add sx atom

4 files changed, 153 insertions, 21 deletions

diff --git a/r600_atom.c b/r600_atom.c
index 976a14e..fad84b5 100644
--- a/r600_atom.c
+++ b/r600_atom.c
@@ -99,16 +99,16 @@ static int r600_cb_emit(struct radeon_device *rdev,
 	cb->pkts[16] = radeon_ib_reloc(ib, cb->bo, cb->placements[0] | cb->placements[1]);
 	cb->pkts[21] = radeon_ib_reloc(ib, cb->bo, cb->placements[0] | cb->placements[1]);
 	cb->pkts[26] = radeon_ib_reloc(ib, cb->bo, cb->placements[0] | cb->placements[1]);
-	cb->pkts[93] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[98] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[111] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[116] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[132] = radeon_ib_reloc(ib, cb->psshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[137] = radeon_ib_reloc(ib, cb->psshader1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[162] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[173] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[180] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
-	cb->pkts[191] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[84] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[89] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[102] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[107] = radeon_ib_reloc(ib, cb->vsshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[123] = radeon_ib_reloc(ib, cb->psshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[128] = radeon_ib_reloc(ib, cb->psshader1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[153] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[164] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[171] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
+	cb->pkts[182] = radeon_ib_reloc(ib, cb->vbo1, RADEON_GEM_DOMAIN_GTT);
 	r = radeon_ib_copy(ib, cb->pkts, atom->npkts);
 	return r;
 }
@@ -1347,6 +1347,108 @@ out_err:
 }
 
