[metal] Refactor runtime ListManager utils (#1444)

* [metal] Refactor runtime ListManager utils * rm NodeManagerData * fix runtime size
taichi-dev · Jul 11, 2020 · f6cf282 · f6cf282
1 parent 0d2e34f
commit f6cf282
Show file tree

Hide file tree

Showing 6 changed files with 96 additions and 91 deletions.
diff --git a/taichi/backends/metal/codegen_metal.cpp b/taichi/backends/metal/codegen_metal.cpp
@@ -946,11 +946,11 @@ class KernelCodegen : public IRVisitor {
     {
       ScopedIndent s2(current_appender());
       emit("const int parent_idx_ = (ii / child_num_slots);");
-      emit("if (parent_idx_ >= num_active(&parent_list)) return;");
+      emit("if (parent_idx_ >= parent_list.num_active()) return;");
       emit("const int child_idx_ = (ii % child_num_slots);");
       emit(
-          "const auto parent_elem_ = get<ListgenElement>(&parent_list, "
-          "parent_idx_);");
+          "const auto parent_elem_ = "
+          "parent_list.get<ListgenElement>(parent_idx_);");
 
       emit("ListgenElement {};", kListgenElemVarName);
       // No need to add mem_offset_in_parent, because place() always starts at 0

diff --git a/taichi/backends/metal/kernel_manager.cpp b/taichi/backends/metal/kernel_manager.cpp
@@ -752,7 +752,7 @@ class KernelManager::Impl {
       ListManager root_lm;
       root_lm.lm_data = rtm_list_head + root_id;
       root_lm.mem_alloc = alloc;
-      append(&root_lm, root_elem);
+      root_lm.append(root_elem);
     }
 
     did_modify_range(runtime_buffer_.get(), /*location=*/0,

diff --git a/taichi/backends/metal/shaders/runtime_kernels.metal.h b/taichi/backends/metal/shaders/runtime_kernels.metal.h
@@ -53,7 +53,7 @@ STR(
       child_list.lm_data =
           (reinterpret_cast<device Runtime *>(runtime_addr)->snode_lists +
            child_snode_id);
-      clear(&child_list);
+      child_list.clear();
     }
 
     kernel void element_listgen(device byte *runtime_addr[[buffer(0)]],
@@ -83,13 +83,13 @@ STR(
       const int max_num_elems = args[2];
       for (int ii = utid_; ii < max_num_elems; ii += grid_size) {
         const int parent_idx = (ii / num_slots);
-        if (parent_idx >= num_active(&parent_list)) {
+        if (parent_idx >= parent_list.num_active()) {
           // Since |parent_idx| increases monotonically, we can return directly
           // once it goes beyond the number of active parent elements.
           return;
         }
         const int child_idx = (ii % num_slots);
-        const auto parent_elem = get<ListgenElement>(&parent_list, parent_idx);
+        const auto parent_elem = parent_list.get<ListgenElement>(parent_idx);
         ListgenElement child_elem;
         child_elem.root_mem_offset = parent_elem.root_mem_offset +
                                      child_idx * child_stride +
@@ -99,7 +99,7 @@ STR(
           refine_coordinates(parent_elem,
                              runtime->snode_extractors[parent_snode_id],
                              child_idx, &child_elem);
-          append(&child_list, child_elem);
+          child_list.append(child_elem);
         }
       }
     }

diff --git a/taichi/backends/metal/shaders/runtime_structs.metal.h b/taichi/backends/metal/shaders/runtime_structs.metal.h
@@ -51,11 +51,6 @@ STR(
       atomic_int chunks[kTaichiNumChunks];
     };
 
-    struct ListManager {
-      device ListManagerData *lm_data;
-      device MemoryAllocator *mem_alloc;
-    };
-
     // This class is very similar to metal::SNodeDescriptor
     struct SNodeMeta {
       enum Type { Root = 0, Dense = 1, Bitmasked = 2, Dynamic = 3 };

diff --git a/taichi/backends/metal/shaders/runtime_utils.metal.h b/taichi/backends/metal/shaders/runtime_utils.metal.h
@@ -31,8 +31,7 @@
 // clang-format off
 METAL_BEGIN_RUNTIME_UTILS_DEF
 STR(
-    using PtrOffset = int32_t;
-    constant constexpr int kAlignment = 8;
+    using PtrOffset = int32_t; constant constexpr int kAlignment = 8;
 
