r600g: atomize sampler state v2

Use atom for sampler state. Does not provide new functionality
or fix any bug. Just a step toward full atom base r600g.

v2: Split seamless on r6xx/r7xx into it's own atom. Make sure it's
    emited after sampler and with a pipeline flush before otherwise
    it does not take effect.

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

7 files changed, 262 insertions, 452 deletions

diff --git a/src/gallium/drivers/r600/evergreen_hw_context.c b/src/gallium/drivers/r600/evergreen_hw_context.c
index 199033f7ae3..6494786b048 100644
--- a/src/gallium/drivers/r600/evergreen_hw_context.c
+++ b/src/gallium/drivers/r600/evergreen_hw_context.c
@@ -575,37 +575,6 @@ static const struct r600_reg cayman_context_reg_list[] = {
 	{R_028EAC_CB_COLOR11_DIM, 0, 0},
 };
 
-/* SHADER SAMPLER BORDER EG/CM */
-static int evergreen_state_sampler_border_init(struct r600_context *ctx, uint32_t offset, unsigned id)
-{
-	struct r600_reg r600_shader_sampler_border[] = {
-		{R_00A400_TD_PS_SAMPLER0_BORDER_INDEX, 0, 0},
-		{R_00A404_TD_PS_SAMPLER0_BORDER_RED, 0, 0},
-		{R_00A408_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0},
-		{R_00A40C_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0},
-		{R_00A410_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0},
-	};
-	unsigned nreg = Elements(r600_shader_sampler_border);
-	unsigned fake_offset = (offset - R_00A400_TD_PS_SAMPLER0_BORDER_INDEX) * 0x100 + 0x40000 + id * 0x1C;
-	struct r600_range *range;
-	struct r600_block *block;
-	int r;
-
-	for (int i = 0; i < nreg; i++) {
-		r600_shader_sampler_border[i].offset -= R_00A400_TD_PS_SAMPLER0_BORDER_INDEX;
-		r600_shader_sampler_border[i].offset += fake_offset;
-	}
-	r = r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, 0);
-	if (r) {
-		return r;
-	}
-	/* set proper offset */
-	range = &ctx->range[CTX_RANGE_ID(r600_shader_sampler_border[0].offset)];
-	block = range->blocks[CTX_BLOCK_ID(r600_shader_sampler_border[0].offset)];
-	block->pm4[1] = (offset - EVERGREEN_CONFIG_REG_OFFSET) >> 2;
-	return 0;
-}
-
 static int evergreen_loop_const_init(struct r600_context *ctx, uint32_t offset)
 {
 	unsigned nreg = 32;
@@ -646,32 +615,6 @@ int evergreen_context_init(struct r600_context *ctx)
 	if (r)
 		goto out_err;
 
-
-	/* PS SAMPLER */
-	for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) {
-		r = r600_state_sampler_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-	/* VS SAMPLER */
-	for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) {
-		r = r600_state_sampler_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-	/* PS SAMPLER BORDER */
-	for (int j = 0; j < 18; j++) {
-		r = evergreen_state_sampler_border_init(ctx, R_00A400_TD_PS_SAMPLER0_BORDER_INDEX, j);
-		if (r)
-			goto out_err;
-	}
-	/* VS SAMPLER BORDER */
-	for (int j = 0; j < 18; j++) {
-		r = evergreen_state_sampler_border_init(ctx, R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, j);
-		if (r)
-			goto out_err;
-	}
-
 	/* PS loop const */
 	evergreen_loop_const_init(ctx, 0);
 	/* VS loop const */
@@ -688,66 +631,6 @@ out_err:
 	return r;
 }
 
-static inline void evergreen_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset, unsigned id)
-{
-	unsigned fake_offset = (offset - R_00A400_TD_PS_SAMPLER0_BORDER_INDEX) * 0x100 + 0x40000 + id * 0x1C;
-	struct r600_range *range;
-	struct r600_block *block;
-	int i;
-	int dirty;
-
-	range = &ctx->range[CTX_RANGE_ID(fake_offset)];
-	block = range->blocks[CTX_BLOCK_ID(fake_offset)];
-	if (state == NULL) {
-		block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
-		LIST_DELINIT(&block->list);
-		LIST_DELINIT(&block->enable_list);
-		return;
-	}
-	if (state->nregs <= 3) {
-		return;
-	}
-
-	dirty = block->status & R600_BLOCK_STATUS_DIRTY;
-	if (block->reg[0] != id) {
-		block->reg[0] = id;
-		dirty |= R600_BLOCK_STATUS_DIRTY;
-	}
-
-	for (i = 1; i < 5; i++) {
-		if (block->reg[i] != state->regs[i + 2].value) {
-			block->reg[i] = state->regs[i + 2].value;
-			dirty |= R600_BLOCK_STATUS_DIRTY;
-		}
-	}
-
-	/* We have to flush the shaders before we change the border color
-	 * registers, or previous draw commands that haven't completed yet
-	 * will end up using the new border color. */
-	if (dirty & R600_BLOCK_STATUS_DIRTY)
-		r600_context_ps_partial_flush(ctx);
-	if (dirty)
-		r600_context_dirty_block(ctx, block, dirty, 4);
-}
-
-void evergreen_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
-{
-	unsigned offset;
-
-	offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*id;
-	r600_context_pipe_state_set_sampler(ctx, state, offset);
-	evergreen_context_pipe_state_set_sampler_border(ctx, state, R_00A400_TD_PS_SAMPLER0_BORDER_INDEX, id);
-}
-
-void evergreen_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
-{
-	unsigned offset;
-
-	offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*(id + 18);
-	r600_context_pipe_state_set_sampler(ctx, state, offset);
-	evergreen_context_pipe_state_set_sampler_border(ctx, state, R_00A414_TD_VS_SAMPLER0_BORDER_INDEX, id);
-}
-
 void evergreen_flush_vgt_streamout(struct r600_context *ctx)
 {
 	struct radeon_winsys_cs *cs = ctx->cs;
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 340036c647c..67ae7d3c04c 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -911,43 +911,47 @@ static void *evergreen_create_rs_state(struct pipe_context *ctx,
 static void *evergreen_create_sampler_state(struct pipe_context *ctx,
 					const struct pipe_sampler_state *state)
 {
-	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
+	struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
 	union util_color uc;
 	unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 2 : 0;
 
