r600g: order atom emission v3

To avoid GPU lockup registers must be emited in a specific order
(no kidding ...). This patch rework atom emission so order in which
atom are emited in respect to each other is always the same. We
don't have any informations on what is the correct order so order
will need to be infered from fglrx command stream.

v2: add comment warning that atom order should not be taken lightly
v3: rebase on top of alphatest atom fix

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

7 files changed, 116 insertions, 82 deletions

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index acf91ba43f3..35333120654 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -583,7 +583,7 @@ void evergreen_init_atom_start_compute_cs(struct r600_context *ctx)
 	/* since all required registers are initialised in the
 	 * start_compute_cs_cmd atom, we can EMIT_EARLY here.
 	 */
-	r600_init_command_buffer(cb, 256, EMIT_EARLY);
+	r600_init_command_buffer(ctx, cb, 1, 256);
 	cb->pkt_flags = RADEON_CP_PACKET3_COMPUTE_MODE;
 
 	switch (ctx->family) {
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index bda8ed5dc2d..9a5183ec6be 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -2161,27 +2161,50 @@ static void cayman_emit_sample_mask(struct r600_context *rctx, struct r600_atom
 
 void evergreen_init_state_functions(struct r600_context *rctx)
 {
-	r600_init_atom(&rctx->cb_misc_state.atom, evergreen_emit_cb_misc_state, 0, 0);
-	r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
-	r600_init_atom(&rctx->db_misc_state.atom, evergreen_emit_db_misc_state, 7, 0);
-	r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
-	r600_init_atom(&rctx->vertex_buffer_state.atom, evergreen_fs_emit_vertex_buffers, 0, 0);
-	r600_init_atom(&rctx->cs_vertex_buffer_state.atom, evergreen_cs_emit_vertex_buffers, 0, 0);
-	r600_init_atom(&rctx->vs_constbuf_state.atom, evergreen_emit_vs_constant_buffers, 0, 0);
-	r600_init_atom(&rctx->ps_constbuf_state.atom, evergreen_emit_ps_constant_buffers, 0, 0);
-	r600_init_atom(&rctx->vs_samplers.views.atom, evergreen_emit_vs_sampler_views, 0, 0);
-	r600_init_atom(&rctx->ps_samplers.views.atom, evergreen_emit_ps_sampler_views, 0, 0);
-	r600_init_atom(&rctx->cs_shader_state.atom, evergreen_emit_cs_shader, 0, 0);
-	r600_init_atom(&rctx->vs_samplers.atom_sampler, evergreen_emit_vs_sampler, 0, 0);
-	r600_init_atom(&rctx->ps_samplers.atom_sampler, evergreen_emit_ps_sampler, 0, 0);
-
-	if (rctx->chip_class == EVERGREEN)
-		r600_init_atom(&rctx->sample_mask.atom, evergreen_emit_sample_mask, 3, 0);
-	else
-		r600_init_atom(&rctx->sample_mask.atom, cayman_emit_sample_mask, 4, 0);
+	unsigned id = 4;
+
+	/* !!!
+	 *  To avoid GPU lockup registers must be emited in a specific order
+	 * (no kidding ...). The order below is important and have been
+	 * partialy infered from analyzing fglrx command stream.
+	 *
+	 * Don't reorder atom without carefully checking the effect (GPU lockup
+	 * or piglit regression).
+	 * !!!
+	 */
+
+	/* shader const */
+	r600_init_atom(rctx, &rctx->vs_constbuf_state.atom, id++, evergreen_emit_vs_constant_buffers, 0);
+	r600_init_atom(rctx, &rctx->ps_constbuf_state.atom, id++, evergreen_emit_ps_constant_buffers, 0);
+	/* shader program */
+	r600_init_atom(rctx, &rctx->cs_shader_state.atom, id++, evergreen_emit_cs_shader, 0);
+	/* sampler */
+	r600_init_atom(rctx, &rctx->vs_samplers.atom_sampler, id++, evergreen_emit_vs_sampler, 0);
+	r600_init_atom(rctx, &rctx->ps_samplers.atom_sampler, id++, evergreen_emit_ps_sampler, 0);
+	/* resources */
+	r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, evergreen_fs_emit_vertex_buffers, 0);
+	r600_init_atom(rctx, &rctx->cs_vertex_buffer_state.atom, id++, evergreen_cs_emit_vertex_buffers, 0);
+	r600_init_atom(rctx, &rctx->vs_samplers.views.atom, id++, evergreen_emit_vs_sampler_views, 0);
+	r600_init_atom(rctx, &rctx->ps_samplers.views.atom, id++, evergreen_emit_ps_sampler_views, 0);
+
+	if (rctx->chip_class == EVERGREEN) {
+		r600_init_atom(rctx, &rctx->sample_mask.atom, id++, evergreen_emit_sample_mask, 3);
+	} else {
+		r600_init_atom(rctx, &rctx->sample_mask.atom, id++, cayman_emit_sample_mask, 4);
+	}
 	rctx->sample_mask.sample_mask = ~0;
 	r600_atom_dirty(rctx, &rctx->sample_mask.atom);
 
