r600g: add cb flushing for extra buffers + depth buffer on r600/evergreen

3 files changed, 49 insertions, 6 deletions

diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
index f9328c6198..3cb9e634f7 100644
--- a/src/gallium/drivers/r600/evergreend.h
+++ b/src/gallium/drivers/r600/evergreend.h
@@ -1865,9 +1865,18 @@
 #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
-#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
+#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/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index 1b2c2655b5..88db69c4f7 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -727,6 +727,7 @@ void evergreen_context_pipe_state_set_vs_sampler(struct r600_context *ctx, struc
 void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 {
 	struct radeon_bo *cb[12];
+	struct radeon_bo *db;
 	unsigned ndwords = 9;
 
 	if (draw->indices) {
@@ -738,6 +739,7 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 	}
 
 	/* find number of color buffer */
+	db = r600_context_reg_bo(ctx, R_028048_DB_Z_READ_BASE);
 	cb[0] = r600_context_reg_bo(ctx, R_028C60_CB_COLOR0_BASE);
 	cb[1] = r600_context_reg_bo(ctx, R_028C9C_CB_COLOR1_BASE);
 	cb[2] = r600_context_reg_bo(ctx, R_028CD8_CB_COLOR2_BASE);
@@ -755,6 +757,8 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 			ndwords += 7;
 		}
 	}
+	if (db)
+		ndwords += 7;
 
 	/* queries need some special values */
 	if (ctx->num_query_running) {
@@ -808,11 +812,15 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT;
 
 	/* flush color buffer */
-	for (int i = 0; i < 8; i++) {
+	for (int i = 0; i < 12; i++) {
 		if (cb[i]) {
 			ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3);
-			ctx->pm4[ctx->pm4_cdwords++] = (S_0085F0_CB0_DEST_BASE_ENA(1) << i) |
-							S_0085F0_CB_ACTION_ENA(1);
+			if (i > 7)
+				ctx->pm4[ctx->pm4_cdwords++] = (S_0085F0_CB8_DEST_BASE_ENA(1) << (i - 8)) |
+					S_0085F0_CB_ACTION_ENA(1);
+			else
+				ctx->pm4[ctx->pm4_cdwords++] = (S_0085F0_CB0_DEST_BASE_ENA(1) << i) |
+					S_0085F0_CB_ACTION_ENA(1);
 			ctx->pm4[ctx->pm4_cdwords++] = (cb[i]->size + 255) >> 8;
 			ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
 			ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A;
@@ -821,6 +829,17 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 			r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], cb[i]);
 		}
 	}
+	if (db) {
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3);
+		ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_DB_DEST_BASE_ENA(1) |
+			S_0085F0_DB_ACTION_ENA(1);
+		ctx->pm4[ctx->pm4_cdwords++] = (db->size + 255) >> 8;
+		ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
+		ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A;
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+		ctx->pm4[ctx->pm4_cdwords++] = 0;
+		r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], db);
+	}
 
 	/* all dirty state have been scheduled in current cs */
 	ctx->pm4_dirty_cdwords = 0;
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 88a86d2cf2..4d7547c975 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -878,6 +878,7 @@ struct radeon_bo *r600_context_reg_bo(struct r600_context *ctx, unsigned offset)
 void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 {
 	struct radeon_bo *cb[8];
+	struct radeon_bo *db;
 	unsigned ndwords = 9;
 
 	if (draw->indices) {
@@ -889,6 +890,7 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 	}
 
 	/* find number of color buffer */
+	db = r600_context_reg_bo(ctx, R_02800C_DB_DEPTH_BASE);
 	cb[0] = r600_context_reg_bo(ctx, R_028040_CB_COLOR0_BASE);
 	cb[1] = r600_context_reg_bo(ctx, R_028044_CB_COLOR1_BASE);
 	cb[2] = r600_context_reg_bo(ctx, R_028048_CB_COLOR2_BASE);
@@ -902,6 +904,8 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 			ndwords += 7;
 		}
 	}
+	if (db)
+		ndwords += 7;
 
 	/* queries need some special values */
 	if (ctx->num_query_running) {
@@ -970,6 +974,17 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 			r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], cb[i]);
 		}
 	}
+	if (db) {
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3);
+		ctx->pm4[ctx->pm4_cdwords++] = S_0085F0_DB_DEST_BASE_ENA(1) |
+			S_0085F0_DB_ACTION_ENA(1);
+		ctx->pm4[ctx->pm4_cdwords++] = (db->size + 255) >> 8;
+		ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
+		ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A;
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+		ctx->pm4[ctx->pm4_cdwords++] = 0;
+		r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], db);
+	}
 
 	/* all dirty state have been scheduled in current cs */
 	ctx->pm4_dirty_cdwords = 0;