nvfx: so->RING_3D: fragtex

5 files changed, 83 insertions, 72 deletions

diff --git a/src/gallium/drivers/nvfx/nv30_fragtex.c b/src/gallium/drivers/nvfx/nv30_fragtex.c
index e0356783ea0..dec073ac900 100644
--- a/src/gallium/drivers/nvfx/nv30_fragtex.c
+++ b/src/gallium/drivers/nvfx/nv30_fragtex.c
@@ -88,21 +88,21 @@ nv30_fragtex_format(uint pipe_format)
 }
 
 
-struct nouveau_stateobj *
-nv30_fragtex_build(struct nvfx_context *nvfx, int unit)
+void
+nv30_fragtex_set(struct nvfx_context *nvfx, int unit)
 {
 	struct nvfx_sampler_state *ps = nvfx->tex_sampler[unit];
 	struct nvfx_miptree *nv30mt = (struct nvfx_miptree *)nvfx->fragment_sampler_views[unit]->texture;
 	struct pipe_resource *pt = &nv30mt->base.base;
 	struct nouveau_bo *bo = nv30mt->base.bo;
 	struct nv30_texture_format *tf;
-	struct nouveau_stateobj *so;
+	struct nouveau_channel* chan = nvfx->screen->base.channel;
 	uint32_t txf, txs;
 	unsigned tex_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
 	tf = nv30_fragtex_format(pt->format);
 	if (!tf)
-		return NULL;
+		return;
 
 	txf  = tf->format;
 	txf |= ((pt->last_level>0) ? NV34TCL_TX_FORMAT_MIPMAP : 0);
@@ -126,23 +126,24 @@ nv30_fragtex_build(struct nvfx_context *nvfx, int unit)
 		break;
 	default:
 		NOUVEAU_ERR("Unknown target %d\n", pt->target);
-		return NULL;
+		return;
 	}
 
 	txs = tf->swizzle;
 
-	so = so_new(1, 8, 2);
-	so_method(so, nvfx->screen->eng3d, NV34TCL_TX_OFFSET(unit), 8);
-	so_reloc (so, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0);
-	so_reloc (so, bo, txf, tex_flags | NOUVEAU_BO_OR,
+	MARK_RING(chan, 9, 2);
+	OUT_RING(chan, RING_3D(NV34TCL_TX_OFFSET(unit), 8));
+	OUT_RELOC(chan, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0);
+	OUT_RELOC(chan, bo, txf, tex_flags | NOUVEAU_BO_OR,
 		      NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1);
-	so_data  (so, ps->wrap);
-	so_data  (so, NV34TCL_TX_ENABLE_ENABLE | ps->en);
-	so_data  (so, txs);
-	so_data  (so, ps->filt | 0x2000 /*voodoo*/);
-	so_data  (so, (pt->width0 << NV34TCL_TX_NPOT_SIZE_W_SHIFT) |
+	OUT_RING(chan, ps->wrap);
+	OUT_RING(chan, NV34TCL_TX_ENABLE_ENABLE | ps->en);
+	OUT_RING(chan, txs);
+	OUT_RING(chan, ps->filt | 0x2000 /*voodoo*/);
+	OUT_RING(chan, (pt->width0 << NV34TCL_TX_NPOT_SIZE_W_SHIFT) |
 		       pt->height0);
-	so_data  (so, ps->bcol);
+	OUT_RING(chan, ps->bcol);
 
-	return so;
+	nvfx->hw_txf[unit] = txf;
+	nvfx->hw_samplers |= (1 << unit);
 }
diff --git a/src/gallium/drivers/nvfx/nv40_fragtex.c b/src/gallium/drivers/nvfx/nv40_fragtex.c
index bffdf0893c3..0068b1ba54a 100644
--- a/src/gallium/drivers/nvfx/nv40_fragtex.c
+++ b/src/gallium/drivers/nvfx/nv40_fragtex.c
@@ -106,15 +106,16 @@ nv40_fragtex_format(uint pipe_format)
 }
 
 
-struct nouveau_stateobj *
-nv40_fragtex_build(struct nvfx_context *nvfx, int unit)
+void
+nv40_fragtex_set(struct nvfx_context *nvfx, int unit)
 {
+	struct nouveau_channel* chan = nvfx->screen->base.channel;
 	struct nvfx_sampler_state *ps = nvfx->tex_sampler[unit];
 	struct nvfx_miptree *nv40mt = (struct nvfx_miptree *)nvfx->fragment_sampler_views[unit]->texture;
 	struct nouveau_bo *bo = nv40mt->base.bo;
 	struct pipe_resource *pt = &nv40mt->base.base;
 	struct nv40_texture_format *tf;
-	struct nouveau_stateobj *so;
+
 	uint32_t txf, txs, txp;
 	unsigned tex_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
@@ -144,7 +145,7 @@ nv40_fragtex_build(struct nvfx_context *nvfx, int unit)
 		break;
 	default:
 		NOUVEAU_ERR("Unknown target %d\n", pt->target);
-		return NULL;
+		return;
 	}
 
 	if (!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR)) {
@@ -156,20 +157,20 @@ nv40_fragtex_build(struct nvfx_context *nvfx, int unit)
 
 	txs = tf->swizzle;
 
