X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fnv50%2Fnv50_vbo.c;h=50472868063f5bfa32c983300e8a5f61be2d98d8;hb=275c4bd3643d773210780cb8d578ca84f2604684;hp=db54380241f681afc474d563dec1a60e00cabd55;hpb=4bced42341d13303ae023957d3001a640cf7ea2d;p=mesa.git

diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index db54380241f..50472868063 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -22,58 +22,32 @@
 
 #include "pipe/p_context.h"
 #include "pipe/p_state.h"
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
 
+#include "nouveau/nouveau_util.h"
 #include "nv50_context.h"
 
-static boolean
-nv50_push_elements_u08(struct nv50_context *, uint8_t *, unsigned);
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *, uint16_t *, unsigned);
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
-
-static boolean
-nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
-
-static INLINE unsigned
-nv50_prim(unsigned mode)
+static INLINE uint32_t
+nv50_vbo_type_to_hw(enum pipe_format format)
 {
-	switch (mode) {
-	case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
-	case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
-	case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
-	case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
-	case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
-	case PIPE_PRIM_TRIANGLE_STRIP:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
-	case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
-	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
-	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
-	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
-	default:
-		break;
-	}
+	const struct util_format_description *desc;
 
-	NOUVEAU_ERR("invalid primitive type %d\n", mode);
-	return NV50TCL_VERTEX_BEGIN_POINTS;
-}
+	desc = util_format_description(format);
+	assert(desc);
 
-static INLINE uint32_t
-nv50_vbo_type_to_hw(unsigned type)
-{
-	switch (type) {
-	case PIPE_FORMAT_TYPE_FLOAT:
+	switch (desc->channel[0].type) {
+	case UTIL_FORMAT_TYPE_FLOAT:
 		return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT;
-	case PIPE_FORMAT_TYPE_UNORM:
-		return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM;
-	case PIPE_FORMAT_TYPE_SNORM:
-		return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM;
-	case PIPE_FORMAT_TYPE_USCALED:
+	case UTIL_FORMAT_TYPE_UNSIGNED:
+		if (desc->channel[0].normalized) {
+			return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM;
+		}
 		return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_USCALED;
-	case PIPE_FORMAT_TYPE_SSCALED:
+	case UTIL_FORMAT_TYPE_SIGNED:
+		if (desc->channel[0].normalized) {
+			return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM;
+		}
 		return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SSCALED;
 	/*
 	case PIPE_FORMAT_TYPE_UINT:
@@ -90,19 +64,19 @@ nv50_vbo_size_to_hw(unsigned size, unsigned nr_c)
 {
 	static const uint32_t hw_values[] = {
 		0, 0, 0, 0,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_8_8_8_8,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_16_16_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8_8,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16_16,
 		0, 0, 0, 0,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32,
-		NV50TCL_VERTEX_ARRAY_ATTRIB_SIZE_32_32_32_32 };
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32,
+		NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 };
 
 	/* we'd also have R11G11B10 and R10G10B10A2 */
 
@@ -120,190 +94,374 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 {
 	uint32_t hw_type, hw_size;
 	enum pipe_format pf = ve->src_format;
-	unsigned size = pf_size_x(pf) << pf_exp2(pf);
+	const struct util_format_description *desc;
+	unsigned size, nr_components;
+
+	desc = util_format_description(pf);
+	assert(desc);
 
-	hw_type = nv50_vbo_type_to_hw(pf_type(pf));
-	hw_size = nv50_vbo_size_to_hw(size, ve->nr_components);
+	size = util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0);
+	nr_components = util_format_get_nr_components(pf);
+
+	hw_type = nv50_vbo_type_to_hw(pf);
+	hw_size = nv50_vbo_size_to_hw(size, nr_components);
 
 	if (!hw_type || !hw_size) {
-		NOUVEAU_ERR("unsupported vbo format: %s\n", pf_name(pf));
+		NOUVEAU_ERR("unsupported vbo format: %s\n", util_format_name(pf));
 		abort();
 		return 0x24e80000;
 	}
 
