X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr600%2Fr600_state_common.c;h=6d4f3efe253c8190300266459221e3258c1690d5;hb=d1cc87c0b64c2221b079015a6c3b593af83f0758;hp=80b28d9e7ca8683956ebcbe2dad22fa81593d92b;hpb=c2b800cf38b299c1ab1c53dc0e4ea00c7acef853;p=mesa.git

diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 80b28d9e7ca..6d4f3efe253 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -24,16 +24,101 @@
  * Authors: Dave Airlie <airlied@redhat.com>
  *          Jerome Glisse <jglisse@redhat.com>
  */
-#include "util/u_blitter.h"
-#include "util/u_memory.h"
-#include "util/u_format.h"
-#include "pipebuffer/pb_buffer.h"
-#include "pipe/p_shader_tokens.h"
-#include "tgsi/tgsi_parse.h"
 #include "r600_formats.h"
-#include "r600_pipe.h"
 #include "r600d.h"
 
+#include "util/u_blitter.h"
+#include "util/u_upload_mgr.h"
+#include "tgsi/tgsi_parse.h"
+#include <byteswap.h>
+
+static void r600_emit_command_buffer(struct r600_context *rctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	struct r600_command_buffer *cb = (struct r600_command_buffer*)atom;
+
+	assert(cs->cdw + cb->atom.num_dw <= RADEON_MAX_CMDBUF_DWORDS);
+	memcpy(cs->buf + cs->cdw, cb->buf, 4 * cb->atom.num_dw);
+	cs->cdw += cb->atom.num_dw;
+}
+
+void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags)
+{
+	cb->atom.emit = r600_emit_command_buffer;
+	cb->atom.num_dw = 0;
+	cb->atom.flags = flags;
+	cb->buf = CALLOC(1, 4 * num_dw);
+	cb->max_num_dw = num_dw;
+}
+
+void r600_release_command_buffer(struct r600_command_buffer *cb)
+{
+	FREE(cb->buf);
+}
+
+static void r600_emit_surface_sync(struct r600_context *rctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	struct r600_surface_sync_cmd *a = (struct r600_surface_sync_cmd*)atom;
+
+	cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_SYNC, 3, 0);
+	cs->buf[cs->cdw++] = a->flush_flags;  /* CP_COHER_CNTL */
+	cs->buf[cs->cdw++] = 0xffffffff;      /* CP_COHER_SIZE */
+	cs->buf[cs->cdw++] = 0;               /* CP_COHER_BASE */
+	cs->buf[cs->cdw++] = 0x0000000A;      /* POLL_INTERVAL */
+
+	a->flush_flags = 0;
+}
+
+static void r600_emit_r6xx_flush_and_inv(struct r600_context *rctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
+	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
+}
+
+void r600_init_atom(struct r600_atom *atom,
+		    void (*emit)(struct r600_context *ctx, struct r600_atom *state),
+		    unsigned num_dw, enum r600_atom_flags flags)
+{
+	atom->emit = emit;
+	atom->num_dw = num_dw;
+	atom->flags = flags;
+}
+
+void r600_init_common_atoms(struct r600_context *rctx)
+{
+	r600_init_atom(&rctx->surface_sync_cmd.atom,	r600_emit_surface_sync,		5, EMIT_EARLY);
+	r600_init_atom(&rctx->r6xx_flush_and_inv_cmd,	r600_emit_r6xx_flush_and_inv,	2, EMIT_EARLY);
+}
+
+unsigned r600_get_cb_flush_flags(struct r600_context *rctx)
+{
+	unsigned flags = 0;
+
+	if (rctx->framebuffer.nr_cbufs) {
+		flags |= S_0085F0_CB_ACTION_ENA(1) |
+			 (((1 << rctx->framebuffer.nr_cbufs) - 1) << S_0085F0_CB0_DEST_BASE_ENA_SHIFT);
+	}
+
+	/* Workaround for broken flushing on some R6xx chipsets. */
+	if (rctx->family == CHIP_RV670 ||
+	    rctx->family == CHIP_RS780 ||
+	    rctx->family == CHIP_RS880) {
+		flags |=  S_0085F0_CB1_DEST_BASE_ENA(1) |
+			  S_0085F0_DEST_BASE_0_ENA(1);
+	}
+	return flags;
+}
+
+void r600_texture_barrier(struct pipe_context *ctx)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+
+	rctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1) | r600_get_cb_flush_flags(rctx);
+	r600_atom_dirty(rctx, &rctx->surface_sync_cmd.atom);
+}
+
 static bool r600_conv_pipe_prim(unsigned pprim, unsigned *prim)
 {
 	static const int prim_conv[] = {
@@ -64,7 +149,7 @@ static bool r600_conv_pipe_prim(unsigned pprim, unsigned *prim)
 /* common state between evergreen and r600 */
 void r600_bind_blend_state(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_blend *blend = (struct r600_pipe_blend *)state;
 	struct r600_pipe_state *rstate;
 
