/* May fail if out of video memory for texture or vbo upload, or on
* fallback conditions.
*/
static bool brw_try_draw_prims( struct gl_context *ctx,
const struct gl_client_array *arrays[],
const struct _mesa_prim *prim,
GLuint nr_prims,
const struct _mesa_index_buffer *ib,
GLuint min_index,
GLuint max_index )
{
struct intel_context *intel = intel_context(ctx);
struct brw_context *brw = brw_context(ctx);
bool retval = true;
GLuint i;
bool fail_next = false;
if (ctx->NewState)
_mesa_update_state( ctx );
/* We have to validate the textures *before* checking for fallbacks;
* otherwise, the software fallback won't be able to rely on the
* texture state, the firstLevel and lastLevel fields won't be
* set in the intel texture object (they'll both be 0), and the
* software fallback will segfault if it attempts to access any
* texture level other than level 0.
*/
brw_validate_textures( brw );
/* Resolves must occur after updating state and finalizing textures but
* before setting up any hardware state for this draw call.
*/
brw_predraw_resolve_buffers(brw);
/* Bind all inputs, derive varying and size information:
*/
brw_merge_inputs( brw, arrays );
brw->ib.ib = ib;
brw->state.dirty.brw |= BRW_NEW_INDICES;
brw->vb.min_index = min_index;
brw->vb.max_index = max_index;
brw->state.dirty.brw |= BRW_NEW_VERTICES;
/* Have to validate state quite late. Will rebuild tnl_program,
* which depends on varying information.
*
* Note this is where brw->vs->prog_data.inputs_read is calculated,
* so can't access it earlier.
*/
intel_prepare_render(intel);
for (i = 0; i < nr_prims; i++) {
int estimated_max_prim_size;
estimated_max_prim_size = 512; /* batchbuffer commands */
estimated_max_prim_size += (BRW_MAX_TEX_UNIT *
(sizeof(struct brw_sampler_state) +
sizeof(struct gen5_sampler_default_color)));
estimated_max_prim_size += 1024; /* gen6 VS push constants */
estimated_max_prim_size += 1024; /* gen6 WM push constants */
estimated_max_prim_size += 512; /* misc. pad */
/* Flush the batch if it's approaching full, so that we don't wrap while
* we've got validated state that needs to be in the same batch as the
* primitives.
*/
intel_batchbuffer_require_space(intel, estimated_max_prim_size, false);
intel_batchbuffer_save_state(intel);
if (intel->gen < 6)
brw_set_prim(brw, &prim[i]);
else
gen6_set_prim(brw, &prim[i]);
retry:
/* Note that before the loop, brw->state.dirty.brw was set to != 0, and
* that the state updated in the loop outside of this block is that in
* *_set_prim or intel_batchbuffer_flush(), which only impacts
* brw->state.dirty.brw.
*/
if (brw->state.dirty.brw) {
intel->no_batch_wrap = true;
brw_upload_state(brw);
if (unlikely(brw->intel.Fallback)) {
intel->no_batch_wrap = false;
retval = false;
goto out;
}
}
if (intel->gen >= 7)
gen7_emit_prim(brw, &prim[i], brw->primitive);
else
brw_emit_prim(brw, &prim[i], brw->primitive);
intel->no_batch_wrap = false;
if (dri_bufmgr_check_aperture_space(&intel->batch.bo, 1)) {
if (!fail_next) {
intel_batchbuffer_reset_to_saved(intel);
intel_batchbuffer_flush(intel);
fail_next = true;
goto retry;
} else {
if (intel_batchbuffer_flush(intel) == -ENOSPC) {
static bool warned = false;
if (!warned) {
fprintf(stderr, "i965: Single primitive emit exceeded"
"available aperture space\n");
warned = true;
}
retval = false;
}
}
}
if (!_mesa_meta_in_progress(ctx))
brw_update_primitive_count(brw, &prim[i]);
}
if (intel->always_flush_batch)
intel_batchbuffer_flush(intel);
out:
brw_state_cache_check_size(brw);
brw_postdraw_set_buffers_need_resolve(brw);
return retval;
}
static inline void
brw_upload_pipeline_state(struct brw_context *brw,
enum brw_pipeline pipeline)
{
struct gl_context *ctx = &brw->ctx;
int i;
static int dirty_count = 0;
struct brw_state_flags state = brw->state.pipelines[pipeline];
unsigned int fb_samples = _mesa_geometric_samples(ctx->DrawBuffer);
brw_select_pipeline(brw, pipeline);
if (0) {
/* Always re-emit all state. */
brw->NewGLState = ~0;
ctx->NewDriverState = ~0ull;
}
if (pipeline == BRW_RENDER_PIPELINE) {
if (brw->fragment_program != ctx->FragmentProgram._Current) {
brw->fragment_program = ctx->FragmentProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_FRAGMENT_PROGRAM;
}
if (brw->tess_eval_program != ctx->TessEvalProgram._Current) {
brw->tess_eval_program = ctx->TessEvalProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_TESS_PROGRAMS;
}
if (brw->tess_ctrl_program != ctx->TessCtrlProgram._Current) {
brw->tess_ctrl_program = ctx->TessCtrlProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_TESS_PROGRAMS;
}
if (brw->geometry_program != ctx->GeometryProgram._Current) {
brw->geometry_program = ctx->GeometryProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_GEOMETRY_PROGRAM;
}
if (brw->vertex_program != ctx->VertexProgram._Current) {
brw->vertex_program = ctx->VertexProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_VERTEX_PROGRAM;
}
}
if (brw->compute_program != ctx->ComputeProgram._Current) {
brw->compute_program = ctx->ComputeProgram._Current;
brw->ctx.NewDriverState |= BRW_NEW_COMPUTE_PROGRAM;
}
if (brw->meta_in_progress != _mesa_meta_in_progress(ctx)) {
brw->meta_in_progress = _mesa_meta_in_progress(ctx);
brw->ctx.NewDriverState |= BRW_NEW_META_IN_PROGRESS;
}
if (brw->num_samples != fb_samples) {
brw->num_samples = fb_samples;
brw->ctx.NewDriverState |= BRW_NEW_NUM_SAMPLES;
}
/* Exit early if no state is flagged as dirty */
merge_ctx_state(brw, &state);
if ((state.mesa | state.brw) == 0)
return;
/* Emit Sandybridge workaround flushes on every primitive, for safety. */
if (brw->gen == 6)
brw_emit_post_sync_nonzero_flush(brw);
brw_upload_programs(brw, pipeline);
merge_ctx_state(brw, &state);
const struct brw_tracked_state *atoms =
brw_get_pipeline_atoms(brw, pipeline);
const int num_atoms = brw->num_atoms[pipeline];
if (unlikely(INTEL_DEBUG)) {
/* Debug version which enforces various sanity checks on the
* state flags which are generated and checked to help ensure
* state atoms are ordered correctly in the list.