-	if (rstate == NULL) {
+	if (ss == NULL) {
 		return NULL;
 	}
 
-	rstate->id = R600_PIPE_STATE_SAMPLER;
+	/* directly into sampler avoid r6xx code to emit useless reg */
+	ss->seamless_cube_map = false;
 	util_pack_color(state->border_color.f, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C000_SQ_TEX_SAMPLER_WORD0_0,
-			S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
-			S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
-			S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
-			S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
-			S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
-			S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
-			S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
-			S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
-			S_03C000_BORDER_COLOR_TYPE(uc.ui ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0), NULL, 0);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C004_SQ_TEX_SAMPLER_WORD1_0,
-			S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
-			S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8)),
-			NULL, 0);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0,
-					S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
-					(state->seamless_cube_map ? 0 : S_03C008_DISABLE_CUBE_WRAP(1)) |
-					S_03C008_TYPE(1),
-					NULL, 0);
-
+	ss->border_color_use = false;
+	/* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
+	ss->tex_sampler_words[0] = S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
+				S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
+				S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
+				S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
+				S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
+				S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
+				S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
+				S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
+				S_03C000_BORDER_COLOR_TYPE(uc.ui ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
+	/* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
+	ss->tex_sampler_words[1] = S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 8)) |
+				S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 8));
+	/* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
+	ss->tex_sampler_words[2] = S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
+				(state->seamless_cube_map ? 0 : S_03C008_DISABLE_CUBE_WRAP(1)) |
+				S_03C008_TYPE(1);
 	if (uc.ui) {
-		r600_pipe_state_add_reg_noblock(rstate, R_00A404_TD_PS_SAMPLER0_BORDER_RED, fui(state->border_color.f[0]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A408_TD_PS_SAMPLER0_BORDER_GREEN, fui(state->border_color.f[1]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A40C_TD_PS_SAMPLER0_BORDER_BLUE, fui(state->border_color.f[2]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A410_TD_PS_SAMPLER0_BORDER_ALPHA, fui(state->border_color.f[3]), NULL, 0);
-	}
-	return rstate;
+		ss->border_color_use = true;
+		/* R_00A400_TD_PS_SAMPLER0_BORDER_RED */
+		ss->border_color[0] = fui(state->border_color.f[0]);
+		/* R_00A404_TD_PS_SAMPLER0_BORDER_GREEN */
+		ss->border_color[1] = fui(state->border_color.f[1]);
+		/* R_00A408_TD_PS_SAMPLER0_BORDER_BLUE */
+		ss->border_color[2] = fui(state->border_color.f[2]);
+		/* R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA */
+		ss->border_color[3] = fui(state->border_color.f[3]);
+	}
+	return ss;
 }
 
 static struct pipe_sampler_view *evergreen_create_sampler_view(struct pipe_context *ctx,
@@ -1092,37 +1096,6 @@ static void evergreen_set_ps_sampler_views(struct pipe_context *ctx, unsigned co
 	r600_set_sampler_views(rctx, &rctx->ps_samplers, count, views);
 }
 
-static void evergreen_bind_samplers(struct r600_context *rctx,
-				    struct r600_textures_info *dst,
-				    unsigned count, void **states,
-				    void (*set_sampler)(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id))
-{
-	struct r600_pipe_sampler_state **rstates = (struct r600_pipe_sampler_state**)states;
-
-	for (int i = 0; i < count; i++) {
-		if (rstates[i] != dst->samplers[i]) {
-			set_sampler(rctx, &rstates[i]->rstate, i);
-		}
-	}
-
-	memcpy(dst->samplers, states, sizeof(void*) * count);
-	dst->n_samplers = count;
-}
-
-static void evergreen_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states)
-{
-	struct r600_context *rctx = (struct r600_context *)ctx;
-	evergreen_bind_samplers(rctx, &rctx->ps_samplers, count, states,
-				evergreen_context_pipe_state_set_ps_sampler);
-}
-
-static void evergreen_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states)
-{
-	struct r600_context *rctx = (struct r600_context *)ctx;
-	evergreen_bind_samplers(rctx, &rctx->vs_samplers, count, states,
-				evergreen_context_pipe_state_set_vs_sampler);
-}
-
 static void evergreen_set_clip_state(struct pipe_context *ctx,
 				const struct pipe_clip_state *state)
 {
@@ -1822,6 +1795,41 @@ static void evergreen_emit_ps_sampler_views(struct r600_context *rctx, struct r6
 	evergreen_emit_sampler_views(rctx, &rctx->ps_samplers.views, R600_MAX_CONST_BUFFERS);
 }
 
