cleanup the batch logic

2 files changed, 64 insertions, 158 deletions

diff --git a/r600_atom.c b/r600_atom.c
index ea04252..dc5cd8a 100644
--- a/r600_atom.c
+++ b/r600_atom.c
@@ -18,7 +18,7 @@
 #include "r600d.h"
 
 /*
- * drm_r600_framebuffer
+ * framebuffer
  */
 int r600_framebuffer_emit(struct r600_winsys *rdev,
 				struct r600_atom *atom,
@@ -203,7 +203,7 @@ int r600_cb_cntl_create(struct r600_winsys *rdev, struct r600_atom *atom, void *
 }
 
 /*
- * r600_pa
+ * rasterizer
  */
 int r600_rasterizer_create(struct r600_winsys *rdev, struct r600_atom *atom, void *data)
 {
@@ -281,7 +281,7 @@ int r600_rasterizer_create(struct r600_winsys *rdev, struct r600_atom *atom, voi
 }
 
 /*
- * pipe_viewport_state
+ * viewport state
  */
 int r600_viewport_create(struct r600_winsys *rdev, struct r600_atom *atom, void *data)
 {
@@ -310,7 +310,7 @@ int r600_viewport_create(struct r600_winsys *rdev, struct r600_atom *atom, void
 }
 
 /*
- * pipe_scissor_state
+ * scissor state
  */
 int r600_scissor_create(struct r600_winsys *rdev, struct r600_atom *atom, void *data)
 {
@@ -331,7 +331,7 @@ int r600_scissor_create(struct r600_winsys *rdev, struct r600_atom *atom, void *
 }
 
 /*
- * r600_blend
+ * blend state
  */
 int r600_blend_create(struct r600_winsys *rdev, struct r600_atom *atom, void *data)
 {
@@ -439,7 +439,7 @@ int r600_dsa_create(struct r600_winsys *rdev, struct r600_atom *atom, void *data
 }
 
 /*
- * r600_vs_shader
+ * vexter shader
  */
 int r600_vs_shader_emit(struct r600_winsys *rdev,
 			struct r600_atom *atom,
@@ -576,7 +576,7 @@ int r600_fs_shader_create(struct r600_winsys *rdev, struct r600_atom *atom, void
 }
 
 /*
- * vertex shader input
+ * shader resource
  */
 int r600_shader_resource_emit(struct r600_winsys *rdev,
 			struct r600_atom *atom,
diff --git a/r600_batch.c b/r600_batch.c
index 3a904ce..4b933af 100644
--- a/r600_batch.c
+++ b/r600_batch.c
@@ -52,121 +52,91 @@ static void r600_emit_flush(struct r600_winsys *rdev,
 						RADEON_GEM_DOMAIN_GTT);
 }
 
-static void r600_emit_resources(struct r600_winsys *rdev,
-				struct radeon_ib *ib,
-				struct radeon_bo *bo,
-				u32 dw0, u32 dw1, u32 dw2, u32 dw3,
-				u32 dw4, u32 dw5, u32 dw6, u32 dw7)
-{
-	ib->ptr[ib->cpkts++] = PKT3(PKT3_SURFACE_SYNC, 3);
-	ib->ptr[ib->cpkts++] = 0x01000000;
-	ib->ptr[ib->cpkts++] = radeon_bo_size(bo) >> 8;
-	ib->ptr[ib->cpkts++] = 0x00000000;
-	ib->ptr[ib->cpkts++] = 0x0000000A;
-	ib->ptr[ib->cpkts++] = PKT3(PKT3_NOP, 0);
-	ib->ptr[ib->cpkts++] = radeon_ib_reloc(ib, bo, RADEON_GEM_DOMAIN_VRAM |
-						RADEON_GEM_DOMAIN_GTT);
-	ib->ptr[ib->cpkts++] = PKT3(PKT3_SET_RESOURCE, 7);
-	ib->ptr[ib->cpkts++] = dw0;
-	ib->ptr[ib->cpkts++] = dw1;
-	ib->ptr[ib->cpkts++] = dw2;
-	ib->ptr[ib->cpkts++] = dw3;
-	ib->ptr[ib->cpkts++] = dw4;
-	ib->ptr[ib->cpkts++] = dw5;
-	ib->ptr[ib->cpkts++] = dw6;
-	ib->ptr[ib->cpkts++] = dw7;
-	ib->ptr[ib->cpkts++] = PKT3(PKT3_NOP, 0);
-	ib->ptr[ib->cpkts++] = radeon_ib_reloc(ib, bo, RADEON_GEM_DOMAIN_VRAM |
-						RADEON_GEM_DOMAIN_GTT);
-}
-
-/*
- * r600_batch
- */
-static int r600_batch_alloc(struct r600_batch **batch)
+void r600_atom_flush_add(struct r600_batch *batch, struct radeon_bo *bo, u32 flags)
 {
-	struct r600_batch *rbatch;
+	int i;
 
