nvfx: so->RING_3D: fb

3 files changed, 117 insertions, 77 deletions

diff --git a/src/gallium/drivers/nvfx/nvfx_context.h b/src/gallium/drivers/nvfx/nvfx_context.h
index 46b5db03648..371c61f54dc 100644
--- a/src/gallium/drivers/nvfx/nvfx_context.h
+++ b/src/gallium/drivers/nvfx/nvfx_context.h
@@ -111,6 +111,12 @@ struct nvfx_vtxelt_state {
 	unsigned num_elements;
 };
 
+struct nvfx_render_target {
+	struct nouveau_bo* bo;
+	unsigned offset;
+	unsigned pitch;
+};
+
 struct nvfx_context {
 	struct pipe_context pipe;
 
@@ -169,6 +175,8 @@ struct nvfx_context {
 	unsigned hw_vtxelt_nr;
 	uint8_t hw_samplers;
 	uint32_t hw_txf[8];
+	struct nvfx_render_target hw_rt[4];
+	struct nvfx_render_target hw_zeta;
 };
 
 static INLINE struct nvfx_context *
@@ -218,6 +226,10 @@ extern void nvfx_draw_elements_swtnl(struct pipe_context *pipe,
 					unsigned ib_size, unsigned mode,
 					unsigned start, unsigned count);
 
+/* nvfx_fb.c */
+void
+nvfx_framebuffer_relocate(struct nvfx_context *nvfx);
+
 /* nvfx_fragprog.c */
 extern void nvfx_fragprog_destroy(struct nvfx_context *,
 				    struct nvfx_fragment_program *);
diff --git a/src/gallium/drivers/nvfx/nvfx_state_emit.c b/src/gallium/drivers/nvfx/nvfx_state_emit.c
index 0709c563505..2bc7927a697 100644
--- a/src/gallium/drivers/nvfx/nvfx_state_emit.c
+++ b/src/gallium/drivers/nvfx/nvfx_state_emit.c
@@ -98,9 +98,7 @@ nvfx_state_emit(struct nvfx_context *nvfx)
 void
 nvfx_state_relocate(struct nvfx_context *nvfx)
 {
-	struct nouveau_channel *chan = nvfx->screen->base.channel;
-	struct nvfx_state *state = &nvfx->state;
-	so_emit_reloc_markers(chan, state->hw[NVFX_STATE_FB]);
+	nvfx_framebuffer_relocate(nvfx);
 	nvfx_fragtex_relocate(nvfx);
 	nvfx_fragprog_relocate(nvfx);
 	if (nvfx->render_mode == HW)
diff --git a/src/gallium/drivers/nvfx/nvfx_state_fb.c b/src/gallium/drivers/nvfx/nvfx_state_fb.c
index c64ba27eaaa..93bc0a3feef 100644
--- a/src/gallium/drivers/nvfx/nvfx_state_fb.c
+++ b/src/gallium/drivers/nvfx/nvfx_state_fb.c
@@ -4,17 +4,14 @@
 
 
 
-static boolean
+static void
 nvfx_state_framebuffer_validate(struct nvfx_context *nvfx)
 {
 	struct pipe_framebuffer_state *fb = &nvfx->framebuffer;
 	struct nouveau_channel *chan = nvfx->screen->base.channel;
-	struct nouveau_grobj *eng3d = nvfx->screen->eng3d;
-	struct nv04_surface *rt[4], *zeta = NULL;
 	uint32_t rt_enable = 0, rt_format = 0;
 	int i, colour_format = 0, zeta_format = 0;
 	int depth_only = 0;
-	struct nouveau_stateobj *so = so_new(18, 24, 10);
 	unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM;
 	unsigned w = fb->width;
 	unsigned h = fb->height;
@@ -26,14 +23,18 @@ nvfx_state_framebuffer_validate(struct nvfx_context *nvfx)
 		assert(fb->nr_cbufs <= 4);
 