+static void evergreen_emit_sampler(struct r600_context *rctx,
+				struct r600_textures_info *texinfo,
+				unsigned resource_id_base,
+				unsigned border_index_reg)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	unsigned i;
+
+	for (i = 0; i < texinfo->n_samplers; i++) {
+
+		if (texinfo->samplers[i] == NULL) {
+			continue;
+		}
+		r600_write_value(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
+		r600_write_value(cs, (resource_id_base + i) * 3);
+		r600_write_array(cs, 3, texinfo->samplers[i]->tex_sampler_words);
+
+		if (texinfo->samplers[i]->border_color_use) {
+			r600_write_config_reg_seq(cs, border_index_reg, 5);
+			r600_write_value(cs, i);
+			r600_write_array(cs, 4, texinfo->samplers[i]->border_color);
+		}
+	}
+}
+
+static void evergreen_emit_vs_sampler(struct r600_context *rctx, struct r600_atom *atom)
+{
+	evergreen_emit_sampler(rctx, &rctx->vs_samplers, 18, R_00A414_TD_VS_SAMPLER0_BORDER_INDEX);
+}
+
+static void evergreen_emit_ps_sampler(struct r600_context *rctx, struct r600_atom *atom)
+{
+	evergreen_emit_sampler(rctx, &rctx->ps_samplers, 0, R_00A400_TD_PS_SAMPLER0_BORDER_INDEX);
+}
+
 void evergreen_init_state_functions(struct r600_context *rctx)
 {
 	r600_init_atom(&rctx->cb_misc_state.atom, evergreen_emit_cb_misc_state, 0, 0);
@@ -1835,6 +1843,8 @@ void evergreen_init_state_functions(struct r600_context *rctx)
 	r600_init_atom(&rctx->vs_samplers.views.atom, evergreen_emit_vs_sampler_views, 0, 0);
 	r600_init_atom(&rctx->ps_samplers.views.atom, evergreen_emit_ps_sampler_views, 0, 0);
 	r600_init_atom(&rctx->cs_shader_state.atom, evergreen_emit_cs_shader, 0, 0);
+	r600_init_atom(&rctx->vs_samplers.atom_sampler, evergreen_emit_vs_sampler, 0, 0);
+	r600_init_atom(&rctx->ps_samplers.atom_sampler, evergreen_emit_ps_sampler, 0, 0);
 
 	rctx->context.create_blend_state = evergreen_create_blend_state;
 	rctx->context.create_depth_stencil_alpha_state = evergreen_create_dsa_state;
@@ -1846,17 +1856,17 @@ void evergreen_init_state_functions(struct r600_context *rctx)
 	rctx->context.create_vs_state = r600_create_shader_state_vs;
 	rctx->context.bind_blend_state = r600_bind_blend_state;
 	rctx->context.bind_depth_stencil_alpha_state = r600_bind_dsa_state;
-	rctx->context.bind_fragment_sampler_states = evergreen_bind_ps_samplers;
+	rctx->context.bind_fragment_sampler_states = r600_bind_ps_samplers;
 	rctx->context.bind_fs_state = r600_bind_ps_shader;
 	rctx->context.bind_rasterizer_state = r600_bind_rs_state;
 	rctx->context.bind_vertex_elements_state = r600_bind_vertex_elements;
-	rctx->context.bind_vertex_sampler_states = evergreen_bind_vs_samplers;
+	rctx->context.bind_vertex_sampler_states = r600_bind_vs_samplers;
 	rctx->context.bind_vs_state = r600_bind_vs_shader;
 	rctx->context.delete_blend_state = r600_delete_state;
 	rctx->context.delete_depth_stencil_alpha_state = r600_delete_state;
 	rctx->context.delete_fs_state = r600_delete_ps_shader;
 	rctx->context.delete_rasterizer_state = r600_delete_rs_state;
-	rctx->context.delete_sampler_state = r600_delete_state;
+	rctx->context.delete_sampler_state = r600_delete_sampler;
 	rctx->context.delete_vertex_elements_state = r600_delete_vertex_element;
 	rctx->context.delete_vs_state = r600_delete_vs_shader;
 	rctx->context.set_blend_color = r600_set_blend_color;
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 14993569d17..1c8bd243256 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -182,6 +182,7 @@ struct r600_so_target {
 
 #define R600_CONTEXT_DRAW_PENDING	(1 << 0)
 #define R600_CONTEXT_DST_CACHES_DIRTY	(1 << 1)
+#define R600_PARTIAL_FLUSH		(1 << 2)
 
 struct r600_context;
 struct r600_screen;
@@ -191,8 +192,6 @@ int r600_context_init(struct r600_context *ctx);
 void r600_context_fini(struct r600_context *ctx);
 void r600_context_pipe_state_emit(struct r600_context *ctx, struct r600_pipe_state *state, unsigned pkt_flags);
 void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_state *state);
-void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
-void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 void r600_context_flush(struct r600_context *ctx, unsigned flags);
 
 void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence,
@@ -208,8 +207,6 @@ void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw, boolean count
 void r600_context_block_emit_dirty(struct r600_context *ctx, struct r600_block *block, unsigned pkt_flags);
 
 int evergreen_context_init(struct r600_context *ctx);
-void evergreen_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
-void evergreen_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id);
 
 void _r600_pipe_state_add_reg_bo(struct r600_context *ctx,
 				 struct r600_pipe_state *state,
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index 017e60d20fd..a5e330f391d 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -241,7 +241,6 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg,
 static const struct r600_reg r600_config_reg_list[] = {
 	{R_008958_VGT_PRIMITIVE_TYPE, 0, 0},
 	{R_008C04_SQ_GPR_RESOURCE_MGMT_1, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
-	{R_009508_TA_CNTL_AUX, REG_FLAG_ENABLE_ALWAYS | REG_FLAG_FLUSH_CHANGE, 0},
 };
 
 static const struct r600_reg r600_ctl_const_list[] = {
@@ -506,39 +505,6 @@ static const struct r600_reg r600_context_reg_list[] = {
 	{R_028A94_VGT_MULTI_PRIM_IB_RESET_EN, 0, 0},
 };
 
