nouveau: rewrite nouveau_stateobj to use BEGIN_RING properly

- The previous solution was hacky and didn't do subchannel autobinding.
- The beheaviour should match what libdrm_nouveau does closely.
- The solution remains statically sized, but when debugging is on it will check
for abuse.

Signed-off-by: Maarten Maathuis <[email protected]>

6 files changed, 25 insertions, 25 deletions

diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 2d0b1818ef6..af0759e5038 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -3579,7 +3579,7 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
-	so = so_new(13, 2);
+	so = so_new(5, 8, 2);
 	so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
 		      NOUVEAU_BO_HIGH, 0, 0);
@@ -3615,7 +3615,7 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
-	so = so_new(64, 2);
+	so = so_new(6, 7, 2);
 	so_method(so, tesla, NV50TCL_FP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
 		      NOUVEAU_BO_HIGH, 0, 0);
@@ -3783,7 +3783,7 @@ nv50_linkage_validate(struct nv50_context *nv50)
 	}
 
 	/* now fill the stateobj */
-	so = so_new(64, 0);
+	so = so_new(6, 58, 0);
 
 	n = (m + 3) / 4;
 	so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1);
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 1778a745179..28e2b35deaa 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -251,7 +251,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		nv50_screen_destroy(pscreen);
 		return NULL;
 	}
-	BIND_RING(chan, screen->m2mf, 1);
 
 	/* 2D object */
 	ret = nouveau_grobj_alloc(chan, 0xbeef502d, NV50_2D, &screen->eng2d);
@@ -260,7 +259,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		nv50_screen_destroy(pscreen);
 		return NULL;
 	}
-	BIND_RING(chan, screen->eng2d, 2);
 
 	/* 3D object */
 	switch (chipset & 0xf0) {
@@ -296,7 +294,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 		nv50_screen_destroy(pscreen);
 		return NULL;
 	}
-	BIND_RING(chan, screen->tesla, 3);
 
 	/* Sync notifier */
 	ret = nouveau_notifier_alloc(chan, 0xbeef0301, 1, &screen->sync);
@@ -307,7 +304,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	}
 
 	/* Static M2MF init */
-	so = so_new(32, 0);
+	so = so_new(1, 3, 0);
 	so_method(so, screen->m2mf, NV04_MEMORY_TO_MEMORY_FORMAT_DMA_NOTIFY, 3);
 	so_data  (so, screen->sync->handle);
 	so_data  (so, chan->vram->handle);
@@ -316,7 +313,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref (NULL, &so);
 
 	/* Static 2D init */
-	so = so_new(64, 0);
+	so = so_new(4, 7, 0);
 	so_method(so, screen->eng2d, NV50_2D_DMA_NOTIFY, 4);
 	so_data  (so, screen->sync->handle);
 	so_data  (so, chan->vram->handle);
@@ -332,7 +329,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref(NULL, &so);
 
 	/* Static tesla init */
-	so = so_new(256, 20);
+	so = so_new(40, 84, 20);
 
 	so_method(so, screen->tesla, NV50TCL_COND_MODE, 1);
 	so_data  (so, NV50TCL_COND_MODE_ALWAYS);
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 30b2b0f91bf..fd0a33d7f7e 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -35,7 +35,7 @@ static void *
 nv50_blend_state_create(struct pipe_context *pipe,
 			const struct pipe_blend_state *cso)
 {
-	struct nouveau_stateobj *so = so_new(64, 0);
+	struct nouveau_stateobj *so = so_new(5, 24, 0);
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_blend_stateobj *bso = CALLOC_STRUCT(nv50_blend_stateobj);
 	unsigned cmask = 0, i;
@@ -280,7 +280,7 @@ static void *
 nv50_rasterizer_state_create(struct pipe_context *pipe,
 			     const struct pipe_rasterizer_state *cso)
 {
-	struct nouveau_stateobj *so = so_new(64, 0);
+	struct nouveau_stateobj *so = so_new(15, 21, 0);
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_rasterizer_stateobj *rso =
 		CALLOC_STRUCT(nv50_rasterizer_stateobj);
@@ -425,7 +425,7 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 {
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_zsa_stateobj *zsa = CALLOC_STRUCT(nv50_zsa_stateobj);
-	struct nouveau_stateobj *so = so_new(64, 0);
+	struct nouveau_stateobj *so = so_new(8, 22, 0);
 
 	so_method(so, tesla, NV50TCL_DEPTH_WRITE_ENABLE, 1);
 	so_data  (so, cso->depth.writemask ? 1 : 0);
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 682786345ef..f83232f43cf 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -33,7 +33,7 @@ static void
 nv50_state_validate_fb(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_stateobj *so = so_new(128, 18);
+	struct nouveau_stateobj *so = so_new(32, 79, 18);
 	struct pipe_framebuffer_state *fb = &nv50->framebuffer;
 	unsigned i, w, h, gw = 0;
 