 	for (i = 0; i < fb->nr_cbufs; i++) {
-		if (colour_format) {
+		if (colour_format)
 			assert(colour_format == fb->cbufs[i]->format);
-		} else {
+		else
 			colour_format = fb->cbufs[i]->format;
-			rt_enable |= (NV34TCL_RT_ENABLE_COLOR0 << i);
-			rt[i] = (struct nv04_surface *)fb->cbufs[i];
-		}
+
+		rt_enable |= (NV34TCL_RT_ENABLE_COLOR0 << i);
+		nvfx->hw_rt[i].bo = nvfx_surface_buffer(fb->cbufs[i]);
+		nvfx->hw_rt[i].offset = fb->cbufs[i]->offset;
+		nvfx->hw_rt[i].pitch = ((struct nv04_surface *)fb->cbufs[i])->pitch;
 	}
+	for(; i < 4; ++i)
+		nvfx->hw_rt[i].bo = 0;
 
 	if (rt_enable & (NV34TCL_RT_ENABLE_COLOR1 |
 			 NV40TCL_RT_ENABLE_COLOR2 | NV40TCL_RT_ENABLE_COLOR3))
@@ -41,20 +42,24 @@ nvfx_state_framebuffer_validate(struct nvfx_context *nvfx)
 
 	if (fb->zsbuf) {
 		zeta_format = fb->zsbuf->format;
-		zeta = (struct nv04_surface *)fb->zsbuf;
+		nvfx->hw_zeta.bo = nvfx_surface_buffer(fb->zsbuf);
+		nvfx->hw_zeta.offset = fb->zsbuf->offset;
+		nvfx->hw_zeta.pitch = ((struct nv04_surface *)fb->zsbuf)->pitch;
 	}
+	else
+		nvfx->hw_zeta.bo = 0;
 
 	if (rt_enable & (NV34TCL_RT_ENABLE_COLOR0 | NV34TCL_RT_ENABLE_COLOR1 |
 		NV40TCL_RT_ENABLE_COLOR2 | NV40TCL_RT_ENABLE_COLOR3)) {
 		/* Render to at least a colour buffer */
-		if (!(rt[0]->base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR)) {
+		if (!(fb->cbufs[0]->texture->flags & NVFX_RESOURCE_FLAG_LINEAR)) {
 			assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
 			for (i = 1; i < fb->nr_cbufs; i++)
-				assert(!(rt[i]->base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR));
+				assert(!(fb->cbufs[i]->texture->flags & NVFX_RESOURCE_FLAG_LINEAR));
 
 			rt_format = NV34TCL_RT_FORMAT_TYPE_SWIZZLED |
-				(log2i(rt[0]->base.width) << NV34TCL_RT_FORMAT_LOG2_WIDTH_SHIFT) |
-				(log2i(rt[0]->base.height) << NV34TCL_RT_FORMAT_LOG2_HEIGHT_SHIFT);
+				(log2i(fb->cbufs[0]->width) << NV34TCL_RT_FORMAT_LOG2_WIDTH_SHIFT) |
+				(log2i(fb->cbufs[0]->height) << NV34TCL_RT_FORMAT_LOG2_HEIGHT_SHIFT);
 		}
 		else
 			rt_format = NV34TCL_RT_FORMAT_TYPE_LINEAR;
@@ -62,17 +67,17 @@ nvfx_state_framebuffer_validate(struct nvfx_context *nvfx)
 		depth_only = 1;
 
 		/* Render to depth buffer only */
-		if (!(zeta->base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR)) {
+		if (!(fb->zsbuf->texture->_usage & NVFX_RESOURCE_FLAG_LINEAR)) {
 			assert(!(fb->width & (fb->width - 1)) && !(fb->height & (fb->height - 1)));
 
 			rt_format = NV34TCL_RT_FORMAT_TYPE_SWIZZLED |
-				(log2i(zeta->base.width) << NV34TCL_RT_FORMAT_LOG2_WIDTH_SHIFT) |
-				(log2i(zeta->base.height) << NV34TCL_RT_FORMAT_LOG2_HEIGHT_SHIFT);
+				(log2i(fb->zsbuf->width) << NV34TCL_RT_FORMAT_LOG2_WIDTH_SHIFT) |
+				(log2i(fb->zsbuf->height) << NV34TCL_RT_FORMAT_LOG2_HEIGHT_SHIFT);
 		}
 		else
 			rt_format = NV34TCL_RT_FORMAT_TYPE_LINEAR;
 	} else {
-		return FALSE;
+		return;
 	}
 