@@ -73,14 +158,89 @@ void r600_bind_blend_state(struct pipe_context *ctx, void *state)
 	rstate = &blend->rstate;
 	rctx->states[rstate->id] = rstate;
 	rctx->cb_target_mask = blend->cb_target_mask;
-	r600_context_pipe_state_set(&rctx->ctx, rstate);
+	/* Replace every bit except MULTIWRITE_ENABLE. */
+	rctx->cb_color_control &= ~C_028808_MULTIWRITE_ENABLE;
+	rctx->cb_color_control |= blend->cb_color_control & C_028808_MULTIWRITE_ENABLE;
+	rctx->dual_src_blend = blend->dual_src_blend;
+	r600_context_pipe_state_set(rctx, rstate);
+}
+
+void r600_set_blend_color(struct pipe_context *ctx,
+			  const struct pipe_blend_color *state)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
+
+	if (rstate == NULL)
+		return;
+
+	rstate->id = R600_PIPE_STATE_BLEND_COLOR;
+	r600_pipe_state_add_reg(rstate, R_028414_CB_BLEND_RED, fui(state->color[0]), NULL, 0);
+	r600_pipe_state_add_reg(rstate, R_028418_CB_BLEND_GREEN, fui(state->color[1]), NULL, 0);
+	r600_pipe_state_add_reg(rstate, R_02841C_CB_BLEND_BLUE, fui(state->color[2]), NULL, 0);
+	r600_pipe_state_add_reg(rstate, R_028420_CB_BLEND_ALPHA, fui(state->color[3]), NULL, 0);
+
+	free(rctx->states[R600_PIPE_STATE_BLEND_COLOR]);
+	rctx->states[R600_PIPE_STATE_BLEND_COLOR] = rstate;
+	r600_context_pipe_state_set(rctx, rstate);
+}
+
+static void r600_set_stencil_ref(struct pipe_context *ctx,
+				 const struct r600_stencil_ref *state)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
+
+	if (rstate == NULL)
+		return;
+
+	rstate->id = R600_PIPE_STATE_STENCIL_REF;
+	r600_pipe_state_add_reg(rstate,
+				R_028430_DB_STENCILREFMASK,
+				S_028430_STENCILREF(state->ref_value[0]) |
+				S_028430_STENCILMASK(state->valuemask[0]) |
+				S_028430_STENCILWRITEMASK(state->writemask[0]),
+				NULL, 0);
+	r600_pipe_state_add_reg(rstate,
+				R_028434_DB_STENCILREFMASK_BF,
+				S_028434_STENCILREF_BF(state->ref_value[1]) |
+				S_028434_STENCILMASK_BF(state->valuemask[1]) |
+				S_028434_STENCILWRITEMASK_BF(state->writemask[1]),
+				NULL, 0);
+
+	free(rctx->states[R600_PIPE_STATE_STENCIL_REF]);
+	rctx->states[R600_PIPE_STATE_STENCIL_REF] = rstate;
+	r600_context_pipe_state_set(rctx, rstate);
+}
+
+void r600_set_pipe_stencil_ref(struct pipe_context *ctx,
+			       const struct pipe_stencil_ref *state)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	struct r600_pipe_dsa *dsa = (struct r600_pipe_dsa*)rctx->states[R600_PIPE_STATE_DSA];
+	struct r600_stencil_ref ref;
+
+	rctx->stencil_ref = *state;
+
+	if (!dsa)
+		return;
+
+	ref.ref_value[0] = state->ref_value[0];
+	ref.ref_value[1] = state->ref_value[1];
+	ref.valuemask[0] = dsa->valuemask[0];
+	ref.valuemask[1] = dsa->valuemask[1];
+	ref.writemask[0] = dsa->writemask[0];
+	ref.writemask[1] = dsa->writemask[1];
+
+	r600_set_stencil_ref(ctx, &ref);
 }
 
 void r600_bind_dsa_state(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_dsa *dsa = state;
 	struct r600_pipe_state *rstate;
+	struct r600_stencil_ref ref;
 
 	if (state == NULL)
 		return;
@@ -88,35 +248,76 @@ void r600_bind_dsa_state(struct pipe_context *ctx, void *state)
 	rctx->states[rstate->id] = rstate;
 	rctx->alpha_ref = dsa->alpha_ref;
 	rctx->alpha_ref_dirty = true;
-	r600_context_pipe_state_set(&rctx->ctx, rstate);
+	r600_context_pipe_state_set(rctx, rstate);
+
+	ref.ref_value[0] = rctx->stencil_ref.ref_value[0];
+	ref.ref_value[1] = rctx->stencil_ref.ref_value[1];
+	ref.valuemask[0] = dsa->valuemask[0];
+	ref.valuemask[1] = dsa->valuemask[1];
+	ref.writemask[0] = dsa->writemask[0];
+	ref.writemask[1] = dsa->writemask[1];
+
+	r600_set_stencil_ref(ctx, &ref);
+
+	if (rctx->db_misc_state.flush_depthstencil_enabled != dsa->is_flush) {
+		rctx->db_misc_state.flush_depthstencil_enabled = dsa->is_flush;
+		r600_atom_dirty(rctx, &rctx->db_misc_state.atom);
+	}
+}
+
+void r600_set_max_scissor(struct r600_context *rctx)
+{
+	/* Set a scissor state such that it doesn't do anything. */
+	struct pipe_scissor_state scissor;
+	scissor.minx = 0;
+	scissor.miny = 0;
+	scissor.maxx = 8192;
+	scissor.maxy = 8192;
+
+	r600_set_scissor_state(rctx, &scissor);
 }
 
 void r600_bind_rs_state(struct pipe_context *ctx, void *state)
 {
 	struct r600_pipe_rasterizer *rs = (struct r600_pipe_rasterizer *)state;
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 
 	if (state == NULL)
 		return;
 
 	rctx->sprite_coord_enable = rs->sprite_coord_enable;
 	rctx->two_side = rs->two_side;
+	rctx->pa_sc_line_stipple = rs->pa_sc_line_stipple;
+	rctx->pa_cl_clip_cntl = rs->pa_cl_clip_cntl;
 
 	rctx->rasterizer = rs;
 
 	rctx->states[rs->rstate.id] = &rs->rstate;
-	r600_context_pipe_state_set(&rctx->ctx, &rs->rstate);
+	r600_context_pipe_state_set(rctx, &rs->rstate);
 
 	if (rctx->chip_class >= EVERGREEN) {
 		evergreen_polygon_offset_update(rctx);
 	} else {
 		r600_polygon_offset_update(rctx);
 	}
+
+	/* Workaround for a missing scissor enable on r600. */
+	if (rctx->chip_class == R600) {
+		if (rs->scissor_enable != rctx->scissor_enable) {
+			rctx->scissor_enable = rs->scissor_enable;
+
+			if (rs->scissor_enable) {
+				r600_set_scissor_state(rctx, &rctx->scissor_state);
+			} else {
+				r600_set_max_scissor(rctx);
+			}
+		}
+	}
 }
 
 void r600_delete_rs_state(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_rasterizer *rs = (struct r600_pipe_rasterizer *)state;
 
 	if (rctx->rasterizer == rs) {
@@ -139,7 +340,7 @@ void r600_sampler_view_destroy(struct pipe_context *ctx,
 
 void r600_delete_state(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_state *rstate = (struct r600_pipe_state *)state;
 
 	if (rctx->states[rstate->id] == rstate) {
@@ -153,22 +354,23 @@ void r600_delete_state(struct pipe_context *ctx, void *state)
 