-	if (pf_swizzle_x(pf) == 2) /* BGRA */
+	if (desc->swizzle[0] == UTIL_FORMAT_SWIZZLE_Z) /* BGRA */
 		hw_size |= (1 << 31); /* no real swizzle bits :-( */
 
 	return (hw_type | hw_size);
 }
 
-boolean
-nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
-		 unsigned count)
+struct instance {
+	struct nouveau_bo *bo;
+	unsigned delta;
+	unsigned stride;
+	unsigned step;
+	unsigned divisor;
+};
+
+static void
+instance_init(struct nv50_context *nv50, struct instance *a, unsigned first)
+{
+	int i;
+
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
+		struct pipe_vertex_buffer *vb;
+
+		a[i].divisor = ve->instance_divisor;
+		if (a[i].divisor) {
+			vb = &nv50->vtxbuf[ve->vertex_buffer_index];
+
+			a[i].bo = nouveau_bo(vb->buffer);
+			a[i].stride = vb->stride;
+			a[i].step = first % a[i].divisor;
+			a[i].delta = vb->buffer_offset + ve->src_offset +
+				     (first * a[i].stride);
+		}
+	}
+}
+
+static void
+instance_step(struct nv50_context *nv50, struct instance *a)
+{
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	int i;
+
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		if (!a[i].divisor)
+			continue;
+
+		BEGIN_RING(chan, tesla,
+			   NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
+		OUT_RELOCh(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		OUT_RELOCl(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		if (++a[i].step == a[i].divisor) {
+			a[i].step = 0;
+			a[i].delta += a[i].stride;
+		}
+	}
+}
+
+void
+nv50_draw_arrays_instanced(struct pipe_context *pipe,
+			   unsigned mode, unsigned start, unsigned count,
+			   unsigned startInstance, unsigned instanceCount)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	boolean ret;
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	nv50_state_validate(nv50);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 10 + 16*3))
+		return;
 
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, NULL, 0, mode, start,
+					     count, startInstance,
+					     instanceCount);
+		return;
+	}
 
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 7 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
 
-	if (nv50->vbo_fifo)
-		ret = nv50_push_arrays(nv50, start, count);
-	else {
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, prim);
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
 		OUT_RING  (chan, start);
 		OUT_RING  (chan, count);
-		ret = TRUE;
-	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+		OUT_RING  (chan, 0);
 
-	return ret;
+		prim |= (1 << 28);
+	}
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
-			      unsigned start, unsigned count)
+void
+nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
+		 unsigned count)
 {
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	nv50_draw_arrays_instanced(pipe, mode, start, count, 0, 1);
+}
 
-	map += start;
+struct inline_ctx {
+	struct nv50_context *nv50;
+	void *map;
+};
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u08(nv50, map, count);
+static void
+inline_elt08(void *priv, unsigned start, unsigned count)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint8_t *map = (uint8_t *)ctx->map + start;
 
 	if (count & 1) {
-		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
 		map++;
-		count--;
+		count &= ~1;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt16(void *priv, unsigned start, unsigned count)
 {
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-
-	map += start;
-
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u16(nv50, map, count);
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint16_t *map = (uint16_t *)ctx->map + start;
 
 	if (count & 1) {
-		BEGIN_RING(chan, tesla, 0x15e8, 1);
+		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
+		count &= ~1;
 		map++;
-		count--;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING(chan, tesla, 0x400015f0, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt32(void *priv, unsigned start, unsigned count)
 {
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
 
-	map += start;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, count);
+	OUT_RINGp    (chan, (uint32_t *)ctx->map + start, count);
+}
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u32(nv50, map, count);
+static void
+inline_edgeflag(void *priv, boolean enabled)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
 
-	while (count) {
-		unsigned nr = count > 2047 ? 2047 : count;
+	BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+	OUT_RING  (chan, enabled ? 1 : 0);
+}
 
-		BEGIN_RING(chan, tesla, 0x400015e8, nr);
-		OUT_RINGp (chan, map, nr);
+static void
+nv50_draw_elements_inline(struct pipe_context *pipe,
+			  struct pipe_buffer *indexBuffer, unsigned indexSize,
+			  unsigned mode, unsigned start, unsigned count,
+			  unsigned startInstance, unsigned instanceCount)
+{
+	struct pipe_screen *pscreen = pipe->screen;
+	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct instance a[16];
+	struct inline_ctx ctx;
+	struct u_split_prim s;
+	boolean nzi = FALSE;
+	unsigned overhead;
+
+	overhead = 16*3; /* potential instance adjustments */
+	overhead += 4; /* Begin()/End() */
+	overhead += 4; /* potential edgeflag disable/reenable */
+	overhead += 3; /* potentially 3 VTX_ELT_U16/U32 packet headers */
+
+	s.priv = &ctx;
+	if (indexSize == 1)
+		s.emit = inline_elt08;
+	else
+	if (indexSize == 2)
+		s.emit = inline_elt16;
+	else
+		s.emit = inline_elt32;
+	s.edge = inline_edgeflag;
+
+	ctx.nv50 = nv50;
+	ctx.map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+	assert(ctx.map);
+	if (!ctx.map)
+		return;
 