 	switch (colour_format) {
@@ -108,115 +113,140 @@ nvfx_state_framebuffer_validate(struct nvfx_context *nvfx)
 	if ((!nvfx->is_nv4x) && colour_bits > zeta_bits) {
 		/* TODO: does this limitation really exist?
 		   TODO: can it be worked around somehow? */
-		return FALSE;
+		assert(0);
 	}
 
 	if ((rt_enable & NV34TCL_RT_ENABLE_COLOR0)
 		|| ((!nvfx->is_nv4x) && depth_only)) {
-		struct nv04_surface *rt0 = (depth_only ? zeta : rt[0]);
+		struct nvfx_render_target *rt0 = (depth_only ? &nvfx->hw_zeta : &nvfx->hw_rt[0]);
 		uint32_t pitch = rt0->pitch;
 
 		if(!nvfx->is_nv4x)
 		{
-			if (zeta) {
-				pitch |= (zeta->pitch << 16);
+			if (nvfx->hw_zeta.bo) {
+				pitch |= (nvfx->hw_zeta.pitch << 16);
 			} else {
 				pitch |= (pitch << 16);
 			}
 		}
 
-		so_method(so, eng3d, NV34TCL_DMA_COLOR0, 1);
-		so_reloc (so, nvfx_surface_buffer(&rt0->base), 0,
+		OUT_RING(chan, RING_3D(NV34TCL_DMA_COLOR0, 1));
+		OUT_RELOC(chan, rt0->bo, 0,
 			      rt_flags | NOUVEAU_BO_OR,
 			      chan->vram->handle, chan->gart->handle);
-		so_method(so, eng3d, NV34TCL_COLOR0_PITCH, 2);
-		so_data  (so, pitch);
-		so_reloc (so, nvfx_surface_buffer(&rt[0]->base),
-			      rt0->base.offset, rt_flags | NOUVEAU_BO_LOW,
+		OUT_RING(chan, RING_3D(NV34TCL_COLOR0_PITCH, 2));
+		OUT_RING(chan, pitch);
+		OUT_RELOC(chan, rt0->bo,
+			      rt0->offset, rt_flags | NOUVEAU_BO_LOW,
 			      0, 0);
 	}
 
 	if (rt_enable & NV34TCL_RT_ENABLE_COLOR1) {
-		so_method(so, eng3d, NV34TCL_DMA_COLOR1, 1);
-		so_reloc (so, nvfx_surface_buffer(&rt[1]->base), 0,
+		OUT_RING(chan, RING_3D(NV34TCL_DMA_COLOR1, 1));
+		OUT_RELOC(chan, nvfx->hw_rt[1].bo, 0,
 			      rt_flags | NOUVEAU_BO_OR,
 			      chan->vram->handle, chan->gart->handle);
-		so_method(so, eng3d, NV34TCL_COLOR1_OFFSET, 2);
-		so_reloc (so, nvfx_surface_buffer(&rt[1]->base),
-			      rt[1]->base.offset, rt_flags | NOUVEAU_BO_LOW,
+		OUT_RING(chan, RING_3D(NV34TCL_COLOR1_OFFSET, 2));
+		OUT_RELOC(chan, nvfx->hw_rt[1].bo,
+				nvfx->hw_rt[1].offset, rt_flags | NOUVEAU_BO_LOW,
 			      0, 0);
-		so_data  (so, rt[1]->pitch);
+		OUT_RING(chan, nvfx->hw_rt[1].pitch);
 	}
 
