radv: add support for cmd predication.

This doesn't get used yet, it just adds support to various PKT3
emissions to enable it later.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 105384d315d7e978c2be5920842f451687af0cb0..a2578126c5259db853cfc20798308b651386a800 100644 (file)
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -1159,7 +1159,7 @@ radv_load_color_clear_regs(struct radv_cmd_buffer *cmd_buffer,
        uint32_t reg = R_028C8C_CB_COLOR0_CLEAR_WORD0 + idx * 0x3c;
        cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
 
-       radeon_emit(cmd_buffer->cs, PKT3(PKT3_COPY_DATA, 4, 0));
+       radeon_emit(cmd_buffer->cs, PKT3(PKT3_COPY_DATA, 4, cmd_buffer->state.predicating));
        radeon_emit(cmd_buffer->cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
                                    COPY_DATA_DST_SEL(COPY_DATA_REG) |
                                    COPY_DATA_COUNT_SEL);
@@ -1168,7 +1168,7 @@ radv_load_color_clear_regs(struct radv_cmd_buffer *cmd_buffer,
        radeon_emit(cmd_buffer->cs, reg >> 2);
        radeon_emit(cmd_buffer->cs, 0);
 
-       radeon_emit(cmd_buffer->cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
+       radeon_emit(cmd_buffer->cs, PKT3(PKT3_PFP_SYNC_ME, 0, cmd_buffer->state.predicating));
        radeon_emit(cmd_buffer->cs, 0);
 }
 
@@ -2639,10 +2639,10 @@ void radv_CmdDraw(
        if (cmd_buffer->state.pipeline->graphics.vtx_emit_num == 3)
                radeon_emit(cmd_buffer->cs, 0);
 
-       radeon_emit(cmd_buffer->cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
+       radeon_emit(cmd_buffer->cs, PKT3(PKT3_NUM_INSTANCES, 0, cmd_buffer->state.predicating));
        radeon_emit(cmd_buffer->cs, instanceCount);
 
-       radeon_emit(cmd_buffer->cs, PKT3(PKT3_DRAW_INDEX_AUTO, 1, 0));
+       radeon_emit(cmd_buffer->cs, PKT3(PKT3_DRAW_INDEX_AUTO, 1, cmd_buffer->state.predicating));
        radeon_emit(cmd_buffer->cs, vertexCount);
        radeon_emit(cmd_buffer->cs, V_0287F0_DI_SRC_SEL_AUTO_INDEX |
                    S_0287F0_USE_OPAQUE(0));
@@ -3294,6 +3294,7 @@ static void write_event(struct radv_cmd_buffer *cmd_buffer,
         * the stage mask. */
 
        si_cs_emit_write_event_eop(cs,
+                                  cmd_buffer->state.predicating,
                                   cmd_buffer->device->physical_device->rad_info.chip_class,
                                   false,
                                   EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
@@ -3345,7 +3346,7 @@ void radv_CmdWaitEvents(VkCommandBuffer commandBuffer,
 
                MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cs, 7);
 
-               si_emit_wait_fence(cs, va, 1, 0xffffffff);
+               si_emit_wait_fence(cs, false, va, 1, 0xffffffff);
                assert(cmd_buffer->cs->cdw <= cdw_max);
        }
 

diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index f9328fb1677d3ef1bccffaa6d7f6adc2d867a501..8cf544276346fe8b494015813d523cf8ff02c779 100644 (file)
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1095,6 +1095,7 @@ VkResult radv_CreateDevice(
                case RADV_QUEUE_GENERAL:
                case RADV_QUEUE_COMPUTE:
                        si_cs_emit_cache_flush(device->flush_cs[family],
+                                              false,
                                               device->physical_device->rad_info.chip_class,
                                               NULL, 0,
                                               family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
@@ -1111,6 +1112,7 @@ VkResult radv_CreateDevice(
                case RADV_QUEUE_GENERAL:
                case RADV_QUEUE_COMPUTE:
                        si_cs_emit_cache_flush(device->flush_shader_cs[family],
+                                              false,
                                               device->physical_device->rad_info.chip_class,
                                               NULL, 0,
                                               family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
@@ -1761,6 +1763,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
 
