nv50: make instanced drawing work with edge flags

And fix some obvious mistakes introduced in the
previous instancing commit.

1 files changed, 162 insertions, 45 deletions

diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index 32386294616..b9cf0754b77 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -40,6 +40,8 @@ nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
 static boolean
 nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
 
+#define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
+
 static INLINE unsigned
 nv50_prim(unsigned mode)
 {
@@ -171,7 +173,7 @@ nv50_upload_vtxbuf(struct nv50_context *nv50, unsigned i)
 	struct pipe_buffer *buf = nscreen->strm_vbuf[i];
 	struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
 	uint8_t *src;
-	unsigned size = MAX2(vb->buffer->size, 4096);
+	unsigned size = align(vb->buffer->size, 4096);
 
 	if (buf && buf->size < size)
 		pipe_buffer_reference(&nscreen->strm_vbuf[i], NULL);
@@ -215,6 +217,74 @@ nv50_upload_user_vbufs(struct nv50_context *nv50)
 	}
 }
 
+static void
+nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	float v[4];
+
+	util_format_read_4f(nv50->vtxelt[i].src_format,
+			    v, 0, data, 0, 0, 0, 1, 1);
+
+	switch (nv50->vtxelt[i].nr_components) {
+	case 4:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_4F_X(i), 4);
+		OUT_RING  (chan, v[0]);
+		OUT_RING  (chan, v[1]);
+		OUT_RING  (chan, v[2]);
+		OUT_RING  (chan, v[3]);
+		break;
+	case 3:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_3F_X(i), 4);
+		OUT_RING  (chan, v[0]);
+		OUT_RING  (chan, v[1]);
+		OUT_RING  (chan, v[2]);
+		break;
+	case 2:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_2F_X(i), 4);
+		OUT_RING  (chan, v[0]);
+		OUT_RING  (chan, v[1]);
+		break;
+	case 1:
+		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_1F(i), 4);
+		OUT_RING  (chan, v[0]);
+		break;
+	default:
+		assert(0);
+		break;
+	}
+}
+
+static unsigned
+init_per_instance_arrays_immd(struct nv50_context *nv50,
+			      unsigned startInstance,
+			      unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_bo *bo;
+	unsigned i, b, count = 0;
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		++count;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+
+		pos[i] = nv50->vtxelt[i].src_offset +
+			nv50->vtxbuf[b].buffer_offset +
+			startInstance * nv50->vtxbuf[b].stride;
+		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
+
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+		if (!bo->map)
+			nouveau_bo_map(bo, NOUVEAU_BO_RD);
+
+		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+	}
+
+	return count;
+}
+
 static unsigned
 init_per_instance_arrays(struct nv50_context *nv50,
 			 unsigned startInstance,
@@ -227,6 +297,10 @@ init_per_instance_arrays(struct nv50_context *nv50,
 	unsigned i, b, count = 0;
 	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
+	if (nv50->vbo_fifo)
+		return init_per_instance_arrays_immd(nv50, startInstance,
+						     pos, step);
+
 	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
 
 	for (i = 0; i < nv50->vtxelt_nr; ++i) {
@@ -248,11 +322,11 @@ init_per_instance_arrays(struct nv50_context *nv50,
 		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
 
 		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 	}
 
-	if (count) {
+	if (count && startInstance) {
 		so_ref (so, &nv50->state.instbuf); /* for flush notify */
 		so_emit(chan, nv50->state.instbuf);
 	}
@@ -262,6 +336,28 @@ init_per_instance_arrays(struct nv50_context *nv50,
 }
 
 static void
+step_per_instance_arrays_immd(struct nv50_context *nv50,
+			      unsigned pos[16], unsigned step[16])
+{
+	struct nouveau_bo *bo;
+	unsigned i, b;
+
+	for (i = 0; i < nv50->vtxelt_nr; ++i) {
+		if (!nv50->vtxelt[i].instance_divisor)
+			continue;
+		if (++step[i] != nv50->vtxelt[i].instance_divisor)
+			continue;
+		b = nv50->vtxelt[i].vertex_buffer_index;
+		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+
+		step[i] = 0;
+		pos[i] += nv50->vtxbuf[b].stride;
+
+		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+	}
+}
+
+static void
 step_per_instance_arrays(struct nv50_context *nv50,
 			 unsigned pos[16], unsigned step[16])
 {
@@ -272,6 +368,11 @@ step_per_instance_arrays(struct nv50_context *nv50,
 	unsigned i, b;
 	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
 