-		count -= nr;
-		map += nr;
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, overhead + 6 + 3))
+		return;
+
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		unsigned max_verts;
+		boolean done;
+
+		u_split_prim_init(&s, mode, start, count);
+		do {
+			if (AVAIL_RING(chan) < (overhead + 6)) {
+				FIRE_RING(chan);
+				if (!nv50_state_validate(nv50, (overhead + 6))) {
+					assert(0);
+					return;
+				}
+			}
+
+			max_verts = AVAIL_RING(chan) - overhead;
+			if (max_verts > 2047)
+				max_verts = 2047;
+			if (indexSize != 4)
+				max_verts <<= 1;
+			instance_step(nv50, a);
+
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+			OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1<<28) : 0));
+			done = u_split_prim_next(&s, max_verts);
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+			OUT_RING  (chan, 0);
+		} while (!done);
+
+		nzi = TRUE;
 	}
-	return TRUE;
+
+	pipe_buffer_unmap(pscreen, indexBuffer);
 }
 
-boolean
-nv50_draw_elements(struct pipe_context *pipe,
-		   struct pipe_buffer *indexBuffer, unsigned indexSize,
-		   unsigned mode, unsigned start, unsigned count)
+void
+nv50_draw_elements_instanced(struct pipe_context *pipe,
+			     struct pipe_buffer *indexBuffer,
+			     unsigned indexSize,
+			     unsigned mode, unsigned start, unsigned count,
+			     unsigned startInstance, unsigned instanceCount)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct pipe_screen *pscreen = pipe->screen;
-	void *map;
-	boolean ret;
-	
-	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	nv50_state_validate(nv50);
-
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 13 + 16*3))
+		return;
 
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-	switch (indexSize) {
-	case 1:
-		ret = nv50_draw_elements_inline_u08(nv50, map, start, count);
-		break;
-	case 2:
-		ret = nv50_draw_elements_inline_u16(nv50, map, start, count);
-		break;
-	case 4:
-		ret = nv50_draw_elements_inline_u32(nv50, map, start, count);
-		break;
-	default:
-		assert(0);
-		ret = FALSE;
-		break;
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, indexBuffer, indexSize,
+					     mode, start, count, startInstance,
+					     instanceCount);
+		return;
+	} else
+	if (!(indexBuffer->usage & PIPE_BUFFER_USAGE_INDEX) || indexSize == 1) {
+		nv50_draw_elements_inline(pipe, indexBuffer, indexSize,
+					  mode, start, count, startInstance,
+					  instanceCount);
+		return;
 	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
 
-	pipe_buffer_unmap(pscreen, indexBuffer);
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 10 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
+
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, prim);
+		if (indexSize == 4) {
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
+			OUT_RING  (chan, count);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       start << 2, count << 2);
+		} else
+		if (indexSize == 2) {
+			unsigned vb_start = (start & ~1);
+			unsigned vb_end = (start + count + 1) & ~1;
+			unsigned dwords = (vb_end - vb_start) >> 1;
+
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, ((start & 1) << 31) | count);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
+			OUT_RING  (chan, dwords);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       vb_start << 1, dwords << 2);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, 0);
+		}
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+		OUT_RING  (chan, 0);
 
-	return ret;
+		prim |= (1 << 28);
+	}
+}
+
+void
+nv50_draw_elements(struct pipe_context *pipe,
+		   struct pipe_buffer *indexBuffer, unsigned indexSize,
+		   unsigned mode, unsigned start, unsigned count)
+{
+	nv50_draw_elements_instanced(pipe, indexBuffer, indexSize,
+				     mode, start, count, 0, 1);
 }
 
 static INLINE boolean
@@ -316,24 +474,23 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 	struct nouveau_stateobj *so;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nouveau_bo *bo = nouveau_bo(vb->buffer);
-	float *v;
+	float v[4];
 	int ret;
-	enum pipe_format pf = ve->src_format;
-
-	if ((pf_type(pf) != PIPE_FORMAT_TYPE_FLOAT) ||
-	    (pf_size_x(pf) << pf_exp2(pf)) != 32)
-		return FALSE;
+	unsigned nr_components = util_format_get_nr_components(ve->src_format);
 
 	ret = nouveau_bo_map(bo, NOUVEAU_BO_RD);
 	if (ret)
 		return FALSE;
-	v = (float *)(bo->map + (vb->buffer_offset + ve->src_offset));
 
