r600g: Unify SURFACE_SYNC packet emission for 3D and compute

Drop the compute specific evergreen_set_buffer_sync() function and
instead use the r600_surface_sync_command atom for emitting SURFACE_SYNC
packets.

3 files changed, 30 insertions, 101 deletions

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index b2af26d..4389021 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -187,7 +187,8 @@ static void evergreen_bind_compute_state(struct pipe_context *ctx_, void *state)
 	res->bo = ctx->cs_shader->shader_code_bo;
 	res->usage = RADEON_USAGE_READ;
 	res->coher_bo_size = ctx->cs_shader->bc.ndw*4;
-	res->flags = COMPUTE_RES_SH_FLUSH;
+
+	r600_inval_shader_cache(ctx);
 
 	/* We can't always determine the
 	 * number of iterations in a loop before it's executed,
@@ -363,15 +364,20 @@ static void compute_emit_cs(struct r600_context *ctx)
 						ctx->cs_shader->resources[i].bo,
 						ctx->cs_shader->resources[i].usage);
 				}
-
-				evergreen_set_buffer_sync(ctx, ctx->cs_shader->resources[i].bo,
-					ctx->cs_shader->resources[i].coher_bo_size,
-					ctx->cs_shader->resources[i].flags,
-					ctx->cs_shader->resources[i].usage);
 			}
 		}
 	}
 
+	/* 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.
+	 */
+	r600_flush_framebuffer(ctx, true /* Flush now */);
+
 #if 0
 	COMPUTE_DBG("cdw: %i\n", cs->cdw);
 	for (i = 0; i < cs->cdw; i++) {
diff --git a/src/gallium/drivers/r600/evergreen_compute_internal.c b/src/gallium/drivers/r600/evergreen_compute_internal.c
index d846cbe..13c82f0 100644
--- a/src/gallium/drivers/r600/evergreen_compute_internal.c
+++ b/src/gallium/drivers/r600/evergreen_compute_internal.c
@@ -220,95 +220,6 @@ void evergreen_emit_ctx_reloc(
 	ctx->cs->buf[ctx->cs->cdw++] = rr;
 }
 
-void evergreen_set_buffer_sync(
-	struct r600_context *ctx,
-	struct r600_resource* bo,
-	int size,
-	int flags,
-	enum radeon_bo_usage usage)
-{
-	assert(bo);
-	int32_t cp_coher_size = 0;
-
-	if (size == 0xffffffff || size == 0) {
-		cp_coher_size = 0xffffffff;
-	}
-	else {
-		cp_coher_size = ((size + 255) >> 8);
-	}
-
-	uint32_t sync_flags = 0;
-
-	if ((flags & COMPUTE_RES_TC_FLUSH) == COMPUTE_RES_TC_FLUSH) {
-		sync_flags |= S_0085F0_TC_ACTION_ENA(1);
-	}
-
-	if ((flags & COMPUTE_RES_VC_FLUSH) == COMPUTE_RES_VC_FLUSH) {
-		sync_flags |= S_0085F0_VC_ACTION_ENA(1);
-	}
-
-	if ((flags & COMPUTE_RES_SH_FLUSH) == COMPUTE_RES_SH_FLUSH) {
-		sync_flags |= S_0085F0_SH_ACTION_ENA(1);
-	}
-
-	if ((flags & COMPUTE_RES_CB_FLUSH(0)) == COMPUTE_RES_CB_FLUSH(0)) {
-		sync_flags |= S_0085F0_CB_ACTION_ENA(1);
-
-		switch((flags >> 8) & 0xF) {
-		case 0:
-			sync_flags |= S_0085F0_CB0_DEST_BASE_ENA(1);
-			break;
-		case 1:
-			sync_flags |= S_0085F0_CB1_DEST_BASE_ENA(1);
-			break;
-		case 2:
-			sync_flags |= S_0085F0_CB2_DEST_BASE_ENA(1);
-			break;
-		case 3:
-			sync_flags |= S_0085F0_CB3_DEST_BASE_ENA(1);
-			break;
-		case 4:
-			sync_flags |= S_0085F0_CB4_DEST_BASE_ENA(1);
-			break;
-		case 5:
-			sync_flags |= S_0085F0_CB5_DEST_BASE_ENA(1);
-			break;
-		case 6:
-			sync_flags |= S_0085F0_CB6_DEST_BASE_ENA(1);
-			break;
-		case 7:
-			sync_flags |= S_0085F0_CB7_DEST_BASE_ENA(1);
-			break;
-		case 8:
-			sync_flags |= S_0085F0_CB8_DEST_BASE_ENA(1);
-			break;
-		case 9:
-			sync_flags |= S_0085F0_CB9_DEST_BASE_ENA(1);
-			break;
-		case 10:
-			sync_flags |= S_0085F0_CB10_DEST_BASE_ENA(1);
-			break;
-		case 11:
-			sync_flags |= S_0085F0_CB11_DEST_BASE_ENA(1);
-			break;
-		default:
-			assert(0);
-		}
-	}
-
-	int32_t poll_interval = 10;
-
-	ctx->cs->buf[ctx->cs->cdw++] = PKT3(PKT3_SURFACE_SYNC, 3, 0);
-	ctx->cs->buf[ctx->cs->cdw++] = sync_flags;
-	ctx->cs->buf[ctx->cs->cdw++] = cp_coher_size;
-	ctx->cs->buf[ctx->cs->cdw++] = 0;
-	ctx->cs->buf[ctx->cs->cdw++] = poll_interval;
-
-	if (cp_coher_size != 0xffffffff) {
-		evergreen_emit_ctx_reloc(ctx, bo, usage);
-	}
-}
-
 int evergreen_compute_get_gpu_format(
 	struct number_type_and_format* fmt,
 	struct r600_resource *bo)
@@ -426,7 +337,13 @@ void evergreen_set_rat(
 	res->bo = bo;
 	res->usage = RADEON_USAGE_READWRITE;
 	res->coher_bo_size = size;
-	res->flags = COMPUTE_RES_CB_FLUSH(id);
+
+	/* XXX We are setting nr_cbufs to 1 so we can get the correct
+         * cb flush flags to be emitted with the SURFACE_SYNC packet.
+         * In the future we should be adding the pipe_surface for this RAT
+         * to pipe->ctx->framebuffer.cbufs.
+         */
+	pipe->ctx->framebuffer.nr_cbufs = 1;
 }
 