 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_vertex_element *v = (struct r600_vertex_element*)state;
 
 	rctx->vertex_elements = v;
 	if (v) {
+		r600_inval_shader_cache(rctx);
 		u_vbuf_bind_vertex_elements(rctx->vbuf_mgr, state,
 						v->vmgr_elements);
 
 		rctx->states[v->rstate.id] = &v->rstate;
-		r600_context_pipe_state_set(&rctx->ctx, &v->rstate);
+		r600_context_pipe_state_set(rctx, &v->rstate);
 	}
 }
 
 void r600_delete_vertex_element(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_vertex_element *v = (struct r600_vertex_element*)state;
 
 	if (rctx->states[v->rstate.id] == &v->rstate) {
@@ -186,7 +388,7 @@ void r600_delete_vertex_element(struct pipe_context *ctx, void *state)
 void r600_set_index_buffer(struct pipe_context *ctx,
 			   const struct pipe_index_buffer *ib)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 
 	u_vbuf_set_index_buffer(rctx->vbuf_mgr, ib);
 }
@@ -194,35 +396,17 @@ void r600_set_index_buffer(struct pipe_context *ctx,
 void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
 			     const struct pipe_vertex_buffer *buffers)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-	int i;
-
-	/* Zero states. */
-	for (i = 0; i < count; i++) {
-		if (!buffers[i].buffer) {
-			if (rctx->chip_class >= EVERGREEN) {
-				evergreen_context_pipe_state_set_fs_resource(&rctx->ctx, NULL, i);
-			} else {
-				r600_context_pipe_state_set_fs_resource(&rctx->ctx, NULL, i);
-			}
-		}
-	}
-	for (; i < rctx->vbuf_mgr->nr_real_vertex_buffers; i++) {
-		if (rctx->chip_class >= EVERGREEN) {
-			evergreen_context_pipe_state_set_fs_resource(&rctx->ctx, NULL, i);
-		} else {
-			r600_context_pipe_state_set_fs_resource(&rctx->ctx, NULL, i);
-		}
-	}
+	struct r600_context *rctx = (struct r600_context *)ctx;
 
 	u_vbuf_set_vertex_buffers(rctx->vbuf_mgr, count, buffers);
+	rctx->vertex_buffers_dirty = true;
 }
 
 void *r600_create_vertex_elements(struct pipe_context *ctx,
 				  unsigned count,
 				  const struct pipe_vertex_element *elements)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_vertex_element *v = CALLOC_STRUCT(r600_vertex_element);
 
 	assert(count < 32);
@@ -260,13 +444,20 @@ void *r600_create_shader_state(struct pipe_context *ctx,
 
 void r600_bind_ps_shader(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 
-	/* TODO delete old shader */
-	rctx->ps_shader = (struct r600_pipe_shader *)state;
-	if (state) {
-		r600_context_pipe_state_set(&rctx->ctx, &rctx->ps_shader->rstate);
+	if (!state) {
+		state = rctx->dummy_pixel_shader;
 	}
+
+	rctx->ps_shader = (struct r600_pipe_shader *)state;
+
+	r600_inval_shader_cache(rctx);
+	r600_context_pipe_state_set(rctx, &rctx->ps_shader->rstate);
+
+	rctx->cb_color_control &= C_028808_MULTIWRITE_ENABLE;
+	rctx->cb_color_control |= S_028808_MULTIWRITE_ENABLE(!!rctx->ps_shader->shader.fs_write_all);
+
 	if (rctx->ps_shader && rctx->vs_shader) {
 		r600_adjust_gprs(rctx);
 	}
@@ -274,12 +465,12 @@ void r600_bind_ps_shader(struct pipe_context *ctx, void *state)
 
 void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 
-	/* TODO delete old shader */
 	rctx->vs_shader = (struct r600_pipe_shader *)state;
 	if (state) {
-		r600_context_pipe_state_set(&rctx->ctx, &rctx->vs_shader->rstate);
+		r600_inval_shader_cache(rctx);
+		r600_context_pipe_state_set(rctx, &rctx->vs_shader->rstate);
 	}
 	if (rctx->ps_shader && rctx->vs_shader) {
 		r600_adjust_gprs(rctx);
@@ -288,7 +479,7 @@ void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
 
 void r600_delete_ps_shader(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_shader *shader = (struct r600_pipe_shader *)state;
 
 	if (rctx->ps_shader == shader) {
@@ -302,7 +493,7 @@ void r600_delete_ps_shader(struct pipe_context *ctx, void *state)
 
 void r600_delete_vs_shader(struct pipe_context *ctx, void *state)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_pipe_shader *shader = (struct r600_pipe_shader *)state;
 
 	if (rctx->vs_shader == shader) {
@@ -314,7 +505,7 @@ void r600_delete_vs_shader(struct pipe_context *ctx, void *state)
 	free(shader);
 }
 
-static void r600_update_alpha_ref(struct r600_pipe_context *rctx)
+static void r600_update_alpha_ref(struct r600_context *rctx)
 {
 	unsigned alpha_ref;
 	struct r600_pipe_state rstate;
@@ -323,96 +514,83 @@ static void r600_update_alpha_ref(struct r600_pipe_context *rctx)
 	rstate.nregs = 0;
 	if (rctx->export_16bpc)
 		alpha_ref &= ~0x1FFF;
-	r600_pipe_state_add_reg(&rstate, R_028438_SX_ALPHA_REF, alpha_ref, 0xFFFFFFFF, NULL, 0);
+	r600_pipe_state_add_reg(&rstate, R_028438_SX_ALPHA_REF, alpha_ref, NULL, 0);
 
-	r600_context_pipe_state_set(&rctx->ctx, &rstate);
+	r600_context_pipe_state_set(rctx, &rstate);
 	rctx->alpha_ref_dirty = false;
 }
 
+void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state)
+{
+	r600_inval_shader_cache(rctx);
+	state->atom.num_dw = rctx->chip_class >= EVERGREEN ? util_bitcount(state->dirty_mask)*20
+							   : util_bitcount(state->dirty_mask)*19;
+	r600_atom_dirty(rctx, &state->atom);
+}
+
 void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
 			      struct pipe_resource *buffer)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
-	struct r600_resource *rbuffer = r600_resource(buffer);
-	struct r600_pipe_resource_state *rstate;
-	uint64_t va_offset;
-	uint32_t offset;
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	struct r600_constbuf_state *state;
+	struct r600_constant_buffer *cb;
+	uint8_t *ptr;
+
+	switch (shader) {
+	case PIPE_SHADER_VERTEX:
+		state = &rctx->vs_constbuf_state;
+		break;
+	case PIPE_SHADER_FRAGMENT:
+		state = &rctx->ps_constbuf_state;
+		break;
+	default:
+		return;
+	}
 
