roll out s/IF_FALSE/UNLESS/ for ET_LOG macros

Separating the big find/replace for the previous diff to ease review; this is just the result of

```
fastmod "(ET_LOG_(MSG_)?AND_RETURN_)IF_FALSE" "\${1}UNLESS"
```

with a manual revert in tensor_util.h for the two backward-compatibility defines.

ghstack-source-id: 6d3b14c
ghstack-comment-id: 2644056878
Pull Request resolved: #8318

Author: swolchok

extension/llm/custom_ops/op_sdpa.cpp

@@ -594,46 +594,46 @@ bool validate_flash_attention_args(
     const Tensor& key,
     const Tensor& value,
     const optional<Tensor>& attn_mask) {
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(query.dim() == 4, "query must be a 4D tensor");
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(key.dim() == 4, "key must be a 4D tensor");
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(value.dim() == 4, "value must be a 4D tensor");
+  ET_LOG_MSG_AND_RETURN_UNLESS(query.dim() == 4, "query must be a 4D tensor");
+  ET_LOG_MSG_AND_RETURN_UNLESS(key.dim() == 4, "key must be a 4D tensor");
+  ET_LOG_MSG_AND_RETURN_UNLESS(value.dim() == 4, "value must be a 4D tensor");
 
   // Sizes
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (query.size(3) == value.size(3)) && (key.size(3) == value.size(3)),
       "scaled_dot_product_attention_flash_attention: Q/K/V should have the same head size");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (query.scalar_type() == ScalarType::Float), "Query must be Float type");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (query.scalar_type() == key.scalar_type()) &&
           (query.scalar_type() == value.scalar_type()),
       "Key and Value must have the same data type as Query");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       !attn_mask.has_value() || attn_mask.value().dim() == 2,
       "Attention mask must be a 2D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       !attn_mask.has_value() ||
           attn_mask.value().scalar_type() == query.scalar_type(),
       "Attention mask must be a 2D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(query.dim_order().data(), query.dim()),
       "key cache must be in contiguous dim order");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(key.dim_order().data(), key.dim()),
       "value cache must be in contiguous dim order");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(value.dim_order().data(), value.dim()),
       "value cache must be in contiguous dim order");
 
   if (attn_mask.has_value()) {
-    ET_LOG_MSG_AND_RETURN_IF_FALSE(
+    ET_LOG_MSG_AND_RETURN_UNLESS(
         is_contiguous_dim_order(
             attn_mask.value().dim_order().data(), attn_mask.value().dim()),
         "value cache must be in contiguous dim order");
@@ -647,21 +647,21 @@ bool validate_cache_params(
     const Tensor& v_cache,
     int64_t start_pos,
     int64_t seq_length) {
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       k_cache.dim() == 4, "kcache must be a 4D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       v_cache.dim() == 4, "v_cache must be a 4D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       start_pos < k_cache.size(1),
       "start_pos must be less than key cache at dim 1");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       start_pos < v_cache.size(1),
       "start_pos must be less than value cache at dim 1");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (start_pos + seq_length) <= k_cache.size(1),
       "start_post + seq_length must be less than max seq length supported by key cache."
       "start pos: %" PRId64 ", seq_length: %" PRId64
@@ -671,7 +671,7 @@ bool validate_cache_params(
       seq_length,
       k_cache.size(1));
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (start_pos + seq_length) <= v_cache.size(1),
       "start_post + seq_length must be less than max seq length supported by key cache."
       "start pos: %" PRId64 ", seq_length: %" PRId64
@@ -682,11 +682,11 @@ bool validate_cache_params(
       v_cache.size(1));
 
   // Make sure they are in contiguous dim order
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(k_cache.dim_order().data(), k_cache.dim()),
       "key cache must be in contiguous dim order");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(v_cache.dim_order().data(), v_cache.dim()),
       "value cache must be in contiguous dim order");
 

