From ab3d55e2e3578db8deba84dcf47a024071486bd8 Mon Sep 17 00:00:00 2001
From: Ben Skeggs <skeggsb@gmail.com>
Date: Thu, 12 Jun 2008 13:11:41 +1000
Subject: [PATCH] nv50: more efficient const upload + fixes (fp/* works now!)

---
 src/gallium/drivers/nv50/nv50_program.c | 48 +++++++++++++++++++++------------
 src/gallium/drivers/nv50/nv50_screen.c  |  4 +--
 2 files changed, 33 insertions(+), 19 deletions(-)

diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index cbbecaf..4c41e5a 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -1420,20 +1420,35 @@ out_cleanup:
 static void
 nv50_program_validate(struct nv50_context *nv50, struct nv50_program *p)
 {
-	struct nv50_program_exec *e;
-
 	if (nv50_program_tx(p) == FALSE)
 		assert(0);
 	p->translated = TRUE;
 }
 
 static void
+nv50_program_upload_data(struct nv50_context *nv50, float *map,
+			 unsigned start, unsigned count)
+{
+	while (count) {
+		unsigned nr = count > 2047 ? 2047 : count;
+
+		BEGIN_RING(tesla, 0x00000f00, 1);
+		OUT_RING  ((NV50_CB_PMISC << 0) | (start << 8));
+		BEGIN_RING(tesla, 0x40000f04, nr);
+		OUT_RINGp (map, nr);
+
+		map += nr;
+		start += nr;
+		count -= nr;
+	}
+}
+
+static void
 nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p)
 {
 	struct nouveau_winsys *nvws = nv50->screen->nvws;
 	struct pipe_winsys *ws = nv50->pipe.winsys;
 	unsigned nr = p->param_nr + p->immd_nr;
-	int i;
 
 	if (!p->data && nr) {
 		struct nouveau_resource *heap = nv50->screen->vp_data_heap;
@@ -1452,20 +1467,15 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p)
 	if (p->param_nr) {
 		float *map = ws->buffer_map(ws, nv50->constbuf[p->type],
 					    PIPE_BUFFER_USAGE_CPU_READ);
-		for (i = 0; i < p->param_nr; i++) {
-			BEGIN_RING(tesla, 0x0f00, 2);
-			OUT_RING  ((NV50_CB_PMISC << 0) |
-				   ((p->data->start + i) << 8));
-			OUT_RING  (fui(map[i]));
-		}
+		nv50_program_upload_data(nv50, map, p->data->start,
+					 p->param_nr);
 		ws->buffer_unmap(ws, nv50->constbuf[p->type]);
 	}
 
-	for (i = 0; i < p->immd_nr; i++) {
-		BEGIN_RING(tesla, 0x0f00, 2);
-		OUT_RING  ((NV50_CB_PMISC << 0) |
-			   ((p->data_start + p->param_nr + i) << 8));
-		OUT_RING  (fui(p->immd[i]));
+	if (p->immd_nr) {
+		nv50_program_upload_data(nv50, p->immd,
+					 p->data->start + p->param_nr,
+					 p->immd_nr);
 	}
 }
 
@@ -1481,11 +1491,15 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 
 	if (p->data && p->data->start != p->data_start) {
 		for (e = p->exec_head; e; e = e->next) {
+			unsigned ei, ci;
+
 			if (e->param.index < 0)
 				continue;
-			e->inst[e->param.shift / 32] &= ~e->param.mask;
-			e->inst[e->param.shift / 32] |= (e->param.index <<
-							 e->param.shift);
+			ei = e->param.shift >> 5;
+			ci = e->param.index + p->data->start;
+
+			e->inst[ei] &= ~e->param.mask;
+			e->inst[ei] |= (ci << e->param.shift);
 		}
 
 		p->data_start = p->data->start;
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index 2417bf8..5f10fe2 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -229,8 +229,8 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_winsys *nvws)
 	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)) {
+	screen->constbuf = ws->buffer_create(ws, 0, 0, 128 * 4 * 4);
+	if (nvws->res_init(&screen->vp_data_heap, 0, 128)) {
 		NOUVEAU_ERR("Error initialising constant buffer\n");
 		nv50_screen_destroy(&screen->pipe);
 		return NULL;
-- 
2.7.4