#include "brw_ff_gs.h"
#include "brw_gs.h"
#include "brw_wm.h"
+#include "brw_cs.h"
+#include "main/framebuffer.h"
static const struct brw_tracked_state *gen4_atoms[] =
{
&brw_vertices,
};
-static const struct brw_tracked_state *gen7_atoms[] =
+static const struct brw_tracked_state *gen7_render_atoms[] =
{
/* Command packets: */
&haswell_cut_index,
};
-static const struct brw_tracked_state *gen8_atoms[] =
+static const struct brw_tracked_state *gen7_compute_atoms[] =
+{
+ &brw_state_base_address,
+ &brw_cs_abo_surfaces,
+ &brw_cs_state,
+};
+
+static const struct brw_tracked_state *gen8_render_atoms[] =
{
/* Command packets: */
&gen8_state_base_address,
&gen8_pma_fix,
};
+static const struct brw_tracked_state *gen8_compute_atoms[] =
+{
+ &gen8_state_base_address,
+ &brw_cs_abo_surfaces,
+ &brw_cs_state,
+};
+
static void
brw_upload_initial_gpu_state(struct brw_context *brw)
{
brw_upload_invariant_state(brw);
+ /* Recommended optimization for Victim Cache eviction in pixel backend. */
+ if (brw->gen >= 9) {
+ BEGIN_BATCH(3);
+ OUT_BATCH(MI_LOAD_REGISTER_IMM | (3 - 2));
+ OUT_BATCH(GEN7_CACHE_MODE_1);
+ OUT_BATCH((GEN9_PARTIAL_RESOLVE_DISABLE_IN_VC << 16) |
+ GEN9_PARTIAL_RESOLVE_DISABLE_IN_VC);
+ ADVANCE_BATCH();
+ }
+
if (brw->gen >= 8) {
gen8_emit_3dstate_sample_pattern(brw);
}
}
+static inline const struct brw_tracked_state *
+brw_get_pipeline_atoms(struct brw_context *brw,
+ enum brw_pipeline pipeline)
+{
+ switch (pipeline) {
+ case BRW_RENDER_PIPELINE:
+ return brw->render_atoms;
+ case BRW_COMPUTE_PIPELINE:
+ return brw->compute_atoms;
+ default:
+ STATIC_ASSERT(BRW_NUM_PIPELINES == 2);
+ unreachable("Unsupported pipeline");
+ return NULL;
+ }
+}
+
+static void
+brw_copy_pipeline_atoms(struct brw_context *brw,
+ enum brw_pipeline pipeline,
+ const struct brw_tracked_state **atoms,
+ int num_atoms)
+{
+ /* This is to work around brw_context::atoms being declared const. We want
+ * it to be const, but it needs to be initialized somehow!
+ */
+ struct brw_tracked_state *context_atoms =
+ (struct brw_tracked_state *) brw_get_pipeline_atoms(brw, pipeline);
+
+ for (int i = 0; i < num_atoms; i++) {
+ context_atoms[i] = *atoms[i];
+ assert(context_atoms[i].dirty.mesa | context_atoms[i].dirty.brw);
+ assert(context_atoms[i].emit);
+ }
+
+ brw->num_atoms[pipeline] = num_atoms;
+}
+
void brw_init_state( struct brw_context *brw )
{
struct gl_context *ctx = &brw->ctx;
- const struct brw_tracked_state **atoms;
- int num_atoms;
- STATIC_ASSERT(ARRAY_SIZE(gen4_atoms) <= ARRAY_SIZE(brw->atoms));
- STATIC_ASSERT(ARRAY_SIZE(gen6_atoms) <= ARRAY_SIZE(brw->atoms));
- STATIC_ASSERT(ARRAY_SIZE(gen7_atoms) <= ARRAY_SIZE(brw->atoms));
- STATIC_ASSERT(ARRAY_SIZE(gen8_atoms) <= ARRAY_SIZE(brw->atoms));
+ /* Force the first brw_select_pipeline to emit pipeline select */
+ brw->last_pipeline = BRW_NUM_PIPELINES;
+
+ STATIC_ASSERT(ARRAY_SIZE(gen4_atoms) <= ARRAY_SIZE(brw->render_atoms));
+ STATIC_ASSERT(ARRAY_SIZE(gen6_atoms) <= ARRAY_SIZE(brw->render_atoms));
+ STATIC_ASSERT(ARRAY_SIZE(gen7_render_atoms) <=
+ ARRAY_SIZE(brw->render_atoms));
+ STATIC_ASSERT(ARRAY_SIZE(gen8_render_atoms) <=
+ ARRAY_SIZE(brw->render_atoms));
+ STATIC_ASSERT(ARRAY_SIZE(gen7_compute_atoms) <=
+ ARRAY_SIZE(brw->compute_atoms));
+ STATIC_ASSERT(ARRAY_SIZE(gen8_compute_atoms) <=
+ ARRAY_SIZE(brw->compute_atoms));
brw_init_caches(brw);
if (brw->gen >= 8) {
- atoms = gen8_atoms;
- num_atoms = ARRAY_SIZE(gen8_atoms);
+ brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
+ gen8_render_atoms,
+ ARRAY_SIZE(gen8_render_atoms));
+ brw_copy_pipeline_atoms(brw, BRW_COMPUTE_PIPELINE,
+ gen8_compute_atoms,
+ ARRAY_SIZE(gen8_compute_atoms));
} else if (brw->gen == 7) {
- atoms = gen7_atoms;
- num_atoms = ARRAY_SIZE(gen7_atoms);
+ brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
+ gen7_render_atoms,
+ ARRAY_SIZE(gen7_render_atoms));
+ brw_copy_pipeline_atoms(brw, BRW_COMPUTE_PIPELINE,
+ gen7_compute_atoms,
+ ARRAY_SIZE(gen7_compute_atoms));
} else if (brw->gen == 6) {
- atoms = gen6_atoms;
- num_atoms = ARRAY_SIZE(gen6_atoms);
+ brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
+ gen6_atoms, ARRAY_SIZE(gen6_atoms));
} else {
- atoms = gen4_atoms;
- num_atoms = ARRAY_SIZE(gen4_atoms);
- }
-
- brw->num_atoms = num_atoms;
-
- /* This is to work around brw_context::atoms being declared const. We want
- * it to be const, but it needs to be initialized somehow!