 	if(nvfx->is_nv4x)
 	{
 		if (rt_enable & NV40TCL_RT_ENABLE_COLOR2) {
-			so_method(so, eng3d, NV40TCL_DMA_COLOR2, 1);
-			so_reloc (so, nvfx_surface_buffer(&rt[2]->base), 0,
+			OUT_RING(chan, RING_3D(NV40TCL_DMA_COLOR2, 1));
+			OUT_RELOC(chan, nvfx->hw_rt[2].bo, 0,
 				      rt_flags | NOUVEAU_BO_OR,
 				      chan->vram->handle, chan->gart->handle);
-			so_method(so, eng3d, NV40TCL_COLOR2_OFFSET, 1);
-			so_reloc (so, nvfx_surface_buffer(&rt[2]->base),
-				      rt[2]->base.offset, rt_flags | NOUVEAU_BO_LOW,
+			OUT_RING(chan, RING_3D(NV40TCL_COLOR2_OFFSET, 1));
+			OUT_RELOC(chan, nvfx->hw_rt[2].bo,
+				      nvfx->hw_rt[2].offset, rt_flags | NOUVEAU_BO_LOW,
 				      0, 0);
-			so_method(so, eng3d, NV40TCL_COLOR2_PITCH, 1);
-			so_data  (so, rt[2]->pitch);
+			OUT_RING(chan, RING_3D(NV40TCL_COLOR2_PITCH, 1));
+			OUT_RING(chan, nvfx->hw_rt[2].pitch);
 		}
 
 		if (rt_enable & NV40TCL_RT_ENABLE_COLOR3) {
-			so_method(so, eng3d, NV40TCL_DMA_COLOR3, 1);
-			so_reloc (so, nvfx_surface_buffer(&rt[3]->base), 0,
+			OUT_RING(chan, RING_3D(NV40TCL_DMA_COLOR3, 1));
+			OUT_RELOC(chan, nvfx->hw_rt[3].bo, 0,
 				      rt_flags | NOUVEAU_BO_OR,
 				      chan->vram->handle, chan->gart->handle);
-			so_method(so, eng3d, NV40TCL_COLOR3_OFFSET, 1);
-			so_reloc (so, nvfx_surface_buffer(&rt[3]->base),
-				      rt[3]->base.offset, rt_flags | NOUVEAU_BO_LOW,
+			OUT_RING(chan, RING_3D(NV40TCL_COLOR3_OFFSET, 1));
+			OUT_RELOC(chan, nvfx->hw_rt[3].bo,
+					nvfx->hw_rt[3].offset, rt_flags | NOUVEAU_BO_LOW,
 				      0, 0);
-			so_method(so, eng3d, NV40TCL_COLOR3_PITCH, 1);
-			so_data  (so, rt[3]->pitch);
+			OUT_RING(chan, RING_3D(NV40TCL_COLOR3_PITCH, 1));
+			OUT_RING(chan, nvfx->hw_rt[3].pitch);
 		}
 	}
 
 	if (zeta_format) {
-		so_method(so, eng3d, NV34TCL_DMA_ZETA, 1);
-		so_reloc (so, nvfx_surface_buffer(&zeta->base), 0,
+		OUT_RING(chan, RING_3D(NV34TCL_DMA_ZETA, 1));
+		OUT_RELOC(chan, nvfx->hw_zeta.bo, 0,
 			      rt_flags | NOUVEAU_BO_OR,
 			      chan->vram->handle, chan->gart->handle);
-		so_method(so, eng3d, NV34TCL_ZETA_OFFSET, 1);
+		OUT_RING(chan, RING_3D(NV34TCL_ZETA_OFFSET, 1));
 		/* TODO: reverse engineer LMA */
-		so_reloc (so, nvfx_surface_buffer(&zeta->base),
-			      zeta->base.offset, rt_flags | NOUVEAU_BO_LOW, 0, 0);
+		OUT_RELOC(chan, nvfx->hw_zeta.bo,
+			     nvfx->hw_zeta.offset, rt_flags | NOUVEAU_BO_LOW, 0, 0);
 	        if(nvfx->is_nv4x) {
-			so_method(so, eng3d, NV40TCL_ZETA_PITCH, 1);
-			so_data  (so, zeta->pitch);
+			OUT_RING(chan, RING_3D(NV40TCL_ZETA_PITCH, 1));
+			OUT_RING(chan, nvfx->hw_zeta.pitch);
 		}
 	}
 
