radv: Implement cache flushing for the MEC.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index 5ac2a148096841bbf645904244c240f306a487a5..4b2624cb8e2ca462835814978e6875e424875685 100644 (file)
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -601,6 +601,16 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
 {
        enum chip_class chip_class = cmd_buffer->device->instance->physicalDevice.rad_info.chip_class;
        unsigned cp_coher_cntl = 0;
+       bool is_compute = cmd_buffer->queue_family_index == RADV_QUEUE_COMPUTE;
+
+       if (is_compute)
+               cmd_buffer->state.flush_bits &= ~(RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+                                                 RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
+                                                 RADV_CMD_FLAG_FLUSH_AND_INV_DB |
+                                                 RADV_CMD_FLAG_FLUSH_AND_INV_DB_META |
+                                                 RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+                                                 RADV_CMD_FLAG_VS_PARTIAL_FLUSH |
+                                                 RADV_CMD_FLAG_VGT_FLUSH);
 
        radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 128);
 
@@ -679,7 +689,8 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
        /* Make sure ME is idle (it executes most packets) before continuing.
         * This prevents read-after-write hazards between PFP and ME.
         */
-       if (cp_coher_cntl || (cmd_buffer->state.flush_bits & RADV_CMD_FLAG_CS_PARTIAL_FLUSH)) {
+       if ((cp_coher_cntl || (cmd_buffer->state.flush_bits & RADV_CMD_FLAG_CS_PARTIAL_FLUSH)) &&
+           !radv_cmd_buffer_uses_mec(cmd_buffer)) {
                radeon_emit(cmd_buffer->cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
                radeon_emit(cmd_buffer->cs, 0);
        }
@@ -688,12 +699,23 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
         * Therefore, it should be last. Done in PFP.
         */
        if (cp_coher_cntl) {
-               /* ACQUIRE_MEM is only required on a compute ring. */
-               radeon_emit(cmd_buffer->cs, PKT3(PKT3_SURFACE_SYNC, 3, 0));
-               radeon_emit(cmd_buffer->cs, cp_coher_cntl);   /* CP_COHER_CNTL */
-               radeon_emit(cmd_buffer->cs, 0xffffffff);      /* CP_COHER_SIZE */
-               radeon_emit(cmd_buffer->cs, 0);               /* CP_COHER_BASE */
-               radeon_emit(cmd_buffer->cs, 0x0000000A);      /* POLL_INTERVAL */
+               if (radv_cmd_buffer_uses_mec(cmd_buffer)) {
+                       radeon_emit(cmd_buffer->cs, PKT3(PKT3_ACQUIRE_MEM, 5, 0) |
+                                                   PKT3_SHADER_TYPE_S(1));
+                       radeon_emit(cmd_buffer->cs, cp_coher_cntl);   /* CP_COHER_CNTL */
+                       radeon_emit(cmd_buffer->cs, 0xffffffff);      /* CP_COHER_SIZE */
+                       radeon_emit(cmd_buffer->cs, 0xff);            /* CP_COHER_SIZE_HI */
+                       radeon_emit(cmd_buffer->cs, 0);               /* CP_COHER_BASE */
+                       radeon_emit(cmd_buffer->cs, 0);               /* CP_COHER_BASE_HI */
+                       radeon_emit(cmd_buffer->cs, 0x0000000A);      /* POLL_INTERVAL */
+               } else {
+                       /* ACQUIRE_MEM is only required on a compute ring. */
+                       radeon_emit(cmd_buffer->cs, PKT3(PKT3_SURFACE_SYNC, 3, 0));
+                       radeon_emit(cmd_buffer->cs, cp_coher_cntl);   /* CP_COHER_CNTL */
+                       radeon_emit(cmd_buffer->cs, 0xffffffff);      /* CP_COHER_SIZE */
+                       radeon_emit(cmd_buffer->cs, 0);               /* CP_COHER_BASE */
+                       radeon_emit(cmd_buffer->cs, 0x0000000A);      /* POLL_INTERVAL */
+               }
        }
 
        cmd_buffer->state.flush_bits = 0;