                if (!i) {
                        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 ac89fc1e8f2a57c7a55a5cdb7fa4eacbc284c941..a1674098845f5a59530006c3ad964d0930ed1cd2 100644 (file)
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -776,6 +776,7 @@ struct radv_cmd_state {
        uint32_t                                      descriptors_dirty;
        uint32_t                                      trace_id;
        uint32_t                                      last_ia_multi_vgt_param;
+       bool predicating;
 };
 
 struct radv_cmd_pool {
@@ -848,6 +849,7 @@ uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
                                   bool instanced_draw, bool indirect_draw,
                                   uint32_t draw_vertex_count);
 void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
+                               bool predicated,
                                enum chip_class chip_class,
                                bool is_mec,
                                unsigned event, unsigned event_flags,
@@ -857,14 +859,17 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
                                uint32_t new_fence);
 
 void si_emit_wait_fence(struct radeon_winsys_cs *cs,
+                       bool predicated,
                        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,
                            enum radv_cmd_flush_bits flush_bits);
 void si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer);
+void si_emit_set_pred(struct radv_cmd_buffer *cmd_buffer, uint64_t va);
 void si_cp_dma_buffer_copy(struct radv_cmd_buffer *cmd_buffer,
                           uint64_t src_va, uint64_t dest_va,
                           uint64_t size);

diff --git a/src/amd/vulkan/radv_query.c b/src/amd/vulkan/radv_query.c
index 03b5af16a55ebf7cfe5f9a74c49bbbeaeb77c25c..3c40774042dd67b43ae4f07a0fdeabc9bb7893fd 100644 (file)
--- a/src/amd/vulkan/radv_query.c
+++ b/src/amd/vulkan/radv_query.c
@@ -992,7 +992,7 @@ void radv_CmdCopyQueryPoolResults(
                                uint64_t avail_va = va + pool->availability_offset + 4 * query;
 
                                /* This waits on the ME. All copies below are done on the ME */
-                               si_emit_wait_fence(cs, avail_va, 1, 0xffffffff);
+                               si_emit_wait_fence(cs, false, avail_va, 1, 0xffffffff);
                        }
                }
                radv_query_shader(cmd_buffer, cmd_buffer->device->meta_state.query.pipeline_statistics_query_pipeline,
@@ -1015,7 +1015,7 @@ void radv_CmdCopyQueryPoolResults(
                                uint64_t avail_va = va + pool->availability_offset + 4 * query;
 
                                /* This waits on the ME. All copies below are done on the ME */
-                               si_emit_wait_fence(cs, avail_va, 1, 0xffffffff);
+                               si_emit_wait_fence(cs, false, avail_va, 1, 0xffffffff);
                        }
                        if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
                                uint64_t avail_va = va + pool->availability_offset + 4 * query;
@@ -1149,6 +1149,7 @@ void radv_CmdEndQuery(
                radeon_emit(cs, va >> 32);
 
                si_cs_emit_write_event_eop(cs,
+                                          false,
                                           cmd_buffer->device->physical_device->rad_info.chip_class,
                                           false,
                                           EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
@@ -1198,11 +1199,13 @@ void radv_CmdWriteTimestamp(
                break;
        default:
                si_cs_emit_write_event_eop(cs,
+                                          false,
                                           cmd_buffer->device->physical_device->rad_info.chip_class,
                                           mec,
                                           V_028A90_BOTTOM_OF_PIPE_TS, 0,
                                           3, query_va, 0, 0);
                si_cs_emit_write_event_eop(cs,
+                                          false,
                                           cmd_buffer->device->physical_device->rad_info.chip_class,
                                           mec,
                                           V_028A90_BOTTOM_OF_PIPE_TS, 0,

diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index bf08214c551a8ef34a16ac48fa7a71e2dc25cd63..ace9e665354a1df6b988b3d0878782dbd6593afe 100644 (file)
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -824,6 +824,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
 }
 
 void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
+                               bool predicated,
                                enum chip_class chip_class,
                                bool is_mec,
                                unsigned event, unsigned event_flags,
@@ -838,7 +839,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
        unsigned is_gfx8_mec = is_mec && chip_class < GFX9;
 
        if (chip_class >= GFX9 || is_gfx8_mec) {
-               radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, is_gfx8_mec ? 5 : 6, 0));
+               radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, is_gfx8_mec ? 5 : 6, predicated));
                radeon_emit(cs, op);
                radeon_emit(cs, EOP_DATA_SEL(data_sel));
                radeon_emit(cs, va);            /* address lo */
@@ -854,7 +855,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
                         * (and optional cache flushes executed) before the timestamp
                         * is written.
                         */
-                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
+                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, predicated));
                        radeon_emit(cs, op);
                        radeon_emit(cs, va);
                        radeon_emit(cs, ((va >> 32) & 0xffff) | EOP_DATA_SEL(data_sel));
@@ -862,7 +863,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
                        radeon_emit(cs, 0); /* unused */
                }
 
