Add a tool for debugging copies

examples/sycl/pvc/CMakeLists.txt

@@ -27,6 +27,11 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 
+cutlass_example_add_executable(
+  copy_reverse_engineer
+  copy_reverse_engineer.cpp
+)
+
 cutlass_example_add_executable(
   pvc_gemm
   pvc_gemm.cpp

examples/sycl/pvc/copy_reverse_engineer.cpp

@@ -0,0 +1,133 @@
+/***************************************************************************************************
+ * Copyright (c) 2025 - 2025 Codeplay Software Ltd. All rights reserved.
+ * SPDX-License-Identifier: BSD-3-Clause
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ **************************************************************************************************/
+
+#include "cutlass/detail/layout.hpp"
+
+#include <cute/tensor.hpp>
+#include <sycl/sycl.hpp>
+#include <syclcompat.hpp>
+
+#include <cutlass/util/device_memory.h>
+#include <syclcompat/syclcompat.hpp>
+#include <cutlass/cutlass.h>
+
+using namespace syclcompat::experimental;
+
+#define SUBGROUP_SIZE (16)
+
+template <class CopyInstruction, class TensorS, class fragment_size>
+void copy_kernel(TensorS S) {
+  using namespace cute;
+  using Element = typename TensorS::value_type;
+
+  // initialization
+  if(thread(0)){
+    for(int i=0; i<size(S); i++){
+      S(i) = static_cast<Element>(i);
+    }
+  }
+  syncthreads();
+
+  using CopyThreadShape = Shape<_1, Int<SUBGROUP_SIZE>>;
+  using traits_load = Copy_Traits<CopyInstruction, decltype(S)>;
+  using Atom_load = Copy_Atom<traits_load, Element>;
+  auto tiled_copy_load = make_tiled_copy(Atom_load{}.with(S),
+    Layout<CopyThreadShape>{},
+    make_layout(shape_div(typename traits_load::BlockShape{}, CopyThreadShape{})));
+
+  auto thr_copy_load = tiled_copy_load.get_slice(cutlass::get_sub_group_local_id());
+
+  using actual_fragment_size = std::conditional_t<std::is_same_v<fragment_size, void>, C<Atom_load::NumValDst>, fragment_size>;
+  Tensor fragment = make_tensor<Element>(make_shape(actual_fragment_size{},_1{},_1{}));
+  clear(fragment);
+
+  static_assert(actual_fragment_size::value >= Atom_load::NumValDst, "fragment is too small to hold all results!");
+  Tensor fragment_copy_view = make_tensor(fragment.data(), make_shape(C<Atom_load::NumValDst>{},_1{},_1{}));
+  auto blk_load_S = tiled_copy_load.get_pvc_tensor(make_coord(0, 0, 0),make_shape(_1{},_1{},_1{}));
+  copy(tiled_copy_load, blk_load_S, fragment_copy_view);
+
+  if(thread(0)){
+    print("fragment: "); print(fragment); print("\n");
+  }
+
+  for(int i=0;i<SUBGROUP_SIZE;i++){
+    if(thread(i)){
+      print("thread "); 
+      if(i<10) print(" ");
+      print(i); print(" ");
+    }
+  }
+  for(int i = 0;i < size(fragment); i++){
+    if(thread(0)){
+      print("\n    ");
+    }
+    for(int j=0;j<SUBGROUP_SIZE;j++){
+      if(thread(j)){
+        if(fragment(i)<10) print(" ");
+        if(fragment(i)<100) print(" ");
+        if(fragment(i)<1000) print(" ");
+        print(static_cast<int>(fragment(i))); print("      ");
+      }
+    }
+  }
+  if(thread(0)){
+    print("\n");
+  }
+}
+
+// by default select fragment size to match copy size. This can be set manually to a bigger value as copy size might be wrong
+template <class CopyInstruction, class dtype, class fragment_size = void>
+void copy(int global_M, int global_N) {
+  using namespace cute;
+
+  auto tensor_shape = make_shape(global_M, global_N);
+  int tensor_size = size(tensor_shape);
+  cutlass::DeviceAllocation<dtype> src(tensor_size);
+
+  Tensor tensor_S = make_tensor(make_gmem_ptr(src.get()), make_layout(tensor_shape, LayoutLeft{}));
+
+  auto gridDim = syclcompat::dim3(SUBGROUP_SIZE);
+  auto blockDim = syclcompat::dim3(SUBGROUP_SIZE);
+  launch<copy_kernel<CopyInstruction, decltype(tensor_S), fragment_size>>(
+      launch_policy{gridDim, blockDim,
+                    kernel_properties{sycl_exp::sub_group_size<SUBGROUP_SIZE>}},
+      tensor_S);
+
+  syclcompat::wait_and_throw();
+}
+
+int main(){
+  // for 16b copies use integers as floating point types could lose precision for bigger indices
+  // for 8b copies you have to work with overflow
+  copy<XE_2D_U16x32x32_LD_V, int16_t>(256, 256);
+  copy<XE_2D_U16x32x32_LD_N, int16_t>(256, 256);
+  return 0;
+}