[refactor] Some renamings (#4959)

taichi/backends/cuda/codegen_cuda.cpp

@@ -70,7 +70,7 @@ class CodeGenLLVMCUDA : public CodeGenLLVM {
             continue;
           }
           arg_buffers[i] = context.get_arg<void *>(i);
-          if (!context.is_device_allocation[i]) {
+          if (!context.is_device_allocations[i]) {
             // Note: both numpy and PyTorch support arrays/tensors with zeros
             // in shapes, e.g., shape=(0) or shape=(100, 0, 200). This makes
             // `arr_sz` zero.

taichi/codegen/codegen_llvm.cpp

@@ -2296,7 +2296,7 @@ FunctionType CodeGenLLVM::compile_module_to_executable() {
     // For taichi ndarrays, context.args saves pointer to its
     // |DeviceAllocation|, CPU backend actually want to use the raw ptr here.
     for (int i = 0; i < (int)args.size(); i++) {
-      if (args[i].is_array && context.is_device_allocation[i] &&
+      if (args[i].is_array && context.is_device_allocations[i] &&
           context.array_runtime_sizes[i] > 0) {
         DeviceAllocation *ptr =
             static_cast<DeviceAllocation *>(context.get_arg<void *>(i));

taichi/program/context.h

@@ -24,14 +24,14 @@ struct RuntimeContext {
   int32 cpu_thread_id;
 
   // Note that I've tried to group `array_runtime_size` and
-  // `is_device_allocation` into a small struct. However, it caused some test
+  // `is_device_allocations` into a small struct. However, it caused some test
   // cases to stuck.
 
   // `array_runtime_size` records the runtime size of the
   // corresponding array arguments.
   uint64 array_runtime_sizes[taichi_max_num_args_total]{0};
-  // `is_device_allocation` is true iff args[i] is a DeviceAllocation*.
-  bool is_device_allocation[taichi_max_num_args_total]{false};
+  // `is_device_allocations` is true iff i-th arg is a `DeviceAllocation*`.
+  bool is_device_allocations[taichi_max_num_args_total]{false};
   // We move the pointer of result buffer from LLVMRuntime to RuntimeContext
   // because each real function need a place to store its result, but
   // LLVMRuntime is shared among functions. So we moved the pointer to
@@ -61,7 +61,7 @@ struct RuntimeContext {
   }
 
   void set_array_is_device_allocation(int i, bool is_device_allocation) {
-    this->is_device_allocation[i] = is_device_allocation;
+    this->is_device_allocations[i] = is_device_allocation;
   }
 
   template <typename T>

taichi/runtime/opengl/opengl_api.cpp

@@ -402,19 +402,19 @@ void DeviceCompiledTaichiKernel::launch(RuntimeContext &ctx,
     int i = item.first;
     TI_ASSERT(args[i].is_array);
     const auto arr_sz = ctx.array_runtime_sizes[i];
-    if (arr_sz == 0 || ctx.is_device_allocation[i]) {
+    if (arr_sz == 0 || ctx.is_device_allocations[i]) {
       continue;
     }
     has_ext_arr = true;
-    if (arr_sz != item.second.total_size ||
+    if (arr_sz != item.second.runtime_size ||
         ext_arr_bufs_[i] == kDeviceNullAllocation) {
       if (ext_arr_bufs_[i] != kDeviceNullAllocation) {
         device_->dealloc_memory(ext_arr_bufs_[i]);
       }
       ext_arr_bufs_[i] = device_->allocate_memory({arr_sz, /*host_write=*/true,
                                                    /*host_read=*/true,
                                                    /*export_sharing=*/false});
-      item.second.total_size = arr_sz;
+      item.second.runtime_size = arr_sz;
     }
     void *host_ptr = (void *)ctx.args[i];
     void *baseptr = device_->map(ext_arr_bufs_[i]);
@@ -471,7 +471,7 @@ void DeviceCompiledTaichiKernel::launch(RuntimeContext &ctx,
     //       On most devices this number is 8. But I need to look up how
     //       to query this information so currently this is thrown from OpenGl.
     for (const auto [arg_id, bind_id] : program_.used.arr_arg_to_bind_idx) {
-      if (ctx.is_device_allocation[arg_id]) {
+      if (ctx.is_device_allocations[arg_id]) {
         DeviceAllocation *ptr =
             static_cast<DeviceAllocation *>((void *)ctx.args[arg_id]);
 
