r600g: simplify flushing

Based on the patch called "simplify and fix flushing and synchronization"
by Jerome Glisse.

Rebased, removed unneded code, simplified more and cleaned up.

Also, SH_ACTION_ENA is not set when changing shaders (hw doesn't seem
to need it). It's only used to flush constant buffers.

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

12 files changed, 210 insertions, 190 deletions

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index 35333120654..1fb63d608a2 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -96,7 +96,7 @@ static void evergreen_cs_set_vertex_buffer(
 	vb->buffer = buffer;
 	vb->user_buffer = NULL;
 
-	r600_inval_vertex_cache(rctx);
+	rctx->flags |= rctx->has_vertex_cache ? R600_CONTEXT_VTX_FLUSH : R600_CONTEXT_TEX_FLUSH;
 	state->enabled_mask |= 1 << vb_index;
 	state->dirty_mask |= 1 << vb_index;
 	r600_atom_dirty(rctx, &state->atom);
@@ -332,8 +332,11 @@ static void compute_emit_cs(struct r600_context *ctx, const uint *block_layout,
 	 */
 	r600_emit_atom(ctx, &ctx->start_compute_cs_cmd.atom);
 
+	ctx->flags |= R600_CONTEXT_CB_FLUSH;
+	r600_flush_emit(ctx);
+
 	/* Emit cb_state */
-        cb_state = ctx->states[R600_PIPE_STATE_FRAMEBUFFER];
+	cb_state = ctx->states[R600_PIPE_STATE_FRAMEBUFFER];
 	r600_context_pipe_state_emit(ctx, cb_state, RADEON_CP_PACKET3_COMPUTE_MODE);
 
 	/* Set CB_TARGET_MASK  XXX: Use cb_misc_state */
@@ -384,15 +387,10 @@ static void compute_emit_cs(struct r600_context *ctx, const uint *block_layout,
 	/* Emit dispatch state and dispatch packet */
 	evergreen_emit_direct_dispatch(ctx, block_layout, grid_layout);
 
-	/* r600_flush_framebuffer() updates the cb_flush_flags and then
-	 * calls r600_emit_atom() on the ctx->surface_sync_cmd.atom, which emits
-	 * a SURFACE_SYNC packet via r600_emit_surface_sync().
-	 *
-	 * XXX r600_emit_surface_sync() hardcodes the CP_COHER_SIZE to
-	 * 0xffffffff, so we will need to add a field to struct
-	 * r600_surface_sync_cmd if we want to manually set this value.
+	/* XXX evergreen_flush_emit() hardcodes the CP_COHER_SIZE to 0xffffffff
 	 */
-	r600_flush_framebuffer(ctx, true /* Flush now */);
+	ctx->flags |= R600_CONTEXT_CB_FLUSH;
+	r600_flush_emit(ctx);
 
 #if 0
 	COMPUTE_DBG("cdw: %i\n", cs->cdw);
@@ -444,7 +442,7 @@ void evergreen_emit_cs_shader(
 	r600_write_value(cs, r600_context_bo_reloc(rctx, shader->shader_code_bo,
 							RADEON_USAGE_READ));
 
-	r600_inval_shader_cache(rctx);
+	rctx->flags |= R600_CONTEXT_SHADERCONST_FLUSH;
 }
 
 static void evergreen_launch_grid(
diff --git a/src/gallium/drivers/r600/evergreen_compute_internal.c b/src/gallium/drivers/r600/evergreen_compute_internal.c
index 50a60d361f2..dc957320685 100644
--- a/src/gallium/drivers/r600/evergreen_compute_internal.c
+++ b/src/gallium/drivers/r600/evergreen_compute_internal.c
@@ -562,7 +562,7 @@ void evergreen_set_tex_resource(
 			     util_format_get_blockwidth(tmp->resource.b.b.format) *
 			     view->base.texture->width0*height*depth;
 
-	r600_inval_texture_cache(pipe->ctx);
+	pipe->ctx->flags |= R600_CONTEXT_TEX_FLUSH;
 
 	evergreen_emit_force_reloc(res);
 	evergreen_emit_force_reloc(res);
@@ -621,7 +621,7 @@ void evergreen_set_const_cache(
 	res->usage = RADEON_USAGE_READ;
 	res->coher_bo_size = size;
 
