tex_buffer[i]) {
OUT_RELOC(state->tex_buffer[i],
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
DRM_BO_MASK_MEM | DRM_BO_FLAG_READ,
state->tex_offset[i]);
}
else {
assert(i == 0);
assert(state == &i915->meta);
OUT_BATCH(0);
}
OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
}
ADVANCE_BATCH();
BEGIN_BATCH(2 + nr * 3, 0);
OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
}
ADVANCE_BATCH();
}
/* Push the state into the sarea and/or texture memory.
*/
static void
i915_emit_state(struct intel_context *intel)
{
struct i915_context *i915 = i915_context(&intel->ctx);
struct i915_hw_state *state = i915->current;
int i;
int ret, count;
GLuint dirty;
GET_CURRENT_CONTEXT(ctx);
BATCH_LOCALS;
/* We don't hold the lock at this point, so want to make sure that
* there won't be a buffer wrap between the state emits and the primitive
* emit header.
*
* It might be better to talk about explicit places where
* scheduling is allowed, rather than assume that it is whenever a
* batchbuffer fills up.
*
* Set the space as LOOP_CLIPRECTS now, since that's what our primitives
* will be emitted under.
*/
intel_batchbuffer_require_space(intel->batch, get_state_size(state) + 8,
LOOP_CLIPRECTS);
count = 0;
again:
dirty = get_dirty(state);
ret = 0;
if (dirty & I915_UPLOAD_BUFFERS) {
ret |= dri_bufmgr_check_aperture_space(state->draw_region->buffer);
if (state->depth_region)
ret |= dri_bufmgr_check_aperture_space(state->depth_region->buffer);
}
if (dirty & I915_UPLOAD_TEX_ALL) {
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
if (state->tex_buffer[i]) {
ret |= dri_bufmgr_check_aperture_space(state->tex_buffer[i]);
}
}
}
if (ret) {
if (count == 0) {
count++;
intel_batchbuffer_flush(intel->batch);
goto again;
} else {
_mesa_error(ctx, GL_OUT_OF_MEMORY, "i915 emit state");
assert(0);
}
}
/* work out list of buffers to emit */
/* Do this here as we may have flushed the batchbuffer above,
* causing more state to be dirty!
*/
dirty = get_dirty(state);
state->emitted |= dirty;
assert(get_dirty(state) == 0);
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
if (dirty & I915_UPLOAD_INVARIENT) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_INVARIENT:\n");
i915_emit_invarient_state(intel);
}
if (dirty & I915_UPLOAD_CTX) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_CTX:\n");
emit(intel, state->Ctx, sizeof(state->Ctx));
}
if (dirty & I915_UPLOAD_BUFFERS) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_BUFFERS:\n");
BEGIN_BATCH(I915_DEST_SETUP_SIZE + 2, IGNORE_CLIPRECTS);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
OUT_RELOC(state->draw_region->buffer,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
state->draw_region->draw_offset);
if (state->depth_region) {
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
OUT_RELOC(state->depth_region->buffer,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
state->depth_region->draw_offset);
}
OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
ADVANCE_BATCH();
}
if (dirty & I915_UPLOAD_STIPPLE) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_STIPPLE:\n");
emit(intel, state->Stipple, sizeof(state->Stipple));
}
if (dirty & I915_UPLOAD_FOG) {
if (INTEL_DEBUG & DEBUG_STATE)
fprintf(stderr, "I915_UPLOAD_FOG:\n");
emit(intel, state->Fog, sizeof(state->Fog));
}
/* Combine all the dirty texture state into a single command to
* avoid lockups on I915 hardware.
*/
if (dirty & I915_UPLOAD_TEX_ALL) {
int nr = 0;
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i))
nr++;
BEGIN_BATCH(2 + nr * 3, IGNORE_CLIPRECTS);
OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
if (state->tex_buffer[i]) {
OUT_RELOC(state->tex_buffer[i],
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
state->tex_offset[i]);
}
else if (state == &i915->meta) {
assert(i == 0);
OUT_BATCH(0);
}
else {
OUT_BATCH(state->tex_offset[i]);
}
OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
}
ADVANCE_BATCH();
BEGIN_BATCH(2 + nr * 3, IGNORE_CLIPRECTS);
OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3 * nr));
OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
for (i = 0; i < I915_TEX_UNITS; i++)
if (dirty & I915_UPLOAD_TEX(i)) {
OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
}
ADVANCE_BATCH();
}