+	util_format_read_4f(ve->src_format, v, 0, (uint8_t *)bo->map +
+			    (vb->buffer_offset + ve->src_offset), 0,
+			    0, 0, 1, 1);
 	so = *pso;
 	if (!so)
-		*pso = so = so_new(nv50->vtxelt_nr * 5, 0);
+		*pso = so = so_new(nv50->vtxelt->num_elements,
+				   nv50->vtxelt->num_elements * 4, 0);
 
-	switch (ve->nr_components) {
+	switch (nr_components) {
 	case 4:
 		so_method(so, tesla, NV50TCL_VTX_ATTR_4F_X(attrib), 4);
 		so_data  (so, fui(v[0]));
@@ -353,6 +510,10 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 		so_data  (so, fui(v[1]));
 		break;
 	case 1:
+		if (attrib == nv50->vertprog->cfg.edgeflag_in) {
+			so_method(so, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+			so_data  (so, v[0] ? 1 : 0);
+		}
 		so_method(so, tesla, NV50TCL_VTX_ATTR_1F(attrib), 1);
 		so_data  (so, fui(v[0]));
 		break;
@@ -366,6 +527,18 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 }
 
 void
+nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso)
+{
+	unsigned i;
+
+	for (i = 0; i < cso->num_elements; ++i) {
+		struct pipe_vertex_element *ve = &cso->pipe[i];
+
+		cso->hw[i] = nv50_vbo_vtxelt_to_hw(ve);
+	}
+}
+
+struct nouveau_stateobj *
 nv50_vbo_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -374,27 +547,32 @@ nv50_vbo_validate(struct nv50_context *nv50)
 
 	/* don't validate if Gallium took away our buffers */
 	if (nv50->vtxbuf_nr == 0)
-		return;
+		return NULL;
+
 	nv50->vbo_fifo = 0;
+	if (nv50->screen->force_push ||
+	    nv50->vertprog->cfg.edgeflag_in < 16)
+		nv50->vbo_fifo = 0xffff;
 
-	for (i = 0; i < nv50->vtxbuf_nr; ++i)
+	for (i = 0; i < nv50->vtxbuf_nr; i++) {
 		if (nv50->vtxbuf[i].stride &&
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
+	}
 
-	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
+	n_ve = MAX2(nv50->vtxelt->num_elements, nv50->state.vtxelt_nr);
 
 	vtxattr = NULL;
-	vtxbuf = so_new(n_ve * 7, nv50->vtxelt_nr * 4);
-	vtxfmt = so_new(n_ve + 1, 0);
+	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt->num_elements * 4);
+	vtxfmt = so_new(1, n_ve, 0);
 	so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
 
-	for (i = 0; i < nv50->vtxelt_nr; i++) {
-		struct pipe_vertex_element *ve = &nv50->vtxelt[i];
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
 		struct pipe_vertex_buffer *vb =
 			&nv50->vtxbuf[ve->vertex_buffer_index];
 		struct nouveau_bo *bo = nouveau_bo(vb->buffer);
-		uint32_t hw = nv50_vbo_vtxelt_to_hw(ve);
+		uint32_t hw = nv50->vtxelt->hw[i];
 
 		if (!vb->stride &&
 		    nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) {
@@ -407,17 +585,20 @@ 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 | vb->stride);
+		so_data  (vtxbuf, 0x20000000 |
+			  (ve->instance_divisor ? 0 : vb->stride));
 		so_reloc (vtxbuf, bo, vb->buffer_offset +
 			  ve->src_offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_GART |
 			  NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0);
@@ -426,7 +607,7 @@ nv50_vbo_validate(struct nv50_context *nv50)
 			  NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0);
 
 		/* vertex array limits */
-		so_method(vtxbuf, tesla, 0x1080 + (i * 8), 2);
+		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i), 2);
 		so_reloc (vtxbuf, bo, vb->buffer->size - 1,
 			  NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD |
 			  NOUVEAU_BO_HIGH, 0, 0);
@@ -440,330 +621,13 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 		so_data  (vtxbuf, 0);
 	}
-	nv50->state.vtxelt_nr = nv50->vtxelt_nr;
+	nv50->state.vtxelt_nr = nv50->vtxelt->num_elements;
 
-	so_ref (vtxfmt, &nv50->state.vtxfmt);
 	so_ref (vtxbuf, &nv50->state.vtxbuf);
 	so_ref (vtxattr, &nv50->state.vtxattr);
 	so_ref (NULL, &vtxbuf);
-	so_ref (NULL, &vtxfmt);
 	so_ref (NULL, &vtxattr);
+	return vtxfmt;
 }
 