-	r600_inval_shader_cache(pipe->ctx);
+	pipe->ctx->flags |= R600_CONTEXT_SHADERCONST_FLUSH;
 }
 
 struct r600_resource* r600_compute_buffer_alloc_vram(
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 405d1b03eeb..7386726ca16 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -1697,7 +1697,12 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
 	if (rstate == NULL)
 		return;
 
-	r600_flush_framebuffer(rctx, false);
+	if (rctx->framebuffer.nr_cbufs) {
+		rctx->flags |= R600_CONTEXT_CB_FLUSH;
+	}
+	if (rctx->framebuffer.zsbuf) {
+		rctx->flags |= R600_CONTEXT_DB_FLUSH;
+	}
 
 	/* unreference old buffer and reference new one */
 	rstate->id = R600_PIPE_STATE_FRAMEBUFFER;
diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
index e4d72f5411e..18e1eb7bc66 100644
--- a/src/gallium/drivers/r600/evergreend.h
+++ b/src/gallium/drivers/r600/evergreend.h
@@ -46,7 +46,8 @@
 #define EVENT_TYPE_PS_PARTIAL_FLUSH            0x10
 #define EVENT_TYPE_ZPASS_DONE                  0x15
 #define EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT   0x16
-#define EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH	0x1f
+#define EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH       0x1f
+#define EVENT_TYPE_FLUSH_AND_INV_DB_META       0x2c
 
 #define		EVENT_TYPE(x)                           ((x) << 0)
 #define		EVENT_INDEX(x)                          ((x) << 8)
@@ -2186,16 +2187,12 @@
 #define   C_0085F0_DB_DEST_BASE_ENA                    0xFFFFBFFF
 #define   S_0085F0_CB8_DEST_BASE_ENA(x)                (((x) & 0x1) << 15)
 #define   G_0085F0_CB8_DEST_BASE_ENA(x)                (((x) >> 15) & 0x1)
-
 #define   S_0085F0_CB9_DEST_BASE_ENA(x)                (((x) & 0x1) << 16)
 #define   G_0085F0_CB9_DEST_BASE_ENA(x)                (((x) >> 16) & 0x1)
-
 #define   S_0085F0_CB10_DEST_BASE_ENA(x)               (((x) & 0x1) << 17)
 #define   G_0085F0_CB10_DEST_BASE_ENA(x)               (((x) >> 17) & 0x1)
-
 #define   S_0085F0_CB11_DEST_BASE_ENA(x)               (((x) & 0x1) << 18)
 #define   G_0085F0_CB11_DEST_BASE_ENA(x)               (((x) >> 18) & 0x1)
-
 #define   S_0085F0_TC_ACTION_ENA(x)                    (((x) & 0x1) << 23)
 #define   G_0085F0_TC_ACTION_ENA(x)                    (((x) >> 23) & 0x1)
 #define   C_0085F0_TC_ACTION_ENA                       0xFF7FFFFF
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index af7b0ca2e74..7c1d77313de 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -180,9 +180,16 @@ struct r600_so_target {
 	unsigned		so_index;
 };
 
-#define R600_CONTEXT_DRAW_PENDING	(1 << 0)
-#define R600_CONTEXT_DST_CACHES_DIRTY	(1 << 1)
-#define R600_PARTIAL_FLUSH		(1 << 2)
+#define R600_CONTEXT_PS_PARTIAL_FLUSH		(1 << 0)
+#define R600_CONTEXT_CB_FLUSH			(1 << 1)
+#define R600_CONTEXT_DB_FLUSH			(1 << 2)
+#define R600_CONTEXT_SHADERCONST_FLUSH		(1 << 3)
+#define R600_CONTEXT_TEX_FLUSH			(1 << 4)
+#define R600_CONTEXT_VTX_FLUSH			(1 << 5)
+#define R600_CONTEXT_STREAMOUT_FLUSH		(1 << 6)
+#define R600_CONTEXT_WAIT_IDLE			(1 << 7)
+#define R600_CONTEXT_FLUSH_AND_INV		(1 << 8)
+#define R600_CONTEXT_HTILE_ERRATA		(1 << 9)
 