- */
- struct brw_tracked_state *context_atoms =
- (struct brw_tracked_state *) &brw->atoms[0];
-
- for (int i = 0; i < num_atoms; i++)
- context_atoms[i] = *atoms[i];
-
- while (num_atoms--) {
- assert((*atoms)->dirty.mesa | (*atoms)->dirty.brw);
- assert((*atoms)->emit);
- atoms++;
+ brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
+ gen4_atoms, ARRAY_SIZE(gen4_atoms));
}
brw_upload_initial_gpu_state(brw);
- brw->state.dirty.mesa = ~0;
- brw->state.dirty.brw = ~0ull;
+ brw->NewGLState = ~0;
+ brw->ctx.NewDriverState = ~0ull;
/* ~0 is a nonsensical value which won't match anything we program, so
* the programming will take effect on the first time around.
*/
brw->pma_stall_bits = ~0;
- /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
+ /* Make sure that brw->ctx.NewDriverState has enough bits to hold all possible
* dirty flags.
*/
- STATIC_ASSERT(BRW_NUM_STATE_BITS <= 8 * sizeof(brw->state.dirty.brw));
+ STATIC_ASSERT(BRW_NUM_STATE_BITS <= 8 * sizeof(brw->ctx.NewDriverState));
ctx->DriverFlags.NewTransformFeedback = BRW_NEW_TRANSFORM_FEEDBACK;
ctx->DriverFlags.NewTransformFeedbackProg = BRW_NEW_TRANSFORM_FEEDBACK;
DEFINE_BIT(BRW_NEW_FF_GS_PROG_DATA),
DEFINE_BIT(BRW_NEW_GS_PROG_DATA),
DEFINE_BIT(BRW_NEW_CLIP_PROG_DATA),
+ DEFINE_BIT(BRW_NEW_CS_PROG_DATA),
DEFINE_BIT(BRW_NEW_URB_FENCE),
DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM),
DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM),
DEFINE_BIT(BRW_NEW_CLIP_VP),
DEFINE_BIT(BRW_NEW_SAMPLER_STATE_TABLE),
DEFINE_BIT(BRW_NEW_VS_ATTRIB_WORKAROUNDS),
+ DEFINE_BIT(BRW_NEW_COMPUTE_PROGRAM),
{0, 0, 0}
};
}
}
-static void
-brw_upload_programs(struct brw_context *brw)
+static inline void
+brw_upload_programs(struct brw_context *brw,
+ enum brw_pipeline pipeline)
{
- brw_upload_vs_prog(brw);
+ if (pipeline == BRW_RENDER_PIPELINE) {
+ brw_upload_vs_prog(brw);
- if (brw->gen < 6)
- brw_upload_ff_gs_prog(brw);
- else
- brw_upload_gs_prog(brw);
+ if (brw->gen < 6)
+ brw_upload_ff_gs_prog(brw);
+ else
+ brw_upload_gs_prog(brw);
- brw_upload_wm_prog(brw);
+ brw_upload_wm_prog(brw);
+ } else if (pipeline == BRW_COMPUTE_PIPELINE) {
+ brw_upload_cs_prog(brw);
+ }
}
-/***********************************************************************
- * Emit all state:
- */
-void brw_upload_state(struct brw_context *brw)
+static inline void
+merge_ctx_state(struct brw_context *brw,
+ struct brw_state_flags *state)
+{
+ state->mesa |= brw->NewGLState;
+ state->brw |= brw->ctx.NewDriverState;
+}
+
+static inline void
+check_and_emit_atom(struct brw_context *brw,
+ struct brw_state_flags *state,
+ const struct brw_tracked_state *atom)
+{
+ if (check_state(state, &atom->dirty)) {
+ atom->emit(brw);
+ merge_ctx_state(brw, state);
+ }
+}
+
+static inline void
+brw_upload_pipeline_state(struct brw_context *brw,
+ enum brw_pipeline pipeline)
{
struct gl_context *ctx = &brw->ctx;
- struct brw_state_flags *state = &brw->state.dirty;
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);
- state->mesa |= brw->NewGLState;
- brw->NewGLState = 0;
-
- state->brw |= ctx->NewDriverState;
- ctx->NewDriverState = 0;
+ brw_select_pipeline(brw, pipeline);
if (0) {
/* Always re-emit all state. */
- state->mesa |= ~0;
- state->brw |= ~0ull;
+ brw->NewGLState = ~0;
+ ctx->NewDriverState = ~0ull;
}
- if (brw->fragment_program != ctx->FragmentProgram._Current) {
- brw->fragment_program = ctx->FragmentProgram._Current;
- brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
- }
+ 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->geometry_program != ctx->GeometryProgram._Current) {
- brw->geometry_program = ctx->GeometryProgram._Current;
- brw->state.dirty.brw |= BRW_NEW_GEOMETRY_PROGRAM;
