Refactor: remove dead gm_heap_ptr_ and slot_states_ptr_ from trb Runtime

src/a2a3/runtime/tensormap_and_ringbuffer/aicpu/aicpu_executor.cpp

@@ -560,9 +560,6 @@ int32_t AicpuExecutor::run(Runtime *runtime) {
                 }
 #endif
 
-                // With multi-ring, slot_states are per-ring inside the scheduler.
-                runtime->set_slot_states_ptr(nullptr);
-
                 // Wire scheduler context to the newly created PTO2Runtime before
                 // releasing scheduler threads from runtime_init_ready_.
                 sched_ctx_.bind_runtime(rt);

src/a2a3/runtime/tensormap_and_ringbuffer/host/runtime_maker.cpp

@@ -514,7 +514,6 @@ static bool ensure_static_arenas(Runtime *runtime, const ArenaSizingConfig &sizi
         LOG_ERROR("Failed to acquire pooled GM heap");
         return false;
     }
-    runtime->set_gm_heap(out->gm_heap);
 
     int64_t t_sm_start = _now_ms();
     out->gm_sm = runtime->host_api.acquire_pooled_gm_sm();

src/a2a3/runtime/tensormap_and_ringbuffer/runtime/runtime.h

@@ -236,7 +236,6 @@ struct alignas(64) DeviceRuntimeLaunchDesc {
     bool serial_orch_sched;
 
     void *gm_sm_ptr_;                        // GM pointer to PTO2 shared memory (device)
-    void *slot_states_ptr_;                  // Pointer to PTO2TaskSlotState array (scheduler-private, for profiling)
     ChipStorageTaskArgs orch_args_storage_;  // Copy of args for device
 
     // Prebuilt-arena fast path (trb only). Set by the host before rtMemcpy'ing
@@ -280,8 +279,6 @@ class Runtime {
     int registered_kernel_func_ids_[RUNTIME_MAX_FUNC_ID];
     int registered_kernel_count_;
 
-    void *gm_heap_ptr_;  // GM heap for orchestrator output buffers (device); host-only bookkeeping
-
 public:
     /**
      * Constructor - zero-initialize all arrays
@@ -319,11 +316,8 @@ class Runtime {
     // =========================================================================
 
     void *get_gm_sm_ptr() const;
-    void *get_gm_heap_ptr() const;
     const ChipStorageTaskArgs &get_orch_args() const;
     void set_gm_sm_ptr(void *p);
-    void set_gm_heap(void *p);
-    void set_slot_states_ptr(void *p);
     void set_orch_args(const ChipStorageTaskArgs &args);
 
     // Prebuilt-arena fast path (trb only). Set by host's

src/a2a3/runtime/tensormap_and_ringbuffer/runtime/shared/runtime.cpp

@@ -40,7 +40,6 @@ Runtime::Runtime() {
     dev.orch_to_sched = false;
     dev.serial_orch_sched = false;
     dev.gm_sm_ptr_ = nullptr;
-    dev.slot_states_ptr_ = nullptr;
     dev.orch_args_storage_.clear();
     dev.prebuilt_arena_base_ = nullptr;
     dev.prebuilt_runtime_offset_ = 0;
@@ -50,7 +49,6 @@ Runtime::Runtime() {
     }
 
     // Initialize host-only tail.
-    gm_heap_ptr_ = nullptr;
     registered_kernel_count_ = 0;
 }
 
@@ -59,11 +57,8 @@ Runtime::Runtime() {
 // =============================================================================
 
 void *Runtime::get_gm_sm_ptr() const { return dev.gm_sm_ptr_; }
-void *Runtime::get_gm_heap_ptr() const { return gm_heap_ptr_; }
 const ChipStorageTaskArgs &Runtime::get_orch_args() const { return dev.orch_args_storage_; }
 void Runtime::set_gm_sm_ptr(void *p) { dev.gm_sm_ptr_ = p; }
-void Runtime::set_gm_heap(void *p) { gm_heap_ptr_ = p; }
-void Runtime::set_slot_states_ptr(void *p) { dev.slot_states_ptr_ = p; }
 void Runtime::set_orch_args(const ChipStorageTaskArgs &args) { dev.orch_args_storage_ = args; }
 
 void Runtime::set_prebuilt_arena(void *arena_base, size_t runtime_off) {

src/a5/runtime/tensormap_and_ringbuffer/aicpu/aicpu_executor.cpp

@@ -558,9 +558,6 @@ int32_t AicpuExecutor::run(Runtime *runtime) {
                 }
 #endif
 