 	/* Note that the state tracker can unbind constant buffers by
 	 * passing NULL here.
 	 */
-	if (buffer == NULL) {
+	if (unlikely(!buffer)) {
+		state->enabled_mask &= ~(1 << index);
+		state->dirty_mask &= ~(1 << index);
+		pipe_resource_reference(&state->cb[index].buffer, NULL);
 		return;
 	}
 
-	r600_upload_const_buffer(rctx, &rbuffer, &offset);
-	va_offset = r600_resource_va(ctx->screen, (void*)rbuffer);
-	va_offset += offset;
-	va_offset >>= 8;
+	cb = &state->cb[index];
+	cb->buffer_size = buffer->width0;
 
-	switch (shader) {
-	case PIPE_SHADER_VERTEX:
-		rctx->vs_const_buffer.nregs = 0;
-		r600_pipe_state_add_reg(&rctx->vs_const_buffer,
-					R_028180_ALU_CONST_BUFFER_SIZE_VS_0 + index * 4,
-					ALIGN_DIVUP(buffer->width0 >> 4, 16),
-					0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vs_const_buffer,
-					R_028980_ALU_CONST_CACHE_VS_0 + index * 4,
-					va_offset, 0xFFFFFFFF, rbuffer, RADEON_USAGE_READ);
-		r600_context_pipe_state_set(&rctx->ctx, &rctx->vs_const_buffer);
-
-		rstate = &rctx->vs_const_buffer_resource[index];
-		if (!rstate->id) {
-			if (rctx->chip_class >= EVERGREEN) {
-				evergreen_pipe_init_buffer_resource(rctx, rstate);
-			} else {
-				r600_pipe_init_buffer_resource(rctx, rstate);
+	ptr = u_vbuf_resource(buffer)->user_ptr;
+
+	if (ptr) {
+		/* Upload the user buffer. */
+		if (R600_BIG_ENDIAN) {
+			uint32_t *tmpPtr;
+			unsigned i, size = buffer->width0;
+
+			if (!(tmpPtr = malloc(size))) {
+				R600_ERR("Failed to allocate BE swap buffer.\n");
+				return;
 			}
-		}
 
-		if (rctx->chip_class >= EVERGREEN) {
-			evergreen_pipe_mod_buffer_resource(ctx, rstate, rbuffer, offset, 16, RADEON_USAGE_READ);
-			evergreen_context_pipe_state_set_vs_resource(&rctx->ctx, rstate, index);
-		} else {
-			r600_pipe_mod_buffer_resource(rstate, rbuffer, offset, 16, RADEON_USAGE_READ);
-			r600_context_pipe_state_set_vs_resource(&rctx->ctx, rstate, index);
-		}
-		break;
-	case PIPE_SHADER_FRAGMENT:
-		rctx->ps_const_buffer.nregs = 0;
-		r600_pipe_state_add_reg(&rctx->ps_const_buffer,
-					R_028140_ALU_CONST_BUFFER_SIZE_PS_0,
-					ALIGN_DIVUP(buffer->width0 >> 4, 16),
-					0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->ps_const_buffer,
-					R_028940_ALU_CONST_CACHE_PS_0,
-					va_offset, 0xFFFFFFFF, rbuffer, RADEON_USAGE_READ);
-		r600_context_pipe_state_set(&rctx->ctx, &rctx->ps_const_buffer);
-
-		rstate = &rctx->ps_const_buffer_resource[index];
-		if (!rstate->id) {
-			if (rctx->chip_class >= EVERGREEN) {
-				evergreen_pipe_init_buffer_resource(rctx, rstate);
-			} else {
-				r600_pipe_init_buffer_resource(rctx, rstate);
+			for (i = 0; i < size / 4; ++i) {
+				tmpPtr[i] = bswap_32(((uint32_t *)ptr)[i]);
 			}
-		}
-		if (rctx->chip_class >= EVERGREEN) {
-			evergreen_pipe_mod_buffer_resource(ctx, rstate, rbuffer, offset, 16, RADEON_USAGE_READ);
-			evergreen_context_pipe_state_set_ps_resource(&rctx->ctx, rstate, index);
+
+			u_upload_data(rctx->vbuf_mgr->uploader, 0, size, tmpPtr, &cb->buffer_offset, &cb->buffer);
+			free(tmpPtr);
 		} else {
-			r600_pipe_mod_buffer_resource(rstate, rbuffer, offset, 16, RADEON_USAGE_READ);
-			r600_context_pipe_state_set_ps_resource(&rctx->ctx, rstate, index);
+			u_upload_data(rctx->vbuf_mgr->uploader, 0, buffer->width0, ptr, &cb->buffer_offset, &cb->buffer);
 		}
-		break;
-	default:
-		R600_ERR("unsupported %d\n", shader);
-		return;
+	} else {
+		/* Setup the hw buffer. */
+		cb->buffer_offset = 0;
+		pipe_resource_reference(&cb->buffer, buffer);
 	}
 
-	if (buffer != &rbuffer->b.b.b)
-		pipe_resource_reference((struct pipe_resource**)&rbuffer, NULL);
+	state->enabled_mask |= 1 << index;
+	state->dirty_mask |= 1 << index;
+	r600_constant_buffers_dirty(rctx, state);
 }
 
 struct pipe_stream_output_target *
@@ -421,7 +599,7 @@ r600_create_so_target(struct pipe_context *ctx,
 		      unsigned buffer_offset,
 		      unsigned buffer_size)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct r600_so_target *t;
 	void *ptr;
 
@@ -438,7 +616,7 @@ r600_create_so_target(struct pipe_context *ctx,
 
 	t->filled_size = (struct r600_resource*)
 		pipe_buffer_create(ctx->screen, PIPE_BIND_CUSTOM, PIPE_USAGE_STATIC, 4);
-	ptr = rctx->ws->buffer_map(t->filled_size->buf, rctx->ctx.cs, PIPE_TRANSFER_WRITE);
+	ptr = rctx->ws->buffer_map(t->filled_size->buf, rctx->cs, PIPE_TRANSFER_WRITE);
 	memset(ptr, 0, t->filled_size->buf->size);
 	rctx->ws->buffer_unmap(t->filled_size->buf);
 
@@ -459,83 +637,30 @@ void r600_set_so_targets(struct pipe_context *ctx,
 			 struct pipe_stream_output_target **targets,
 			 unsigned append_bitmask)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	unsigned i;
 
 	/* Stop streamout. */
-	if (rctx->ctx.num_so_targets) {
-		r600_context_streamout_end(&rctx->ctx);
+	if (rctx->num_so_targets) {
+		r600_context_streamout_end(rctx);
 	}
 