 /*
+ * r600_sx
+ */
+struct r600_sx {
+	struct radeon_atom	atom;
+	u32			pkts[32];
+	struct r600_atoms	*atoms;
+	struct radeon_bo	*buffer;
+	u32			placements[2];
+};
+
+static void r600_sx_release(struct kref *kref)
+{
+	struct radeon_atom *atom = container_of(kref, struct radeon_atom, kref);
+	struct r600_sx *sx = container_of(atom, struct r600_sx, atom);
+
+	mutex_lock(&sx->atoms->mutex);
+	list_del_init(&sx->atom.list);
+	mutex_unlock(&sx->atoms->mutex);
+	if (sx->buffer)
+		radeon_bo_unref(sx->buffer);
+	kfree(sx);
+}
+
+static int r600_sx_emit(struct radeon_device *rdev,
+			struct radeon_atom *atom,
+			void *data,
+			struct radeon_ib *ib)
+{
+	struct r600_sx *sx = (struct r600_sx *)atom;
+	int r;
+
+	if (sx->buffer)
+		sx->pkts[6] = radeon_ib_reloc(ib, sx->buffer, sx->placements[0] | sx->placements[1]);
+	r = radeon_ib_copy(ib, sx->pkts, atom->npkts);
+	return r;
+}
+
+static int r600_sx_create(struct radeon_device *rdev,
+				struct r600_atoms *atoms,
+				struct drm_radeon_atom *patom,
+				struct radeon_atom **atom)
+{
+	struct drm_r600_sx psx;
+	struct r600_sx *sx;
+	u32 size;
+	int r;
+
+	sx = kmalloc(sizeof(*sx), GFP_KERNEL);
+	if (sx == NULL)
+		return -ENOMEM;
+	/* make sure structure properly initialized */
+	memset(sx, 0, sizeof(*sx));
+	r = radeon_atom_init(&sx->atom, &atoms->idr, &r600_sx_release,
+			&r600_sx_emit, &r600_atom_process_default,
+			sx->pkts);
+	if (r)
+		goto out_err;
+	/* KERNEL use get user data */
+	memcpy(&psx, (void*)(unsigned long)patom->data, sizeof(struct drm_r600_sx));
+	sx->buffer = psx.buffer;
+	sx->placements[0] = psx.placements[0];
+	sx->placements[1] = psx.placements[1];
+	if (sx->buffer) {
+		radeon_bo_ref(sx->buffer);
+		size = radeon_bo_size(sx->buffer);
+		if (psx.offset > size) {
+			dev_err(rdev->dev, "offset into sx past bo end\n");
+			r = -EINVAL;
+			goto out_err;
+		}
+		size -= psx.offset;
+		/* SX_EXPORT_BUFFER_SIZES */
+		sx->pkts[sx->atom.npkts++] = PKT3(PKT3_SET_CONFIG_REG, 3);
+		sx->pkts[sx->atom.npkts++] = 0x00000403;
+		sx->pkts[sx->atom.npkts++] = psx.sx_export_buffer_sizes;
+		sx->pkts[sx->atom.npkts++] = psx.offset;
+		sx->pkts[sx->atom.npkts++] = size;
+		sx->pkts[sx->atom.npkts++] = PKT3(PKT3_NOP, 0);
+		sx->pkts[sx->atom.npkts++] = 0x00000000;
+	}
+	/* SX_MISC */
+	sx->pkts[sx->atom.npkts++] = PKT3(PKT3_SET_CONTEXT_REG, 1);
+	sx->pkts[sx->atom.npkts++] = 0x000000D4;
+	sx->pkts[sx->atom.npkts++] = psx.sx_misc;
+	/* SX_ALPHA_TEST_CONTROL */
+	sx->pkts[sx->atom.npkts++] = PKT3(PKT3_SET_CONTEXT_REG, 1);
+	sx->pkts[sx->atom.npkts++] = 0x00000104;
+	sx->pkts[sx->atom.npkts++] = psx.sx_alpha_test_control;
+	/* SX_ALPHA_REF */
+	sx->pkts[sx->atom.npkts++] = PKT3(PKT3_SET_CONTEXT_REG, 1);
+	sx->pkts[sx->atom.npkts++] = 0x0000010E;
+	sx->pkts[sx->atom.npkts++] = psx.sx_alpha_ref;
+	*atom = &sx->atom;
+fprintf(stderr, "%s %d pkts\n", __func__, (*atom)->npkts);
+	return 0;
+out_err:
+	radeon_atom_put(&sx->atom);
+	*atom = NULL;
+	return r;
+}
+
+/*
  * r600 atom core functions
  */
 int r600_atom_create(struct radeon_device *rdev,
@@ -1437,6 +1539,13 @@ int r600_atom_create(struct radeon_device *rdev,
 		atom->type = patom->type;
 		list_add_tail(&atom->list, &atoms->spi_atoms);
 		break;
+	case R600_ATOM_SX:
+		r = r600_sx_create(rdev, atoms, patom, &atom);
+		if (r)
+			return r;
+		atom->type = patom->type;
+		list_add_tail(&atom->list, &atoms->sx_atoms);
+		break;
 	default:
 		dev_err(rdev->dev, "unknown R600 atom type 0x%08X\n", patom->type);
 		return -EINVAL;
@@ -1541,7 +1650,7 @@ int r600_batches_queue(struct radeon_device *rdev,
 		batch->pa == NULL || batch->vport == NULL ||
 		batch->tp == NULL || batch->cb == NULL ||
 		batch->db_cntl == NULL || batch->vgt == NULL ||
-		batch->spi == NULL) {
+		batch->spi == NULL || batch->sx == NULL) {
 		kfree(rbatch);
 		dev_err(rdev->dev, "invalid batch\n");
 		return -EINVAL;
@@ -1571,6 +1680,8 @@ int r600_batches_queue(struct radeon_device *rdev,
 	kref_get(&batch->vgt->kref);
 	rbatch->atoms[i++] = batch->spi;
 	kref_get(&batch->spi->kref);
+	rbatch->atoms[i++] = batch->sx;
+	kref_get(&batch->sx->kref);
 	rbatch->atoms[i++] = batch->cb;
 	kref_get(&batch->cb->kref);
 	db_cntl = container_of(batch->db_cntl, struct r600_db_cntl, atom);
@@ -1693,6 +1804,7 @@ int r600_atoms_init(struct radeon_device *rdev, struct r600_atoms *atoms)
 	INIT_LIST_HEAD(&atoms->pa_atoms);
 	INIT_LIST_HEAD(&atoms->spi_atoms);
 	INIT_LIST_HEAD(&atoms->sq_atoms);
+	INIT_LIST_HEAD(&atoms->sx_atoms);
 	INIT_LIST_HEAD(&atoms->tp_atoms);
 	INIT_LIST_HEAD(&atoms->vgt_atoms);
 	INIT_LIST_HEAD(&atoms->vport_atoms);
@@ -1779,15 +1891,6 @@ void r600_tflat(struct radeon_atom *atom)
     WPKT(PKT3(PKT3_SET_CONTEXT_REG, 1));
     WPKT(0x000001E8);
     WPKT(0x00000003);
-    WPKT(PKT3(PKT3_SET_CONTEXT_REG, 1));
-    WPKT(0x000000D4);
-    WPKT(0x00000000);
-    WPKT(PKT3(PKT3_SET_CONTEXT_REG, 1));
-    WPKT(0x00000104);
-    WPKT(0x00000000);
-    WPKT(PKT3(PKT3_SET_CONTEXT_REG, 1));
-    WPKT(0x0000010E);
-    WPKT(0x00000000);
     WPKT(PKT3(PKT3_SURFACE_SYNC, 3));
     WPKT(0x08000000);
     WPKT(0x00000001);
diff --git a/r600_atom_api.h b/r600_atom_api.h
index c21d21a..e676c5a 100644
--- a/r600_atom_api.h
+++ b/r600_atom_api.h
@@ -38,6 +38,7 @@ struct drm_radeon_atom {
 #define R600_ATOM_DB_CNTL	9
 #define R600_ATOM_VGT		10
 #define R600_ATOM_SPI		11
+#define R600_ATOM_SX		12
 