@@ -507,7 +507,7 @@ void DeviceCompiledTaichiKernel::launch(RuntimeContext &ctx,
     for (auto &item : program_.arr_args) {
       int i = item.first;
       const auto arr_sz = ctx.array_runtime_sizes[i];
-      if (arr_sz > 0 && !ctx.is_device_allocation[i]) {
+      if (arr_sz > 0 && !ctx.is_device_allocations[i]) {
         uint8_t *baseptr = (uint8_t *)device_->map(ext_arr_bufs_[i]);
         memcpy((void *)ctx.args[i], baseptr, arr_sz);
         device_->unmap(ext_arr_bufs_[i]);

taichi/runtime/opengl/opengl_api.h

@@ -66,7 +66,7 @@ struct CompiledArrayArg {
   bool is_scalar{false};
   std::vector<int> element_shape;
   size_t shape_offset_in_bytes_in_args_buf{0};
-  size_t total_size{0};  // Runtime information
+  size_t runtime_size{0};  // Runtime information
 
   TI_IO_DEF(field_dim,
             is_scalar,

taichi/runtime/vulkan/runtime.cpp

@@ -79,7 +79,7 @@ class HostDeviceContextBlitter {
       char *device_ptr = device_base + arg.offset_in_mem;
       do {
         if (arg.is_array) {
-          if (!host_ctx_->is_device_allocation[i] && ext_arr_size.at(i)) {
+          if (!host_ctx_->is_device_allocations[i] && ext_arr_size.at(i)) {
             // Only need to blit ext arrs (host array)
             DeviceAllocation buffer = ext_arrays.at(i);
             char *const device_arr_ptr =
@@ -150,7 +150,7 @@ class HostDeviceContextBlitter {
       for (int i = 0; i < ctx_attribs_->args().size(); ++i) {
         const auto &arg = ctx_attribs_->args()[i];
         if (arg.is_array) {
-          if (!host_ctx_->is_device_allocation[i] && ext_arr_size.at(i)) {
+          if (!host_ctx_->is_device_allocations[i] && ext_arr_size.at(i)) {
             require_sync = true;
           }
         }
@@ -166,7 +166,7 @@ class HostDeviceContextBlitter {
     for (int i = 0; i < ctx_attribs_->args().size(); ++i) {
       const auto &arg = ctx_attribs_->args()[i];
       if (arg.is_array) {
-        if (!host_ctx_->is_device_allocation[i] && ext_arr_size.at(i)) {
+        if (!host_ctx_->is_device_allocations[i] && ext_arr_size.at(i)) {
           // Only need to blit ext arrs (host array)
           DeviceAllocation buffer = ext_arrays.at(i);
           char *const device_arr_ptr =
@@ -455,7 +455,7 @@ void VkRuntime::launch_kernel(KernelHandle handle, RuntimeContext *host_ctx) {
     const auto &args = ti_kernel->ti_kernel_attribs().ctx_attribs.args();
     for (auto &arg : args) {
       if (arg.is_array) {
-        if (host_ctx->is_device_allocation[i]) {
+        if (host_ctx->is_device_allocations[i]) {
           // NDArray
           if (host_ctx->args[i]) {
             any_arrays[i] = *(DeviceAllocation *)(host_ctx->args[i]);
@@ -546,7 +546,7 @@ void VkRuntime::launch_kernel(KernelHandle handle, RuntimeContext *host_ctx) {
   // Dealloc external arrays
   for (auto pair : any_arrays) {
     if (pair.second != kDeviceNullAllocation) {
-      if (!host_ctx->is_device_allocation[pair.first]) {
+      if (!host_ctx->is_device_allocations[pair.first]) {
         device_->dealloc_memory(pair.second);
       }
     }