-/* SHADER SAMPLER R600/R700/EG/CM */
-int r600_state_sampler_init(struct r600_context *ctx, uint32_t offset)
-{
-	struct r600_reg r600_shader_sampler[] = {
-		{R_03C000_SQ_TEX_SAMPLER_WORD0_0, 0, 0},
-		{R_03C004_SQ_TEX_SAMPLER_WORD1_0, 0, 0},
-		{R_03C008_SQ_TEX_SAMPLER_WORD2_0, 0, 0},
-	};
-	unsigned nreg = Elements(r600_shader_sampler);
-
-	for (int i = 0; i < nreg; i++) {
-		r600_shader_sampler[i].offset += offset;
-	}
-	return r600_context_add_block(ctx, r600_shader_sampler, nreg, PKT3_SET_SAMPLER, R600_SAMPLER_OFFSET);
-}
-
-/* SHADER SAMPLER BORDER R600/R700 */
-static int r600_state_sampler_border_init(struct r600_context *ctx, uint32_t offset)
-{
-	struct r600_reg r600_shader_sampler_border[] = {
-		{R_00A400_TD_PS_SAMPLER0_BORDER_RED, 0, 0},
-		{R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, 0, 0},
-		{R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, 0, 0},
-		{R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, 0, 0},
-	};
-	unsigned nreg = Elements(r600_shader_sampler_border);
-
-	for (int i = 0; i < nreg; i++) {
-		r600_shader_sampler_border[i].offset += offset;
-	}
-	return r600_context_add_block(ctx, r600_shader_sampler_border, nreg, PKT3_SET_CONFIG_REG, R600_CONFIG_REG_OFFSET);
-}
-
 static int r600_loop_const_init(struct r600_context *ctx, uint32_t offset)
 {
 	unsigned nreg = 32;
@@ -631,32 +597,6 @@ int r600_context_init(struct r600_context *ctx)
 	if (r)
 		goto out_err;
 
-	/* PS SAMPLER BORDER */
-	for (int j = 0, offset = 0; j < 18; j++, offset += 0x10) {
-		r = r600_state_sampler_border_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-
-	/* VS SAMPLER BORDER */
-	for (int j = 0, offset = 0x200; j < 18; j++, offset += 0x10) {
-		r = r600_state_sampler_border_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-	/* PS SAMPLER */
-	for (int j = 0, offset = 0; j < 18; j++, offset += 0xC) {
-		r = r600_state_sampler_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-	/* VS SAMPLER */
-	for (int j = 0, offset = 0xD8; j < 18; j++, offset += 0xC) {
-		r = r600_state_sampler_init(ctx, offset);
-		if (r)
-			goto out_err;
-	}
-
 	/* PS loop const */
 	r600_loop_const_init(ctx, 0);
 	/* VS loop const */
@@ -836,89 +776,6 @@ void r600_context_pipe_state_set(struct r600_context *ctx, struct r600_pipe_stat
 	}
 }
 
-void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
-{
-	struct r600_range *range;
-	struct r600_block *block;
-	int i;
-	int dirty;
-
-	range = &ctx->range[CTX_RANGE_ID(offset)];
-	block = range->blocks[CTX_BLOCK_ID(offset)];
-	if (state == NULL) {
-		block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
-		LIST_DELINIT(&block->list);
-		LIST_DELINIT(&block->enable_list);
-		return;
-	}
-	dirty = block->status & R600_BLOCK_STATUS_DIRTY;
-
-	for (i = 0; i < 3; i++) {
-		if (block->reg[i] != state->regs[i].value) {
-			block->reg[i] = state->regs[i].value;
-			dirty |= R600_BLOCK_STATUS_DIRTY;
-		}
-	}
-
-	if (dirty)
-		r600_context_dirty_block(ctx, block, dirty, 2);
-}
-
-static inline void r600_context_pipe_state_set_sampler_border(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset)
-{
-	struct r600_range *range;
-	struct r600_block *block;
-	int i;
-	int dirty;
-
-	range = &ctx->range[CTX_RANGE_ID(offset)];
-	block = range->blocks[CTX_BLOCK_ID(offset)];
-	if (state == NULL) {
-		block->status &= ~(R600_BLOCK_STATUS_ENABLED | R600_BLOCK_STATUS_DIRTY);
-		LIST_DELINIT(&block->list);
-		LIST_DELINIT(&block->enable_list);
-		return;
-	}
-	if (state->nregs <= 3) {
-		return;
-	}
-	dirty = block->status & R600_BLOCK_STATUS_DIRTY;
-	for (i = 0; i < 4; i++) {
-		if (block->reg[i] != state->regs[i + 3].value) {
-			block->reg[i] = state->regs[i + 3].value;
-			dirty |= R600_BLOCK_STATUS_DIRTY;
-		}
-	}
-
-	/* We have to flush the shaders before we change the border color
-	 * registers, or previous draw commands that haven't completed yet
-	 * will end up using the new border color. */
-	if (dirty & R600_BLOCK_STATUS_DIRTY)
-		r600_context_ps_partial_flush(ctx);
-	if (dirty)
-		r600_context_dirty_block(ctx, block, dirty, 3);
-}
-
-void r600_context_pipe_state_set_ps_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
-{
-	unsigned offset;
-
-	offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*id;
-	r600_context_pipe_state_set_sampler(ctx, state, offset);
-	offset = R_00A400_TD_PS_SAMPLER0_BORDER_RED + 16*id;
-	r600_context_pipe_state_set_sampler_border(ctx, state, offset);
-}
-
-void r600_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned id)
-{
-	unsigned offset;
-
-	offset = R_03C000_SQ_TEX_SAMPLER_WORD0_0 + 12*(id + 18);
-	r600_context_pipe_state_set_sampler(ctx, state, offset);
-	offset = R_00A600_TD_VS_SAMPLER0_BORDER_RED + 16*id;
-	r600_context_pipe_state_set_sampler_border(ctx, state, offset);
-}
-
 /**
  * @param pkt_flags should be set to RADEON_CP_PACKET3_COMPUTE_MODE if this
  * block will be used for compute shaders.
@@ -1091,6 +948,12 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 	r600_atom_dirty(ctx, &ctx->alphatest_state.atom);
 	r600_atom_dirty(ctx, &ctx->cb_misc_state.atom);
 	r600_atom_dirty(ctx, &ctx->db_misc_state.atom);
+	/* reemit sampler, will only matter if atom_sampler.num_dw != 0 */
+	r600_atom_dirty(ctx, &ctx->vs_samplers.atom_sampler);
+	r600_atom_dirty(ctx, &ctx->ps_samplers.atom_sampler);
+	if (ctx->chip_class <= R700) {
+		r600_atom_dirty(ctx, &ctx->seamless_cube_map.atom);
+	}
 