-               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
+               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, predicated));
                radeon_emit(cs, op);
                radeon_emit(cs, va);
                radeon_emit(cs, ((va >> 32) & 0xffff) | EOP_DATA_SEL(data_sel));
@@ -873,10 +874,11 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
 
 void
 si_emit_wait_fence(struct radeon_winsys_cs *cs,
+                  bool predicated,
                   uint64_t va, uint32_t ref,
                   uint32_t mask)
 {
-       radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
+       radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, predicated));
        radeon_emit(cs, WAIT_REG_MEM_EQUAL | WAIT_REG_MEM_MEM_SPACE(1));
        radeon_emit(cs, va);
        radeon_emit(cs, va >> 32);
@@ -887,12 +889,14 @@ si_emit_wait_fence(struct radeon_winsys_cs *cs,
 
 static void
 si_emit_acquire_mem(struct radeon_winsys_cs *cs,
-                    bool is_mec, bool is_gfx9,
+                    bool is_mec,
+                   bool predicated,
+                   bool is_gfx9,
                     unsigned cp_coher_cntl)
 {
        if (is_mec || is_gfx9) {
                uint32_t hi_val = is_gfx9 ? 0xffffff : 0xff;
-               radeon_emit(cs, PKT3(PKT3_ACQUIRE_MEM, 5, 0) |
+               radeon_emit(cs, PKT3(PKT3_ACQUIRE_MEM, 5, predicated) |
                                            PKT3_SHADER_TYPE_S(is_mec));
                radeon_emit(cs, cp_coher_cntl);   /* CP_COHER_CNTL */
                radeon_emit(cs, 0xffffffff);      /* CP_COHER_SIZE */
@@ -902,7 +906,7 @@ si_emit_acquire_mem(struct radeon_winsys_cs *cs,
                radeon_emit(cs, 0x0000000A);      /* POLL_INTERVAL */
        } else {
                /* ACQUIRE_MEM is only required on a compute ring. */
-               radeon_emit(cs, PKT3(PKT3_SURFACE_SYNC, 3, 0));
+               radeon_emit(cs, PKT3(PKT3_SURFACE_SYNC, 3, predicated));
                radeon_emit(cs, cp_coher_cntl);   /* CP_COHER_CNTL */
                radeon_emit(cs, 0xffffffff);      /* CP_COHER_SIZE */
                radeon_emit(cs, 0);               /* CP_COHER_BASE */
@@ -912,6 +916,7 @@ 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,
@@ -942,6 +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,
                                                           chip_class,
                                                           is_mec,
                                                           V_028A90_FLUSH_AND_INV_CB_DATA_TS,
@@ -955,27 +961,27 @@ 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, 0));
+               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                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, 0));
+               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META) | EVENT_INDEX(0));
        }
 
        if (!flush_cb_db) {
                if (flush_bits & RADV_CMD_FLAG_PS_PARTIAL_FLUSH) {
-                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                        radeon_emit(cs, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
                } else if (flush_bits & RADV_CMD_FLAG_VS_PARTIAL_FLUSH) {
-                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+                       radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                        radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
                }
        }
 
        if (flush_bits & RADV_CMD_FLAG_CS_PARTIAL_FLUSH) {
-               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                radeon_emit(cs, EVENT_TYPE(V_028A90_CS_PARTIAL_FLUSH) | EVENT_INDEX(4));
        }
 
@@ -1022,14 +1028,14 @@ 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, chip_class, false, cb_db_event, tc_flags, 1,
+               si_cs_emit_write_event_eop(cs, predicated, chip_class, false, cb_db_event, tc_flags, 1,
                                           flush_va, old_fence, *flush_cnt);
-               si_emit_wait_fence(cs, flush_va, *flush_cnt, 0xffffffff);
+               si_emit_wait_fence(cs, predicated, flush_va, *flush_cnt, 0xffffffff);
        }
 