 struct drm_r600_cb {
 	u32			pitch;
@@ -262,6 +263,18 @@ struct drm_r600_spi {
 	u32			spi_vs_out_id_9;
 };
 
+/* sx - shader export */
+struct drm_r600_sx {
+	u32			sx_alpha_ref;
+	u32			sx_alpha_test_control;
+	u32			sx_export_buffer_sizes;
+	u32			sx_misc;
+	u32			offset;
+	struct radeon_bo	*buffer;
+	u32			placements[2];
+};
+
+
 struct drm_r600_batch {
 	struct radeon_atom	*vs_constants;
 	struct radeon_atom	*ps_constants;
@@ -275,6 +288,7 @@ struct drm_r600_batch {
 	struct radeon_atom	*db_cntl;
 	struct radeon_atom	*vgt;
 	struct radeon_atom	*spi;
+	struct radeon_atom	*sx;
 };
 
 
diff --git a/radeon_atom.h b/radeon_atom.h
index 58d150d..36805b8 100644
--- a/radeon_atom.h
+++ b/radeon_atom.h
@@ -53,7 +53,7 @@ struct radeon_atom {
 };
 
 /* R600 */
-#define R600_BATCH_NATOMS	12
+#define R600_BATCH_NATOMS	13
 struct r600_batch {
 	struct list_head	list;
 	struct list_head	pre_flushes;
@@ -82,6 +82,7 @@ struct r600_atoms {
 	struct list_head	pa_atoms;
 	struct list_head	spi_atoms;
 	struct list_head	sq_atoms;
+	struct list_head	sx_atoms;
 	struct list_head	tp_atoms;
 	struct list_head	vgt_atoms;
 	struct list_head	vport_atoms;
diff --git a/test.c b/test.c
index 12479f7..ab7448d 100644
--- a/test.c
+++ b/test.c
@@ -64,6 +64,7 @@ int r600_tri_flat(struct radeon *radeon)
 	struct drm_r600_constants vs_constants;
 	struct drm_r600_vgt vgt;
 	struct drm_r600_spi spi;
+	struct drm_r600_sx sx;
 	struct drm_r600_batch batch;
 	struct drm_radeon_atom atom;
 	int r;
@@ -319,6 +320,19 @@ int r600_tri_flat(struct radeon *radeon)
 	r = radeon_atom_create(rdev, &atom, &batch.spi);
 	if (r)
 		return r;
+	/* sx */
+	sx.sx_alpha_ref = 0x00000000;
+	sx.sx_alpha_test_control = 0x00000000;
+	sx.sx_export_buffer_sizes = 0x00000000;
+	sx.sx_misc = 0x00000000;
+	sx.offset = 0x00000000;
+	sx.buffer = NULL;
+	atom.type = R600_ATOM_SX;
+	atom.id = 0;
+	atom.data = (uint64_t)(uintptr_t)&sx;
+	r = radeon_atom_create(rdev, &atom, &batch.sx);
+	if (r)
+		return r;
 
 	/* batch */
 	r =  radeon_batches_queue(rdev, &batch);