extension/llm/custom_ops/op_tile_crop.cpp

@@ -19,12 +19,12 @@ bool check_tile_crop_out_args(
     const Tensor& in,
     int64_t tile_size,
     Tensor& out) {
-  ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, out));
-  ET_LOG_AND_RETURN_IF_FALSE(tensor_is_rank(in, 3));
-  ET_LOG_AND_RETURN_IF_FALSE(tensor_is_rank(out, 4));
-  ET_LOG_AND_RETURN_IF_FALSE(tile_size > 0);
-  ET_LOG_AND_RETURN_IF_FALSE(in.size(in.dim() - 1) % tile_size == 0);
-  ET_LOG_AND_RETURN_IF_FALSE(in.size(in.dim() - 2) % tile_size == 0);
+  ET_LOG_AND_RETURN_UNLESS(tensors_have_same_dtype(in, out));
+  ET_LOG_AND_RETURN_UNLESS(tensor_is_rank(in, 3));
+  ET_LOG_AND_RETURN_UNLESS(tensor_is_rank(out, 4));
+  ET_LOG_AND_RETURN_UNLESS(tile_size > 0);
+  ET_LOG_AND_RETURN_UNLESS(in.size(in.dim() - 1) % tile_size == 0);
+  ET_LOG_AND_RETURN_UNLESS(in.size(in.dim() - 2) % tile_size == 0);
   return true;
 }
 

extension/llm/custom_ops/op_update_cache.cpp

@@ -25,17 +25,17 @@ bool validate_cache_params(
     const Tensor& quantized_cache,
     int64_t start_pos,
     int64_t seq_length) {
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       quantized_cache.dim() == 4, "quantized cache must be a 4D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       quantized_value.dim() == 4, "quantized_value must be a 4D tensor");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       start_pos < quantized_cache.size(1),
       "start_pos must be less than cache size at dim 1");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       (start_pos + seq_length) <= quantized_cache.size(1),
       "start_post + seq_length must be less than max seq length supported by cache."
       "start pos: %" PRId64 ", seq_length: %" PRId64
@@ -46,12 +46,12 @@ bool validate_cache_params(
       quantized_cache.size(1));
 
   // Make sure they are in contiguous dim order
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(
           quantized_cache.dim_order().data(), quantized_cache.dim()),
       "quantized cache must be in contiguous dim order");
 
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       is_contiguous_dim_order(
           quantized_value.dim_order().data(), quantized_value.dim()),
       "quantized value must be in contiguous dim order");

extension/parallel/thread_parallel.cpp

@@ -53,9 +53,9 @@ bool parallel_for(
     const int64_t end,
     const int64_t grain_size,
     const std::function<void(int64_t, int64_t)>& f) {
-  ET_LOG_AND_RETURN_IF_FALSE(begin >= 0 && end >= 0);
-  ET_LOG_AND_RETURN_IF_FALSE(end >= begin);
-  ET_LOG_AND_RETURN_IF_FALSE(grain_size > 0);
+  ET_LOG_AND_RETURN_UNLESS(begin >= 0 && end >= 0);
+  ET_LOG_AND_RETURN_UNLESS(end >= begin);
+  ET_LOG_AND_RETURN_UNLESS(grain_size > 0);
   int64_t num_tasks = 0, chunk_size = 0;
   std::tie(num_tasks, chunk_size) =
       calc_num_tasks_and_chunk_size(begin, end, grain_size);

kernels/aten/cpu/op__empty_dim_order.cpp

@@ -44,13 +44,13 @@ inline bool _check__empty_out_dim_order(
   }
 
   // dim order size shall equal to input dim
-  ET_LOG_AND_RETURN_IF_FALSE(dim_order_ref.size() == out.dim());
+  ET_LOG_AND_RETURN_UNLESS(dim_order_ref.size() == out.dim());
 
-  ET_LOG_AND_RETURN_IF_FALSE(
+  ET_LOG_AND_RETURN_UNLESS(
       is_channels_last_dim_order(dim_order_ref.data(), dim_order_ref.size()) ||
       is_contiguous_dim_order(dim_order_ref.data(), dim_order_ref.size()));
 
-  ET_LOG_AND_RETURN_IF_FALSE(kMaxNumOfDimensions >= out.dim());
+  ET_LOG_AND_RETURN_UNLESS(kMaxNumOfDimensions >= out.dim());
   executorch::aten::StridesType target_strides[kMaxNumOfDimensions];
   dim_order_to_stride_nocheck(
       out.sizes().data(),
@@ -59,7 +59,7 @@ inline bool _check__empty_out_dim_order(
       target_strides);
 
   for (size_t i = 0; i < dim_order_ref.size(); i++) {
-    ET_LOG_AND_RETURN_IF_FALSE(target_strides[i] == out.strides()[i]);
+    ET_LOG_AND_RETURN_UNLESS(target_strides[i] == out.strides()[i]);
   }
 
   return true;

kernels/aten/cpu/op__to_dim_order_copy.cpp

@@ -47,43 +47,43 @@ bool check__to_dim_order_copy_args(
     executorch::aten::OptionalArrayRef<int64_t> dim_order,
     Tensor& out) {
   // Right now we only support blocking data transfer
-  ET_LOG_AND_RETURN_IF_FALSE(non_blocking == false);
+  ET_LOG_AND_RETURN_UNLESS(non_blocking == false);
 