@@ -299,7 +299,7 @@ nv50_state_validate(struct nv50_context *nv50)
 		so_ref(nv50->rasterizer->so, &nv50->state.rast);
 
 	if (nv50->dirty & NV50_NEW_BLEND_COLOUR) {
-		so = so_new(5, 0);
+		so = so_new(1, 4, 0);
 		so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
 		so_data  (so, fui(nv50->blend_colour.color[0]));
 		so_data  (so, fui(nv50->blend_colour.color[1]));
@@ -310,7 +310,7 @@ nv50_state_validate(struct nv50_context *nv50)
 	}
 
 	if (nv50->dirty & NV50_NEW_STIPPLE) {
-		so = so_new(33, 0);
+		so = so_new(1, 32, 0);
 		so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
 		for (i = 0; i < 32; i++)
 			so_data(so, util_bswap32(nv50->stipple.stipple[i]));
@@ -327,7 +327,7 @@ nv50_state_validate(struct nv50_context *nv50)
 			goto scissor_uptodate;
 		nv50->state.scissor_enabled = rast->scissor;
 
-		so = so_new(3, 0);
+		so = so_new(1, 2, 0);
 		so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
 		if (nv50->state.scissor_enabled) {
 			so_data(so, (s->maxx << 16) | s->minx);
@@ -356,7 +356,7 @@ scissor_uptodate:
 			goto viewport_uptodate;
 		nv50->state.viewport_bypass = bypass;
 
-		so = so_new(14, 0);
+		so = so_new(5, 9, 0);
 		if (!bypass) {
 			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
 			so_data  (so, fui(nv50->viewport.translate[0]));
@@ -400,7 +400,8 @@ viewport_uptodate:
 		for (i = 0; i < PIPE_SHADER_TYPES; ++i)
 			nr += nv50->sampler_nr[i];
 
-		so = so_new(nr * 8 + 24 * PIPE_SHADER_TYPES + 2, 4);
+		so = so_new(1+ 5 * PIPE_SHADER_TYPES, 1+ 19 * PIPE_SHADER_TYPES
+					+ nr * 8, PIPE_SHADER_TYPES * 2);
 
 		nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX);
 		nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT);
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index c4ca096d6ac..bef548b7286 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -199,16 +199,18 @@ nv50_tex_validate(struct nv50_context *nv50)
 {
 	struct nouveau_stateobj *so;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	unsigned p, push, nrlc;
+	unsigned p, start, push, nrlc;
 
-	for (nrlc = 0, push = 0, p = 0; p < PIPE_SHADER_TYPES; ++p) {
+	for (nrlc = 0, start = 0, push = 0, p = 0; p < PIPE_SHADER_TYPES; ++p) {
+		start += MAX2(nv50->miptree_nr[p], nv50->state.miptree_nr[p]);
 		push += MAX2(nv50->miptree_nr[p], nv50->state.miptree_nr[p]);
 		nrlc += nv50->miptree_nr[p];
 	}
-	push = push * 11 + 23 * PIPE_SHADER_TYPES + 4;
+	start = start * 2 + 4 * PIPE_SHADER_TYPES + 2;
+	push = push * 9 + 19 * PIPE_SHADER_TYPES + 2;
 	nrlc = nrlc * 2 + 2 * PIPE_SHADER_TYPES;
 
-	so = so_new(push, nrlc);
+	so = so_new(start, push, nrlc);
 
 	if (nv50_validate_textures(nv50, so, PIPE_SHADER_VERTEX) == FALSE ||
 	    nv50_validate_textures(nv50, so, PIPE_SHADER_FRAGMENT) == FALSE) {
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index df18c2dd20a..f2e510fba61 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -354,7 +354,7 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 
 	so = *pso;
 	if (!so)
-		*pso = so = so_new(nv50->vtxelt_nr * 5, 0);
+		*pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
 
 	switch (ve->nr_components) {
 	case 4:
@@ -415,8 +415,8 @@ nv50_vbo_validate(struct nv50_context *nv50)
 	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
 
 	vtxattr = NULL;
-	vtxbuf = so_new(n_ve * 7, nv50->vtxelt_nr * 4);
-	vtxfmt = so_new(n_ve + 1, 0);
+	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt_nr * 4);
+	vtxfmt = so_new(1, n_ve, 0);
 	so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
 
 	for (i = 0; i < nv50->vtxelt_nr; i++) {