     [[maybe_unused]] PtrOffset mtl_memalloc_alloc(device MemoryAllocator * ma,
                                                   int32_t size) {
@@ -46,86 +45,98 @@ STR(
           return reinterpret_cast<device char *>(ma + 1) + offs;
         }
 
-    [[maybe_unused]] int num_active(thread ListManager *l) {
-      return atomic_load_explicit(&(l->lm_data->next),
-                                  metal::memory_order_relaxed);
-    }
+    struct ListManager {
+      device ListManagerData *lm_data;
+      device MemoryAllocator *mem_alloc;
 
-    [[maybe_unused]] void clear(thread ListManager *l) {
-      atomic_store_explicit(&(l->lm_data->next), 0,
-                            metal::memory_order_relaxed);
-    }
+      inline int num_active() {
+        return atomic_load_explicit(&(lm_data->next),
+                                    metal::memory_order_relaxed);
+      }
 
-    [[maybe_unused]] PtrOffset mtl_listmgr_ensure_chunk(thread ListManager *l,
-                                                        int i) {
-      device ListManagerData *list = l->lm_data;
-      PtrOffset offs = 0;
-      const int kChunkBytes =
-          (list->element_stride << list->log2_num_elems_per_chunk);
-
-      while (true) {
-        int stored = 0;
-        // If chunks[i] is unallocated, i.e. 0, mark it as 1 to prevent others
-        // from requesting memory again. Once allocated, set chunks[i] to the
-        // actual address offset, which is guaranteed to be greater than 1.
-        const bool is_me = atomic_compare_exchange_weak_explicit(
-            list->chunks + i, &stored, 1, metal::memory_order_relaxed,
-            metal::memory_order_relaxed);
-        if (is_me) {
-          offs = mtl_memalloc_alloc(l->mem_alloc, kChunkBytes);
-          atomic_store_explicit(list->chunks + i, offs,
-                                metal::memory_order_relaxed);
-          break;
-        } else if (stored > 1) {
-          offs = stored;
-          break;
-        }
-        // |stored| == 1, just spin
+      inline void resize(int sz) {
+        atomic_store_explicit(&(lm_data->next), sz,
+                              metal::memory_order_relaxed);
       }
-      return offs;
-    }
 
-    [[maybe_unused]] device char *mtl_listmgr_get_elem_from_chunk(
-        thread ListManager *l,
-        int i,
-        PtrOffset chunk_ptr_offs) {
-      device ListManagerData *list = l->lm_data;
-      device char *chunk_ptr = reinterpret_cast<device char *>(
-          mtl_memalloc_to_ptr(l->mem_alloc, chunk_ptr_offs));
-      const uint32_t mask = ((1 << list->log2_num_elems_per_chunk) - 1);
-      return chunk_ptr + ((i & mask) * list->element_stride);
-    }
+      inline void clear() {
+        resize(0);
+      }
 
-    [[maybe_unused]] device char *append(thread ListManager *l) {
-      device ListManagerData *list = l->lm_data;
-      const int elem_idx = atomic_fetch_add_explicit(
-          &list->next, 1, metal::memory_order_relaxed);
-      const int chunk_idx = elem_idx >> list->log2_num_elems_per_chunk;
-      const PtrOffset chunk_ptr_offs = mtl_listmgr_ensure_chunk(l, chunk_idx);
-      return mtl_listmgr_get_elem_from_chunk(l, elem_idx, chunk_ptr_offs);
-    }
+      struct ReserveElemResult {
+        int elem_idx;
+        PtrOffset chunk_ptr_offs;
+      };
+
+      ReserveElemResult reserve_new_elem() {
+        const int elem_idx = atomic_fetch_add_explicit(
+            &lm_data->next, 1, metal::memory_order_relaxed);
+        const int chunk_idx = elem_idx >> lm_data->log2_num_elems_per_chunk;
+        const PtrOffset chunk_ptr_offs = ensure_chunk(chunk_idx);
+        return {elem_idx, chunk_ptr_offs};
+      }
+
+      device char *append() {
+        auto reserved = reserve_new_elem();
+        return get_elem_from_chunk(reserved.elem_idx, reserved.chunk_ptr_offs);
+      }
 
-    template <typename T>
-    [[maybe_unused]] void append(thread ListManager *l, thread const T &elem) {
-      device char *ptr = append(l);
-      thread char *elem_ptr = (thread char *)(&elem);
+      template <typename T>
+      void append(thread const T &elem) {
+        device char *ptr = append();
+        thread char *elem_ptr = (thread char *)(&elem);
 