 	ctx->vertex_buffer_state.dirty_mask = ctx->vertex_buffer_state.enabled_mask;
 	r600_vertex_buffers_dirty(ctx);
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index f41fa8fc7f6..04641838e0b 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -240,15 +240,21 @@ struct r600_pipe_shader {
 };
 
 struct r600_pipe_sampler_state {
-	struct r600_pipe_state		rstate;
-	boolean seamless_cube_map;
+	uint32_t			tex_sampler_words[3];
+	uint32_t			border_color[4];
+	bool				border_color_use;
+	bool				seamless_cube_map;
 };
 
 /* needed for blitter save */
 #define NUM_TEX_UNITS 16
 
-struct r600_samplerview_state
-{
+struct r600_seamless_cube_map {
+	struct r600_atom		atom;
+	bool				enabled;
+};
+
+struct r600_samplerview_state {
 	struct r600_atom		atom;
 	struct r600_pipe_sampler_view	*views[NUM_TEX_UNITS];
 	uint32_t			enabled_mask;
@@ -258,17 +264,16 @@ struct r600_samplerview_state
 
 struct r600_textures_info {
 	struct r600_samplerview_state	views;
-
+	struct r600_atom		atom_sampler;
 	struct r600_pipe_sampler_state	*samplers[NUM_TEX_UNITS];
 	unsigned			n_samplers;
-	bool				samplers_dirty;
 	bool				is_array_sampler[NUM_TEX_UNITS];
 };
 
 struct r600_fence {
 	struct pipe_reference		reference;
 	unsigned			index; /* in the shared bo */
-	struct r600_resource            *sleep_bo;
+	struct r600_resource		*sleep_bo;
 	struct list_head		head;
 };
 
@@ -367,6 +372,7 @@ struct r600_context {
 	struct r600_constbuf_state	ps_constbuf_state;
 	struct r600_textures_info	vs_samplers;
 	struct r600_textures_info	ps_samplers;
+	struct r600_seamless_cube_map	seamless_cube_map;
 	struct r600_cs_shader_state	cs_shader_state;
 
 	struct radeon_winsys_cs	*cs;
@@ -502,7 +508,6 @@ void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader
 /* r600_state.c */
 void r600_set_scissor_state(struct r600_context *rctx,
 			    const struct pipe_scissor_state *state);
-void r600_update_sampler_states(struct r600_context *rctx);
 void r600_init_state_functions(struct r600_context *rctx);
 void r600_init_atom_start_cs(struct r600_context *rctx);
 void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
@@ -550,6 +555,8 @@ void r600_set_sampler_views(struct r600_context *rctx,
 			    struct r600_textures_info *dst,
 			    unsigned count,
 			    struct pipe_sampler_view **views);
+void r600_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states);
+void r600_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states);
 void *r600_create_vertex_elements(struct pipe_context *ctx,
 				  unsigned count,
 				  const struct pipe_vertex_element *elements);
@@ -563,6 +570,7 @@ void r600_bind_rs_state(struct pipe_context *ctx, void *state);
 void r600_delete_rs_state(struct pipe_context *ctx, void *state);
 void r600_sampler_view_destroy(struct pipe_context *ctx,
 			       struct pipe_sampler_view *state);
+void r600_delete_sampler(struct pipe_context *ctx, void *state);
 void r600_delete_state(struct pipe_context *ctx, void *state);
 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
 void *r600_create_shader_state_ps(struct pipe_context *ctx,
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 842808a990b..9c7ff7febb5 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -932,7 +932,6 @@ static void *r600_create_sampler_state(struct pipe_context *ctx,
 					const struct pipe_sampler_state *state)
 {
 	struct r600_pipe_sampler_state *ss = CALLOC_STRUCT(r600_pipe_sampler_state);
-	struct r600_pipe_state *rstate;
 	union util_color uc;
 	unsigned aniso_flag_offset = state->max_anisotropy > 1 ? 4 : 0;
 