 void evergreen_set_lds(
@@ -689,7 +606,13 @@ void evergreen_set_vtx_resource(
 	}
 
 	res->coher_bo_size = size;
-	res->flags = COMPUTE_RES_TC_FLUSH | COMPUTE_RES_VC_FLUSH;
+
+	r600_inval_vertex_cache(pipe->ctx);
+	/* XXX: Do we really need to invalidate the texture cache here?
+	 * r600_inval_vertex_cache() will invalidate the texture cache
+	 * if the chip does not have a vertex cache.
+	 */
+	r600_inval_texture_cache(pipe->ctx);
 }
 
 void evergreen_set_tex_resource(
@@ -761,7 +684,8 @@ void evergreen_set_tex_resource(
 	res->usage = RADEON_USAGE_READ;
 
 	res->coher_bo_size = tmp->offset[0] + util_format_get_blockwidth(tmp->real_format)*view->base.texture->width0*height*depth;
-	res->flags = COMPUTE_RES_TC_FLUSH;
+
+	r600_inval_texture_cache(pipe->ctx);
 
 	evergreen_emit_force_reloc(res);
 	evergreen_emit_force_reloc(res);
@@ -819,7 +743,8 @@ void evergreen_set_const_cache(
 	res->bo = cbo;
 	res->usage = RADEON_USAGE_READ;
 	res->coher_bo_size = size;
-	res->flags = COMPUTE_RES_SH_FLUSH;
+
+	r600_inval_shader_cache(pipe->ctx);
 }
 
 struct r600_resource* r600_compute_buffer_alloc_vram(
diff --git a/src/gallium/drivers/r600/evergreen_compute_internal.h b/src/gallium/drivers/r600/evergreen_compute_internal.h
index 340ff4b..75ada51 100644
--- a/src/gallium/drivers/r600/evergreen_compute_internal.h
+++ b/src/gallium/drivers/r600/evergreen_compute_internal.h
@@ -96,8 +96,6 @@ void evergreen_emit_ctx_reloc(struct r600_context *ctx, struct r600_resource *bo
 void evergreen_reg_set(struct evergreen_compute_resource* res, unsigned index, unsigned value);
 void evergreen_emit_force_reloc(struct evergreen_compute_resource* res);
 
-void evergreen_set_buffer_sync(struct r600_context *ctx, struct r600_resource* bo, int size, int flags, enum radeon_bo_usage usage);
-
 struct evergreen_compute_resource* get_empty_res(struct r600_pipe_compute*, enum evergreen_compute_resources res_code, int index);
 int get_compute_resource_num(void);