nv50: use "real" constbufs for shaders + tcb uploads

7 files changed, 90 insertions, 45 deletions

diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index d4d716b94bf..c4a8a4c064e 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -23,6 +23,14 @@
 #define NOUVEAU_MSG(fmt, args...) \
 	fprintf(stderr, "nouveau: "fmt, ##args);
 
+/* Constant buffer assignment */
+#define NV50_CB_PMISC		0
+#define NV50_CB_PVP		1
+#define NV50_CB_PFP		2
+#define NV50_CB_PGP		3
+#define NV50_CB_TIC		4
+#define NV50_CB_TSC		5
+
 #define NV50_NEW_BLEND		(1 << 0)
 #define NV50_NEW_ZSA		(1 << 1)
 #define NV50_NEW_BLEND_COLOUR	(1 << 2)
@@ -32,8 +40,10 @@
 #define NV50_NEW_RASTERIZER	(1 << 6)
 #define NV50_NEW_FRAMEBUFFER	(1 << 7)
 #define NV50_NEW_VERTPROG	(1 << 8)
-#define NV50_NEW_FRAGPROG	(1 << 9)
-#define NV50_NEW_ARRAYS		(1 << 10)
+#define NV50_NEW_VERTPROG_CB	(1 << 9)
+#define NV50_NEW_FRAGPROG	(1 << 10)
+#define NV50_NEW_FRAGPROG_CB	(1 << 11)
+#define NV50_NEW_ARRAYS		(1 << 12)
 
 struct nv50_blend_stateobj {
 	struct pipe_blend_state pipe;
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 30953b7d8af..0a436469238 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -179,6 +179,10 @@ emit(struct nv50_pc *pc, unsigned op, struct nv50_reg *dst,
 
 		if (src1) {
 			if (src1->type == P_CONST || src1->type == P_IMMD) {
+				if (src1->type == P_IMMD)
+					inst[1] |= (NV50_CB_PMISC << 22);
+				else
+					inst[1] |= (NV50_CB_PVP << 22);
 				inst[0] |= 0x00800000; /* src1 is const */
 				/*XXX: does src1 come from "src2" now? */
 				alloc_reg(pc, src1);
@@ -196,6 +200,10 @@ emit(struct nv50_pc *pc, unsigned op, struct nv50_reg *dst,
 
 		if (src2) {
 			if (src2->type == P_CONST || src2->type == P_IMMD) {
+				if (src2->type == P_IMMD)
+					inst[1] |= (NV50_CB_PMISC << 22);
+				else
+					inst[1] |= (NV50_CB_PVP << 22);
 				inst[0] |= 0x01000000; /* src2 is const */
 				inst[1] |= (src2->hw << 14);
 			} else {
@@ -526,7 +534,7 @@ nv50_program_tx_prep(struct nv50_pc *pc)
 	}
 
 	if (pc->immd_nr) {
-		int rid = pc->param_nr * 4;
+		int rid = 0;
 
 		pc->immd = calloc(pc->immd_nr * 4, sizeof(struct nv50_reg));
 		if (!pc->immd)
@@ -581,7 +589,6 @@ nv50_program_tx(struct nv50_program *p)
 		}
 	}
 
-	p->param_nr = pc->param_nr * 4;
 	p->immd_nr = pc->immd_nr * 4;
 	p->immd = pc->immd_buf;
 
@@ -654,23 +661,9 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	memcpy(map, p->insns, p->insns_nr * 4);
 	ws->buffer_unmap(ws, p->buffer);
 
-	if (p->param_nr) {
-		float *cb;
-
-		cb = ws->buffer_map(ws, nv50->constbuf[PIPE_SHADER_VERTEX],
-				    PIPE_BUFFER_USAGE_CPU_READ);
-		for (i = 0; i < p->param_nr; i++) {
-			BEGIN_RING(tesla, 0x0f00, 2);
-			OUT_RING  (i << 8);
-			OUT_RING  (fui(cb[i]));
-		}
-		ws->buffer_unmap(ws, nv50->constbuf[PIPE_SHADER_VERTEX]);
-	}
-
-
 	for (i = 0; i < p->immd_nr; i++) {
 		BEGIN_RING(tesla, 0x0f00, 2);
-		OUT_RING  ((p->param_nr + i) << 8);
+		OUT_RING  ((NV50_CB_PMISC << 16) | (i << 8));
 		OUT_RING  (fui(p->immd[i]));
 	}
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index fc3eeed9133..6c0810a9cfe 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -203,14 +203,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 		return NULL;
 	}
 
-	/* Static constant buffer */
-	screen->constbuf = ws->buffer_create(ws, 0, 0, 256 * 4 * 4);
-	if (nvws->res_init(&screen->vp_data_heap, 0, 256)) {
-		NOUVEAU_ERR("Error initialising constant buffer\n");
-		nv50_screen_destroy(&screen->pipe);
-		return NULL;
-	}
-
 	/* Static tesla init */
 	so = so_new(256, 20);
 
@@ -245,37 +237,56 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 	so_method(so, screen->tesla, 0x16b8, 1);
 	so_data  (so, 8);
 
