radv: remove predication on cache flushes

This can lead to a situation where cache flushes could get conditionally
disabled while still clearing the flush_bits, and thus flushes due to
application pipeline barriers may never get executed.

Fixes: a6c2001ace (radv: add support for cmd predication.)
Signed-off-by: Dave Airlie <airlied@redhat.com>

4 files changed, 13 insertions, 18 deletions

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 1280a18652..48fe09da53 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -446,7 +446,7 @@ radv_cmd_buffer_after_draw(struct radv_cmd_buffer *cmd_buffer,
 		}
 
 		/* Force wait for graphics or compute engines to be idle. */
-		si_cs_emit_cache_flush(cmd_buffer->cs, false,
+		si_cs_emit_cache_flush(cmd_buffer->cs,
 				       cmd_buffer->device->physical_device->rad_info.chip_class,
 				       ptr, va,
 				       radv_cmd_buffer_uses_mec(cmd_buffer),
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 2ce667fd21..aea723cfbc 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1771,7 +1771,6 @@ radv_get_preamble_cs(struct radv_queue *queue,
 
 		if (i == 0) {
 			si_cs_emit_cache_flush(cs,
-					       false,
 			                       queue->device->physical_device->rad_info.chip_class,
 					       NULL, 0,
 			                       queue->queue_family_index == RING_COMPUTE &&
@@ -1783,7 +1782,6 @@ radv_get_preamble_cs(struct radv_queue *queue,
 			                       RADV_CMD_FLAG_INV_GLOBAL_L2);
 		} else if (i == 1) {
 			si_cs_emit_cache_flush(cs,
-					       false,
 			                       queue->device->physical_device->rad_info.chip_class,
 					       NULL, 0,
 			                       queue->queue_family_index == RING_COMPUTE &&
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 503881fc83..6239e33183 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1021,7 +1021,6 @@ void si_emit_wait_fence(struct radeon_winsys_cs *cs,
 			uint64_t va, uint32_t ref,
 			uint32_t mask);
 void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
-			    bool predicated,
 			    enum chip_class chip_class,
 			    uint32_t *fence_ptr, uint64_t va,
 			    bool is_mec,
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index d9c78bf117..06e8442100 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -917,7 +917,6 @@ si_emit_acquire_mem(struct radeon_winsys_cs *cs,
 
 void
 si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
-		       bool predicated,
                        enum chip_class chip_class,
 		       uint32_t *flush_cnt,
 		       uint64_t flush_va,
@@ -948,7 +947,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 			/* Necessary for DCC */
 			if (chip_class >= VI) {
 				si_cs_emit_write_event_eop(cs,
-							   predicated,
+							   false,
 							   chip_class,
 							   is_mec,
 							   V_028A90_FLUSH_AND_INV_CB_DATA_TS,
@@ -962,12 +961,12 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 	}
 
 	if (flush_bits & RADV_CMD_FLAG_FLUSH_AND_INV_CB_META) {
-		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
+		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
 		radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META) | EVENT_INDEX(0));
 	}
 
 	if (flush_bits & RADV_CMD_FLAG_FLUSH_AND_INV_DB_META) {
-		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
+		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
 		radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META) | EVENT_INDEX(0));
 	}
 
@@ -980,7 +979,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 	}
 
 	if (flush_bits & RADV_CMD_FLAG_CS_PARTIAL_FLUSH) {
-		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
+		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
 		radeon_emit(cs, EVENT_TYPE(V_028A90_CS_PARTIAL_FLUSH) | EVENT_INDEX(4));
 	}
 
@@ -1037,14 +1036,14 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 		assert(flush_cnt);
 		uint32_t old_fence = (*flush_cnt)++;
 
-		si_cs_emit_write_event_eop(cs, predicated, chip_class, false, cb_db_event, tc_flags, 1,
+		si_cs_emit_write_event_eop(cs, false, chip_class, false, cb_db_event, tc_flags, 1,
 					   flush_va, old_fence, *flush_cnt);
-		si_emit_wait_fence(cs, predicated, flush_va, *flush_cnt, 0xffffffff);
+		si_emit_wait_fence(cs, false, flush_va, *flush_cnt, 0xffffffff);
 	}
 
 	/* VGT state sync */
 	if (flush_bits & RADV_CMD_FLAG_VGT_FLUSH) {
-		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
+		radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
 		radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
 	}
 
@@ -1057,13 +1056,13 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 			    RADV_CMD_FLAG_INV_GLOBAL_L2 |
 			    RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2))) &&
 	    !is_mec) {
-		radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, predicated));
+		radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
 		radeon_emit(cs, 0);
 	}
 
 	if ((flush_bits & RADV_CMD_FLAG_INV_GLOBAL_L2) ||
 	    (chip_class <= CIK && (flush_bits & RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2))) {
-		si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9,
+		si_emit_acquire_mem(cs, is_mec, false, chip_class >= GFX9,
 				    cp_coher_cntl |
 				    S_0085F0_TC_ACTION_ENA(1) |
 				    S_0085F0_TCL1_ACTION_ENA(1) |
@@ -1077,7 +1076,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 			 *
 			 * WB doesn't work without NC.
 			 */
-			si_emit_acquire_mem(cs, is_mec, predicated,
+			si_emit_acquire_mem(cs, is_mec, false,
 					    chip_class >= GFX9,
 					    cp_coher_cntl |
 					    S_0301F0_TC_WB_ACTION_ENA(1) |
@@ -1086,7 +1085,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 		}
 		if (flush_bits & RADV_CMD_FLAG_INV_VMEM_L1) {
 			si_emit_acquire_mem(cs, is_mec,
-					    predicated, chip_class >= GFX9,
+					    false, chip_class >= GFX9,
 					    cp_coher_cntl |
 					    S_0085F0_TCL1_ACTION_ENA(1));
 			cp_coher_cntl = 0;
@@ -1097,7 +1096,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
 	 * Therefore, it should be last. Done in PFP.
 	 */
 	if (cp_coher_cntl)
-		si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9, cp_coher_cntl);
+		si_emit_acquire_mem(cs, is_mec, false, chip_class >= GFX9, cp_coher_cntl);
 }
 
 void
@@ -1127,7 +1126,6 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
 		ptr = &cmd_buffer->gfx9_fence_idx;
 	}
 	si_cs_emit_cache_flush(cmd_buffer->cs,
-			       cmd_buffer->state.predicating,
 	                       cmd_buffer->device->physical_device->rad_info.chip_class,
 			       ptr, va,
 	                       radv_cmd_buffer_uses_mec(cmd_buffer),