   // dim_order is set, the target dim_order will be either contiguous or
   // channels_last memory format
   if (dim_order.has_value()) {
     executorch::aten::ArrayRef<int64_t> dim_order_ref = dim_order.value();
 
     // dim order size shall equal to input dim
-    ET_LOG_AND_RETURN_IF_FALSE(dim_order_ref.size() == input.dim());
+    ET_LOG_AND_RETURN_UNLESS(dim_order_ref.size() == input.dim());
 
-    ET_LOG_AND_RETURN_IF_FALSE(
+    ET_LOG_AND_RETURN_UNLESS(
         is_channels_last_dim_order(
             dim_order.value().data(), dim_order.value().size()) ||
         is_contiguous_dim_order(
             dim_order.value().data(), dim_order.value().size()));
 
     // Out Aten tensor shall have same memory format stride as dim_order
     const size_t kMaxNumOfDimensions = 16;
-    ET_LOG_AND_RETURN_IF_FALSE(kMaxNumOfDimensions >= out.dim());
+    ET_LOG_AND_RETURN_UNLESS(kMaxNumOfDimensions >= out.dim());
     executorch::aten::StridesType target_strides[kMaxNumOfDimensions];
     dim_order_to_stride_nocheck(
         out.sizes().data(),
         dim_order_ref.data(),
         dim_order_ref.size(),
         target_strides);
-    ET_LOG_AND_RETURN_IF_FALSE(out.dim() == dim_order_ref.size());
+    ET_LOG_AND_RETURN_UNLESS(out.dim() == dim_order_ref.size());
     for (size_t i = 0; i < dim_order_ref.size(); i++) {
-      ET_LOG_AND_RETURN_IF_FALSE(target_strides[i] == out.strides()[i]);
+      ET_LOG_AND_RETURN_UNLESS(target_strides[i] == out.strides()[i]);
     }
 
   } else { // dim_order is not set, preserve the dim order of input
 
     auto out_strides = out.strides();
     auto input_strides = input.strides();
-    ET_LOG_AND_RETURN_IF_FALSE(input_strides.size() == out_strides.size());
+    ET_LOG_AND_RETURN_UNLESS(input_strides.size() == out_strides.size());
     for (size_t i = 0; i < input_strides.size(); i++) {
-      ET_LOG_AND_RETURN_IF_FALSE(input_strides[i] == out_strides[i]);
+      ET_LOG_AND_RETURN_UNLESS(input_strides[i] == out_strides[i]);
     }
   }
   return true;

kernels/aten/cpu/util/copy_ops_util.cpp

@@ -22,43 +22,43 @@ bool check__to_dim_order_copy_args(
     executorch::aten::OptionalArrayRef<int64_t> dim_order,
     Tensor& out) {
   // Right now we only support blocking data transfer
-  ET_LOG_AND_RETURN_IF_FALSE(non_blocking == false);
+  ET_LOG_AND_RETURN_UNLESS(non_blocking == false);
 
   // dim_order is set, the target dim_order will be either contiguous or
   // channels_last memory format
   if (dim_order.has_value()) {
     executorch::aten::ArrayRef<int64_t> dim_order_ref = dim_order.value();
 
     // dim order size shall equal to input dim
-    ET_LOG_AND_RETURN_IF_FALSE(dim_order_ref.size() == input.dim());
+    ET_LOG_AND_RETURN_UNLESS(dim_order_ref.size() == input.dim());
 
-    ET_LOG_AND_RETURN_IF_FALSE(
+    ET_LOG_AND_RETURN_UNLESS(
         is_channels_last_dim_order(
             dim_order.value().data(), dim_order.value().size()) ||
         is_contiguous_dim_order(
             dim_order.value().data(), dim_order.value().size()));
 
