nvfx: add rewritten swtnl support

The old swtnl code was broken by the new shader linkage support for
GLSL.

This is a rewrite of swtnl support, which should instead work properly,
be faster and more closer to the much more tested hardware pipeline.

1 files changed, 54 insertions, 182 deletions

diff --git a/src/gallium/drivers/nvfx/nvfx_draw.c b/src/gallium/drivers/nvfx/nvfx_draw.c
index 2601d5b8e2e..4bf38a9c181 100644
--- a/src/gallium/drivers/nvfx/nvfx_draw.c
+++ b/src/gallium/drivers/nvfx/nvfx_draw.c
@@ -1,6 +1,5 @@
 #include "pipe/p_shader_tokens.h"
 #include "util/u_inlines.h"
-#include "tgsi/tgsi_ureg.h"
 
 #include "util/u_pack_color.h"
 
@@ -11,11 +10,6 @@
 #include "nvfx_context.h"
 #include "nvfx_resource.h"
 
-/* Simple, but crappy, swtnl path, hopefully we wont need to hit this very
- * often at all.  Uses "quadro style" vertex submission + a fixed vertex
- * layout to avoid the need to generate a vertex program or vtxfmt.
- */
-
 struct nvfx_render_stage {
 	struct draw_stage stage;
 	struct nvfx_context *nvfx;
@@ -28,58 +22,18 @@ nvfx_render_stage(struct draw_stage *stage)
 	return (struct nvfx_render_stage *)stage;
 }
 
-static INLINE void
-nvfx_render_vertex(struct nvfx_context *nvfx, const struct vertex_header *v)
+static void
+nvfx_render_flush(struct draw_stage *stage, unsigned flags)
 {
-	struct nvfx_screen *screen = nvfx->screen;
-	struct nouveau_channel *chan = screen->base.channel;
-	struct nouveau_grobj *eng3d = screen->eng3d;
-	unsigned i;
+	struct nvfx_render_stage *rs = nvfx_render_stage(stage);
+	struct nvfx_context *nvfx = rs->nvfx;
+	struct nouveau_channel *chan = nvfx->screen->base.channel;
 
-	for (i = 0; i < nvfx->swtnl.nr_attribs; i++) {
-		unsigned idx = nvfx->swtnl.draw[i];
-		unsigned hw = nvfx->swtnl.hw[i];
-
-		WAIT_RING(chan, 5);
-		switch (nvfx->swtnl.emit[i]) {
-		case EMIT_OMIT:
-			break;
-		case EMIT_1F:
-			nvfx_emit_vtx_attr(chan, hw, v->data[idx], 1);
-			break;
-		case EMIT_2F:
-			nvfx_emit_vtx_attr(chan, hw, v->data[idx], 2);
-			break;
-		case EMIT_3F:
-			nvfx_emit_vtx_attr(chan, hw, v->data[idx], 3);
-			break;
-		case EMIT_4F:
-			nvfx_emit_vtx_attr(chan, hw, v->data[idx], 4);
-			break;
-		case 0xff:
-			BEGIN_RING(chan, eng3d, NV34TCL_VTX_ATTR_4F_X(hw), 4);
-			OUT_RING  (chan, fui(v->data[idx][0] / v->data[idx][3]));
-			OUT_RING  (chan, fui(v->data[idx][1] / v->data[idx][3]));
-			OUT_RING  (chan, fui(v->data[idx][2] / v->data[idx][3]));
-			OUT_RING  (chan, fui(1.0f / v->data[idx][3]));
-			break;
-		case EMIT_4UB:
-			BEGIN_RING(chan, eng3d, NV34TCL_VTX_ATTR_4UB(hw), 1);
-			OUT_RING  (chan, pack_ub4(float_to_ubyte(v->data[idx][0]),
-					    float_to_ubyte(v->data[idx][1]),
-					    float_to_ubyte(v->data[idx][2]),
-					    float_to_ubyte(v->data[idx][3])));
-		case EMIT_4UB_BGRA:
-			BEGIN_RING(chan, eng3d, NV34TCL_VTX_ATTR_4UB(hw), 1);
-			OUT_RING  (chan, pack_ub4(float_to_ubyte(v->data[idx][2]),
-					    float_to_ubyte(v->data[idx][1]),
-					    float_to_ubyte(v->data[idx][0]),
-					    float_to_ubyte(v->data[idx][3])));
-			break;
-		default:
-			assert(0);
-			break;
-		}
+	if (rs->prim != NV34TCL_VERTEX_BEGIN_END_STOP) {
+		assert(AVAIL_RING(chan) >= 2);
+		OUT_RING(chan, RING_3D(NV34TCL_VERTEX_BEGIN_END, 1));
+		OUT_RING(chan, NV34TCL_VERTEX_BEGIN_END_STOP);
+		rs->prim = NV34TCL_VERTEX_BEGIN_END_STOP;
 	}
 }
 