-	*batch = NULL;
-	rbatch = malloc(sizeof(*rbatch));
-	if (rbatch == NULL)
-		return -ENOMEM;
-	INIT_LIST_HEAD(&rbatch->list);
-	INIT_LIST_HEAD(&rbatch->pre_flushes);
-	INIT_LIST_HEAD(&rbatch->post_flushes);
-	rbatch->nemit_atoms = 0;
-	*batch = rbatch;
-	return 0;
+	for (i = 0; i < batch->nflush; i++) {
+		if (batch->flush[i].bo == bo) {
+			batch->flush[i].flags |= flags;
+			return;
+		}
+	}
+	batch->flush[batch->nflush].bo = bo;
+	batch->flush[batch->nflush].flags = flags;
+	batch->nflush++;
 }
 
 /*
  * r600_batches
  */
-static void r600_batches_clear_locked(struct r600_winsys *rdev, struct r600_batches *batches)
+static void r600_batches_clear_locked(struct r600_winsys *rw, struct r600_batches *batches)
 {
 	struct r600_batch *batch, *n;
-	int i;
+	int i, j;
 
-	list_for_each_entry_safe(batch, n, &batches->batches, list) {
-		for (i = 0; i < R600_BATCH_NATOMS; i++) {
-			if (batch->atoms[i])
-				r600_atom_destroy(batch->atoms[i]);
+	for (i = 0; i < batches->nbatch; i++) {
+		for (j = 0; j < R600_BATCH_NATOMS; j++) {
+			r600_atom_destroy(batches->batch[i].atoms[j]);
 		}
-		r600_atom_flush_cleanup(&batch->pre_flushes);
-		r600_atom_flush_cleanup(&batch->post_flushes);
-		list_del(&batch->list);
-		free(batch);
 	}
-	INIT_LIST_HEAD(&batches->batches);
-	r700_batches_states_default(rdev, batches);
+	batches->ib->cpkts = 0;
+	batches->nbatch = 0;
+	r700_batches_states_default(rw, batches);
 	batches->npkts = batches->ib->cpkts;
 }
 
-static int r600_batches_flush_locked(struct r600_winsys *rdev, struct r600_batches *batches)
+int r600_batches_flush(struct r600_winsys *rw)
 {
-	struct r600_batch *batch;
-	struct r600_atom_flush *flush;
-	int r, i;
+	struct r600_batches *batches =  &rw->batches;
+	int r, i, j;
 