-	so = so_new(2, 9, 2);
-	so_method(so, nvfx->screen->eng3d, NV34TCL_TX_OFFSET(unit), 8);
-	so_reloc (so, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0);
-	so_reloc (so, bo, txf, tex_flags | NOUVEAU_BO_OR,
-		      NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1);
-	so_data  (so, ps->wrap);
-	so_data  (so, NV40TCL_TEX_ENABLE_ENABLE | ps->en);
-	so_data  (so, txs);
-	so_data  (so, ps->filt | tf->sign | 0x2000 /*voodoo*/);
-	so_data  (so, (pt->width0 << NV34TCL_TX_NPOT_SIZE_W_SHIFT) |
-		       pt->height0);
-	so_data  (so, ps->bcol);
-	so_method(so, nvfx->screen->eng3d, NV40TCL_TEX_SIZE1(unit), 1);
-	so_data  (so, (pt->depth0 << NV40TCL_TEX_SIZE1_DEPTH_SHIFT) | txp);
-
-	return so;
+	MARK_RING(chan, 11 + 2 * !unit, 2);
+	OUT_RING(chan, RING_3D(NV34TCL_TX_OFFSET(unit), 8));
+	OUT_RELOC(chan, bo, 0, tex_flags | NOUVEAU_BO_LOW, 0, 0);
+	OUT_RELOC(chan, bo, txf, tex_flags | NOUVEAU_BO_OR,
+			NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1);
+	OUT_RING(chan, ps->wrap);
+	OUT_RING(chan, NV40TCL_TEX_ENABLE_ENABLE | ps->en);
+	OUT_RING(chan, txs);
+	OUT_RING(chan, ps->filt | tf->sign | 0x2000 /*voodoo*/);
+	OUT_RING(chan, (pt->width0 << NV34TCL_TX_NPOT_SIZE_W_SHIFT) | pt->height0);
+	OUT_RING(chan, ps->bcol);
+	OUT_RING(chan, RING_3D(NV40TCL_TEX_SIZE1(unit), 1));
+	OUT_RING(chan, (pt->depth0 << NV40TCL_TEX_SIZE1_DEPTH_SHIFT) | txp);
+
+	nvfx->hw_txf[unit] = txf;
+	nvfx->hw_samplers |= (1 << unit);
 }
diff --git a/src/gallium/drivers/nvfx/nvfx_context.h b/src/gallium/drivers/nvfx/nvfx_context.h
index 2022f6e5121..7f6e5500e54 100644
--- a/src/gallium/drivers/nvfx/nvfx_context.h
+++ b/src/gallium/drivers/nvfx/nvfx_context.h
@@ -167,6 +167,8 @@ struct nvfx_context {
 
 	unsigned vbo_bo;
 	unsigned hw_vtxelt_nr;
+	uint8_t hw_samplers;
+	uint32_t hw_txf[8];
 };
 
 static INLINE struct nvfx_context *
@@ -220,23 +222,23 @@ extern void nvfx_draw_elements_swtnl(struct pipe_context *pipe,
 extern void nvfx_fragprog_destroy(struct nvfx_context *,
 				    struct nvfx_fragment_program *);
 
+/* nvfx_fragtex.c */
+extern void
+nvfx_fragtex_relocate(struct nvfx_context *nvfx);
+
 /* nv30_fragtex.c */
 extern void
 nv30_sampler_state_init(struct pipe_context *pipe,
 			  struct nvfx_sampler_state *ps,
 			  const struct pipe_sampler_state *cso);
-extern void nv30_fragtex_bind(struct nvfx_context *);
-extern struct nouveau_stateobj *
-nv30_fragtex_build(struct nvfx_context *nvfx, int unit);
+extern void nv30_fragtex_set(struct nvfx_context *nvfx, int unit);
 
 /* nv40_fragtex.c */
 extern void
 nv40_sampler_state_init(struct pipe_context *pipe,
 			  struct nvfx_sampler_state *ps,
 			  const struct pipe_sampler_state *cso);
-extern void nv40_fragtex_bind(struct nvfx_context *);
-extern struct nouveau_stateobj *
-nv40_fragtex_build(struct nvfx_context *nvfx, int unit);
+extern void nv40_fragtex_set(struct nvfx_context *nvfx, int unit);
 
 /* nvfx_state.c */
 extern void nvfx_init_state_functions(struct nvfx_context *nvfx);
diff --git a/src/gallium/drivers/nvfx/nvfx_fragtex.c b/src/gallium/drivers/nvfx/nvfx_fragtex.c
index 84e4eb10042..e239235c3f5 100644
--- a/src/gallium/drivers/nvfx/nvfx_fragtex.c
+++ b/src/gallium/drivers/nvfx/nvfx_fragtex.c
@@ -1,43 +1,58 @@
 #include "nvfx_context.h"
+#include "nvfx_resource.h"
 
 static boolean
 nvfx_fragtex_validate(struct nvfx_context *nvfx)
 {
-	struct nvfx_fragment_program *fp = nvfx->fragprog;
-	struct nvfx_state *state = &nvfx->state;
-	struct nouveau_stateobj *so;
+	struct nouveau_channel* chan = nvfx->screen->base.channel;
 	unsigned samplers, unit;
 