@@ -941,31 +940,36 @@ static void *r600_create_sampler_state(struct pipe_context *ctx,
 	}
 
 	ss->seamless_cube_map = state->seamless_cube_map;
-	rstate = &ss->rstate;
-	rstate->id = R600_PIPE_STATE_SAMPLER;
+	ss->border_color_use = false;
 	util_pack_color(state->border_color.f, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C000_SQ_TEX_SAMPLER_WORD0_0,
-					S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
-					S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
-					S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
-					S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
-					S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
-					S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
-					S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
-					S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
-					S_03C000_BORDER_COLOR_TYPE(uc.ui ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0), NULL, 0);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C004_SQ_TEX_SAMPLER_WORD1_0,
-					S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
-					S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
-					S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6)), NULL, 0);
-	r600_pipe_state_add_reg_noblock(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0, S_03C008_TYPE(1), NULL, 0);
+	/* R_03C000_SQ_TEX_SAMPLER_WORD0_0 */
+	ss->tex_sampler_words[0] = S_03C000_CLAMP_X(r600_tex_wrap(state->wrap_s)) |
+				S_03C000_CLAMP_Y(r600_tex_wrap(state->wrap_t)) |
+				S_03C000_CLAMP_Z(r600_tex_wrap(state->wrap_r)) |
+				S_03C000_XY_MAG_FILTER(r600_tex_filter(state->mag_img_filter) | aniso_flag_offset) |
+				S_03C000_XY_MIN_FILTER(r600_tex_filter(state->min_img_filter) | aniso_flag_offset) |
+				S_03C000_MIP_FILTER(r600_tex_mipfilter(state->min_mip_filter)) |
+				S_03C000_MAX_ANISO(r600_tex_aniso_filter(state->max_anisotropy)) |
+				S_03C000_DEPTH_COMPARE_FUNCTION(r600_tex_compare(state->compare_func)) |
+				S_03C000_BORDER_COLOR_TYPE(uc.ui ? V_03C000_SQ_TEX_BORDER_COLOR_REGISTER : 0);
+	/* R_03C004_SQ_TEX_SAMPLER_WORD1_0 */
+	ss->tex_sampler_words[1] = S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
+				S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
+				S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6));
+	/* R_03C008_SQ_TEX_SAMPLER_WORD2_0 */
+	ss->tex_sampler_words[2] = S_03C008_TYPE(1);
 	if (uc.ui) {
-		r600_pipe_state_add_reg_noblock(rstate, R_00A400_TD_PS_SAMPLER0_BORDER_RED, fui(state->border_color.f[0]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, fui(state->border_color.f[1]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A408_TD_PS_SAMPLER0_BORDER_BLUE, fui(state->border_color.f[2]), NULL, 0);
-		r600_pipe_state_add_reg_noblock(rstate, R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA, fui(state->border_color.f[3]), NULL, 0);
-	}
-	return rstate;
+		ss->border_color_use = true;
+		/* R_00A400_TD_PS_SAMPLER0_BORDER_RED */
+		ss->border_color[0] = fui(state->border_color.f[0]);
+		/* R_00A404_TD_PS_SAMPLER0_BORDER_GREEN */
+		ss->border_color[1] = fui(state->border_color.f[1]);
+		/* R_00A408_TD_PS_SAMPLER0_BORDER_BLUE */
+		ss->border_color[2] = fui(state->border_color.f[2]);
+		/* R_00A40C_TD_PS_SAMPLER0_BORDER_ALPHA */
+		ss->border_color[3] = fui(state->border_color.f[3]);
+	}
+	return ss;
 }
 
 static struct pipe_sampler_view *r600_create_sampler_view(struct pipe_context *ctx,
@@ -1086,93 +1090,6 @@ static void r600_set_ps_sampler_views(struct pipe_context *ctx, unsigned count,
 	r600_set_sampler_views(rctx, &rctx->ps_samplers, count, views);
 }
 