-      for (int i = 0; i < l->lm_data->element_stride; ++i) {
-        *ptr = *elem_ptr;
-        ++ptr;
-        ++elem_ptr;
+        for (int i = 0; i < lm_data->element_stride; ++i) {
+          *ptr = *elem_ptr;
+          ++ptr;
+          ++elem_ptr;
+        }
       }
-    }
 
-    template <typename T>
-    [[maybe_unused]] T get(thread ListManager *l, int i) {
-      device ListManagerData *list = l->lm_data;
-      const int chunk_idx = i >> list->log2_num_elems_per_chunk;
-      const PtrOffset chunk_ptr_offs = atomic_load_explicit(
-          list->chunks + chunk_idx, metal::memory_order_relaxed);
-      return *reinterpret_cast<device T *>(
-          mtl_listmgr_get_elem_from_chunk(l, i, chunk_ptr_offs));
-    }
+      template <typename T>
+      T get(int i) {
+        const int chunk_idx = i >> lm_data->log2_num_elems_per_chunk;
+        const PtrOffset chunk_ptr_offs = atomic_load_explicit(
+            lm_data->chunks + chunk_idx, metal::memory_order_relaxed);
+        return *reinterpret_cast<device T *>(
+            get_elem_from_chunk(i, chunk_ptr_offs));
+      }
+
+     private:
+      PtrOffset ensure_chunk(int i) {
+        PtrOffset offs = 0;
+        const int kChunkBytes =
+            (lm_data->element_stride << lm_data->log2_num_elems_per_chunk);
+
+        while (true) {
+          int stored = 0;
+          // If chunks[i] is unallocated, i.e. 0, mark it as 1 to prevent others
+          // from requesting memory again. Once allocated, set chunks[i] to the
+          // actual address offset, which is guaranteed to be greater than 1.
+          const bool is_me = atomic_compare_exchange_weak_explicit(
+              lm_data->chunks + i, &stored, 1, metal::memory_order_relaxed,
+              metal::memory_order_relaxed);
+          if (is_me) {
+            offs = mtl_memalloc_alloc(mem_alloc, kChunkBytes);
+            atomic_store_explicit(lm_data->chunks + i, offs,
+                                  metal::memory_order_relaxed);
+            break;
+          } else if (stored > 1) {
+            offs = stored;
+            break;
+          }
+          // |stored| == 1, just spin
+        }
+        return offs;
+      }
+
+      device char *get_elem_from_chunk(int i, PtrOffset chunk_ptr_offs) {
+        device char *chunk_ptr = reinterpret_cast<device char *>(
+            mtl_memalloc_to_ptr(mem_alloc, chunk_ptr_offs));
+        const uint32_t mask = ((1 << lm_data->log2_num_elems_per_chunk) - 1);
+        return chunk_ptr + ((i & mask) * lm_data->element_stride);
+      }
+    };
 
     [[maybe_unused]] int is_active(device byte *addr, SNodeMeta meta, int i) {
       if (meta.type == SNodeMeta::Root || meta.type == SNodeMeta::Dense) {
@@ -207,8 +218,7 @@ STR(
       device auto *n_ptr = reinterpret_cast<device atomic_int *>(
           addr + (meta.num_slots * meta.element_stride));
       return atomic_load_explicit(n_ptr, metal::memory_order_relaxed);
-    }
-)
+    })
 METAL_END_RUNTIME_UTILS_DEF
 // clang-format on
 

diff --git a/taichi/backends/metal/struct_metal.cpp b/taichi/backends/metal/struct_metal.cpp
@@ -26,7 +26,7 @@ namespace shaders {
 
 }  // namespace shaders
 
-constexpr size_t kListManagerSize = sizeof(shaders::ListManager);
+constexpr size_t kListManagerDataSize = sizeof(shaders::ListManagerData);
 constexpr size_t kSNodeMetaSize = sizeof(shaders::SNodeMeta);
 constexpr size_t kSNodeExtractorsSize = sizeof(shaders::SNodeExtractors);
 
@@ -226,8 +226,8 @@ class StructCompiler {
   }
 
   size_t compute_runtime_size() {
-    size_t result = (max_snodes_) *
-                    (kSNodeMetaSize + kSNodeExtractorsSize + kListManagerSize);
+    size_t result = (max_snodes_) * (kSNodeMetaSize + kSNodeExtractorsSize +
+                                     kListManagerDataSize);
     result += sizeof(uint32_t) * kNumRandSeeds;
     return result;
   }