 	/* Set the new targets. */
 	for (i = 0; i < num_targets; i++) {
-		pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->ctx.so_targets[i], targets[i]);
+		pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->so_targets[i], targets[i]);
 	}
-	for (; i < rctx->ctx.num_so_targets; i++) {
-		pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->ctx.so_targets[i], NULL);
+	for (; i < rctx->num_so_targets; i++) {
+		pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->so_targets[i], NULL);
 	}
 
-	rctx->ctx.num_so_targets = num_targets;
-	rctx->ctx.streamout_start = num_targets != 0;
-	rctx->ctx.streamout_append_bitmask = append_bitmask;
-}
-
-static void r600_vertex_buffer_update(struct r600_pipe_context *rctx)
-{
-	struct r600_pipe_resource_state *rstate;
-	struct r600_resource *rbuffer;
-	struct pipe_vertex_buffer *vertex_buffer;
-	unsigned i, count, offset;
-
-	if (rctx->vertex_elements->vbuffer_need_offset) {
-		/* one resource per vertex elements */
-		count = rctx->vertex_elements->count;
-	} else {
-		/* bind vertex buffer once */
-		count = rctx->vbuf_mgr->nr_real_vertex_buffers;
-	}
-
-	for (i = 0 ; i < count; i++) {
-		rstate = &rctx->fs_resource[i];
-
-		if (rctx->vertex_elements->vbuffer_need_offset) {
-			/* one resource per vertex elements */
-			unsigned vbuffer_index;
-			vbuffer_index = rctx->vertex_elements->elements[i].vertex_buffer_index;
-			vertex_buffer = &rctx->vbuf_mgr->real_vertex_buffer[vbuffer_index];
-			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
-			offset = rctx->vertex_elements->vbuffer_offset[i];
-		} else {
-			/* bind vertex buffer once */
-			vertex_buffer = &rctx->vbuf_mgr->real_vertex_buffer[i];
-			rbuffer = (struct r600_resource*)vertex_buffer->buffer;
-			offset = 0;
-		}
-		if (vertex_buffer == NULL || rbuffer == NULL)
-			continue;
-		offset += vertex_buffer->buffer_offset;
-
-		if (!rstate->id) {
-			if (rctx->chip_class >= EVERGREEN) {
-				evergreen_pipe_init_buffer_resource(rctx, rstate);
-			} else {
-				r600_pipe_init_buffer_resource(rctx, rstate);
-			}
-		}
-
-		if (rctx->chip_class >= EVERGREEN) {
-			evergreen_pipe_mod_buffer_resource(&rctx->context, rstate, rbuffer, offset, vertex_buffer->stride, RADEON_USAGE_READ);
-			evergreen_context_pipe_state_set_fs_resource(&rctx->ctx, rstate, i);
-		} else {
-			r600_pipe_mod_buffer_resource(rstate, rbuffer, offset, vertex_buffer->stride, RADEON_USAGE_READ);
-			r600_context_pipe_state_set_fs_resource(&rctx->ctx, rstate, i);
-		}
-	}
+	rctx->num_so_targets = num_targets;
+	rctx->streamout_start = num_targets != 0;
+	rctx->streamout_append_bitmask = append_bitmask;
 }
 
 static int r600_shader_rebuild(struct pipe_context * ctx, struct r600_pipe_shader * shader)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	int r;
 
 	r600_pipe_shader_destroy(ctx, shader);
@@ -543,38 +668,14 @@ static int r600_shader_rebuild(struct pipe_context * ctx, struct r600_pipe_shade
 	if (r) {
 		return r;
 	}
-	r600_context_pipe_state_set(&rctx->ctx, &shader->rstate);
+	r600_context_pipe_state_set(rctx, &shader->rstate);
 
 	return 0;
 }
 
-static void r600_update_derived_state(struct r600_pipe_context *rctx)
+static void r600_update_derived_state(struct r600_context *rctx)
 {
 	struct pipe_context * ctx = (struct pipe_context*)rctx;
-	struct r600_pipe_state rstate;
-	unsigned user_clip_plane_enable;
-	unsigned clip_dist_enable;
-
-	if (rctx->vs_shader->shader.clip_dist_write || rctx->vs_shader->shader.vs_prohibit_ucps)
-		user_clip_plane_enable = 0;
-	else
-		user_clip_plane_enable = rctx->rasterizer->clip_plane_enable & 0x3F;
-
-	clip_dist_enable = rctx->rasterizer->clip_plane_enable & rctx->vs_shader->shader.clip_dist_write;
-	rstate.nregs = 0;
-
-	if (user_clip_plane_enable != rctx->user_clip_plane_enable) {
-		r600_pipe_state_add_reg(&rstate, R_028810_PA_CL_CLIP_CNTL, user_clip_plane_enable , 0x3F, NULL, 0);
-		rctx->user_clip_plane_enable = user_clip_plane_enable;
-	}
-
-	if (clip_dist_enable != rctx->clip_dist_enable) {
-		r600_pipe_state_add_reg(&rstate, R_02881C_PA_CL_VS_OUT_CNTL, clip_dist_enable, 0xFF, NULL, 0);
-		rctx->clip_dist_enable = clip_dist_enable;
-	}
-
-	if (rstate.nregs)
-		r600_context_pipe_state_set(&rctx->ctx, &rstate);
 
 	if (!rctx->blitter->running) {
 		if (rctx->have_depth_fb || rctx->have_depth_texture)
@@ -595,43 +696,82 @@ static void r600_update_derived_state(struct r600_pipe_context *rctx)
 		r600_update_alpha_ref(rctx);
 	}
 
-	if (rctx->ps_shader && rctx->sprite_coord_enable &&
-		(rctx->ps_shader->sprite_coord_enable != rctx->sprite_coord_enable)) {
+	if (rctx->ps_shader && ((rctx->sprite_coord_enable &&
+		(rctx->ps_shader->sprite_coord_enable != rctx->sprite_coord_enable)) ||
+		(rctx->rasterizer && rctx->rasterizer->flatshade != rctx->ps_shader->flatshade))) {
 
 		if (rctx->chip_class >= EVERGREEN)
 			evergreen_pipe_shader_ps(ctx, rctx->ps_shader);
 		else
 			r600_pipe_shader_ps(ctx, rctx->ps_shader);
 
-		r600_context_pipe_state_set(&rctx->ctx, &rctx->ps_shader->rstate);
+		r600_context_pipe_state_set(rctx, &rctx->ps_shader->rstate);
 	}
 