@@ -92,42 +46,61 @@ nvfx_render_prim(struct draw_stage *stage, struct prim_header *prim,
 
 	struct nvfx_screen *screen = nvfx->screen;
 	struct nouveau_channel *chan = screen->base.channel;
-	struct nouveau_grobj *eng3d = screen->eng3d;
-	unsigned i;
+	boolean no_elements = nvfx->vertprog->draw_no_elements;
+	unsigned num_attribs = nvfx->vertprog->draw_elements;
 
-	/* Ensure there's room for 4xfloat32 + potentially 3 begin/end */
-	if (AVAIL_RING(chan) < ((count * 20) + 6)) {
-		if (rs->prim != NV34TCL_VERTEX_BEGIN_END_STOP) {
-			NOUVEAU_ERR("AIII, missed flush\n");
-			assert(0);
-		}
+	/* we need to account the flush as well here even if it is done afterthis
+	 * function
+	 */
+	if (AVAIL_RING(chan) < ((1 + count * num_attribs * 4) + 6 + 64)) {
+		nvfx_render_flush(stage, 0);
 		FIRE_RING(chan);
 		nvfx_state_emit(nvfx);
+
+		assert(AVAIL_RING(chan) >= ((1 + count * num_attribs * 4) + 6 + 64));
 	}
 
 	/* Switch primitive modes if necessary */
 	if (rs->prim != mode) {
 		if (rs->prim != NV34TCL_VERTEX_BEGIN_END_STOP) {
-			BEGIN_RING(chan, eng3d, NV34TCL_VERTEX_BEGIN_END, 1);
-			OUT_RING  (chan, NV34TCL_VERTEX_BEGIN_END_STOP);
+			OUT_RING(chan, RING_3D(NV34TCL_VERTEX_BEGIN_END, 1));
+			OUT_RING(chan, NV34TCL_VERTEX_BEGIN_END_STOP);
+		}
+
+		/* XXX: any command a lot of times seems to (mostly) fix corruption that would otherwise happen */
+		/* this seems to cause issues on nv3x, and also be unneeded there */
+		if(nvfx->is_nv4x)
+		{
+			int i;
+			for(i = 0; i < 32; ++i)
+			{
+				OUT_RING(chan, RING_3D(0x1dac, 1));
+				OUT_RING(chan, 0);
+			}
 		}
 
-		BEGIN_RING(chan, eng3d, NV34TCL_VERTEX_BEGIN_END, 1);
+		OUT_RING(chan, RING_3D(NV34TCL_VERTEX_BEGIN_END, 1));
 		OUT_RING  (chan, mode);
 		rs->prim = mode;
 	}
 
-	/* Emit vertex data */
-	for (i = 0; i < count; i++)
-		nvfx_render_vertex(nvfx, prim->v[i]);
-
-	/* If it's likely we'll need to empty the push buffer soon, finish
-	 * off the primitive now.
-	 */
-	if (AVAIL_RING(chan) < ((count * 20) + 6)) {
-		BEGIN_RING(chan, eng3d, NV34TCL_VERTEX_BEGIN_END, 1);
-		OUT_RING  (chan, NV34TCL_VERTEX_BEGIN_END_STOP);
-		rs->prim = NV34TCL_VERTEX_BEGIN_END_STOP;
+	OUT_RING(chan, RING_3D_NI(NV34TCL_VERTEX_DATA, num_attribs * 4 * count));
+	if(no_elements) {
+		OUT_RING(chan, 0);
+		OUT_RING(chan, 0);
+		OUT_RING(chan, 0);
+		OUT_RING(chan, 0);
+	} else {
+		for (unsigned i = 0; i < count; ++i)
+		{
+			struct vertex_header* v = prim->v[i];
+			/* TODO: disable divide where it's causing the problem, and remove this hack */
+			OUT_RING(chan, fui(v->data[0][0] / v->data[0][3]));
+			OUT_RING(chan, fui(v->data[0][1] / v->data[0][3]));
+			OUT_RING(chan, fui(v->data[0][2] / v->data[0][3]));
+			OUT_RING(chan, fui(1.0f / v->data[0][3]));
+			OUT_RINGp(chan, &v->data[1][0], 4 * (num_attribs - 1));
+		}
 	}
 }
 
@@ -150,24 +123,10 @@ nvfx_render_tri(struct draw_stage *draw, struct prim_header *prim)
 }
 
 static void
-nvfx_render_flush(struct draw_stage *draw, unsigned flags)
-{
-	struct nvfx_render_stage *rs = nvfx_render_stage(draw);
-	struct nvfx_context *nvfx = rs->nvfx;
-	struct nvfx_screen *screen = nvfx->screen;
-	struct nouveau_channel *chan = screen->base.channel;
-	struct nouveau_grobj *eng3d = screen->eng3d;
-
-	if (rs->prim != NV34TCL_VERTEX_BEGIN_END_STOP) {
-		BEGIN_RING(chan, eng3d, NV34TCL_VERTEX_BEGIN_END, 1);
-		OUT_RING  (chan, NV34TCL_VERTEX_BEGIN_END_STOP);
-		rs->prim = NV34TCL_VERTEX_BEGIN_END_STOP;
-	}
-}
-
-static void
 nvfx_render_reset_stipple_counter(struct draw_stage *draw)
 {
+	/* this doesn't really seem to work, but it matters rather little */
+	nvfx_render_flush(draw, 0);
 }
 
 static void
@@ -176,40 +135,11 @@ nvfx_render_destroy(struct draw_stage *draw)
 	FREE(draw);
 }
 