-static void r600_set_seamless_cubemap(struct r600_context *rctx, boolean enable)
-{
-	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
-	if (rstate == NULL)
-		return;
-
-	rstate->id = R600_PIPE_STATE_SEAMLESS_CUBEMAP;
-	r600_pipe_state_add_reg(rstate, R_009508_TA_CNTL_AUX,
-				(enable ? 0 : S_009508_DISABLE_CUBE_WRAP(1)) |
-				S_009508_DISABLE_CUBE_ANISO(1) |
-				S_009508_SYNC_GRADIENT(1) |
-				S_009508_SYNC_WALKER(1) |
-				S_009508_SYNC_ALIGNER(1));
-
-	free(rctx->states[R600_PIPE_STATE_SEAMLESS_CUBEMAP]);
-	rctx->states[R600_PIPE_STATE_SEAMLESS_CUBEMAP] = rstate;
-	r600_context_pipe_state_set(rctx, rstate);
-}
-
-static void r600_bind_samplers(struct r600_context *rctx,
-			       struct r600_textures_info *dst,
-			       unsigned count, void **states)
-{
-	memcpy(dst->samplers, states, sizeof(void*) * count);
-	dst->n_samplers = count;
-	dst->samplers_dirty = true;
-}
-
-static void r600_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states)
-{
-	struct r600_context *rctx = (struct r600_context *)ctx;
-	r600_bind_samplers(rctx, &rctx->vs_samplers, count, states);
-}
-
-static void r600_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states)
-{
-	struct r600_context *rctx = (struct r600_context *)ctx;
-	r600_bind_samplers(rctx, &rctx->ps_samplers, count, states);
-}
-
-static void r600_update_samplers(struct r600_context *rctx,
-				 struct r600_textures_info *tex,
-				 void (*set_sampler)(struct r600_context*, struct r600_pipe_state*, unsigned))
-{
-	unsigned i;
-
-	if (tex->samplers_dirty) {
-		int seamless = -1;
-		for (i = 0; i < tex->n_samplers; i++) {
-			if (!tex->samplers[i])
-				continue;
-
-			/* TEX_ARRAY_OVERRIDE must be set for array textures to disable
-			 * filtering between layers.
-			 * Don't update TEX_ARRAY_OVERRIDE if we don't have the sampler view. */
-			if (tex->views.views[i]) {
-				if (tex->views.views[i]->base.texture->target == PIPE_TEXTURE_1D_ARRAY ||
-				    tex->views.views[i]->base.texture->target == PIPE_TEXTURE_2D_ARRAY) {
-					tex->samplers[i]->rstate.regs[0].value |= S_03C000_TEX_ARRAY_OVERRIDE(1);
-					tex->is_array_sampler[i] = true;
-				} else {
-					tex->samplers[i]->rstate.regs[0].value &= C_03C000_TEX_ARRAY_OVERRIDE;
-					tex->is_array_sampler[i] = false;
-				}
-			}
-
-			set_sampler(rctx, &tex->samplers[i]->rstate, i);
-
-			if (tex->samplers[i])
-				seamless = tex->samplers[i]->seamless_cube_map;
-		}
-
-		if (seamless != -1)
-			r600_set_seamless_cubemap(rctx, seamless);
-
-		tex->samplers_dirty = false;
-	}
-}
-
-void r600_update_sampler_states(struct r600_context *rctx)
-{
-	r600_update_samplers(rctx, &rctx->vs_samplers,
-			     r600_context_pipe_state_set_vs_sampler);
-	r600_update_samplers(rctx, &rctx->ps_samplers,
-			     r600_context_pipe_state_set_ps_sampler);
-}
-
 static void r600_set_clip_state(struct pipe_context *ctx,
 				const struct pipe_clip_state *state)
 {
@@ -1764,8 +1681,79 @@ static void r600_emit_ps_sampler_views(struct r600_context *rctx, struct r600_at
 	r600_emit_sampler_views(rctx, &rctx->ps_samplers.views, R600_MAX_CONST_BUFFERS);
 }
 
+static void r600_emit_sampler(struct r600_context *rctx,
+				struct r600_textures_info *texinfo,
+				unsigned resource_id_base,
+				unsigned border_color_reg)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	unsigned i;
+
+	for (i = 0; i < texinfo->n_samplers; i++) {
+
+		if (texinfo->samplers[i] == NULL) {
+			continue;
+		}
+
+		/* TEX_ARRAY_OVERRIDE must be set for array textures to disable
+		 * filtering between layers.
+		 * Don't update TEX_ARRAY_OVERRIDE if we don't have the sampler view.
+		 */
+		if (texinfo->views.views[i]) {
+			if (texinfo->views.views[i]->base.texture->target == PIPE_TEXTURE_1D_ARRAY ||
+			    texinfo->views.views[i]->base.texture->target == PIPE_TEXTURE_2D_ARRAY) {
+				texinfo->samplers[i]->tex_sampler_words[0] |= S_03C000_TEX_ARRAY_OVERRIDE(1);
+				texinfo->is_array_sampler[i] = true;
+			} else {
+				texinfo->samplers[i]->tex_sampler_words[0] &= C_03C000_TEX_ARRAY_OVERRIDE;
+				texinfo->is_array_sampler[i] = false;
+			}
+		}
+
+		r600_write_value(cs, PKT3(PKT3_SET_SAMPLER, 3, 0));
+		r600_write_value(cs, (resource_id_base + i) * 3);
+		r600_write_array(cs, 3, texinfo->samplers[i]->tex_sampler_words);
+
+		if (texinfo->samplers[i]->border_color_use) {
+			unsigned offset;
+
+			offset = border_color_reg;
+			offset += i * 16;
+			r600_write_config_reg_seq(cs, offset, 4);
+			r600_write_array(cs, 4, texinfo->samplers[i]->border_color);
+		}
+	}
+}
+
+static void r600_emit_vs_sampler(struct r600_context *rctx, struct r600_atom *atom)
+{
+	r600_emit_sampler(rctx, &rctx->vs_samplers, 18, R_00A600_TD_VS_SAMPLER0_BORDER_RED);
+}
+
+static void r600_emit_ps_sampler(struct r600_context *rctx, struct r600_atom *atom)
+{
+	r600_emit_sampler(rctx, &rctx->ps_samplers, 0, R_00A400_TD_PS_SAMPLER0_BORDER_RED);
+}
+
+static void r600_emit_seamless_cube_map(struct r600_context *rctx, struct r600_atom *atom)
+{
+	struct radeon_winsys_cs *cs = rctx->cs;
+	unsigned tmp;
+
+	tmp = S_009508_DISABLE_CUBE_ANISO(1) |
+		S_009508_SYNC_GRADIENT(1) |
+		S_009508_SYNC_WALKER(1) |
+		S_009508_SYNC_ALIGNER(1);
+	if (!rctx->seamless_cube_map.enabled) {
+		tmp |= S_009508_DISABLE_CUBE_WRAP(1);
+	}
+	r600_write_config_reg(cs, R_009508_TA_CNTL_AUX, tmp);
+}
+
 void r600_init_state_functions(struct r600_context *rctx)
 {
+	r600_init_atom(&rctx->seamless_cube_map.atom, r600_emit_seamless_cube_map, 3, 0);
+	r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
 	r600_init_atom(&rctx->cb_misc_state.atom, r600_emit_cb_misc_state, 0, 0);
 	r600_atom_dirty(rctx, &rctx->cb_misc_state.atom);
 	r600_init_atom(&rctx->db_misc_state.atom, r600_emit_db_misc_state, 4, 0);
@@ -1775,6 +1763,11 @@ void r600_init_state_functions(struct r600_context *rctx)
 	r600_init_atom(&rctx->ps_constbuf_state.atom, r600_emit_ps_constant_buffers, 0, 0);
 	r600_init_atom(&rctx->vs_samplers.views.atom, r600_emit_vs_sampler_views, 0, 0);
 	r600_init_atom(&rctx->ps_samplers.views.atom, r600_emit_ps_sampler_views, 0, 0);
+	/* sampler must be emited before TA_CNTL_AUX otherwise DISABLE_CUBE_WRAP change
+	 * does not take effect
+	 */
+	r600_init_atom(&rctx->vs_samplers.atom_sampler, r600_emit_vs_sampler, 0, EMIT_EARLY);
+	r600_init_atom(&rctx->ps_samplers.atom_sampler, r600_emit_ps_sampler, 0, EMIT_EARLY);
 