+	/* Shared constant buffer */
+	screen->constbuf = ws->buffer_create(ws, 0, 0, 256 * 4 * 4);
+	if (nvws->res_init(&screen->vp_data_heap, 0, 256)) {
+		NOUVEAU_ERR("Error initialising constant buffer\n");
+		nv50_screen_destroy(&screen->pipe);
+		return NULL;
+	}
+
 	so_method(so, screen->tesla, 0x1280, 3);
 	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
 	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-	so_data  (so, 0x00001000);
+	so_data  (so, (NV50_CB_PMISC << 16) | 0x00001000);
+
+	/* Texture sampler/image unit setup - we abuse the constant buffer
+	 * upload mechanism for the moment to upload data to the tex config
+	 * blocks.  At some point we *may* want to go the NVIDIA way of doing
+	 * things?
+	 */
+	screen->tic = ws->buffer_create(ws, 0, 0, 32 * 8 * 4);
 	so_method(so, screen->tesla, 0x1280, 3);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-	so_data  (so, 0x00014000);
-	so_method(so, screen->tesla, 0x1280, 3);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_data  (so, (NV50_CB_TIC << 16) | 0x0800);
+	so_method(so, screen->tesla, 0x1574, 3);
+	so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tic, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-	so_data  (so, 0x00024000);
+	so_data  (so, 0x00000800);
+
+	screen->tsc = ws->buffer_create(ws, 0, 0, 32 * 8 * 4);
 	so_method(so, screen->tesla, 0x1280, 3);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-	so_data  (so, 0x00034000);
-	so_method(so, screen->tesla, 0x1280, 3);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_data  (so, (NV50_CB_TSC << 16) | 0x0800);
+	so_method(so, screen->tesla, 0x155c, 3);
+	so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
-	so_reloc (so, screen->constbuf, 0, NOUVEAU_BO_VRAM |
+	so_reloc (so, screen->tsc, 0, NOUVEAU_BO_VRAM |
 		  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
-	so_data  (so, 0x00040100);
+	so_data  (so, 0x00000800);
+
 
+	/* Vertex array limits - max them out */
 	for (i = 0; i < 16; i++) {
 		so_method(so, screen->tesla, 0x1080 + (i * 8), 2);
 		so_data  (so, 0x000000ff);
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index d63dd485085..5acb5003ba4 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -15,6 +15,9 @@ struct nv50_screen {
 
 	struct pipe_buffer *constbuf;
 	struct nouveau_resource *vp_data_heap;
+
+	struct pipe_buffer *tic;
+	struct pipe_buffer *tsc;
 };
 
 static INLINE struct nv50_screen *
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index fd10a383782..ba3d04cede3 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -406,11 +406,11 @@ nv50_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index,
 
 	if (shader == PIPE_SHADER_VERTEX) {
 		nv50->constbuf[PIPE_SHADER_VERTEX] = buf->buffer;
-		nv50->dirty |= NV50_NEW_VERTPROG;
+		nv50->dirty |= NV50_NEW_VERTPROG_CB;
 	} else
 	if (shader == PIPE_SHADER_FRAGMENT) {
 		nv50->constbuf[PIPE_SHADER_FRAGMENT] = buf->buffer;
-		nv50->dirty |= NV50_NEW_FRAGPROG;
+		nv50->dirty |= NV50_NEW_FRAGPROG_CB;
 	}
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_state.h b/src/gallium/drivers/nv50/nv50_state.h
index be0c75ad6ea..9e3876871bd 100644
--- a/src/gallium/drivers/nv50/nv50_state.h
+++ b/src/gallium/drivers/nv50/nv50_state.h
@@ -20,7 +20,6 @@ struct nv50_program {
 
 	struct pipe_buffer *buffer;
 
-	unsigned param_nr;
 	float *immd;
 	unsigned immd_nr;
 
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 4a548378b71..05395c6df7c 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -168,6 +168,35 @@ nv50_state_validate(struct nv50_context *nv50)
 		so_ref(NULL, &so);
 	}
 
+	if (nv50->dirty & NV50_NEW_VERTPROG_CB) {
+		so = so_new(4, 2);
+		so_method(so, tesla, 0x1280, 3);
+		so_reloc (so, nv50->constbuf[PIPE_SHADER_VERTEX], 0,
+			  NOUVEAU_BO_HIGH | NOUVEAU_BO_RD | NOUVEAU_BO_VRAM,
+			  0, 0);
+		so_reloc (so, nv50->constbuf[PIPE_SHADER_VERTEX], 0,
+			  NOUVEAU_BO_LOW | NOUVEAU_BO_RD | NOUVEAU_BO_VRAM,
+			  0, 0);
+		so_data  (so, (NV50_CB_PVP << 16) | 0x1000);
+		so_emit(nvws, so);
+		so_ref(NULL, &so);
+	}
+
+	if (nv50->dirty & NV50_NEW_FRAGPROG_CB) {
+		so = so_new(4, 2);
+		so_method(so, tesla, 0x1280, 3);
+		so_reloc (so, nv50->constbuf[PIPE_SHADER_FRAGMENT], 0,
+			  NOUVEAU_BO_HIGH | NOUVEAU_BO_RD | NOUVEAU_BO_VRAM,
+			  0, 0);
+		so_reloc (so, nv50->constbuf[PIPE_SHADER_FRAGMENT], 0,
+			  NOUVEAU_BO_LOW | NOUVEAU_BO_RD | NOUVEAU_BO_VRAM,
+			  0, 0);
+		so_data  (so, (NV50_CB_PFP << 16) | 0x1000);
+		so_emit(nvws, so);
+		so_ref(NULL, &so);
+	}
+
+
 	if (nv50->dirty & NV50_NEW_ARRAYS)
 		nv50_vbo_validate(nv50);