 struct r600_context;
 struct r600_screen;
@@ -196,10 +203,7 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags);
 
 void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fence,
                              unsigned offset, unsigned value);
-void r600_inval_shader_cache(struct r600_context *ctx);
-void r600_inval_texture_cache(struct r600_context *ctx);
-void r600_inval_vertex_cache(struct r600_context *ctx);
-void r600_flush_framebuffer(struct r600_context *ctx, bool flush_now);
+void r600_flush_emit(struct r600_context *ctx);
 
 void r600_context_streamout_begin(struct r600_context *ctx);
 void r600_context_streamout_end(struct r600_context *ctx);
diff --git a/src/gallium/drivers/r600/r600_hw_context.c b/src/gallium/drivers/r600/r600_hw_context.c
index 0ec13e5bade..d40f6b60933 100644
--- a/src/gallium/drivers/r600/r600_hw_context.c
+++ b/src/gallium/drivers/r600/r600_hw_context.c
@@ -114,19 +114,6 @@ err:
 	return;
 }
 
-void r600_context_ps_partial_flush(struct r600_context *ctx)
-{
-	struct radeon_winsys_cs *cs = ctx->cs;
-
-	if (!(ctx->flags & R600_CONTEXT_DRAW_PENDING))
-		return;
-
-	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
-	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
-
-	ctx->flags &= ~R600_CONTEXT_DRAW_PENDING;
-}
-
 static void r600_init_block(struct r600_context *ctx,
 			    struct r600_block *block,
 			    const struct r600_reg *reg, int index, int nreg,
@@ -665,7 +652,7 @@ void r600_need_cs_space(struct r600_context *ctx, unsigned num_dw,
 	}
 
 	/* Count in framebuffer cache flushes at the end of CS. */
-	num_dw += 7; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
+	num_dw += 44; /* one SURFACE_SYNC and CACHE_FLUSH_AND_INV (r6xx-only) */
 
 	/* Save 16 dwords for the fence mechanism. */
 	num_dw += 16;
@@ -693,7 +680,7 @@ void r600_context_dirty_block(struct r600_context *ctx,
 		LIST_ADDTAIL(&block->list,&ctx->dirty);
 
 		if (block->flags & REG_FLAG_FLUSH_CHANGE) {
-			r600_context_ps_partial_flush(ctx);
+			ctx->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
 		}
 	}
 }
@@ -861,54 +848,138 @@ out:
 	LIST_DELINIT(&block->list);
 }
 
-void r600_inval_shader_cache(struct r600_context *ctx)
+void r600_flush_emit(struct r600_context *rctx)
 {
-	ctx->surface_sync_cmd.flush_flags |= S_0085F0_SH_ACTION_ENA(1);
-	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
-}
+	struct radeon_winsys_cs *cs = rctx->cs;
 
-void r600_inval_texture_cache(struct r600_context *ctx)
-{
-	ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
-	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
-}
+	if (!rctx->flags) {
+		return;
+	}
 
-void r600_inval_vertex_cache(struct r600_context *ctx)
-{
-	if (ctx->has_vertex_cache) {
-		ctx->surface_sync_cmd.flush_flags |= S_0085F0_VC_ACTION_ENA(1);
-	} else {
-		/* Some GPUs don't have the vertex cache and must use the texture cache instead. */
-		ctx->surface_sync_cmd.flush_flags |= S_0085F0_TC_ACTION_ENA(1);
+	if (rctx->flags & R600_CONTEXT_PS_PARTIAL_FLUSH) {
+		cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
+		cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
 	}
-	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
-}
 
-void r600_flush_framebuffer(struct r600_context *ctx, bool flush_now)
-{
-	if (!(ctx->flags & R600_CONTEXT_DST_CACHES_DIRTY))
-		return;
+	if (rctx->flags & R600_CONTEXT_FLUSH_AND_INV) {
+		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);
+
+		/* DB flushes are special due to errata with hyperz, we need to
+		 * insert a no-op, so that the cache has time to really flush.
+		 */
+		if (rctx->chip_class <= R700 &&
+		    rctx->flags & R600_CONTEXT_HTILE_ERRATA) {
+			cs->buf[cs->cdw++] = PKT3(PKT3_NOP, 31, 0);
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+			cs->buf[cs->cdw++] = 0xdeadcafe;
+		}
+	}
 