     // Out Aten tensor shall have same memory format stride as dim_order
     const size_t kMaxNumOfDimensions = 16;
-    ET_LOG_AND_RETURN_IF_FALSE(kMaxNumOfDimensions >= out.dim());
+    ET_LOG_AND_RETURN_UNLESS(kMaxNumOfDimensions >= out.dim());
     executorch::aten::StridesType target_strides[kMaxNumOfDimensions];
     dim_order_to_stride_nocheck(
         out.sizes().data(),
         dim_order_ref.data(),
         dim_order_ref.size(),
         target_strides);
-    ET_LOG_AND_RETURN_IF_FALSE(out.dim() == dim_order_ref.size());
+    ET_LOG_AND_RETURN_UNLESS(out.dim() == dim_order_ref.size());
     for (size_t i = 0; i < dim_order_ref.size(); i++) {
-      ET_LOG_AND_RETURN_IF_FALSE(target_strides[i] == out.strides()[i]);
+      ET_LOG_AND_RETURN_UNLESS(target_strides[i] == out.strides()[i]);
     }
 
   } else { // dim_order is not set, preserve the dim order of input
 
     auto out_strides = out.strides();
     auto input_strides = input.strides();
-    ET_LOG_AND_RETURN_IF_FALSE(input_strides.size() == out_strides.size());
+    ET_LOG_AND_RETURN_UNLESS(input_strides.size() == out_strides.size());
     for (size_t i = 0; i < input_strides.size(); i++) {
-      ET_LOG_AND_RETURN_IF_FALSE(input_strides[i] == out_strides[i]);
+      ET_LOG_AND_RETURN_UNLESS(input_strides[i] == out_strides[i]);
     }
   }
   return true;

kernels/optimized/cpu/op_bmm.cpp

@@ -31,46 +31,46 @@ namespace {
 // Verifies that the parameters are valid.
 bool check_bmm_out_args(const Tensor& self, const Tensor& mat2, Tensor& out) {
   // Ensure dimensions is 3 for all input and out
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.dim() == mat2.dim(),
       "self.dim() %zd != mat2.dim() %zd",
       self.dim(),
       mat2.dim());
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.dim() == out.dim(),
       "self.dim() %zd != out.dim() %zd",
       self.dim(),
       out.dim());
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.dim() == 3, "self.dim() %zd != 3", self.dim());
   // Ensure batch larger than or equals to 0
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.size(0) >= 0, "self.size(0) %zd < 0", self.size(0));
   // Ensure batches are the same
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.size(0) == mat2.size(0),
       "self.size(0) %zd != mat2.size(0) %zd",
       self.size(0),
       mat2.size(0));
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.size(0) == out.size(0),
       "self.size(0) %zd != out.size(0) %zd",
       self.size(0),
       out.size(0));
   // Ensure the out size is compatible with input tensors
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       mat2.size(2) == out.size(2),
       "mat2.size(2) %zd != out.size(2) %zd",
       mat2.size(2),
       out.size(2));
-  ET_LOG_MSG_AND_RETURN_IF_FALSE(
+  ET_LOG_MSG_AND_RETURN_UNLESS(
       self.size(1) == out.size(1),
       "self.size(1) %zd != out.size(1) %zd",
       self.size(1),
       out.size(1));
 
   // Ensure that all tensors share a dtype
-  ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(self, mat2, out));
+  ET_LOG_AND_RETURN_UNLESS(tensors_have_same_dtype(self, mat2, out));
 
   return true;
 }

kernels/portable/cpu/op__empty_dim_order.cpp

@@ -30,20 +30,20 @@ bool _check__empty_out_dim_order(OptionalIntArrayRef dim_order, Tensor& out) {
     // out tensor's dim order shall equal to input dim order
     IntArrayRef dim_order_ref = dim_order.value();
 
-    ET_LOG_AND_RETURN_IF_FALSE(
+    ET_LOG_AND_RETURN_UNLESS(
         is_channels_last_dim_order(
             dim_order.value().data(), dim_order.value().size()) ||
         is_contiguous_dim_order(
             dim_order.value().data(), dim_order.value().size()));
 
     // Out tensor shall have same dim order as dim_order
-    ET_LOG_AND_RETURN_IF_FALSE(out_dim_order.size() == dim_order_ref.size());
+    ET_LOG_AND_RETURN_UNLESS(out_dim_order.size() == dim_order_ref.size());
     for (size_t i = 0; i < dim_order_ref.size(); i++) {
-      ET_LOG_AND_RETURN_IF_FALSE(out_dim_order[i] == dim_order_ref[i]);
+      ET_LOG_AND_RETURN_UNLESS(out_dim_order[i] == dim_order_ref[i]);
     }
   } else { // dim_order is not set, out tensor should be contiguous memory
            // format
-    ET_LOG_AND_RETURN_IF_FALSE(
+    ET_LOG_AND_RETURN_UNLESS(
         is_contiguous_dim_order(out_dim_order.data(), out_dim_order.size()));
   }
   return true;