#13655: Initial FD refactor to support sub devices

Support multiple dispatch entries for worker->dispatch sync
Update dispatch d/s to have a semaphore per dispatch entry to enable syncing on specific worker counts
Update LaunchMessageRingBufferState and WorkerConfigBufferMgr to be tracked per sub_device
Update various FD commands to support syncing on multiple sub devices:
- ERB, EWB, ERE will be updated to take in a list of sub devices for blocking on in the future. Currently will sync all sub_devices
- Trace will currently track all sub devices. Potential to track specific sub devices (could be automatic) in the future
- EP is currently hardcoded to sub device 0. This will be updated to determine the used sub devices in the future

tests/tt_metal/tools/profiler/test_device_profiler.py

@@ -167,11 +167,11 @@ def test_dispatch_cores():
     REF_COUNT_DICT = {
         "grayskull": {
             "Tensix CQ Dispatch": 16,
-            "Tensix CQ Prefetch": 24,
+            "Tensix CQ Prefetch": 25,
         },
         "wormhole_b0": {
             "Tensix CQ Dispatch": 16,
-            "Tensix CQ Prefetch": 24,
+            "Tensix CQ Prefetch": 25,
         },
     }
 

tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/test_dispatcher.cpp

@@ -477,6 +477,7 @@ int main(int argc, char **argv) {
              0,    // prefetch_downstream_buffer_pages
              num_compute_cores, // max_write_packed_cores
              0,
+             dispatch_constants::DISPATCH_MESSAGE_ENTRIES,
              0,
              0,
              0,

tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/test_prefetcher.cpp

@@ -1912,6 +1912,7 @@ void configure_for_single_chip(Device *device,
          prefetch_downstream_buffer_pages,
          num_compute_cores, // max_write_packed_cores
          0,
+         dispatch_constants::DISPATCH_MESSAGE_ENTRIES,
          0,
          0,
          0,
@@ -1932,6 +1933,7 @@ void configure_for_single_chip(Device *device,
         dispatch_compile_args[12] = dispatch_downstream_cb_sem;
         dispatch_compile_args[13] = dispatch_h_cb_sem;
         dispatch_compile_args[14] = dispatch_d_preamble_size;
+        dispatch_compile_args[21] = dispatch_constants::DISPATCH_MESSAGE_ENTRIES;
         CoreCoord phys_dispatch_d_downstream_core =
             packetized_path_en_g ? phys_dispatch_relay_mux_core : phys_dispatch_h_core;
         configure_kernel_variant<true, false>(program,
@@ -1952,6 +1954,7 @@ void configure_for_single_chip(Device *device,
         dispatch_compile_args[12] = dispatch_h_cb_sem;
         dispatch_compile_args[13] = dispatch_downstream_cb_sem;
         dispatch_compile_args[14] = 0; // preamble size
+        dispatch_compile_args[21] = 1; // unused: dispatch_d only. max_num_worker_sems is used for array sizing, set to 1
         CoreCoord phys_dispatch_h_upstream_core =
             packetized_path_en_g ? phys_dispatch_relay_demux_core : phys_dispatch_core;
         configure_kernel_variant<false, true>(program,
@@ -2655,6 +2658,7 @@ void configure_for_multi_chip(Device *device,
          prefetch_downstream_buffer_pages,
          num_compute_cores,
          0,
+         dispatch_constants::DISPATCH_MESSAGE_ENTRIES,
          0,
          0,
          0,
@@ -2675,6 +2679,7 @@ void configure_for_multi_chip(Device *device,
         dispatch_compile_args[12] = dispatch_downstream_cb_sem;
         dispatch_compile_args[13] = dispatch_h_cb_sem;
         dispatch_compile_args[14] = dispatch_d_preamble_size;
+        dispatch_compile_args[21] = dispatch_constants::DISPATCH_MESSAGE_ENTRIES;
         CoreCoord phys_dispatch_d_downstream_core =
             packetized_path_en_g ? phys_dispatch_relay_mux_core : phys_dispatch_h_core;
         configure_kernel_variant<true, false>(program_r,
@@ -2694,6 +2699,7 @@ void configure_for_multi_chip(Device *device,
         dispatch_compile_args[12] = dispatch_h_cb_sem;
         dispatch_compile_args[13] = dispatch_downstream_cb_sem;
         dispatch_compile_args[14] = 0; // preamble size
+        dispatch_compile_args[21] = 1; // unused: dispatch_d only. max_num_worker_sems is used for array sizing, set to 1
         CoreCoord phys_dispatch_h_upstream_core =
             packetized_path_en_g ? phys_dispatch_relay_demux_core : phys_dispatch_core;
         configure_kernel_variant<false, true>(program,