-                // With multi-ring, slot_states are per-ring inside the scheduler.
-                runtime->set_slot_states_ptr(nullptr);
-
                 // Wire scheduler context to the newly created PTO2Runtime before
                 // releasing scheduler threads from runtime_init_ready_.
                 sched_ctx_.bind_runtime(rt);

src/a5/runtime/tensormap_and_ringbuffer/host/runtime_maker.cpp

@@ -514,7 +514,6 @@ static bool ensure_static_arenas(Runtime *runtime, const ArenaSizingConfig &sizi
         LOG_ERROR("Failed to acquire pooled GM heap");
         return false;
     }
-    runtime->set_gm_heap(out->gm_heap);
 
     int64_t t_sm_start = _now_ms();
     out->gm_sm = runtime->host_api.acquire_pooled_gm_sm();

src/a5/runtime/tensormap_and_ringbuffer/runtime/runtime.h

@@ -250,7 +250,6 @@ struct alignas(64) DeviceRuntimeLaunchDesc {
     bool serial_orch_sched;
 
     void *gm_sm_ptr_;                        // GM pointer to PTO2 shared memory (device)
-    void *slot_states_ptr_;                  // Pointer to PTO2TaskSlotState array (scheduler-private, for profiling)
     ChipStorageTaskArgs orch_args_storage_;  // Copy of args for device
 
     // Prebuilt-arena fast path (trb only). Set by the host before rtMemcpy'ing
@@ -294,8 +293,6 @@ class Runtime {
     int registered_kernel_func_ids_[RUNTIME_MAX_FUNC_ID];
     int registered_kernel_count_;
 
-    void *gm_heap_ptr_;  // GM heap for orchestrator output buffers (device); host-only bookkeeping
-
 public:
     /**
      * Constructor - zero-initialize all arrays
@@ -333,11 +330,8 @@ class Runtime {
     // =========================================================================
 
     void *get_gm_sm_ptr() const;
-    void *get_gm_heap_ptr() const;
     const ChipStorageTaskArgs &get_orch_args() const;
     void set_gm_sm_ptr(void *p);
-    void set_gm_heap(void *p);
-    void set_slot_states_ptr(void *p);
     void set_orch_args(const ChipStorageTaskArgs &args);
 
     // Prebuilt-arena fast path (trb only). Set by host's

src/a5/runtime/tensormap_and_ringbuffer/runtime/shared/runtime.cpp

@@ -40,7 +40,6 @@ Runtime::Runtime() {
     dev.orch_to_sched = false;
     dev.serial_orch_sched = false;
     dev.gm_sm_ptr_ = nullptr;
-    dev.slot_states_ptr_ = nullptr;
     dev.orch_args_storage_.clear();
     dev.prebuilt_arena_base_ = nullptr;
     dev.prebuilt_runtime_offset_ = 0;
@@ -50,7 +49,6 @@ Runtime::Runtime() {
     }
 
     // Initialize host-only tail.
-    gm_heap_ptr_ = nullptr;
     registered_kernel_count_ = 0;
 }
 
@@ -59,11 +57,8 @@ Runtime::Runtime() {
 // =============================================================================
 
 void *Runtime::get_gm_sm_ptr() const { return dev.gm_sm_ptr_; }
-void *Runtime::get_gm_heap_ptr() const { return gm_heap_ptr_; }
 const ChipStorageTaskArgs &Runtime::get_orch_args() const { return dev.orch_args_storage_; }
 void Runtime::set_gm_sm_ptr(void *p) { dev.gm_sm_ptr_ = p; }
-void Runtime::set_gm_heap(void *p) { gm_heap_ptr_ = p; }
-void Runtime::set_slot_states_ptr(void *p) { dev.slot_states_ptr_ = p; }
 void Runtime::set_orch_args(const ChipStorageTaskArgs &args) { dev.orch_args_storage_ = args; }
 
 void Runtime::set_prebuilt_arena(void *arena_base, size_t runtime_off) {