+	r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, evergreen_emit_cb_misc_state, 0);
+	r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
+
+	r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
+	r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
+
+	r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, evergreen_emit_db_misc_state, 7);
+	r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
+
+
 	rctx->context.create_blend_state = evergreen_create_blend_state;
 	rctx->context.create_depth_stencil_alpha_state = evergreen_create_dsa_state;
 	rctx->context.create_fs_state = r600_create_shader_state_ps;
@@ -2230,7 +2253,7 @@ static void cayman_init_atom_start_cs(struct r600_context *rctx)
 {
 	struct r600_command_buffer *cb = &rctx->start_cs_cmd;
 
-	r600_init_command_buffer(cb, 256, EMIT_EARLY);
+	r600_init_command_buffer(rctx, cb, 0, 256);
 
 	/* This must be first. */
 	r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
@@ -2608,7 +2631,7 @@ void evergreen_init_atom_start_cs(struct r600_context *rctx)
 		return;
 	}
 
-	r600_init_command_buffer(cb, 256, EMIT_EARLY);
+	r600_init_command_buffer(rctx, cb, 0, 256);
 
 	/* This must be first. */
 	r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index 7b7b6b1ffac..0ec13e5bade 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -633,15 +633,17 @@ out_err:
 void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
 			boolean count_draw_in)
 {
-	struct r600_atom *state;
-
 	/* The number of dwords we already used in the CS so far. */
 	num_dw += ctx->cs->cdw;
 
 	if (count_draw_in) {
+		unsigned i;
+
 		/* The number of dwords all the dirty states would take. */
-		LIST_FOR_EACH_ENTRY(state, &ctx->dirty_states, head) {
-			num_dw += state->num_dw;
+		for (i = 0; i < R600_MAX_ATOM; i++) {
+			if (ctx->atoms[i] && ctx->atoms[i]->dirty) {
+				num_dw += ctx->atoms[i]->num_dw;
+			}
 		}
 
 		num_dw += ctx->pm4_dirty_cdwords;
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 286c676f85b..9e6c28d523b 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -224,7 +224,6 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	rctx->family = rscreen->family;
 	rctx->chip_class = rscreen->chip_class;
 
-	LIST_INITHEAD(&rctx->dirty_states);
 	LIST_INITHEAD(&rctx->active_timer_queries);
 	LIST_INITHEAD(&rctx->active_nontimer_queries);
 	LIST_INITHEAD(&rctx->dirty);
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 721334d1e84..ff720e95cfe 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -35,6 +35,8 @@
 #include "r600_resource.h"
 #include "evergreen_compute.h"
 
+#define R600_MAX_ATOM						17
+
 #define R600_MAX_CONST_BUFFERS 2
 #define R600_MAX_CONST_BUFFER_SIZE 4096
 