-	samplers = state->fp_samplers & ~fp->samplers;
+	samplers = nvfx->dirty_samplers;
+	if(!samplers)
+		return FALSE;
+
 	while (samplers) {
 		unit = ffs(samplers) - 1;
 		samplers &= ~(1 << unit);
 
-		so = so_new(1, 1, 0);
-		so_method(so, nvfx->screen->eng3d, NV34TCL_TX_ENABLE(unit), 1);
-		so_data  (so, 0);
-		so_ref(so, &nvfx->state.hw[NVFX_STATE_FRAGTEX0 + unit]);
-		so_ref(NULL, &so);
-		state->dirty |= (1ULL << (NVFX_STATE_FRAGTEX0 + unit));
+		if(nvfx->fragment_sampler_views[unit] && nvfx->tex_sampler[unit]) {
+			if(!nvfx->is_nv4x)
+				nv30_fragtex_set(nvfx, unit);
+			else
+				nv40_fragtex_set(nvfx, unit);
+		} else {
+			WAIT_RING(chan, 2);
+			/* this is OK for nv40 too */
+			OUT_RING(chan, RING_3D(NV34TCL_TX_ENABLE(unit), 1));
+			OUT_RING(chan, 0);
+			nvfx->hw_samplers &= ~(1 << unit);
+		}
 	}
+	nvfx->dirty_samplers = 0;
+	return FALSE;
+}
+
+void
+nvfx_fragtex_relocate(struct nvfx_context *nvfx)
+{
+	struct nouveau_channel* chan = nvfx->screen->base.channel;
+	unsigned samplers, unit;
+	unsigned tex_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
-	samplers = nvfx->dirty_samplers & fp->samplers;
+	samplers = nvfx->hw_samplers;
 	while (samplers) {
 		unit = ffs(samplers) - 1;
 		samplers &= ~(1 << unit);
 
-		if(!nvfx->is_nv4x)
-			so = nv30_fragtex_build(nvfx, unit);
-		else
-			so = nv40_fragtex_build(nvfx, unit);
+		struct nvfx_miptree* mt = (struct nvfx_miptree*)nvfx->fragment_sampler_views[unit]->texture;
+		struct nouveau_bo *bo = mt->base.bo;
 
-		so_ref(so, &nvfx->state.hw[NVFX_STATE_FRAGTEX0 + unit]);
-		so_ref(NULL, &so);
-		state->dirty |= (1ULL << (NVFX_STATE_FRAGTEX0 + unit));
+		MARK_RING(chan, 3, 3);
+		OUT_RELOC(chan, bo, RING_3D(NV34TCL_TX_OFFSET(unit), 2), tex_flags | NOUVEAU_BO_DUMMY, 0, 0);
+		OUT_RELOC(chan, bo, 0, tex_flags | NOUVEAU_BO_LOW | NOUVEAU_BO_DUMMY, 0, 0);
+		OUT_RELOC(chan, bo, nvfx->hw_txf[unit], tex_flags | NOUVEAU_BO_OR | NOUVEAU_BO_DUMMY,
+				NV34TCL_TX_FORMAT_DMA0, NV34TCL_TX_FORMAT_DMA1);
 	}
-
-	nvfx->state.fp_samplers = fp->samplers;
-	return FALSE;
 }
 
 struct nvfx_state_entry nvfx_state_fragtex = {
diff --git a/src/gallium/drivers/nvfx/nvfx_state_emit.c b/src/gallium/drivers/nvfx/nvfx_state_emit.c
index cb6ae89e2ab..8adf59d8e3d 100644
--- a/src/gallium/drivers/nvfx/nvfx_state_emit.c
+++ b/src/gallium/drivers/nvfx/nvfx_state_emit.c
@@ -100,16 +100,8 @@ nvfx_state_relocate(struct nvfx_context *nvfx)
 {
 	struct nouveau_channel *chan = nvfx->screen->base.channel;
 	struct nvfx_state *state = &nvfx->state;
-	unsigned i, samplers;
-
 	so_emit_reloc_markers(chan, state->hw[NVFX_STATE_FB]);
-	for (i = 0, samplers = state->fp_samplers; i < 16 && samplers; i++) {
-		if (!(samplers & (1 << i)))
-			continue;
-		so_emit_reloc_markers(chan,
-				      state->hw[NVFX_STATE_FRAGTEX0+i]);
-		samplers &= ~(1ULL << i);
-	}
+	nvfx_fragtex_relocate(nvfx);
 	so_emit_reloc_markers(chan, state->hw[NVFX_STATE_FRAGPROG]);
 	if (nvfx->render_mode == HW)
 		nvfx_vbo_relocate(nvfx);