#14427: use dispatch buffer base

dispatch buffer base is the correct addr to use for the queue start addr. hal does not account for dispatch dev addr and can easily result in overlapping memory regions
tenstorrent · Jan 16, 2025 · ef631eb · ef631eb
1 parent c21e4bd
commit ef631eb
Show file tree

Hide file tree

Showing 6 changed files with 13 additions and 11 deletions.
diff --git a/tt_metal/impl/dispatch/kernel_config/demux.cpp b/tt_metal/impl/dispatch/kernel_config/demux.cpp
@@ -2,6 +2,7 @@
 //
 // SPDX-License-Identifier: Apache-2.0
 #include "demux.hpp"
+#include "common/logger.hpp"
 #include "dispatch.hpp"
 #include "eth_tunneler.hpp"
 
@@ -11,12 +12,12 @@
 using namespace tt::tt_metal;
 
 void DemuxKernel::GenerateStaticConfigs() {
+    auto& my_dispatch_constants = dispatch_constants::get(GetCoreType());
     uint16_t channel =
         tt::Cluster::instance().get_assigned_channel_for_device(servicing_device_id_);  // TODO: this can be mmio
     logical_core_ = dispatch_core_manager::instance().demux_core(servicing_device_id_, channel, placement_cq_id_);
     static_config_.endpoint_id_start_index = 0xD1;
-    static_config_.rx_queue_start_addr_words =
-        hal.get_dev_addr(HalProgrammableCoreType::TENSIX, HalL1MemAddrType::UNRESERVED) >> 4;
+    static_config_.rx_queue_start_addr_words = my_dispatch_constants.dispatch_buffer_base() >> 4;
     static_config_.rx_queue_size_words = 0x10000 >> 4;
     static_config_.demux_fan_out = downstream_kernels_.size();
 

diff --git a/tt_metal/impl/dispatch/kernel_config/eth_router.cpp b/tt_metal/impl/dispatch/kernel_config/eth_router.cpp
@@ -45,8 +45,7 @@ void EthRouterKernel::GenerateStaticConfigs() {
     } else {
         uint16_t channel = tt::Cluster::instance().get_assigned_channel_for_device(device_->id());
         logical_core_ = dispatch_core_manager::instance().demux_d_core(device_->id(), channel, placement_cq_id_);
-        static_config_.rx_queue_start_addr_words =
-            hal.get_dev_addr(HalProgrammableCoreType::TENSIX, HalL1MemAddrType::UNRESERVED) >> 4;
+        static_config_.rx_queue_start_addr_words = my_dispatch_constants.dispatch_buffer_base() >> 4;
         static_config_.rx_queue_size_words = 0x8000 >> 4;
 
         static_config_.kernel_status_buf_addr_arg = 0;

diff --git a/tt_metal/impl/dispatch/kernel_config/eth_tunneler.cpp b/tt_metal/impl/dispatch/kernel_config/eth_tunneler.cpp
@@ -27,7 +27,7 @@ void EthTunnelerKernel::GenerateStaticConfigs() {
         logical_core_ = dispatch_core_manager::instance().us_tunneler_core_local(device_->id(), channel, cq_id_);
     }
     static_config_.endpoint_id_start_index = 0xDACADACA;
-    static_config_.in_queue_start_addr_words = 0x19000 >> 4;
+    static_config_.in_queue_start_addr_words = 0x19A00 >> 4;
     static_config_.in_queue_size_words = 0x4000 >> 4;
     static_config_.kernel_status_buf_addr_arg = 0x39000;
     static_config_.kernel_status_buf_size_bytes = 0x7000;

diff --git a/tt_metal/impl/dispatch/kernel_config/fd_kernel.cpp b/tt_metal/impl/dispatch/kernel_config/fd_kernel.cpp
@@ -119,9 +119,9 @@ void FDKernel::configure_kernel_variant(
     if (tt::llrt::RunTimeOptions::get_instance().watcher_dispatch_disabled()) {
         defines["FORCE_WATCHER_OFF"] = "1";
     }
-    // if (!tt::DPrintServerReadsDispatchCores(device_)) {
-    //     defines["FORCE_DPRINT_OFF"] = "1";
-    // }
+    if (!tt::DPrintServerReadsDispatchCores(device_)) {
+        defines["FORCE_DPRINT_OFF"] = "1";
+    }
     defines.insert(defines_in.begin(), defines_in.end());
 
     if (GetCoreType() == CoreType::WORKER) {

diff --git a/tt_metal/impl/dispatch/kernels/vc_eth_tunneler.cpp b/tt_metal/impl/dispatch/kernels/vc_eth_tunneler.cpp
@@ -219,7 +219,7 @@ using output_queue_cb_mode_sequence = CBModeTypeSequence<false,
                                                          false,
                                                          false>;
 
-#define SWITCH_THRESHOLD 32
+#define SWITCH_THRESHOLD 16
 void kernel_main() {
     rtos_context_switch_ptr = (void (*)())RtosTable[0];
 

diff --git a/tt_metal/impl/dispatch/kernels/vc_packet_router.cpp b/tt_metal/impl/dispatch/kernels/vc_packet_router.cpp
@@ -264,8 +264,10 @@ void kernel_main() {
                 bool full_packet_sent;
                 const auto words_sent = output_queues[sequence_i].template forward_data_from_input<remote_tx_network_type[sequence_i], output_depacketize[sequence_i], remote_rx_network_type[sequence_i], input_packetize[sequence_i]>(0, full_packet_sent, input_queues[sequence_i].get_end_of_cmd());
                 data_words_sent += words_sent;
-                if ((words_sent > 0) && (timeout_cycles > 0)) {
-                    progress_timestamp = get_timestamp_32b();
+                if constexpr (timeout_cycles > 0) {
+                    if (words_sent > 0) {
+                        progress_timestamp = get_timestamp_32b();
+                    }
                 }
             }