-	ctx->surface_sync_cmd.flush_flags |=
-		r600_get_cb_flush_flags(ctx) |
-		(ctx->framebuffer.zsbuf ? S_0085F0_DB_ACTION_ENA(1) | S_0085F0_DB_DEST_BASE_ENA(1) : 0);
+	if (rctx->flags & (R600_CONTEXT_CB_FLUSH |
+			   R600_CONTEXT_DB_FLUSH |
+		           R600_CONTEXT_SHADERCONST_FLUSH |
+		           R600_CONTEXT_TEX_FLUSH |
+		           R600_CONTEXT_VTX_FLUSH |
+		           R600_CONTEXT_STREAMOUT_FLUSH)) {
+		/* anything left (cb, vtx, shader, streamout) can be flushed
+		 * using the surface sync packet
+		 */
+		unsigned flags = 0;
+
+		if (rctx->flags & R600_CONTEXT_CB_FLUSH) {
+			flags |= S_0085F0_CB_ACTION_ENA(1) |
+				 S_0085F0_CB0_DEST_BASE_ENA(1) |
+				 S_0085F0_CB1_DEST_BASE_ENA(1) |
+				 S_0085F0_CB2_DEST_BASE_ENA(1) |
+				 S_0085F0_CB3_DEST_BASE_ENA(1) |
+				 S_0085F0_CB4_DEST_BASE_ENA(1) |
+				 S_0085F0_CB5_DEST_BASE_ENA(1) |
+				 S_0085F0_CB6_DEST_BASE_ENA(1) |
+				 S_0085F0_CB7_DEST_BASE_ENA(1);
+
+			if (rctx->chip_class >= EVERGREEN) {
+				flags |= S_0085F0_CB8_DEST_BASE_ENA(1) |
+					 S_0085F0_CB9_DEST_BASE_ENA(1) |
+					 S_0085F0_CB10_DEST_BASE_ENA(1) |
+					 S_0085F0_CB11_DEST_BASE_ENA(1);
+			}
 
-	if (flush_now) {
-		r600_emit_atom(ctx, &ctx->surface_sync_cmd.atom);
-	} else {
-		r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
-	}
+			/* RV670 errata
+			 * (CB1_DEST_BASE_ENA is also required, which is
+			 * included unconditionally above). */
+			if (rctx->family == CHIP_RV670 ||
+			    rctx->family == CHIP_RS780 ||
+			    rctx->family == CHIP_RS880) {
+				flags |= S_0085F0_DEST_BASE_0_ENA(1);
+			}
+		}
 
-	/* Also add a complete cache flush to work around broken flushing on R6xx. */
-	if (ctx->chip_class == R600) {
-		if (flush_now) {
-			r600_emit_atom(ctx, &ctx->r6xx_flush_and_inv_cmd);
-		} else {
-			r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
+		if (rctx->flags & R600_CONTEXT_STREAMOUT_FLUSH) {
+			flags |= S_0085F0_SO0_DEST_BASE_ENA(1) |
+				 S_0085F0_SO1_DEST_BASE_ENA(1) |
+				 S_0085F0_SO2_DEST_BASE_ENA(1) |
+				 S_0085F0_SO3_DEST_BASE_ENA(1) |
+				 S_0085F0_SMX_ACTION_ENA(1);
+
+			/* RV670 errata */
+			if (rctx->family == CHIP_RV670 ||
+			    rctx->family == CHIP_RS780 ||
+			    rctx->family == CHIP_RS880) {
+				flags |= S_0085F0_DEST_BASE_0_ENA(1);
+			}
 		}
+
+		flags |= (rctx->flags & R600_CONTEXT_DB_FLUSH) ? S_0085F0_DB_ACTION_ENA(1) |
+								 S_0085F0_DB_DEST_BASE_ENA(1): 0;
+		flags |= (rctx->flags & R600_CONTEXT_SHADERCONST_FLUSH) ? S_0085F0_SH_ACTION_ENA(1) : 0;
+		flags |= (rctx->flags & R600_CONTEXT_TEX_FLUSH) ? S_0085F0_TC_ACTION_ENA(1) : 0;
+		flags |= (rctx->flags & R600_CONTEXT_VTX_FLUSH) ? S_0085F0_VC_ACTION_ENA(1) : 0;
+
+		cs->buf[cs->cdw++] = PKT3(PKT3_SURFACE_SYNC, 3, 0);
+		cs->buf[cs->cdw++] = 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 */
+	}
+
+	if (rctx->flags & R600_CONTEXT_WAIT_IDLE) {
+		/* wait for things to settle */
+		r600_write_config_reg(cs, R_008040_WAIT_UNTIL, S_008040_WAIT_3D_IDLE(1));
 	}
 