+	if (nv50->vbo_fifo) {
+		step_per_instance_arrays_immd(nv50, pos, step);
+		return;
+	}
+
 	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
 
 	for (i = 0; i < nv50->vtxelt_nr; ++i) {
@@ -287,8 +388,8 @@ step_per_instance_arrays(struct nv50_context *nv50,
 		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
 
 		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
+		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 	}
 
 	so_ref (so, &nv50->state.instbuf); /* for flush notify */
@@ -297,6 +398,16 @@ step_per_instance_arrays(struct nv50_context *nv50,
 	so_emit(chan, nv50->state.instbuf);
 }
 
+static INLINE void
+nv50_unmap_vbufs(struct nv50_context *nv50)
+{
+        unsigned i;
+
+        for (i = 0; i < nv50->vtxbuf_nr; ++i)
+                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
+                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
+}
+
 void
 nv50_draw_arrays_instanced(struct pipe_context *pipe,
 			   unsigned mode, unsigned start, unsigned count,
@@ -308,7 +419,8 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
 	unsigned i, nz_divisors;
 	unsigned step[16], pos[16];
 
-	nv50_upload_user_vbufs(nv50);
+	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
+		nv50_upload_user_vbufs(nv50);
 
 	nv50_state_validate(nv50);
 
@@ -320,9 +432,14 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
 
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
 	OUT_RING  (chan, nv50_prim(mode));
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-	OUT_RING  (chan, start);
-	OUT_RING  (chan, count);
+
+	if (nv50->vbo_fifo)
+		nv50_push_arrays(nv50, start, count);
+	else {
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
+		OUT_RING  (chan, start);
+		OUT_RING  (chan, count);
+	}
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 	OUT_RING  (chan, 0);
 
@@ -332,12 +449,18 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
 
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
 		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-		OUT_RING  (chan, start);
-		OUT_RING  (chan, count);
+
+		if (nv50->vbo_fifo)
+			nv50_push_arrays(nv50, start, count);
+		else {
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
+			OUT_RING  (chan, start);
+			OUT_RING  (chan, count);
+		}
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
 	}
+	nv50_unmap_vbufs(nv50);
 
 	so_ref(NULL, &nv50->state.instbuf);
 }
@@ -372,6 +495,8 @@ nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 	OUT_RING  (chan, 0);
 
+	nv50_unmap_vbufs(nv50);
+
         /* XXX: not sure what to do if ret != TRUE: flush and retry?
          */
         assert(ret);
@@ -400,7 +525,7 @@ nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -433,7 +558,7 @@ nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
 		unsigned nr = count > 2046 ? 2046 : count;
 		int i;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x40000000, nr >> 1);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
 		for (i = 0; i < nr; i += 2)
 			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
 
@@ -458,7 +583,7 @@ nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
 	while (count) {
 		unsigned nr = count > 2047 ? 2047 : count;
 
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x40000000, nr);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
 		OUT_RINGp (chan, map, nr);
 
 		count -= nr;
@@ -502,7 +627,8 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 
 	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
 
-	nv50_upload_user_vbufs(nv50);
+	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
+		nv50_upload_user_vbufs(nv50);
 
 	nv50_state_validate(nv50);
 
@@ -513,7 +639,7 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 	OUT_RING  (chan, startInstance);
 
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
+	OUT_RING  (chan, nv50_prim(mode));
 
 	nv50_draw_elements_inline(nv50, map, indexSize, start, count);
 