tests/tt_metal/tt_metal/test_kernels/dataflow/dram_copy.cpp

@@ -34,7 +34,7 @@ void kernel_main() {
         tt_l1_ptr mailboxes_t* const mailboxes = (tt_l1_ptr mailboxes_t*)(MEM_MAILBOX_BASE);
 #endif
         uint64_t dispatch_addr = NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                                             NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                                             NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
         noc_fast_atomic_increment(noc_index, NCRISC_AT_CMD_BUF, dispatch_addr, NOC_UNICAST_WRITE_VC, 1, 31, false);
 #endif
 

tests/tt_metal/tt_metal/test_kernels/misc/watcher_asserts.cpp

@@ -41,7 +41,7 @@ void MAIN {
 #endif
         uint64_t dispatch_addr =
             NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                        NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                        NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
         noc_fast_atomic_increment(noc_index, NCRISC_AT_CMD_BUF, dispatch_addr, NOC_UNICAST_WRITE_VC, 1, 31 /*wrap*/, false /*linked*/);
     }
 #else

tt_metal/hw/firmware/src/brisc.cc

@@ -374,7 +374,7 @@ int main() {
                 // For future proofing, the noc_index value is initialized to 0, to ensure an invalid NOC txn is not issued.
                 uint64_t dispatch_addr =
                     NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                    NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                    NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
                 mailboxes->go_message.signal = RUN_MSG_DONE;
                 // Notify dispatcher that this has been done
                 DEBUG_SANITIZE_NOC_ADDR(noc_index, dispatch_addr, 4);
@@ -453,7 +453,7 @@ int main() {
                 launch_msg_address->kernel_config.enables = 0;
                 uint64_t dispatch_addr =
                     NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                        NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                        NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
                 DEBUG_SANITIZE_NOC_ADDR(noc_index, dispatch_addr, 4);
                 noc_fast_atomic_increment(
                     noc_index,

tt_metal/hw/firmware/src/erisc.cc

@@ -83,7 +83,7 @@ void __attribute__((noinline)) Application(void) {
                 launch_msg_address->kernel_config.enables = 0;
                 uint64_t dispatch_addr =
                     NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                                NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                                NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
                 internal_::notify_dispatch_core_done(dispatch_addr);
                 mailboxes->launch_msg_rd_ptr = (launch_msg_rd_ptr + 1) & (launch_msg_buffer_num_entries - 1);
                 // Only executed if watcher is enabled. Ensures that we don't report stale data due to invalid launch messages in the ring buffer
@@ -94,9 +94,9 @@ void __attribute__((noinline)) Application(void) {
         } else if (go_message_signal == RUN_MSG_RESET_READ_PTR) {
             // Reset the launch message buffer read ptr
             mailboxes->launch_msg_rd_ptr = 0;
-            int64_t dispatch_addr =
+            uint64_t dispatch_addr =
                 NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                            NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR);
+                            NOC_Y(mailboxes->go_message.master_y), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
             mailboxes->go_message.signal = RUN_MSG_DONE;
             internal_::notify_dispatch_core_done(dispatch_addr);
         } else {

tt_metal/hw/firmware/src/idle_erisc.cc

@@ -145,7 +145,7 @@ int main() {
                 launch_msg_address->kernel_config.enables = 0;
                 uint64_t dispatch_addr =
                     NOC_XY_ADDR(NOC_X(mailboxes->go_message.master_x),
-                        NOC_Y(mailboxes->go_message.master_x), DISPATCH_MESSAGE_ADDR);
+                        NOC_Y(mailboxes->go_message.master_x), DISPATCH_MESSAGE_ADDR + mailboxes->go_message.dispatch_message_offset);
                 DEBUG_SANITIZE_NOC_ADDR(noc_index, dispatch_addr, 4);
                 noc_fast_atomic_increment(noc_index, NCRISC_AT_CMD_BUF, dispatch_addr, NOC_UNICAST_WRITE_VC, 1, 31 /*wrap*/, false /*linked*/);
                 mailboxes->launch_msg_rd_ptr = (launch_msg_rd_ptr + 1) & (launch_msg_buffer_num_entries - 1);

tt_metal/hw/inc/dev_msgs.h

@@ -109,7 +109,7 @@ struct kernel_config_msg_t {
 } __attribute__((packed));
 
 struct go_msg_t {
-    volatile uint8_t pad;
+    volatile uint8_t dispatch_message_offset;
     volatile uint8_t master_x;
     volatile uint8_t master_y;
     volatile uint8_t signal; // INIT, GO, DONE, RESET_RD_PTR