-typedef void (*pfn_push)(struct nouveau_channel *, void *);
-
-struct nv50_vbo_emitctx
-{
-	pfn_push push[16];
-	void *map[16];
-	unsigned stride[16];
-	unsigned nr_ve;
-	unsigned vtx_dwords;
-	unsigned vtx_max;
-};
-
-static INLINE void
-emit_vtx_next(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i) {
-		emit->push[i](chan, emit->map[i]);
-		emit->map[i] += emit->stride[i];
-	}
-}
-
-static INLINE void
-emit_vtx(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit,
-	 uint32_t vi)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i)
-		emit->push[i](chan, emit->map[i] + emit->stride[i] * vi);
-}
-
-static INLINE boolean
-nv50_map_vbufs(struct nv50_context *nv50)
-{
-	int i;
-
-	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
-		struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-		unsigned size, delta;
-
-		if (nouveau_bo(vb->buffer)->map)
-			continue;
-
-		size = vb->stride * (vb->max_index + 1);
-		delta = vb->buffer_offset;
-
-		if (!size)
-			size = vb->buffer->size - vb->buffer_offset;
-
-		if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
-					 delta, size, NOUVEAU_BO_RD))
-			break;
-	}
-
-	if (i == nv50->vtxbuf_nr)
-		return TRUE;
-	for (; i >= 0; --i)
-		nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-	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)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b32_2(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-}
-
-static void
-emit_b32_3(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b32_4(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-	OUT_RING(chan, v[3]);
-}
 
-static void
-emit_b16_1(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b16_3(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, (v[1] << 16) | v[0]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b08_1(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b08_3(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
-}
-
-static boolean
-emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
-	     unsigned start)
-{
-	unsigned i;
-
-	if (nv50_map_vbufs(nv50) == FALSE)
-		return FALSE;
-
-	emit->nr_ve = 0;
-	emit->vtx_dwords = 0;
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		struct pipe_vertex_element *ve;
-		struct pipe_vertex_buffer *vb;
-		unsigned n, type, size;
-
-		ve = &nv50->vtxelt[i];
-		vb = &nv50->vtxbuf[ve->vertex_buffer_index];
-		if (!(nv50->vbo_fifo & (1 << i)))
-			continue;
-		n = emit->nr_ve++;
-
-		emit->stride[n] = vb->stride;
-		emit->map[n] = nouveau_bo(vb->buffer)->map +
-			(start * vb->stride + ve->src_offset);
-
-		type = pf_type(ve->src_format);
-		size = pf_size_x(ve->src_format) << pf_exp2(ve->src_format);
-
-		assert(ve->nr_components > 0 && ve->nr_components <= 4);
-
-		/* It shouldn't be necessary to push the implicit 1s
-		 * for case 3 and size 8 cases 1, 2, 3.
-		 */
-		switch (size) {
-		default:
-			NOUVEAU_ERR("unsupported vtxelt size: %u\n", size);
-			return FALSE;
-		case 32:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b32_1; break;
-			case 2: emit->push[n] = emit_b32_2; break;
-			case 3: emit->push[n] = emit_b32_3; break;
-			case 4: emit->push[n] = emit_b32_4; break;
-			}
-			emit->vtx_dwords += ve->nr_components;
-			break;
-		case 16:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b16_1; break;
-			case 2: emit->push[n] = emit_b32_1; break;
-			case 3: emit->push[n] = emit_b16_3; break;
-			case 4: emit->push[n] = emit_b32_2; break;
-			}
-			emit->vtx_dwords += (ve->nr_components + 1) >> 1;
-			break;
-		case 8:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b08_1; break;
-			case 2: emit->push[n] = emit_b16_1; break;
-			case 3: emit->push[n] = emit_b08_3; break;
-			case 4: emit->push[n] = emit_b32_1; break;
-			}
-			emit->vtx_dwords += 1;
-			break;
-		}
-	}
-
-	emit->vtx_max = 512 / emit->vtx_dwords;
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, start) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx_next(chan, &emit);
-
-		count -= nr;
-	}
-	nv50_unmap_vbufs(nv50);
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-	nv50_unmap_vbufs(nv50);
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-	nv50_unmap_vbufs(nv50);
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_DATA | 0x40000000, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-	nv50_unmap_vbufs(nv50);
-
-	return TRUE;
-}