Enable (very) basic compute kernels (#37328)

### Ticket
N/A

### Problem description
Quasar wasn't running any compute kernels

### What's changed
This PR adds compute FW and support to run compute kernels on Neo0
triscs. It only alows to run basic code on hte trisc and doesn't support
any real compute workloads. That work is included on a different branch
that has LLK API changes etc. The goal of this PR is to unblock others
that needs to test features that relate to triscs.

### Checklist

- [x] [![All post-commit
tests](https://github.com/tenstorrent/tt-metal/actions/workflows/all-post-commit-workflows.yaml/badge.svg?branch=ayaacob/compute_fw)](https://github.com/tenstorrent/tt-metal/actions/runs/21767356986)
- [x] [![Blackhole Post
commit](https://github.com/tenstorrent/tt-metal/actions/workflows/blackhole-post-commit.yaml/badge.svg?branch=ayaacob/compute_fw)](https://github.com/tenstorrent/tt-metal/actions/runs/21767228933)
- [x] New/Existing tests provide coverage for changes - added a test
that run a simple compute kernel that adds numbers on RISC-V and prints.

---------

Co-authored-by: Nathan Sidwell <nsidwell@tenstorrent.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

Author: 3 people

tests/tt_metal/tt_metal/debug_tools/watcher/test_pause.cpp

@@ -136,7 +136,7 @@ void RunTest(MeshWatcherFixture* fixture, const std::shared_ptr<distributed::Mes
     for (uint32_t x = xy_start.x; x <= xy_end.x; x++) {
         for (uint32_t y = xy_start.y; y <= xy_end.y; y++) {
             CoreCoord virtual_core = device->worker_core_from_logical_core({x, y});
-            for (const auto& risc_str : {" brisc", "ncrisc", "trisc0", "trisc1", "trisc2"}) {
+            for (const auto& risc_str : {"BRISC", "NCRISC", "TRISC0", "TRISC1", "TRISC2"}) {
                 std::string expected = fmt::format("{}:{}", virtual_core.str(), risc_str);
                 expected_strings.push_back(expected);
             }

tests/tt_metal/tt_metal/debug_tools/watcher/test_stack_usage.cpp

@@ -20,7 +20,7 @@ using namespace tt::tt_metal;
 namespace {
 void RunOneTest(
     MeshWatcherFixture* fixture, const std::shared_ptr<distributed::MeshDevice>& mesh_device, unsigned free) {
-    static const char* const names[] = {"brisc", "ncrisc", "trisc0", "trisc1", "trisc2", "aerisc", "ierisc"};
+    static const char* const names[] = {"BRISC", "NCRISC", "TRISC0", "TRISC1", "TRISC2", "aerisc", "ierisc"};
     const std::string path = "tests/tt_metal/tt_metal/test_kernels/misc/watcher_stack.cpp";
     auto msg = [&](std::vector<std::string> &msgs, const char *cpu, unsigned free) {
         if (msgs.empty()) {

tests/tt_metal/tt_metal/sources.cmake

@@ -24,6 +24,7 @@ set(UNIT_TESTS_LEGACY_SRC
     test_multi_core_kernel.cpp
     test_multi_dm_add_two_ints.cpp
     test_multiple_programs.cpp
+    test_quasar_basic_trisc.cpp
     test_sdpa_reduce_c.cpp
     test_single_dm_l1_write.cpp
     test_stress_noc_mcast.cpp

tests/tt_metal/tt_metal/test_kernels/compute/risc_math.cpp

@@ -0,0 +1,31 @@
+// SPDX-FileCopyrightText: © 2026 Tenstorrent AI ULC
+//
+// SPDX-License-Identifier: Apache-2.0
+
+#include "api/debug/dprint.h"
+
+void kernel_main() {
+#ifdef TRISC_PACK
+    int32_t A = 1;
+    int32_t B = 2;
+
+    DPRINT << "TEST packer" << ENDL();
+    DPRINT << A + B << ENDL();
+#endif
+
+#ifdef TRISC_UNPACK
+    int32_t A = 2;
+    int32_t B = 2;
+
+    DPRINT << "TEST unpacker" << ENDL();
+    DPRINT << A + B << ENDL();
+#endif
+
+#ifdef TRISC_MATH
+    int32_t A = 3;
+    int32_t B = 2;
+
+    DPRINT << "TEST math" << ENDL();
+    DPRINT << A + B << ENDL();
+#endif
+}

tests/tt_metal/tt_metal/test_kernels/dataflow/reader_bmm_8bank.cpp

@@ -9,8 +9,8 @@
 
 void kernel_main() {
     // same arg indices as in reader_binary_diff_lengths for compat
-    uint32_t src0_addr = get_arg_val<uint32_t>(0);
-    uint32_t src1_addr = get_arg_val<uint32_t>(1);
+    uintptr_t src0_addr = get_arg_val<uint32_t>(0);
+    uintptr_t src1_addr = get_arg_val<uint32_t>(1);
     uint32_t Mt = get_arg_val<uint32_t>(2);
     uint32_t Kt = get_arg_val<uint32_t>(3);
     uint32_t Nt = get_arg_val<uint32_t>(4);

