disambiguate semaphore lookup by including core type arg (#15133)

When looking up semaphores, we now also specify which core type we are
specifically looking at.

There are multiple core types, each with there own logical core
mappings. Logical core locations can alias between core types, even
though they map to unique noc locations.

tt_metal/impl/program/program.cpp

@@ -117,7 +117,7 @@ class Program_ {
 
     std::vector<CoreRange> circular_buffers_unique_coreranges() const;
 
-    std::vector<std::reference_wrapper<const Semaphore>> semaphores_on_core(const CoreCoord &core) const;
+    std::vector<std::reference_wrapper<const Semaphore>> semaphores_on_core(const CoreCoord &core, CoreType core_type) const;
 
     size_t num_semaphores () const;
     void init_semaphores ( const Device & device, const CoreCoord &logical_core, uint32_t programmable_core_type_index) const;
@@ -830,18 +830,18 @@ void detail::Program_::validate_circular_buffer_region(const Device *device) {
     }
 }
 
-size_t Program::num_semaphores(const CoreCoord &core) const { return semaphores_on_core(core).size(); }
+size_t Program::num_semaphores(const CoreCoord &core, CoreType core_type) const { return semaphores_on_core(core, core_type).size(); }
 
 size_t detail::Program_::num_semaphores() const { return semaphores_.size(); }
 
 size_t Program::num_semaphores() const { return pimpl_->num_semaphores(); }
 
 void detail::Program_::init_semaphores(const Device &device, const CoreCoord &logical_core, uint32_t programmable_core_type_index) const {
-    auto semaphores_on_core = this->semaphores_on_core(logical_core);
 
     uint64_t kernel_config_base = hal.get_dev_addr(programmable_core_type_index, HalL1MemAddrType::KERNEL_CONFIG);
     uint64_t addr = kernel_config_base + this->program_configs_[programmable_core_type_index].sem_offset;
     CoreType core_type = hal.get_core_type(programmable_core_type_index);
+    auto semaphores_on_core = this->semaphores_on_core(logical_core, core_type);
     for (auto semaphore : semaphores_on_core) {
         llrt::write_hex_vec_to_core(
             device.id(),
@@ -1772,18 +1772,18 @@ void detail::Program_::release_buffers() { owned_buffer_pool = {}; }
 
 void Program::release_buffers() { pimpl_->release_buffers(); }
 
-std::vector<std::reference_wrapper<const Semaphore>> detail::Program_::semaphores_on_core(const CoreCoord &core) const {
+std::vector<std::reference_wrapper<const Semaphore>> detail::Program_::semaphores_on_core(const CoreCoord &core, CoreType core_type) const {
     std::vector<std::reference_wrapper<const Semaphore>> semaphores;
     for (const Semaphore &s : this->semaphores_) {
-        if (s.initialized_on_logical_core(core)) {
+        if (s.initialized_on_logical_core(core) && s.core_type() == core_type) {
             semaphores.emplace_back(std::cref(s));
         }
     }
     return semaphores;
 }
 
-std::vector<std::reference_wrapper<const Semaphore>> Program::semaphores_on_core(const CoreCoord &core) const {
-    return pimpl_->semaphores_on_core(core);
+std::vector<std::reference_wrapper<const Semaphore>> Program::semaphores_on_core(const CoreCoord &core, CoreType core_type) const {
+    return pimpl_->semaphores_on_core(core, core_type);
 }
 
 bool detail::Program_::is_finalized() const { return this->finalized_; }

tt_metal/impl/program/program.hpp

@@ -135,9 +135,9 @@ class Program {
 
     std::vector<CoreRange> circular_buffers_unique_coreranges() const;
 
-    std::vector<std::reference_wrapper<const Semaphore>> semaphores_on_core(const CoreCoord &core) const;
+    std::vector<std::reference_wrapper<const Semaphore>> semaphores_on_core(const CoreCoord &core, CoreType core_type) const;
 
-    size_t num_semaphores ( const CoreCoord & core ) const;
+    size_t num_semaphores ( const CoreCoord & core, CoreType core_type ) const;
     size_t num_semaphores () const;
     void init_semaphores ( const Device & device, const CoreCoord &logical_core, uint32_t programmable_core_type_index) const;
     // XXXXX TODO: this should return a const reference

tt_metal/tt_metal.cpp

@@ -162,13 +162,13 @@ void ConfigureKernelGroup(
     }
 }
 
-std::optional<uint32_t> get_semaphore_id(const Program& program, const CoreRange& core_range) {
+std::optional<uint32_t> get_semaphore_id(const Program &program, const CoreRange& core_range, CoreType core_type) {
     std::optional<uint32_t> semaphore_id = std::nullopt;
     std::vector<uint32_t> semaphore_histogram(NUM_SEMAPHORES, 0);
     for (auto x = core_range.start_coord.x; x <= core_range.end_coord.x; x++) {
         for (auto y = core_range.start_coord.y; y <= core_range.end_coord.y; y++) {
             CoreCoord logical_core(x, y);
-            auto semaphores = program.semaphores_on_core(logical_core);
+            auto semaphores = program.semaphores_on_core(logical_core, core_type);
             if (semaphores.size() == NUM_SEMAPHORES) {
                 TT_THROW(
                     "Cannot add semaphore on core {}. Max number of semaphores ({}) reached!",
@@ -1158,7 +1158,7 @@ uint32_t CreateSemaphore(
             for (const auto& core_range : crs.ranges()) {
                 CoreCoord start_core = core_range.start_coord;
                 CoreCoord end_core = core_range.end_coord;
-                std::optional<uint32_t> semaphore_id_candidate = get_semaphore_id(program, core_range);
+                std::optional<uint32_t> semaphore_id_candidate = get_semaphore_id(program, core_range, core_type);
                 if (!semaphore_id.has_value()) {
                     semaphore_id = semaphore_id_candidate;
                 } else {