        /* VGT state sync */
        if (flush_bits & RADV_CMD_FLAG_VGT_FLUSH) {
-               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+               radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
                radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
        }
 
@@ -1042,13 +1048,13 @@ 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, 0));
+               radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, predicated));
                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, chip_class >= GFX9,
+               si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9,
                                    cp_coher_cntl |
                                    S_0085F0_TC_ACTION_ENA(1) |
                                    S_0085F0_TCL1_ACTION_ENA(1) |
@@ -1062,14 +1068,16 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
                         *
                         * WB doesn't work without NC.
                         */
-                       si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9,
+                       si_emit_acquire_mem(cs, is_mec, predicated,
+                                           chip_class >= GFX9,
                                            cp_coher_cntl |
                                            S_0301F0_TC_WB_ACTION_ENA(1) |
                                            S_0301F0_TC_NC_ACTION_ENA(1));
                        cp_coher_cntl = 0;
                }
                if (flush_bits & RADV_CMD_FLAG_INV_VMEM_L1) {
-                       si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9,
+                       si_emit_acquire_mem(cs, is_mec,
+                                           predicated, chip_class >= GFX9,
                                            cp_coher_cntl |
                                            S_0085F0_TCL1_ACTION_ENA(1));
                        cp_coher_cntl = 0;
@@ -1080,7 +1088,7 @@ 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, chip_class >= GFX9, cp_coher_cntl);
+               si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9, cp_coher_cntl);
 }
 
 void
@@ -1110,6 +1118,7 @@ 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),
@@ -1120,6 +1129,19 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
        cmd_buffer->state.flush_bits = 0;
 }
 
+void
+si_emit_set_pred(struct radv_cmd_buffer *cmd_buffer, uint64_t va)
+{
+       uint32_t val = 0;
+
+       if (va)
+               val = (((va >> 32) & 0xff) |
+                      PRED_OP(PREDICATION_OP_BOOL64)|
+                      PREDICATION_DRAW_VISIBLE);
+       radeon_emit(cmd_buffer->cs, PKT3(PKT3_SET_PREDICATION, 1, 0));
+       radeon_emit(cmd_buffer->cs, va);
+       radeon_emit(cmd_buffer->cs, val);
+}
 
 /* Set this if you want the 3D engine to wait until CP DMA is done.
  * It should be set on the last CP DMA packet. */
@@ -1193,7 +1215,7 @@ static void si_emit_cp_dma(struct radv_cmd_buffer *cmd_buffer,
                header |= S_411_SRC_SEL(V_411_SRC_ADDR_TC_L2);
 
        if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
-               radeon_emit(cs, PKT3(PKT3_DMA_DATA, 5, 0));
+               radeon_emit(cs, PKT3(PKT3_DMA_DATA, 5, cmd_buffer->state.predicating));
                radeon_emit(cs, header);
                radeon_emit(cs, src_va);                /* SRC_ADDR_LO [31:0] */
                radeon_emit(cs, src_va >> 32);          /* SRC_ADDR_HI [31:0] */
@@ -1203,7 +1225,7 @@ static void si_emit_cp_dma(struct radv_cmd_buffer *cmd_buffer,
        } else {
                assert(!(flags & CP_DMA_USE_L2));
                header |= S_411_SRC_ADDR_HI(src_va >> 32);
-               radeon_emit(cs, PKT3(PKT3_CP_DMA, 4, 0));
+               radeon_emit(cs, PKT3(PKT3_CP_DMA, 4, cmd_buffer->state.predicating));
                radeon_emit(cs, src_va);                        /* SRC_ADDR_LO [31:0] */
                radeon_emit(cs, header);                        /* SRC_ADDR_HI [15:0] + flags. */
                radeon_emit(cs, dst_va);                        /* DST_ADDR_LO [31:0] */
@@ -1217,7 +1239,7 @@ static void si_emit_cp_dma(struct radv_cmd_buffer *cmd_buffer,
         * should precede it.
         */
        if ((flags & CP_DMA_SYNC) && cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL) {
-               radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
+               radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, cmd_buffer->state.predicating));
                radeon_emit(cs, 0);
        }