tests/tt_metal/tt_metal/test_kernels/dataflow/writer_bmm_8bank.cpp

@@ -9,7 +9,7 @@
 
 void kernel_main() {
     // same arg indices as in reader_bmm_8bank for reuse
-    uint32_t dst_addr = get_arg_val<uint32_t>(0);
+    uintptr_t dst_addr = get_arg_val<uint32_t>(0);
     uint32_t Mt = get_arg_val<uint32_t>(2);
     uint32_t Nt = get_arg_val<uint32_t>(4);
     uint32_t batch = get_arg_val<uint32_t>(7);

tests/tt_metal/tt_metal/test_quasar_basic_trisc.cpp

@@ -0,0 +1,55 @@
+// SPDX-FileCopyrightText: © 2026 Tenstorrent AI ULC
+//
+// SPDX-License-Identifier: Apache-2.0
+
+#include "common/device_fixture.hpp"
+
+#include <tt-metalium/device.hpp>
+#include <tt-metalium/distributed.hpp>
+#include <tt-metalium/host_api.hpp>
+#include <tt-metalium/experimental/host_api.hpp>
+#include <tt-metalium/tt_metal.hpp>
+
+#ifndef OVERRIDE_KERNEL_PREFIX
+#define OVERRIDE_KERNEL_PREFIX ""
+#endif
+
+using namespace tt;
+using namespace tt::tt_metal;
+
+// This test requires simulator environment
+TEST_F(MeshDeviceSingleCardFixture, QuasarBasicTrisc) {
+    // Skip if simulator is not available
+    char* env_var = std::getenv("TT_METAL_SIMULATOR");
+    if (env_var == nullptr) {
+        GTEST_SKIP() << "This test can only be run using a simulator. Set TT_METAL_SIMULATOR environment variable.";
+    }
+
+    auto mesh_device = devices_[0];
+
+    env_var = std::getenv("TT_METAL_DPRINT_CORES");
+    if (env_var == nullptr) {
+        std::cerr << "WARNING: Please set the environment variable TT_METAL_DPRINT_CORES to 0,0 to see the output of "
+                     "the Data Movement kernels."
+                  << std::endl;
+        std::cerr << "WARNING: For example, export TT_METAL_DPRINT_CORES=0,0" << std::endl;
+    }
+
+    // We are going to use the first device (0) and the first core (0, 0) on the device.
+    constexpr CoreCoord core = {0, 0};
+    // Command queue lets us submit work (execute programs and read/write buffers) to the device.
+    distributed::MeshCommandQueue& cq = mesh_device->mesh_command_queue();
+    // Prepare a workload and a device coordinate range that spans the mesh.
+    distributed::MeshWorkload workload;
+    distributed::MeshCoordinateRange device_range = distributed::MeshCoordinateRange(mesh_device->shape());
+    Program program = CreateProgram();
+
+    CreateKernel(
+        program,
+        OVERRIDE_KERNEL_PREFIX "tests/tt_metal/tt_metal/test_kernels/compute/risc_math.cpp",
+        core,
+        ComputeConfig{.compile_args = {}});
+
+    workload.add_program(device_range, std::move(program));
+    distributed::EnqueueMeshWorkload(cq, workload, true);
+}

tests/tt_metal/tt_metal/test_single_dm_l1_write.cpp

@@ -61,7 +61,7 @@ TEST_F(MeshDeviceSingleCardFixture, SingleDmL1Write) {
         OVERRIDE_KERNEL_PREFIX "tests/tt_metal/tt_metal/test_kernels/dataflow/simple_l1_write.cpp",
         core,
         experimental::quasar::QuasarDataMovementConfig{
-            .num_processors_per_cluster = 1, .named_compile_args = named_compile_time_args});
+            .num_processors_per_cluster = 2, .named_compile_args = named_compile_time_args});
 
     // Set Runtime Arguments for the Data Movement Kernel (memory address to write to)
     SetRuntimeArgs(program, data_movement_kernel_0, core, {address});

tt_metal/CMakeLists.txt

@@ -160,6 +160,7 @@ file(
     GLOB_RECURSE TT_LLK_HEADERS
     third_party/tt_llk/tt_llk_wormhole_b0/**/*.h
     third_party/tt_llk/tt_llk_blackhole/**/*.h
+    third_party/tt_llk/tt_llk_quasar/**/*.h
 )
 
 target_sources(

tt_metal/hw/CMakeLists.txt

@@ -457,6 +457,9 @@ foreach(ARCH_CPU IN LISTS TLS_ARCH_CPUS)
                 -mcpu=${CPU}
                 ${GPP_FLAGS_common}
             )
+            if(CPU STREQUAL "tt-qsr32")
+                list(APPEND GPP_FLAGS -DCOMPILE_FOR_TRISC)
+            endif()
             add_custom_command(
                 OUTPUT
                     ${HW_OBJ_DIR}/${HW_OBJ_FILE}