*/
struct brw_state_flags examined, prev;
memset(&examined, 0, sizeof(examined));
prev = state;
for (i = 0; i < num_atoms; i++) {
const struct brw_tracked_state *atom = &atoms[i];
struct brw_state_flags generated;
check_and_emit_atom(brw, &state, atom);
accumulate_state(&examined, &atom->dirty);
/* generated = (prev ^ state)
* if (examined & generated)
* fail;
*/
xor_states(&generated, &prev, &state);
assert(!check_state(&examined, &generated));
prev = state;
}
}
else {
for (i = 0; i < num_atoms; i++) {
const struct brw_tracked_state *atom = &atoms[i];
check_and_emit_atom(brw, &state, atom);
}
}
if (unlikely(INTEL_DEBUG & DEBUG_STATE)) {
STATIC_ASSERT(ARRAY_SIZE(brw_bits) == BRW_NUM_STATE_BITS + 1);
brw_update_dirty_count(mesa_bits, state.mesa);
brw_update_dirty_count(brw_bits, state.brw);
if (dirty_count++ % 1000 == 0) {
brw_print_dirty_count(mesa_bits);
brw_print_dirty_count(brw_bits);
fprintf(stderr, "\n");
}
}
}
/***********************************************************************
* Emit all state:
*/
void brw_upload_state(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
struct brw_state_flags *state = &brw->state.dirty;
int i;
static int dirty_count = 0;
state->mesa |= brw->NewGLState;
brw->NewGLState = 0;
state->brw |= ctx->NewDriverState;
ctx->NewDriverState = 0;
if (0) {
/* Always re-emit all state. */
state->mesa |= ~0;
state->brw |= ~0ull;
}
if (brw->fragment_program != ctx->FragmentProgram._Current) {
brw->fragment_program = ctx->FragmentProgram._Current;
brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
}
if (brw->geometry_program != ctx->GeometryProgram._Current) {
brw->geometry_program = ctx->GeometryProgram._Current;
brw->state.dirty.brw |= BRW_NEW_GEOMETRY_PROGRAM;
}
if (brw->vertex_program != ctx->VertexProgram._Current) {
brw->vertex_program = ctx->VertexProgram._Current;
brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
}
if (brw->meta_in_progress != _mesa_meta_in_progress(ctx)) {
brw->meta_in_progress = _mesa_meta_in_progress(ctx);
brw->state.dirty.brw |= BRW_NEW_META_IN_PROGRESS;
}
if (brw->num_samples != ctx->DrawBuffer->Visual.samples) {
brw->num_samples = ctx->DrawBuffer->Visual.samples;
brw->state.dirty.brw |= BRW_NEW_NUM_SAMPLES;
}
if ((state->mesa | state->brw) == 0)
return;
if (unlikely(INTEL_DEBUG)) {
/* Debug version which enforces various sanity checks on the
* state flags which are generated and checked to help ensure
* state atoms are ordered correctly in the list.
*/
struct brw_state_flags examined, prev;
memset(&examined, 0, sizeof(examined));
prev = *state;
for (i = 0; i < brw->num_atoms; i++) {
const struct brw_tracked_state *atom = brw->atoms[i];
struct brw_state_flags generated;
if (check_state(state, &atom->dirty)) {
atom->emit(brw);
}
accumulate_state(&examined, &atom->dirty);
/* generated = (prev ^ state)
* if (examined & generated)
* fail;
*/
xor_states(&generated, &prev, state);
assert(!check_state(&examined, &generated));
prev = *state;
}
}
else {
for (i = 0; i < brw->num_atoms; i++) {
const struct brw_tracked_state *atom = brw->atoms[i];
if (check_state(state, &atom->dirty)) {
atom->emit(brw);
}
}
}
if (unlikely(INTEL_DEBUG & DEBUG_STATE)) {
STATIC_ASSERT(ARRAY_SIZE(brw_bits) == BRW_NUM_STATE_BITS + 1);
brw_update_dirty_count(mesa_bits, state->mesa);
brw_update_dirty_count(brw_bits, state->brw);
if (dirty_count++ % 1000 == 0) {
brw_print_dirty_count(mesa_bits);
brw_print_dirty_count(brw_bits);
fprintf(stderr, "\n");
}
}
}