@@ -44,23 +46,14 @@
 #define R600_BIG_ENDIAN 0
 #endif
 
-enum r600_atom_flags {
-	/* When set, atoms are added at the beginning of the dirty list
-	 * instead of the end. */
-	EMIT_EARLY = (1 << 0)
-};
-
 /* This encapsulates a state or an operation which can emitted into the GPU
  * command stream. It's not limited to states only, it can be used for anything
  * that wants to write commands into the CS (e.g. cache flushes). */
 struct r600_atom {
 	void (*emit)(struct r600_context *ctx, struct r600_atom *state);
-
+	unsigned		id;
 	unsigned		num_dw;
-	enum r600_atom_flags	flags;
 	bool			dirty;
-
-	struct list_head	head;
 };
 
 /* This is an atom containing GPU commands that never change.
@@ -372,8 +365,8 @@ struct r600_context {
 	unsigned default_ps_gprs, default_vs_gprs;
 
 	/* States based on r600_atom. */
-	struct list_head		dirty_states;
 	struct r600_command_buffer	start_cs_cmd; /* invariant state mostly */
+	struct r600_atom		*atoms[R600_MAX_ATOM];
 	/** Compute specific registers initializations.  The start_cs_cmd atom
 	 *  must be emitted before start_compute_cs_cmd. */
         struct r600_command_buffer      start_compute_cs_cmd;
@@ -464,20 +457,11 @@ static INLINE void r600_emit_atom(struct r600_context *rctx, struct r600_atom *a
 {
 	atom->emit(rctx, atom);
 	atom->dirty = false;
-	if (atom->head.next && atom->head.prev)
-		LIST_DELINIT(&atom->head);
 }
 
 static INLINE void r600_atom_dirty(struct r600_context *rctx, struct r600_atom *state)
 {
-	if (!state->dirty) {
-		if (state->flags & EMIT_EARLY) {
-			LIST_ADD(&state->head, &rctx->dirty_states);
-		} else {
-			LIST_ADDTAIL(&state->head, &rctx->dirty_states);
-		}
-		state->dirty = true;
-	}
+	state->dirty = true;
 }
 
 /* evergreen_state.c */
@@ -587,9 +571,10 @@ void r600_translate_index_buffer(struct r600_context *r600,
 				 unsigned count);
 
 /* r600_state_common.c */
-void r600_init_atom(struct r600_atom *atom,
+void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom);
+void r600_init_atom(struct r600_context *rctx, struct r600_atom *atom, unsigned id,
 		    void (*emit)(struct r600_context *ctx, struct r600_atom *state),
-		    unsigned num_dw, enum r600_atom_flags flags);
+		    unsigned num_dw);
 void r600_init_common_atoms(struct r600_context *rctx);
 unsigned r600_get_cb_flush_flags(struct r600_context *rctx);
 void r600_texture_barrier(struct pipe_context *ctx);
@@ -772,7 +757,7 @@ static INLINE void eg_store_loop_const(struct r600_command_buffer *cb, unsigned
 	r600_store_value(cb, value);
 }
 
-void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags);
+void r600_init_command_buffer(struct r600_context *rctx, struct r600_command_buffer *cb, unsigned id, unsigned num_dw);
 void r600_release_command_buffer(struct r600_command_buffer *cb);
 
 /*
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 53bbdd9da7d..ccafdc6733b 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -2031,27 +2031,48 @@ static void r600_emit_sample_mask(struct r600_context *rctx, struct r600_atom *a
 