+	if (rctx->dual_src_blend)
+		rctx->cb_shader_mask = rctx->ps_shader->ps_cb_shader_mask | rctx->fb_cb_shader_mask;
+	else
+		rctx->cb_shader_mask = rctx->fb_cb_shader_mask;
+}
+
+static unsigned r600_conv_prim_to_gs_out(unsigned mode)
+{
+	static const int prim_conv[] = {
+		V_028A6C_OUTPRIM_TYPE_POINTLIST,
+		V_028A6C_OUTPRIM_TYPE_LINESTRIP,
+		V_028A6C_OUTPRIM_TYPE_LINESTRIP,
+		V_028A6C_OUTPRIM_TYPE_LINESTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_LINESTRIP,
+		V_028A6C_OUTPRIM_TYPE_LINESTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP,
+		V_028A6C_OUTPRIM_TYPE_TRISTRIP
+	};
+	assert(mode < Elements(prim_conv));
+
+	return prim_conv[mode];
 }
 
 void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 {
-	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_context *rctx = (struct r600_context *)ctx;
 	struct pipe_draw_info info = *dinfo;
-	struct r600_draw rdraw = {};
 	struct pipe_index_buffer ib = {};
 	unsigned prim, mask, ls_mask = 0;
+	struct r600_block *dirty_block = NULL, *next_block = NULL;
+	struct r600_atom *state = NULL, *next_state = NULL;
+	struct radeon_winsys_cs *cs = rctx->cs;
+	uint64_t va;
+	uint8_t *ptr;
 
 	if ((!info.count && (info.indexed || !info.count_from_stream_output)) ||
 	    (info.indexed && !rctx->vbuf_mgr->index_buffer.buffer) ||
 	    !r600_conv_pipe_prim(info.mode, &prim)) {
+		assert(0);
 		return;
 	}
 
-	if (!rctx->ps_shader || !rctx->vs_shader)
+	if (!rctx->vs_shader) {
+		assert(0);
 		return;
+	}
 
 	r600_update_derived_state(rctx);
 
-	u_vbuf_draw_begin(rctx->vbuf_mgr, &info);
-	r600_vertex_buffer_update(rctx);
-
-	rdraw.vgt_num_indices = info.count;
-	rdraw.vgt_num_instances = info.instance_count;
+	/* Update vertex buffers. */
+	if ((u_vbuf_draw_begin(rctx->vbuf_mgr, &info) & U_VBUF_BUFFERS_UPDATED) ||
+	    rctx->vertex_buffers_dirty) {
+		r600_inval_vertex_cache(rctx);
+		rctx->vertex_buffer_state.num_dw = (rctx->chip_class >= EVERGREEN ? 12 : 10) *
+						   rctx->vbuf_mgr->nr_real_vertex_buffers;
+		r600_atom_dirty(rctx, &rctx->vertex_buffer_state);
+		rctx->vertex_buffers_dirty = FALSE;
+	}
 
 	if (info.indexed) {
 		/* Initialize the index buffer struct. */
@@ -642,84 +782,122 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 		/* Translate or upload, if needed. */
 		r600_translate_index_buffer(rctx, &ib, info.count);
 
-		if (u_vbuf_resource(ib.buffer)->user_ptr) {
-			r600_upload_index_buffer(rctx, &ib, info.count);
+		ptr = u_vbuf_resource(ib.buffer)->user_ptr;
+		if (ptr) {
+			u_upload_data(rctx->vbuf_mgr->uploader, 0, info.count * ib.index_size,
+				      ptr, &ib.offset, &ib.buffer);
 		}
-
-		/* Initialize the r600_draw struct with index buffer info. */
-		if (ib.index_size == 4) {
-			rdraw.vgt_index_type = VGT_INDEX_32 |
-				(R600_BIG_ENDIAN ? VGT_DMA_SWAP_32_BIT : 0);
-		} else {
-			rdraw.vgt_index_type = VGT_INDEX_16 |
-				(R600_BIG_ENDIAN ? VGT_DMA_SWAP_16_BIT : 0);
-		}
-		rdraw.indices = (struct r600_resource*)ib.buffer;
-		rdraw.indices_bo_offset = ib.offset;
-		rdraw.vgt_draw_initiator = V_0287F0_DI_SRC_SEL_DMA;
 	} else {
 		info.index_bias = info.start;
-		rdraw.vgt_draw_initiator = V_0287F0_DI_SRC_SEL_AUTO_INDEX;
 		if (info.count_from_stream_output) {
-			rdraw.vgt_draw_initiator |= S_0287F0_USE_OPAQUE(1);
-
-			r600_context_draw_opaque_count(&rctx->ctx, (struct r600_so_target*)info.count_from_stream_output);
+			r600_context_draw_opaque_count(rctx, (struct r600_so_target*)info.count_from_stream_output);
 		}
 	}
 
-	rctx->ctx.vs_so_stride_in_dw = rctx->vs_shader->so.stride;
-
 	mask = (1ULL << ((unsigned)rctx->framebuffer.nr_cbufs * 4)) - 1;
 
 	if (rctx->vgt.id != R600_PIPE_STATE_VGT) {
 		rctx->vgt.id = R600_PIPE_STATE_VGT;
 		rctx->vgt.nregs = 0;
-		r600_pipe_state_add_reg(&rctx->vgt, R_008958_VGT_PRIMITIVE_TYPE, prim, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028238_CB_TARGET_MASK, rctx->cb_target_mask & mask, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028400_VGT_MAX_VTX_INDX, ~0, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028404_VGT_MIN_VTX_INDX, 0, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028408_VGT_INDX_OFFSET, info.index_bias, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, info.restart_index, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, info.primitive_restart, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_03CFF4_SQ_VTX_START_INST_LOC, info.start_instance, 0xFFFFFFFF, NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028A0C_PA_SC_LINE_STIPPLE,
-					0,
-					S_028A0C_AUTO_RESET_CNTL(3), NULL, 0);
-		r600_pipe_state_add_reg(&rctx->vgt, R_028814_PA_SU_SC_MODE_CNTL,
-					0,
-					S_028814_PROVOKING_VTX_LAST(1), NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_008958_VGT_PRIMITIVE_TYPE, prim, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028A6C_VGT_GS_OUT_PRIM_TYPE, 0, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028238_CB_TARGET_MASK, rctx->cb_target_mask & mask, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_02823C_CB_SHADER_MASK, 0, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028408_VGT_INDX_OFFSET, info.index_bias, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX, info.restart_index, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, info.primitive_restart, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_03CFF4_SQ_VTX_START_INST_LOC, info.start_instance, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028A0C_PA_SC_LINE_STIPPLE, 0, NULL, 0);
+		if (rctx->chip_class <= R700)
+			r600_pipe_state_add_reg(&rctx->vgt, R_028808_CB_COLOR_CONTROL, rctx->cb_color_control, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_02881C_PA_CL_VS_OUT_CNTL, 0, NULL, 0);
+		r600_pipe_state_add_reg(&rctx->vgt, R_028810_PA_CL_CLIP_CNTL, 0, NULL, 0);
+
+		if (rctx->chip_class <= R700)
+			r600_pipe_state_add_reg(&rctx->vgt, R_0280A4_CB_COLOR1_INFO, 0, NULL, 0);	
+		else
+			r600_pipe_state_add_reg(&rctx->vgt, 0x28CAC, 0, NULL, 0);	
 	}
 