 	rctx->context.create_blend_state = r600_create_blend_state;
 	rctx->context.create_depth_stencil_alpha_state = r600_create_dsa_state;
@@ -1796,7 +1789,7 @@ void r600_init_state_functions(struct r600_context *rctx)
 	rctx->context.delete_depth_stencil_alpha_state = r600_delete_state;
 	rctx->context.delete_fs_state = r600_delete_ps_shader;
 	rctx->context.delete_rasterizer_state = r600_delete_rs_state;
-	rctx->context.delete_sampler_state = r600_delete_state;
+	rctx->context.delete_sampler_state = r600_delete_sampler;
 	rctx->context.delete_vertex_elements_state = r600_delete_vertex_element;
 	rctx->context.delete_vs_state = r600_delete_vs_shader;
 	rctx->context.set_blend_color = r600_set_blend_color;
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index f91f2654f58..4a75c144d8a 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -369,6 +369,59 @@ void r600_sampler_view_destroy(struct pipe_context *ctx,
 	FREE(resource);
 }
 
+static void r600_bind_samplers(struct r600_context *rctx,
+			       struct r600_textures_info *dst,
+			       unsigned count, void **states)
+{
+	int seamless_cube_map = -1;
+	unsigned i;
+
+	memcpy(dst->samplers, states, sizeof(void*) * count);
+	dst->n_samplers = count;
+	dst->atom_sampler.num_dw = 0;
+
+	for (i = 0; i < count; i++) {
+		struct r600_pipe_sampler_state *sampler = states[i];
+
+		if (sampler == NULL) {
+			continue;
+		}
+		if (sampler->border_color_use) {
+			dst->atom_sampler.num_dw += 11;
+			rctx->flags |= R600_PARTIAL_FLUSH;
+		} else {
+			dst->atom_sampler.num_dw += 5;
+		}
+		seamless_cube_map = sampler->seamless_cube_map;
+	}
+	if (rctx->chip_class <= R700 && seamless_cube_map != -1 && seamless_cube_map != rctx->seamless_cube_map.enabled) {
+		/* change in TA_CNTL_AUX need a pipeline flush */
+		rctx->flags |= R600_PARTIAL_FLUSH;
+		rctx->seamless_cube_map.enabled = seamless_cube_map;
+		r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
+	}
+	if (dst->atom_sampler.num_dw) {
+		r600_atom_dirty(rctx, &dst->atom_sampler);
+	}
+}
+
+void r600_bind_vs_samplers(struct pipe_context *ctx, unsigned count, void **states)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	r600_bind_samplers(rctx, &rctx->vs_samplers, count, states);
+}
+
+void r600_bind_ps_samplers(struct pipe_context *ctx, unsigned count, void **states)
+{
+	struct r600_context *rctx = (struct r600_context *)ctx;
+	r600_bind_samplers(rctx, &rctx->ps_samplers, count, states);
+}
+
+void r600_delete_sampler(struct pipe_context *ctx, void *state)
+{
+	free(state);
+}
+
 void r600_delete_state(struct pipe_context *ctx, void *state)
 {
 	struct r600_context *rctx = (struct r600_context *)ctx;
@@ -532,7 +585,7 @@ void r600_set_sampler_views(struct r600_context *rctx,
 			if (rctx->chip_class <= R700 &&
 			    (rviews[i]->base.texture->target == PIPE_TEXTURE_1D_ARRAY ||
 			     rviews[i]->base.texture->target == PIPE_TEXTURE_2D_ARRAY) != dst->is_array_sampler[i]) {
-				dst->samplers_dirty = true;
+				r600_atom_dirty(rctx, &dst->atom_sampler);
 			}
 
 			pipe_sampler_view_reference((struct pipe_sampler_view **)&dst->views.views[i], views[i]);
@@ -935,10 +988,6 @@ static void r600_update_derived_state(struct r600_context *rctx)
 		}
 	}
 
-	if (rctx->chip_class < EVERGREEN) {
-		r600_update_sampler_states(rctx);
-	}
-
 	r600_shader_select(ctx, rctx->ps_shader, &ps_dirty);
 
 	if (rctx->ps_shader && ((rctx->sprite_coord_enable &&
@@ -1011,6 +1060,13 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 
 	r600_update_derived_state(rctx);
 
+	/* partial flush triggered by border color change */
+	if (rctx->flags & R600_PARTIAL_FLUSH) {
+		rctx->flags &= ~R600_PARTIAL_FLUSH;
+		r600_write_value(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+		r600_write_value(cs, EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
+	}
+
 	if (info.indexed) {
 		/* Initialize the index buffer struct. */
 		pipe_resource_reference(&ib.buffer, rctx->index_buffer.buffer);