-	ctx->flags &= ~R600_CONTEXT_DST_CACHES_DIRTY;
+	/* everything is properly flushed */
+	rctx->flags = 0;
 }
 
 void r600_context_flush(struct r600_context *ctx, unsigned flags)
@@ -937,10 +1008,18 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 		streamout_suspended = true;
 	}
 
-	r600_flush_framebuffer(ctx, true);
-
 	/* partial flush is needed to avoid lockups on some chips with user fences */
-	r600_context_ps_partial_flush(ctx);
+	ctx->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
+
+	/* flush the framebuffer */
+	ctx->flags |= R600_CONTEXT_CB_FLUSH | R600_CONTEXT_DB_FLUSH;
+
+	/* R6xx errata */
+	if (ctx->chip_class == R600) {
+		ctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
+	}
+
+	r600_flush_emit(ctx);
 
 	/* old kernels and userspace don't set SX_MISC, so we must reset it to 0 here */
 	if (ctx->chip_class <= R700) {
@@ -959,10 +1038,6 @@ void r600_context_flush(struct r600_context *ctx, unsigned flags)
 	/* Begin a new CS. */
 	r600_emit_atom(ctx, &ctx->start_cs_cmd.atom);
 
-	/* Invalidate caches. */
-	r600_inval_texture_cache(ctx);
-	r600_flush_framebuffer(ctx, false);
-
 	/* Re-emit states. */
 	r600_atom_dirty(ctx, &ctx->alphatest_state.atom);
 	r600_atom_dirty(ctx, &ctx->cb_misc_state.atom);
@@ -1024,7 +1099,10 @@ void r600_context_emit_fence(struct r600_context *ctx, struct r600_resource *fen
 	va = r600_resource_va(&ctx->screen->screen, (void*)fence_bo);
 	va = va + (offset << 2);
 
-	r600_context_ps_partial_flush(ctx);
+	ctx->flags &= ~R600_CONTEXT_PS_PARTIAL_FLUSH;
+	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
+	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
+
 	cs->buf[cs->cdw++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
 	cs->buf[cs->cdw++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
 	cs->buf[cs->cdw++] = va & 0xFFFFFFFFUL;       /* ADDRESS_LO */
@@ -1185,7 +1263,7 @@ void r600_context_streamout_end(struct r600_context *ctx)
 {
 	struct radeon_winsys_cs *cs = ctx->cs;
 	struct r600_so_target **t = ctx->so_targets;
-	unsigned i, flush_flags = 0;
+	unsigned i;
 	uint64_t va;
 
 	if (ctx->chip_class >= EVERGREEN) {
@@ -1212,7 +1290,6 @@ void r600_context_streamout_end(struct r600_context *ctx)
 				r600_context_bo_reloc(ctx,  t[i]->filled_size,
 						      RADEON_USAGE_WRITE);
 
-			flush_flags |= S_0085F0_SO0_DEST_BASE_ENA(1) << i;
 		}
 	}
 
@@ -1221,22 +1298,11 @@ void r600_context_streamout_end(struct r600_context *ctx)
 	} else {
 		r600_set_streamout_enable(ctx, 0);
 	}
+	ctx->flags |= R600_CONTEXT_STREAMOUT_FLUSH;
 