@@ -532,6 +658,7 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
 	}
+	nv50_unmap_vbufs(nv50);
 
 	so_ref(NULL, &nv50->state.instbuf);
 }
@@ -564,6 +691,8 @@ nv50_draw_elements(struct pipe_context *pipe,
 	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 	OUT_RING  (chan, 0);
 
+	nv50_unmap_vbufs(nv50);
+
 	pipe_buffer_unmap(pscreen, indexBuffer);
 }
 
@@ -644,7 +773,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
 
-	if (nv50->vertprog->cfg.edgeflag_in < 16)
+	if (NV50_USING_LOATHED_EDGEFLAG(nv50))
 		nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
 
 	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
@@ -672,14 +801,16 @@ nv50_vbo_validate(struct nv50_context *nv50)
 			nv50->vbo_fifo &= ~(1 << i);
 			continue;
 		}
-		so_data(vtxfmt, hw | i);
 
 		if (nv50->vbo_fifo) {
+			so_data  (vtxfmt, hw |
+				  (ve->instance_divisor ? (1 << 4) : i));
 			so_method(vtxbuf, tesla,
 				  NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 			so_data  (vtxbuf, 0);
 			continue;
 		}
+		so_data(vtxfmt, hw | i);
 
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3);
 		so_data  (vtxbuf, 0x20000000 |
@@ -721,7 +852,7 @@ typedef void (*pfn_push)(struct nouveau_channel *, void *);
 struct nv50_vbo_emitctx
 {
 	pfn_push push[16];
-	void *map[16];
+	uint8_t *map[16];
 	unsigned stride[16];
 	unsigned nr_ve;
 	unsigned vtx_dwords;
@@ -759,19 +890,18 @@ nv50_map_vbufs(struct nv50_context *nv50)
 
 	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
 		struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-		unsigned size, delta;
+		unsigned size = vb->stride * (vb->max_index + 1) + 16;
 
 		if (nouveau_bo(vb->buffer)->map)
 			continue;
 
-		size = vb->stride * (vb->max_index + 1);
-		delta = vb->buffer_offset;
-
+		size = vb->stride * (vb->max_index + 1) + 16;
+		size = MIN2(size, vb->buffer->size);
 		if (!size)
-			size = vb->buffer->size - vb->buffer_offset;
+			size = vb->buffer->size;
 
 		if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
-					 delta, size, NOUVEAU_BO_RD))
+					 0, size, NOUVEAU_BO_RD))
 			break;
 	}
 
@@ -782,16 +912,6 @@ nv50_map_vbufs(struct nv50_context *nv50)
 	return FALSE;
 }
 
-static INLINE void
-nv50_unmap_vbufs(struct nv50_context *nv50)
-{
-        unsigned i;
-
-        for (i = 0; i < nv50->vtxbuf_nr; ++i)
-                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
-                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-}
-
 static void
 emit_b32_1(struct nouveau_channel *chan, void *data)
 {
@@ -886,12 +1006,13 @@ emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
 
 		ve = &nv50->vtxelt[i];
 		vb = &nv50->vtxbuf[ve->vertex_buffer_index];
-		if (!(nv50->vbo_fifo & (1 << i)))
+		if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
 			continue;
 		n = emit->nr_ve++;
 
 		emit->stride[n] = vb->stride;
-		emit->map[n] = nouveau_bo(vb->buffer)->map +
+		emit->map[n] = (uint8_t *)nouveau_bo(vb->buffer)->map +
+			vb->buffer_offset +
 			(start * vb->stride + ve->src_offset);
 
 		desc = util_format_description(ve->src_format);
@@ -981,13 +1102,12 @@ nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx_next(chan, &emit);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -1008,13 +1128,12 @@ nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -1035,13 +1154,12 @@ nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }
@@ -1062,13 +1180,12 @@ nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
 
 		set_edgeflag(chan, tesla, &emit, *map);
 
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
+		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
 		for (i = 0; i < nr; ++i)
 			emit_vtx(chan, &emit, *map++);
 
 		count -= nr;
 	}
-	nv50_unmap_vbufs(nv50);
 
 	return TRUE;
 }