 void r600_init_state_functions(struct r600_context *rctx)
 {
-	r600_init_atom(&rctx->seamless_cube_map.atom, r600_emit_seamless_cube_map, 3, 0);
-	r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
-	r600_init_atom(&rctx->cb_misc_state.atom, r600_emit_cb_misc_state, 0, 0);
-	r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
-	r600_init_atom(&rctx->db_misc_state.atom, r600_emit_db_misc_state, 4, 0);
-	r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
-	r600_init_atom(&rctx->vertex_buffer_state.atom, r600_emit_vertex_buffers, 0, 0);
-	r600_init_atom(&rctx->vs_constbuf_state.atom, r600_emit_vs_constant_buffers, 0, 0);
-	r600_init_atom(&rctx->ps_constbuf_state.atom, r600_emit_ps_constant_buffers, 0, 0);
-	r600_init_atom(&rctx->vs_samplers.views.atom, r600_emit_vs_sampler_views, 0, 0);
-	r600_init_atom(&rctx->ps_samplers.views.atom, r600_emit_ps_sampler_views, 0, 0);
+	unsigned id = 4;
+
+	/* !!!
+	 *  To avoid GPU lockup registers must be emited in a specific order
+	 * (no kidding ...). The order below is important and have been
+	 * partialy infered from analyzing fglrx command stream.
+	 *
+	 * Don't reorder atom without carefully checking the effect (GPU lockup
+	 * or piglit regression).
+	 * !!!
+	 */
+
+	/* shader const */
+	r600_init_atom(rctx, &rctx->vs_constbuf_state.atom, id++, r600_emit_vs_constant_buffers, 0);
+	r600_init_atom(rctx, &rctx->ps_constbuf_state.atom, id++, r600_emit_ps_constant_buffers, 0);
+
 	/* sampler must be emited before TA_CNTL_AUX otherwise DISABLE_CUBE_WRAP change
-	 * does not take effect
+	 * does not take effect (TA_CNTL_AUX emited by r600_emit_seamless_cube_map)
 	 */
-	r600_init_atom(&rctx->vs_samplers.atom_sampler, r600_emit_vs_sampler, 0, EMIT_EARLY);
-	r600_init_atom(&rctx->ps_samplers.atom_sampler, r600_emit_ps_sampler, 0, EMIT_EARLY);
+	r600_init_atom(rctx, &rctx->vs_samplers.atom_sampler, id++, r600_emit_vs_sampler, 0);
+	r600_init_atom(rctx, &rctx->ps_samplers.atom_sampler, id++, r600_emit_ps_sampler, 0);
+	/* resource */
+	r600_init_atom(rctx, &rctx->vs_samplers.views.atom, id++, r600_emit_vs_sampler_views, 0);
+	r600_init_atom(rctx, &rctx->ps_samplers.views.atom, id++, r600_emit_ps_sampler_views, 0);
+	r600_init_atom(rctx, &rctx->vertex_buffer_state.atom, id++, r600_emit_vertex_buffers, 0);
+
+	r600_init_atom(rctx, &rctx->seamless_cube_map.atom, id++, r600_emit_seamless_cube_map, 3);
+	r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
 
-	r600_init_atom(&rctx->sample_mask.atom, r600_emit_sample_mask, 3, 0);
+	r600_init_atom(rctx, &rctx->sample_mask.atom, id++, r600_emit_sample_mask, 3);
 	rctx->sample_mask.sample_mask = ~0;
 	r600_atom_dirty(rctx, &rctx->sample_mask.atom);
 
+	r600_init_atom(rctx, &rctx->cb_misc_state.atom, id++, r600_emit_cb_misc_state, 0);
+	r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
+
+	r600_init_atom(rctx, &rctx->alphatest_state.atom, id++, r600_emit_alphatest_state, 6);
+	r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
+
+	r600_init_atom(rctx, &rctx->db_misc_state.atom, id++, r600_emit_db_misc_state, 4);
+	r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
+
 	rctx->context.create_blend_state = r600_create_blend_state;
 	rctx->context.create_depth_stencil_alpha_state = r600_create_dsa_state;
 	rctx->context.create_fs_state = r600_create_shader_state_ps;
@@ -2157,7 +2178,7 @@ void r600_init_atom_start_cs(struct r600_context *rctx)
 	struct r600_command_buffer *cb = &rctx->start_cs_cmd;
 	uint32_t tmp;
 
-	r600_init_command_buffer(cb, 256, EMIT_EARLY);
+	r600_init_command_buffer(rctx, cb, 0, 256);
 