-	list_for_each_entry(batch, &batches->batches, list) {
-		list_for_each_entry(flush, &batch->pre_flushes, list) {
-			r600_emit_flush(rdev, batches->ib, flush->bo, flush->flags);
+	for (i = 0; i < batches->nbatch; i++) {
+		for (j = 0; j < batches->batch[i].nflush; j++) {
+			r600_emit_flush(rw, batches->ib,
+					batches->batch[i].flush[j].bo,
+					batches->batch[i].flush[j].flags);
 		}
-		for (i = 0; i < batch->nemit_atoms; i++) {
-			r = batch->emit_atoms[i]->emit(rdev, batch->emit_atoms[i], batch, batches->ib);
+		for (j = 0; j < batches->batch[i].nemit_atoms; j++) {
+			r = batches->batch[i].emit_atoms[j]->emit(rw,
+							batches->batch[i].emit_atoms[j],
+							&batches->batch[i],
+							batches->ib);
 			if (r)
 				goto out_err;
 		}
-		r = r600_draw_cmd_emit(batches->ib, &batch->drm);
+		r = r600_draw_cmd_emit(batches->ib, &batches->batch[i].drm);
 		/* flush + wait until */
 		batches->ib->ptr[batches->ib->cpkts++] = PKT3(PKT3_EVENT_WRITE, 0);
 		batches->ib->ptr[batches->ib->cpkts++] = 0x00000016;
 		batches->ib->ptr[batches->ib->cpkts++] = PKT3(PKT3_SET_CONFIG_REG, 1);
 		batches->ib->ptr[batches->ib->cpkts++] = 0x00000010;
 		batches->ib->ptr[batches->ib->cpkts++] = 0x00028000;
-		list_for_each_entry(flush, &batch->post_flushes, list) {
-			r600_emit_flush(rdev, batches->ib, flush->bo, flush->flags);
-		}
 	}
-	r = radeon_ib_schedule(rdev, batches->ib);
+	printf("ib %d dw\n", batches->ib->cpkts);
+	r = radeon_ib_schedule(rw, batches->ib);
 out_err:
 	/* FIXME helper function */
 	batches->ib->cpkts = 0;
 	batches->ib->nrelocs = 0;
-	r600_batches_clear_locked(rdev, batches);
+	r600_batches_clear_locked(rw, batches);
 	return r;
 }
 
-int r600_batches_queue(struct r600_winsys *rdev, struct r600_request *rq)
+int r600_batches_queue(struct r600_winsys *rw, struct r600_request *rq)
 {
 	struct drm_r600_batch *batch = rq->data;
 	struct r600_batch *rbatch;
-	struct r600_batches *batches =  &rdev->batches;
+	struct r600_batches *batches =  &rw->batches;
 	int r, i, j;
 
-	r = r600_batch_alloc(&rbatch);
-	if (r)
-		return r;
+	if (batches->nbatch >= R600_MAX_BATCH) {
+		r = r600_batches_flush(rw);
+		if (r)
+			return r;
+	}
+	rbatch = &batches->batch[batches->nbatch];
+	memset(rbatch, 0, sizeof(struct r600_batch));
 	i = 0;
 	if (batch->blend == NULL || batch->cb_cntl == NULL ||
 		batch->rasterizer == NULL || batch->viewport == NULL ||
@@ -194,9 +164,7 @@ int r600_batches_queue(struct r600_winsys *rdev, struct r600_request *rq)
 	}
 	memcpy(&rbatch->drm, batch, sizeof(struct drm_r600_batch));
 reprocess:
-	r600_atom_flush_cleanup(&rbatch->pre_flushes);
-	r600_atom_flush_cleanup(&rbatch->post_flushes);
-	rbatch->nflushes = 0;
+	rbatch->nflush = 0;
 	rbatch->npkts = 0;
 	/* flush + wait until = 5dw */
 	rbatch->npkts += 5;
@@ -204,18 +172,14 @@ reprocess:
 	for (i = 0; i < R600_BATCH_NATOMS; i++) {
 		if (rbatch->atoms[i]) {
 			for (j = 0; j < rbatch->atoms[i]->nbo; j++) {
-				r = r600_atom_flush_add(&rbatch->pre_flushes, rbatch->atoms[i]->bo[j],  rbatch->atoms[i]->flags[j]);
-				if (r < 0)
-					goto out_err;
+				r600_atom_flush_add(rbatch, rbatch->atoms[i]->bo[j],  rbatch->atoms[i]->flags[j]);
 			}
-			rbatch->nflushes += rbatch->atoms[i]->nbo;
 			rbatch->emit_atoms[rbatch->nemit_atoms++] = rbatch->atoms[i];
+			rbatch->npkts += rbatch->atoms[i]->npkts;
 		}
 	}
 	/* add flush */
-	rbatch->npkts += rbatch->nflushes * 7;
-	/* FIXME shader flush should be conditional only if we change shaders */
-	rbatch->npkts += 7;
+	rbatch->npkts += rbatch->nflush * 7;
 	/* if batch is bigger than ib size it's an invalid one, this should
 	 * not happen
 	 */
@@ -226,7 +190,7 @@ reprocess:
 	}
 	/* flush or not ? */
 	if (batches->npkts + rbatch->npkts > batches->ib->length_dw) {
-		r = r600_batches_flush_locked(rdev, batches);
+		r = r600_batches_flush(rw);
 		if (r)
 			goto out_err;
 		goto reprocess;
@@ -237,9 +201,9 @@ reprocess:
 			batches->last_id[i] = rbatch->atoms[i]->id;
 		}
 	}
-	r600_winsys_set_bo_list(rdev, rq->nbo, rq->bo);
+	printf("batch %d dw batches with %d dw\n", rbatch->npkts, batches->npkts);
 	batches->npkts += rbatch->npkts;
-	list_add_tail(&rbatch->list, &batches->batches);
+	batches->nbatch++;
 	return 0;
 out_err:
 	for (i = 0; i < R600_BATCH_NATOMS; i++) {
@@ -250,35 +214,25 @@ out_err:
 	return r;
 }
 