-	/* This is needed to fix cache flushes on r600. */
+	/* R6xx errata */
 	if (ctx->chip_class == R600) {
-		if (ctx->family == CHIP_RV670 ||
-		    ctx->family == CHIP_RS780 ||
-		    ctx->family == CHIP_RS880) {
-			flush_flags |= S_0085F0_DEST_BASE_0_ENA(1);
-		}
-
-		r600_atom_dirty(ctx, &ctx->r6xx_flush_and_inv_cmd);
+		ctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
 	}
-
-	/* Flush streamout caches. */
-	ctx->surface_sync_cmd.flush_flags |=
-		S_0085F0_SMX_ACTION_ENA(1) | flush_flags;
-	r600_atom_dirty(ctx, &ctx->surface_sync_cmd.atom);
-
 	ctx->num_cs_dw_streamout_end = 0;
 }
diff --git a/src/gallium/drivers/r600/r600_hw_context_priv.h b/src/gallium/drivers/r600/r600_hw_context_priv.h
index 8d0ebc9d52d..12a9750e8b7 100644
--- a/src/gallium/drivers/r600/r600_hw_context_priv.h
+++ b/src/gallium/drivers/r600/r600_hw_context_priv.h
@@ -28,7 +28,8 @@
 
 #include "r600_pipe.h"
 
-#define R600_MAX_DRAW_CS_DWORDS 16
+/* the number of CS dwords for flushing and drawing */
+#define R600_MAX_DRAW_CS_DWORDS 64
 
 /* these flags are used in register flags and added into block flags */
 #define REG_FLAG_NEED_BO 1
@@ -54,9 +55,6 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg,
 void r600_context_dirty_block(struct r600_context *ctx, struct r600_block *block,
 			      int dirty, int index);
 int r600_setup_block_table(struct r600_context *ctx);
-int r600_state_sampler_init(struct r600_context *ctx, uint32_t offset);
-void r600_context_pipe_state_set_sampler(struct r600_context *ctx, struct r600_pipe_state *state, unsigned offset);
-void r600_context_ps_partial_flush(struct r600_context *ctx);
 
 /*
  * evergreen_hw_context.c
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 9e6c28d523b..d0dd4d54aa6 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -242,8 +242,6 @@ static struct pipe_context *r600_create_context(struct pipe_screen *screen, void
 	rctx->context.create_video_decoder = vl_create_decoder;
 	rctx->context.create_video_buffer = vl_video_buffer_create;
 
-	r600_init_common_atoms(rctx);
-
 	switch (rctx->chip_class) {
 	case R600:
 	case R700:
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index ff720e95cfe..ee021ed8980 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -65,11 +65,6 @@ struct r600_command_buffer {
 	unsigned pkt_flags;
 };
 
-struct r600_surface_sync_cmd {
-	struct r600_atom atom;
-	unsigned flush_flags; /* CP_COHER_CNTL */
-};
-
 struct r600_db_misc_state {
 	struct r600_atom atom;
 	bool occlusion_query_enabled;
@@ -370,8 +365,6 @@ struct r600_context {
 	/** Compute specific registers initializations.  The start_cs_cmd atom
 	 *  must be emitted before start_compute_cs_cmd. */
         struct r600_command_buffer      start_compute_cs_cmd;
-	struct r600_surface_sync_cmd	surface_sync_cmd;
-	struct r600_atom		r6xx_flush_and_inv_cmd;
 	struct r600_alphatest_state	alphatest_state;
 	struct r600_cb_misc_state	cb_misc_state;
 	struct r600_db_misc_state	db_misc_state;
@@ -575,8 +568,6 @@ void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom
 void r600_init_atom(struct r600_context *rctx, struct r600_atom *atom, unsigned id,
 		    void (*emit)(struct r600_context *ctx, struct r600_atom *state),
 		    unsigned num_dw);
-void r600_init_common_atoms(struct r600_context *rctx);
-unsigned r600_get_cb_flush_flags(struct r600_context *rctx);
 void r600_texture_barrier(struct pipe_context *ctx);
 void r600_set_index_buffer(struct pipe_context *ctx,
 			   const struct pipe_index_buffer *ib);
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index b363dc1ba98..579bcee9634 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -1600,13 +1600,23 @@ static void r600_set_framebuffer_state(struct pipe_context *ctx,
 	if (rstate == NULL)
 		return;
 
-	r600_flush_framebuffer(rctx, false);
+	if (rctx->framebuffer.nr_cbufs) {
+		rctx->flags |= R600_CONTEXT_CB_FLUSH;
+	}
+	if (rctx->framebuffer.zsbuf) {
+		rctx->flags |= R600_CONTEXT_DB_FLUSH;
+	}
+	/* R6xx errata */
+	if (rctx->chip_class == R600) {
+		rctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
+	}
 
