#0: Handle 0 volume case in layout change code (#14578)

#0: handle 0 volume case in layout change code
tenstorrent · Nov 1, 2024 · c42a534 · c42a534
1 parent 5c31e3e
commit c42a534
Showing 1 changed file with 21 additions and 2 deletions.
diff --git a/tt_metal/common/test_tiles.hpp b/tt_metal/common/test_tiles.hpp
@@ -32,6 +32,10 @@ std::vector<T> convert_to_tile_layout(
     std::optional<tt::stl::Span<const uint32_t>> face_shape = std::nullopt) {
     ZoneScoped;
     std::vector<T> result;
+    if(data.size() == 0) {
+        return result;
+    }
+
     result.reserve(data.size());
     auto tile_H = tile_shape.has_value() ? tile_shape.value()[0] : tt::constants::TILE_HEIGHT;
     auto tile_W = tile_shape.has_value() ? tile_shape.value()[1] : tt::constants::TILE_WIDTH;
@@ -86,6 +90,9 @@ std::vector<T> convert_to_flat_layout(
     std::optional<tt::stl::Span<const uint32_t>> face_shape = std::nullopt) {
     ZoneScoped;
     std::vector<T> result;
+    if(data.size() == 0) {
+        return result;
+    }
     result.reserve(data.size());
     auto tile_H = tile_shape.has_value() ? tile_shape.value()[0] : tt::constants::TILE_HEIGHT;
     auto tile_W = tile_shape.has_value() ? tile_shape.value()[1] : tt::constants::TILE_WIDTH;
@@ -123,9 +130,13 @@ inline std::vector<T> untilize_nchw(const BufferType<T>& in, tt::stl::Span<const
     auto tile_H = tile_shape.has_value() ? tile_shape.value()[0] : tt::constants::TILE_HEIGHT;
     auto tile_W = tile_shape.has_value() ? tile_shape.value()[1] : tt::constants::TILE_WIDTH;
 
+    std::vector<T> result;
+    if(in.size() == 0) {
+        return result;
+    }
+
     TT_ASSERT(shape[shape.size() - 2] % tile_H == 0 && shape[shape.size() - 1] % tile_W == 0);
 
-    std::vector<T> result;
     // Untilize into row major
     int H = shape[shape.size() - 2], W = shape[shape.size() - 1];
     auto batch_size = 1;
@@ -164,6 +175,11 @@ inline std::uint32_t round_up_to_tile(int val, int tile_val) { return (val + til
 template <typename T, template <typename...> typename BufferType>
 inline std::vector<T> tilize_nchw(const BufferType<T>& in_rowmajor, tt::stl::Span<const uint32_t> shape, std::optional<tt::stl::Span<const uint32_t>> tile_shape = std::nullopt) {
     ZoneScoped;
+    std::vector<T> tilized_result;
+    if(in_rowmajor.size() == 0) {
+        return tilized_result;
+    }
+
     int H = shape[shape.size() - 2], W = shape[shape.size() - 1];
     auto batch_size = 1;
     for (int i = 0; i < shape.size() - 2; i++) {
@@ -174,7 +190,6 @@ inline std::vector<T> tilize_nchw(const BufferType<T>& in_rowmajor, tt::stl::Spa
     auto tile_W = tile_shape.has_value() ? tile_shape.value()[1] : tt::constants::TILE_WIDTH;
     int OH = round_up_to_tile(H, tile_H);
     int OW = round_up_to_tile(W, tile_W);
-    std::vector<T> tilized_result;
     tilized_result.resize(batch_size * OH * OW);
     std::fill(tilized_result.begin(), tilized_result.end(), 0);
     int out_index = 0;
@@ -230,6 +245,10 @@ inline std::vector<T> convert_layout(
     std::optional<tt::stl::Span<const uint32_t>> tile_shape = std::nullopt,
     std::optional<const tt::stl::Span<const uint32_t>> face_shape = std::nullopt) {
     ZoneScoped;
+    if(inp.size() == 0) {
+        return std::vector<T>();
+    }
+
     switch (inL) {
         case tests::utils::TensorLayoutType::TILED_SWIZZLED:
             if (outL == tests::utils::TensorLayoutType::TILED_NFACES) {