-static struct nvfx_vertex_program *
-nvfx_create_drawvp(struct nvfx_context *nvfx)
-{
-	struct ureg_program *ureg;
-	uint i;
-
-	ureg = ureg_create( TGSI_PROCESSOR_VERTEX );
-	if (ureg == NULL)
-		return NULL;
-
-	ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0), ureg_DECL_vs_input(ureg, 0));
-	ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0), ureg_DECL_vs_input(ureg, 3));
-	ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 1), ureg_DECL_vs_input(ureg, 4));
-	ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_BCOLOR, 0), ureg_DECL_vs_input(ureg, 3));
-	ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_BCOLOR, 1), ureg_DECL_vs_input(ureg, 4));
-	ureg_MOV(ureg,
-		   ureg_writemask(ureg_DECL_output(ureg, TGSI_SEMANTIC_FOG, 1), TGSI_WRITEMASK_X),
-		   ureg_DECL_vs_input(ureg, 5));
-	for (i = 0; i < 8; ++i)
-		ureg_MOV(ureg, ureg_DECL_output(ureg, TGSI_SEMANTIC_GENERIC, i), ureg_DECL_vs_input(ureg, 8 + i));
-
-	ureg_END( ureg );
-
-	return ureg_create_shader_and_destroy( ureg, &nvfx->pipe );
-}
-
 struct draw_stage *
 nvfx_draw_render_stage(struct nvfx_context *nvfx)
 {
 	struct nvfx_render_stage *render = CALLOC_STRUCT(nvfx_render_stage);
 
-	if (!nvfx->swtnl.vertprog)
-		nvfx->swtnl.vertprog = nvfx_create_drawvp(nvfx);
-
 	render->nvfx = nvfx;
 	render->stage.draw = nvfx->draw;
 	render->stage.point = nvfx_render_point;
@@ -231,6 +161,7 @@ nvfx_draw_vbo_swtnl(struct pipe_context *pipe, const struct pipe_draw_info* info
 
 	if (!nvfx_state_validate_swtnl(nvfx))
 		return;
+
 	nvfx_state_emit(nvfx);
 
 	/* these must be passed without adding the offsets */
@@ -256,62 +187,3 @@ nvfx_draw_vbo_swtnl(struct pipe_context *pipe, const struct pipe_draw_info* info
 
 	draw_flush(nvfx->draw);
 }
-
-static INLINE void
-emit_attrib(struct nvfx_context *nvfx, unsigned hw, unsigned emit,
-	    unsigned semantic, unsigned index)
-{
-	unsigned draw_out = draw_find_shader_output(nvfx->draw, semantic, index);
-	unsigned a = nvfx->swtnl.nr_attribs++;
-
-	nvfx->swtnl.hw[a] = hw;
-	nvfx->swtnl.emit[a] = emit;
-	nvfx->swtnl.draw[a] = draw_out;
-}
-
-void
-nvfx_vtxfmt_validate(struct nvfx_context *nvfx)
-{
-	struct nvfx_pipe_fragment_program *pfp = nvfx->fragprog;
-	unsigned colour = 0, texcoords = 0, fog = 0, i;
-
-	/* Determine needed fragprog inputs */
-	for (i = 0; i < pfp->info.num_inputs; i++) {
-		switch (pfp->info.input_semantic_name[i]) {
-		case TGSI_SEMANTIC_POSITION:
-			break;
-		case TGSI_SEMANTIC_COLOR:
-			colour |= (1 << pfp->info.input_semantic_index[i]);
-			break;
-		case TGSI_SEMANTIC_GENERIC:
-			texcoords |= (1 << pfp->info.input_semantic_index[i]);
-			break;
-		case TGSI_SEMANTIC_FOG:
-			fog = 1;
-			break;
-		default:
-			assert(0);
-		}
-	}
-
-	nvfx->swtnl.nr_attribs = 0;
-
-	/* Map draw vtxprog output to hw attribute IDs */
-	for (i = 0; i < 2; i++) {
-		if (!(colour & (1 << i)))
-			continue;
-		emit_attrib(nvfx, 3 + i, EMIT_4F, TGSI_SEMANTIC_COLOR, i);
-	}
-
-	for (i = 0; i < 8; i++) {
-		if (!(texcoords & (1 << i)))
-			continue;
-		emit_attrib(nvfx, 8 + i, EMIT_4F, TGSI_SEMANTIC_GENERIC, i);
-	}
-
-	if (fog) {
-		emit_attrib(nvfx, 5, EMIT_1F, TGSI_SEMANTIC_FOG, 0);
-	}
-
-	emit_attrib(nvfx, 0, 0xff, TGSI_SEMANTIC_POSITION, 0);
-}