 	rctx->vgt.nregs = 0;
 	r600_pipe_state_mod_reg(&rctx->vgt, prim);
+	r600_pipe_state_mod_reg(&rctx->vgt, r600_conv_prim_to_gs_out(info.mode));
 	r600_pipe_state_mod_reg(&rctx->vgt, rctx->cb_target_mask & mask);
-	r600_pipe_state_mod_reg(&rctx->vgt, ~0);
-	r600_pipe_state_mod_reg(&rctx->vgt, 0);
+	r600_pipe_state_mod_reg(&rctx->vgt, rctx->cb_shader_mask);
 	r600_pipe_state_mod_reg(&rctx->vgt, info.index_bias);
 	r600_pipe_state_mod_reg(&rctx->vgt, info.restart_index);
 	r600_pipe_state_mod_reg(&rctx->vgt, info.primitive_restart);
-	r600_pipe_state_mod_reg(&rctx->vgt, 0);
 	r600_pipe_state_mod_reg(&rctx->vgt, info.start_instance);
 
 	if (prim == V_008958_DI_PT_LINELIST)
 		ls_mask = 1;
 	else if (prim == V_008958_DI_PT_LINESTRIP) 
 		ls_mask = 2;
-	r600_pipe_state_mod_reg(&rctx->vgt, S_028A0C_AUTO_RESET_CNTL(ls_mask));
+	r600_pipe_state_mod_reg(&rctx->vgt, S_028A0C_AUTO_RESET_CNTL(ls_mask) | rctx->pa_sc_line_stipple);
+	if (rctx->chip_class <= R700)
+		r600_pipe_state_mod_reg(&rctx->vgt, rctx->cb_color_control);
+	r600_pipe_state_mod_reg(&rctx->vgt,
+				rctx->vs_shader->pa_cl_vs_out_cntl |
+				(rctx->rasterizer->clip_plane_enable & rctx->vs_shader->shader.clip_dist_write));
+	r600_pipe_state_mod_reg(&rctx->vgt,
+				rctx->pa_cl_clip_cntl |
+				(rctx->vs_shader->shader.clip_dist_write ||
+				 rctx->vs_shader->shader.vs_prohibit_ucps ?
+				 0 : rctx->rasterizer->clip_plane_enable & 0x3F));
+
+	if (rctx->dual_src_blend) {
+		r600_pipe_state_mod_reg(&rctx->vgt, 
+					rctx->color0_format);
+	}
 
-	if (info.mode == PIPE_PRIM_QUADS || info.mode == PIPE_PRIM_QUAD_STRIP || info.mode == PIPE_PRIM_POLYGON) {
-		r600_pipe_state_mod_reg(&rctx->vgt, S_028814_PROVOKING_VTX_LAST(1));
+	r600_context_pipe_state_set(rctx, &rctx->vgt);
+
+	/* Emit states (the function expects that we emit at most 17 dwords here). */
+	r600_need_cs_space(rctx, 0, TRUE);
+
+	LIST_FOR_EACH_ENTRY_SAFE(state, next_state, &rctx->dirty_states, head) {
+		r600_emit_atom(rctx, state);
+	}
+	LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &rctx->dirty,list) {
+		r600_context_block_emit_dirty(rctx, dirty_block);
 	}
+	LIST_FOR_EACH_ENTRY_SAFE(dirty_block, next_block, &rctx->resource_dirty,list) {
+		r600_context_block_resource_emit_dirty(rctx, dirty_block);
+	}
+	rctx->pm4_dirty_cdwords = 0;
 
-	r600_context_pipe_state_set(&rctx->ctx, &rctx->vgt);
+	/* Enable stream out if needed. */
+	if (rctx->streamout_start) {
+		r600_context_streamout_begin(rctx);
+		rctx->streamout_start = FALSE;
+	}
 
-	if (rctx->chip_class >= EVERGREEN) {
-		evergreen_context_draw(&rctx->ctx, &rdraw);
+	/* draw packet */
+	cs->buf[cs->cdw++] = PKT3(PKT3_INDEX_TYPE, 0, rctx->predicate_drawing);
+	cs->buf[cs->cdw++] = ib.index_size == 4 ?
+				(VGT_INDEX_32 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_32_BIT : 0)) :
+				(VGT_INDEX_16 | (R600_BIG_ENDIAN ? VGT_DMA_SWAP_16_BIT : 0));
+	cs->buf[cs->cdw++] = PKT3(PKT3_NUM_INSTANCES, 0, rctx->predicate_drawing);
+	cs->buf[cs->cdw++] = info.instance_count;
+	if (info.indexed) {
+		va = r600_resource_va(ctx->screen, ib.buffer);
+		va += ib.offset;
+		cs->buf[cs->cdw++] = PKT3(PKT3_DRAW_INDEX, 3, rctx->predicate_drawing);
+		cs->buf[cs->cdw++] = va;
+		cs->buf[cs->cdw++] = (va >> 32UL) & 0xFF;
+		cs->buf[cs->cdw++] = info.count;
+		cs->buf[cs->cdw++] = V_0287F0_DI_SRC_SEL_DMA;
+		cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 0, rctx->predicate_drawing);
+		cs->buf[cs->cdw++] = r600_context_bo_reloc(rctx, (struct r600_resource*)ib.buffer, RADEON_USAGE_READ);
 	} else {
-		r600_context_draw(&rctx->ctx, &rdraw);
+		cs->buf[cs->cdw++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, rctx->predicate_drawing);
+		cs->buf[cs->cdw++] = info.count;
+		cs->buf[cs->cdw++] = V_0287F0_DI_SRC_SEL_AUTO_INDEX |
+					(info.count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0);
 	}
 