 	/* R6xx requires this packet at the start of each command buffer */
 	if (rctx->chip_class == R600) {
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 26af6f609a5..e67eba8921c 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -44,11 +44,9 @@ static void r600_emit_command_buffer(struct r600_context *rctx, struct r600_atom
 	cs->cdw += cb->atom.num_dw;
 }
 
-void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags)
+void r600_init_command_buffer(struct r600_context *rctx, struct r600_command_buffer *cb, unsigned id, unsigned num_dw)
 {
-	cb->atom.emit = r600_emit_command_buffer;
-	cb->atom.num_dw = 0;
-	cb->atom.flags = flags;
+	r600_init_atom(rctx, &cb->atom, id, r600_emit_command_buffer, 0);
 	cb->buf = CALLOC(1, 4 * num_dw);
 	cb->max_num_dw = num_dw;
 }
@@ -79,16 +77,22 @@ static void r600_emit_r6xx_flush_and_inv(struct r600_context *rctx, struct r600_
 	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
 }
 
-void r600_init_atom(struct r600_atom *atom,
+void r600_init_atom(struct r600_context *rctx,
+		    struct r600_atom *atom,
+		    unsigned id,
 		    void (*emit)(struct r600_context *ctx, struct r600_atom *state),
-		    unsigned num_dw, enum r600_atom_flags flags)
+		    unsigned num_dw)
 {
+	assert(id < R600_MAX_ATOM);
+	assert(rctx->atoms[id] == NULL);
+	rctx->atoms[id] = atom;
+	atom->id = id;
 	atom->emit = emit;
 	atom->num_dw = num_dw;
-	atom->flags = flags;
+	atom->dirty = false;
 }
 
-static void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom)
+void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom)
 {
 	struct radeon_winsys_cs *cs = rctx->cs;
 	struct r600_alphatest_state *a = (struct r600_alphatest_state*)atom;
@@ -106,10 +110,8 @@ static void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_ato
 
 void r600_init_common_atoms(struct r600_context *rctx)
 {
-	r600_init_atom(&rctx->surface_sync_cmd.atom,	r600_emit_surface_sync,		5, EMIT_EARLY);
-	r600_init_atom(&rctx->r6xx_flush_and_inv_cmd,	r600_emit_r6xx_flush_and_inv,	2, EMIT_EARLY);
-	r600_init_atom(&rctx->alphatest_state.atom,	r600_emit_alphatest_state,	6, 0);
-	r600_atom_dirty(rctx, &rctx->alphatest_state.atom);
+	r600_init_atom(rctx, &rctx->r6xx_flush_and_inv_cmd, 2, r600_emit_r6xx_flush_and_inv, 2);
+	r600_init_atom(rctx, &rctx->surface_sync_cmd.atom, 3, r600_emit_surface_sync, 5);
 }
 
 unsigned r600_get_cb_flush_flags(struct r600_context *rctx)
@@ -1127,9 +1129,8 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct pipe_draw_info info = *dinfo;
 	struct pipe_index_buffer ib = {};
-	unsigned prim, ls_mask = 0;
+	unsigned prim, ls_mask = 0, i;
 	struct r600_block *dirty_block = NULL, *next_block = NULL;
-	struct r600_atom *state = NULL, *next_state = NULL;
 	struct radeon_winsys_cs *cs = rctx->cs;
 	uint64_t va;
 	uint8_t *ptr;
@@ -1221,8 +1222,11 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 	/* Emit states (the function expects that we emit at most 17 dwords here). */
 	r600_need_cs_space(rctx, 0, TRUE);
 
-	LIST_FOR_EACH_ENTRY_SAFE(state, next_state, &rctx->dirty_states, head) {
-		r600_emit_atom(rctx, state);
+	for (i = 0; i < R600_MAX_ATOM; i++) {
+		if (rctx->atoms[i] == NULL || !rctx->atoms[i]->dirty) {
+			continue;
+		}
+		r600_emit_atom(rctx, rctx->atoms[i]);
 	}
 	LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &rctx->dirty,list) {
 		r600_context_block_emit_dirty(rctx, dirty_block, 0 /* pkt_flags */);