-static int r600_batches_init(struct r600_winsys *rdev, struct r600_batches *batches)
+static int r600_batches_init(struct r600_winsys *rw, struct r600_batches *batches)
 {
 	int r;
 
 	memset(batches, 0 , sizeof(struct r600_batches));
-	INIT_LIST_HEAD(&batches->batches);
-	r = radeon_ib_get(rdev, &batches->ib);
+	r = radeon_ib_get(rw, &batches->ib);
 	if (r)
 		return r;
-	r700_batches_states_default(rdev, batches);
-	batches->npkts = batches->ib->cpkts;
+	r600_batches_clear_locked(rw, batches);
 	return 0;
 }
 
-static void r600_batches_cleanup_locked(struct r600_winsys *rdev, struct r600_batches *batches)
+static void r600_batches_cleanup_locked(struct r600_winsys *rw, struct r600_batches *batches)
 {
-	r600_batches_clear_locked(rdev, batches);
+	r600_batches_clear_locked(rw, batches);
 	radeon_ib_free(batches->ib);
 	batches->ib = NULL;
 }
 
-int r600_batches_flush(struct r600_winsys *rdev)
-{
-	int r;
-
-	r = r600_batches_flush_locked(rdev, &rdev->batches);
-	return r;
-}
-
 int r600_atoms_init(struct r600_winsys *rdev)
 {
 	rdev->npipes = 2;
@@ -354,60 +308,12 @@ struct r600_atom *r600_atom_create(struct r600_winsys *rdev, struct r600_request
 	return atom;
 }
 
-void r600_atom_flush_cleanup(struct list_head *flushes)
-{
-	struct r600_atom_flush *i, *n;
-
-	list_for_each_entry_safe(i, n, flushes, list) {
-		list_del(&i->list);
-		free(i);
-	}
-}
-
-int r600_atom_flush_add(struct list_head *flushes, struct radeon_bo *bo, u32 flags)
-{
-	struct r600_atom_flush *i;
-
-	list_for_each_entry(i, flushes, list) {
-		if (i->bo->handle == bo->handle) {
-			i->flags |= flags;
-			return 0;
-		}
-	}
-	i = malloc(sizeof(*i));
-	if (i == NULL)
-		return -ENOMEM;
-	i->bo = bo;
-	i->flags = flags;
-	list_add_tail(&i->list, flushes);
-	return 1;
-}
-
 int r600_atom_emit_default(struct r600_winsys *rdev, struct r600_atom *atom,
 				void *data, struct radeon_ib *ib)
 {
 	return radeon_ib_copy(ib, atom->pkts, atom->npkts);
 }
 
-struct radeon_bo *radeon_bo_lookup(struct r600_winsys *rdev, u32 handle)
-{
-	int i;
-
-	for (i = 0; i < rdev->nbo; i++) {
-		if (rdev->bo[i] && rdev->bo[i]->handle == handle) {
-			radeon_bo_ref(rdev->bo[i]);
-			return rdev->bo[i];
-		}
-	}
-	return NULL;
-}
-
-void r600_winsys_set_bo_list(struct r600_winsys *rdev, u32 nbo, struct radeon_bo **bo)
-{
-	memcpy(rdev->bo, bo, sizeof(void*) * nbo);
-	rdev->nbo = nbo;
-}
-
 void r600_atom_ref(struct r600_atom *atom)
 {
 	if (atom == NULL)