 	/* unreference old buffer and reference new one */
 	rstate->id = R600_PIPE_STATE_FRAMEBUFFER;
 
 	util_copy_framebuffer_state(&rctx->framebuffer, state);
 
+
 	/* Colorbuffers. */
 	rctx->export_16bpc = true;
 	rctx->nr_cbufs = state->nr_cbufs;
@@ -2125,14 +2135,7 @@ void r600_adjust_gprs(struct r600_context *rctx)
 	unsigned tmp;
 	int diff;
 
-	/* XXX: Following call moved from r600_bind_[ps|vs]_shader,
-	 * it seems eg+ doesn't need it, r6xx/7xx probably need it only for
-	 * adjusting the GPR allocation?
-	 * Do we need this if we aren't really changing config below? */
-	r600_inval_shader_cache(rctx);
-
-	if (rctx->ps_shader->current->shader.bc.ngpr > rctx->default_ps_gprs)
-	{
+	if (rctx->ps_shader->current->shader.bc.ngpr > rctx->default_ps_gprs) {
 		diff = rctx->ps_shader->current->shader.bc.ngpr - rctx->default_ps_gprs;
 		num_vs_gprs -= diff;
 		num_ps_gprs += diff;
diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index 079934074a5..b8a0e484e89 100644
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -56,27 +56,6 @@ 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_context *rctx,
 		    struct r600_atom *atom,
 		    unsigned id,
@@ -108,37 +87,16 @@ void r600_emit_alphatest_state(struct r600_context *rctx, struct r600_atom *atom
 	r600_write_context_reg(cs, R_028438_SX_ALPHA_REF, alpha_ref);
 }
 
-void r600_init_common_atoms(struct r600_context *rctx)
-{
-	r600_init_atom(rctx, &rctx->r6xx_flush_and_inv_cmd, 2, r600_emit_r6xx_flush_and_inv, 2);
-	r600_init_atom(rctx, &rctx->surface_sync_cmd.atom, 3, r600_emit_surface_sync, 5);
-}
-
-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);
+	rctx->flags |= R600_CONTEXT_CB_FLUSH | R600_CONTEXT_TEX_FLUSH;
+
+	/* R6xx errata */
+	if (rctx->chip_class == R600) {
+		rctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
+	}
 }
 
 static bool r600_conv_pipe_prim(unsigned pprim, unsigned *prim)
@@ -424,7 +382,7 @@ static void r600_bind_samplers(struct pipe_context *pipe,
 		}
 		if (sampler->border_color_use) {
 			dst->atom_sampler.num_dw += 11;
-			rctx->flags |= R600_PARTIAL_FLUSH;
+			rctx->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
 		} else {
 			dst->atom_sampler.num_dw += 5;
 		}
@@ -432,7 +390,7 @@ static void r600_bind_samplers(struct pipe_context *pipe,
 	}
 	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->flags |= R600_CONTEXT_PS_PARTIAL_FLUSH;
 		rctx->seamless_cube_map.enabled = seamless_cube_map;
 		r600_atom_dirty(rctx, &rctx->seamless_cube_map.atom);
 	}
@@ -477,8 +435,6 @@ void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
 