+	rctx->flags |= R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING;
+
 	if (rctx->framebuffer.zsbuf)
 	{
 		struct pipe_resource *tex = rctx->framebuffer.zsbuf->texture;
@@ -732,8 +910,8 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 
 void _r600_pipe_state_add_reg(struct r600_context *ctx,
 			      struct r600_pipe_state *state,
-			      u32 offset, u32 value, u32 mask,
-			      u32 range_id, u32 block_id,
+			      uint32_t offset, uint32_t value,
+			      uint32_t range_id, uint32_t block_id,
 			      struct r600_resource *bo,
 			      enum radeon_bo_usage usage)
 {
@@ -748,7 +926,6 @@ void _r600_pipe_state_add_reg(struct r600_context *ctx,
 	state->regs[state->nregs].id = (offset - block->start_offset) >> 2;
 
 	state->regs[state->nregs].value = value;
-	state->regs[state->nregs].mask = mask;
 	state->regs[state->nregs].bo = bo;
 	state->regs[state->nregs].bo_usage = usage;
 
@@ -757,7 +934,7 @@ void _r600_pipe_state_add_reg(struct r600_context *ctx,
 }
 
 void r600_pipe_state_add_reg_noblock(struct r600_pipe_state *state,
-				     u32 offset, u32 value, u32 mask,
+				     uint32_t offset, uint32_t value,
 				     struct r600_resource *bo,
 				     enum radeon_bo_usage usage)
 {
@@ -766,10 +943,121 @@ void r600_pipe_state_add_reg_noblock(struct r600_pipe_state *state,
 	state->regs[state->nregs].id = offset;
 	state->regs[state->nregs].block = NULL;
 	state->regs[state->nregs].value = value;
-	state->regs[state->nregs].mask = mask;
 	state->regs[state->nregs].bo = bo;
 	state->regs[state->nregs].bo_usage = usage;
 
 	state->nregs++;
 	assert(state->nregs < R600_BLOCK_MAX_REG);
 }
+
+uint32_t r600_translate_stencil_op(int s_op)
+{
+	switch (s_op) {
+	case PIPE_STENCIL_OP_KEEP:
+		return V_028800_STENCIL_KEEP;
+	case PIPE_STENCIL_OP_ZERO:
+		return V_028800_STENCIL_ZERO;
+	case PIPE_STENCIL_OP_REPLACE:
+		return V_028800_STENCIL_REPLACE;
+	case PIPE_STENCIL_OP_INCR:
+		return V_028800_STENCIL_INCR;
+	case PIPE_STENCIL_OP_DECR:
+		return V_028800_STENCIL_DECR;
+	case PIPE_STENCIL_OP_INCR_WRAP:
+		return V_028800_STENCIL_INCR_WRAP;
+	case PIPE_STENCIL_OP_DECR_WRAP:
+		return V_028800_STENCIL_DECR_WRAP;
+	case PIPE_STENCIL_OP_INVERT:
+		return V_028800_STENCIL_INVERT;
+	default:
+		R600_ERR("Unknown stencil op %d", s_op);
+		assert(0);
+		break;
+	}
+	return 0;
+}
+
+uint32_t r600_translate_fill(uint32_t func)
+{
+	switch(func) {
+	case PIPE_POLYGON_MODE_FILL:
+		return 2;
+	case PIPE_POLYGON_MODE_LINE:
+		return 1;
+	case PIPE_POLYGON_MODE_POINT:
+		return 0;
+	default:
+		assert(0);
+		return 0;
+	}
+}
+
+unsigned r600_tex_wrap(unsigned wrap)
+{
+	switch (wrap) {
+	default:
+	case PIPE_TEX_WRAP_REPEAT:
+		return V_03C000_SQ_TEX_WRAP;
+	case PIPE_TEX_WRAP_CLAMP:
+		return V_03C000_SQ_TEX_CLAMP_HALF_BORDER;
+	case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+		return V_03C000_SQ_TEX_CLAMP_LAST_TEXEL;
+	case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+		return V_03C000_SQ_TEX_CLAMP_BORDER;
+	case PIPE_TEX_WRAP_MIRROR_REPEAT:
+		return V_03C000_SQ_TEX_MIRROR;
+	case PIPE_TEX_WRAP_MIRROR_CLAMP:
+		return V_03C000_SQ_TEX_MIRROR_ONCE_HALF_BORDER;
+	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+		return V_03C000_SQ_TEX_MIRROR_ONCE_LAST_TEXEL;
+	case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+		return V_03C000_SQ_TEX_MIRROR_ONCE_BORDER;
+	}
+}
+
+unsigned r600_tex_filter(unsigned filter)
+{
+	switch (filter) {
+	default:
+	case PIPE_TEX_FILTER_NEAREST:
+		return V_03C000_SQ_TEX_XY_FILTER_POINT;
+	case PIPE_TEX_FILTER_LINEAR:
+		return V_03C000_SQ_TEX_XY_FILTER_BILINEAR;
+	}
+}
+
+unsigned r600_tex_mipfilter(unsigned filter)
+{
+	switch (filter) {
+	case PIPE_TEX_MIPFILTER_NEAREST:
+		return V_03C000_SQ_TEX_Z_FILTER_POINT;
+	case PIPE_TEX_MIPFILTER_LINEAR:
+		return V_03C000_SQ_TEX_Z_FILTER_LINEAR;
+	default:
+	case PIPE_TEX_MIPFILTER_NONE:
+		return V_03C000_SQ_TEX_Z_FILTER_NONE;
+	}
+}
+
+unsigned r600_tex_compare(unsigned compare)
+{
+	switch (compare) {
+	default:
+	case PIPE_FUNC_NEVER:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_NEVER;
+	case PIPE_FUNC_LESS:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_LESS;
+	case PIPE_FUNC_EQUAL:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_EQUAL;
+	case PIPE_FUNC_LEQUAL:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_LESSEQUAL;
+	case PIPE_FUNC_GREATER:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATER;
+	case PIPE_FUNC_NOTEQUAL:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_NOTEQUAL;
+	case PIPE_FUNC_GEQUAL:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL;
+	case PIPE_FUNC_ALWAYS:
+		return V_03C000_SQ_TEX_DEPTH_COMPARE_ALWAYS;
+	}
+}