-	so_method(so, eng3d, NV34TCL_RT_ENABLE, 1);
-	so_data  (so, rt_enable);
-	so_method(so, eng3d, NV34TCL_RT_HORIZ, 3);
-	so_data  (so, (w << 16) | 0);
-	so_data  (so, (h << 16) | 0);
-	so_data  (so, rt_format);
-	so_method(so, eng3d, NV34TCL_VIEWPORT_HORIZ, 2);
-	so_data  (so, (w << 16) | 0);
-	so_data  (so, (h << 16) | 0);
-	so_method(so, eng3d, NV34TCL_VIEWPORT_CLIP_HORIZ(0), 2);
-	so_data  (so, ((w - 1) << 16) | 0);
-	so_data  (so, ((h - 1) << 16) | 0);
-	so_method(so, eng3d, 0x1d88, 1);
-	so_data  (so, (1 << 12) | h);
+	OUT_RING(chan, RING_3D(NV34TCL_RT_ENABLE, 1));
+	OUT_RING(chan, rt_enable);
+	OUT_RING(chan, RING_3D(NV34TCL_RT_HORIZ, 3));
+	OUT_RING(chan, (w << 16) | 0);
+	OUT_RING(chan, (h << 16) | 0);
+	OUT_RING(chan, rt_format);
+	OUT_RING(chan, RING_3D(NV34TCL_VIEWPORT_HORIZ, 2));
+	OUT_RING(chan, (w << 16) | 0);
+	OUT_RING(chan, (h << 16) | 0);
+	OUT_RING(chan, RING_3D(NV34TCL_VIEWPORT_CLIP_HORIZ(0), 2));
+	OUT_RING(chan, ((w - 1) << 16) | 0);
+	OUT_RING(chan, ((h - 1) << 16) | 0);
+	OUT_RING(chan, RING_3D(0x1d88, 1));
+	OUT_RING(chan, (1 << 12) | h);
 
 	if(!nvfx->is_nv4x) {
 		/* Wonder why this is needed, context should all be set to zero on init */
 		/* TODO: we can most likely remove this, after putting it in context init */
-		so_method(so, eng3d, NV34TCL_VIEWPORT_TX_ORIGIN, 1);
-		so_data  (so, 0);
+		OUT_RING(chan, RING_3D(NV34TCL_VIEWPORT_TX_ORIGIN, 1));
+		OUT_RING(chan, 0);
 	}
+}
+
+void
+nvfx_framebuffer_relocate(struct nvfx_context *nvfx)
+{
+	struct nouveau_channel *chan = nvfx->screen->base.channel;
+	unsigned rt_flags = NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM;
+	rt_flags |= NOUVEAU_BO_DUMMY;
+	MARK_RING(chan, 20, 20);
+
+#define DO_(var, pfx, name) \
+	if(var.bo) { \
+		OUT_RELOC(chan, var.bo, RING_3D(pfx##TCL_DMA_##name, 1), rt_flags, 0, 0); \
+		OUT_RELOC(chan, var.bo, 0, \
+			rt_flags | NOUVEAU_BO_OR, \
+			chan->vram->handle, chan->gart->handle); \
+		OUT_RELOC(chan, var.bo, RING_3D(pfx##TCL_##name##_OFFSET, 1), rt_flags, 0, 0); \
+		OUT_RELOC(chan, var.bo, \
+			var.offset, rt_flags | NOUVEAU_BO_LOW, \
+			0, 0); \
+	}
+
+#define DO(pfx, num) DO_(nvfx->hw_rt[num], pfx, COLOR##num)
+	DO(NV34, 0);
+	DO(NV34, 1);
+	DO(NV40, 2);
+	DO(NV40, 3);
 
-	so_ref(so, &nvfx->state.hw[NVFX_STATE_FB]);
-	so_ref(NULL, &so);
-	return TRUE;
+	DO_(nvfx->hw_zeta, NV34, ZETA);
 }
 
 struct nvfx_state_entry nvfx_state_framebuffer = {