 	rctx->vertex_elements = v;
 	if (v) {
-		r600_inval_shader_cache(rctx);
-
 		rctx->states[v->rstate.id] = &v->rstate;
 		r600_context_pipe_state_set(rctx, &v->rstate);
 	}
@@ -515,7 +471,7 @@ void r600_set_index_buffer(struct pipe_context *ctx,
 void r600_vertex_buffers_dirty(struct r600_context *rctx)
 {
 	if (rctx->vertex_buffer_state.dirty_mask) {
-		r600_inval_vertex_cache(rctx);
+		rctx->flags |= rctx->has_vertex_cache ? R600_CONTEXT_VTX_FLUSH : R600_CONTEXT_TEX_FLUSH;
 		rctx->vertex_buffer_state.atom.num_dw = (rctx->chip_class >= EVERGREEN ? 12 : 11) *
 					       util_bitcount(rctx->vertex_buffer_state.dirty_mask);
 		r600_atom_dirty(rctx, &rctx->vertex_buffer_state.atom);
@@ -570,7 +526,7 @@ void r600_sampler_views_dirty(struct r600_context *rctx,
 			      struct r600_samplerview_state *state)
 {
 	if (state->dirty_mask) {
-		r600_inval_texture_cache(rctx);
+		rctx->flags |= R600_CONTEXT_TEX_FLUSH;
 		state->atom.num_dw = (rctx->chip_class >= EVERGREEN ? 14 : 13) *
 				     util_bitcount(state->dirty_mask);
 		r600_atom_dirty(rctx, &state->atom);
@@ -898,7 +854,7 @@ void r600_delete_vs_shader(struct pipe_context *ctx, void *state)
 void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state)
 {
 	if (state->dirty_mask) {
-		r600_inval_shader_cache(rctx);
+		rctx->flags |= R600_CONTEXT_SHADERCONST_FLUSH;
 		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);
@@ -1148,13 +1104,6 @@ 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);
@@ -1221,6 +1170,7 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 
 	/* Emit states (the function expects that we emit at most 17 dwords here). */
 	r600_need_cs_space(rctx, 0, TRUE);
+	r600_flush_emit(rctx);
 
 	for (i = 0; i < R600_MAX_ATOM; i++) {
 		if (rctx->atoms[i] == NULL || !rctx->atoms[i]->dirty) {
@@ -1275,8 +1225,6 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *dinfo)
 					(info.count_from_stream_output ? S_0287F0_USE_OPAQUE(1) : 0);
 	}
 
-	rctx->flags |= R600_CONTEXT_DST_CACHES_DIRTY | R600_CONTEXT_DRAW_PENDING;
-
 	/* Set the depth buffer as dirty. */
 	if (rctx->framebuffer.zsbuf) {
 		struct pipe_surface *surf = rctx->framebuffer.zsbuf;
diff --git a/src/gallium/drivers/r600/r600d.h b/src/gallium/drivers/r600/r600d.h
index 28423e1f518..4bd77161ca9 100644
--- a/src/gallium/drivers/r600/r600d.h
+++ b/src/gallium/drivers/r600/r600d.h
@@ -3341,9 +3341,21 @@
 #define   S_0085F0_DB_DEST_BASE_ENA(x)                 (((x) & 0x1) << 14)
 #define   G_0085F0_DB_DEST_BASE_ENA(x)                 (((x) >> 14) & 0x1)
 #define   C_0085F0_DB_DEST_BASE_ENA                    0xFFFFBFFF
+/* r600 only start */
 #define   S_0085F0_CR_DEST_BASE_ENA(x)                 (((x) & 0x1) << 15)
 #define   G_0085F0_CR_DEST_BASE_ENA(x)                 (((x) >> 15) & 0x1)
 #define   C_0085F0_CR_DEST_BASE_ENA                    0xFFFF7FFF
+/* r600 only end */
+/* evergreen only start */
+#define   S_0085F0_CB8_DEST_BASE_ENA(x)                (((x) & 0x1) << 15)
+#define   G_0085F0_CB8_DEST_BASE_ENA(x)                (((x) >> 15) & 0x1)
+#define   S_0085F0_CB9_DEST_BASE_ENA(x)                (((x) & 0x1) << 16)
+#define   G_0085F0_CB9_DEST_BASE_ENA(x)                (((x) >> 16) & 0x1)
+#define   S_0085F0_CB10_DEST_BASE_ENA(x)               (((x) & 0x1) << 17)
+#define   G_0085F0_CB10_DEST_BASE_ENA(x)               (((x) >> 17) & 0x1)
+#define   S_0085F0_CB11_DEST_BASE_ENA(x)               (((x) & 0x1) << 18)
+#define   G_0085F0_CB11_DEST_BASE_ENA(x)               (((x) >> 18) & 0x1)
+/* evergreen only end */
 #define   S_0085F0_TC_ACTION_ENA(x)                    (((x) & 0x1) << 23)
 #define   G_0085F0_TC_ACTION_ENA(x)                    (((x) >> 23) & 0x1)
 #define   C_0085F0_TC_ACTION_ENA                       0xFF7FFFFF