+ 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->vertex_program != ctx->VertexProgram._Current) {
- brw->vertex_program = ctx->VertexProgram._Current;
- brw->state.dirty.brw |= 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->state.dirty.brw |= BRW_NEW_META_IN_PROGRESS;
+ brw->ctx.NewDriverState |= 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 (brw->num_samples != fb_samples) {
+ brw->num_samples = fb_samples;
+ brw->ctx.NewDriverState |= BRW_NEW_NUM_SAMPLES;
}
- if ((state->mesa | state->brw) == 0)
+ /* 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)
intel_emit_post_sync_nonzero_flush(brw);
- brw_upload_programs(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
*/
struct brw_state_flags examined, prev;
memset(&examined, 0, sizeof(examined));
- prev = *state;
+ prev = state;
- for (i = 0; i < brw->num_atoms; i++) {
- const struct brw_tracked_state *atom = &brw->atoms[i];
+ for (i = 0; i < num_atoms; i++) {
+ const struct brw_tracked_state *atom = &atoms[i];
struct brw_state_flags generated;
- if (check_state(state, &atom->dirty)) {
- atom->emit(brw);
- }
+ check_and_emit_atom(brw, &state, atom);
accumulate_state(&examined, &atom->dirty);
* if (examined & generated)
* fail;
*/
- xor_states(&generated, &prev, state);
+ xor_states(&generated, &prev, &state);
assert(!check_state(&examined, &generated));
- prev = *state;
+ prev = state;
}
}
else {
- for (i = 0; i < brw->num_atoms; i++) {
- const struct brw_tracked_state *atom = &brw->atoms[i];
+ for (i = 0; i < num_atoms; i++) {
+ const struct brw_tracked_state *atom = &atoms[i];
- if (check_state(state, &atom->dirty)) {
- atom->emit(brw);
- }
+ 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);
+ 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);
}
}
+/***********************************************************************
+ * Emit all state:
+ */
+void brw_upload_render_state(struct brw_context *brw)
+{
+ brw_upload_pipeline_state(brw, BRW_RENDER_PIPELINE);
+}
+
+static inline void
+brw_pipeline_state_finished(struct brw_context *brw,
+ enum brw_pipeline pipeline)
+{
+ /* Save all dirty state into the other pipelines */
+ for (int i = 0; i < BRW_NUM_PIPELINES; i++) {
+ if (i != pipeline) {
+ brw->state.pipelines[i].mesa |= brw->NewGLState;
+ brw->state.pipelines[i].brw |= brw->ctx.NewDriverState;
+ } else {
+ memset(&brw->state.pipelines[i], 0, sizeof(struct brw_state_flags));
+ }
+ }
+
+ brw->NewGLState = 0;
+ brw->ctx.NewDriverState = 0ull;
+}
/**
* Clear dirty bits to account for the fact that the state emitted by
- * brw_upload_state() has been committed to the hardware. This is a separate
- * call from brw_upload_state() because it's possible that after the call to
- * brw_upload_state(), we will discover that we've run out of aperture space,
- * and need to rewind the batch buffer to the state it had before the
- * brw_upload_state() call.
+ * brw_upload_render_state() has been committed to the hardware. This is a
+ * separate call from brw_upload_render_state() because it's possible that
+ * after the call to brw_upload_render_state(), we will discover that we've
+ * run out of aperture space, and need to rewind the batch buffer to the state
+ * it had before the brw_upload_render_state() call.
*/
void
-brw_clear_dirty_bits(struct brw_context *brw)
+brw_render_state_finished(struct brw_context *brw)
+{
+ brw_pipeline_state_finished(brw, BRW_RENDER_PIPELINE);
+}
+
+void
+brw_upload_compute_state(struct brw_context *brw)
+{
+ brw_upload_pipeline_state(brw, BRW_COMPUTE_PIPELINE);
+}
+
+void
+brw_compute_state_finished(struct brw_context *brw)
{
- struct brw_state_flags *state = &brw->state.dirty;
- memset(state, 0, sizeof(*state));
+ brw_pipeline_state_finished(brw, BRW_COMPUTE_PIPELINE);
}