cutlass/test/unit/cute/hopper/tma_load_testbed.hpp

/***************************************************************************************************
 * Copyright (c) 2017 - 2024 NVIDIA CORPORATION & AFFILIATES. 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.
 *
 **************************************************************************************************/

#pragma once

#include "cutlass_unit_test.h"

#include <iostream>
#include <cstdint>

#include <thrust/host_vector.h>
#include <thrust/device_vector.h>

#include <cute/tensor.hpp>

namespace cutlass::test {

template <class ElementType, class SmemLayout>
struct SharedStorage
{
  cute::ArrayEngine<ElementType, cute::cosize_v<SmemLayout>> smem;
  alignas(16) cute::uint64_t tma_load_mbar[1];
};

#if CUDA_12_0_SM90_FEATURES_SUPPORTED

template <class T, class TiledCopy, class CTA_Tiler, class GmemLayout, class SmemLayout>
__global__ void
tma_test_device_cute(T const* g_in, T* g_out,
                     CUTE_GRID_CONSTANT TiledCopy const tma, CTA_Tiler cta_tiler,
                     GmemLayout gmem_layout, SmemLayout smem_layout)
{
  using namespace cute;
  CUTE_STATIC_ASSERT_V(product_each(shape(cta_tiler)) == product_each(shape(smem_layout)));

  // Use Shared Storage structure to allocate and distribute aligned SMEM addresses
  extern __shared__ char shared_memory[];
  using SharedStorage = SharedStorage<T, SmemLayout>;
  SharedStorage& shared_storage = *reinterpret_cast<SharedStorage*>(shared_memory);

  // Construct SMEM tensor
  Tensor sA = make_tensor(make_smem_ptr(shared_storage.smem.begin()), smem_layout);  // (CTA_TILE_M,CTA_TILE_N,...)
  // Shared memory barriers use 64bits in SMEM for synchronization
  uint64_t* tma_load_mbar = shared_storage.tma_load_mbar;

  // TMA requires special handling of strides to deal with coord codomain mapping
  // Represent the full tensors -- get these from TMA
  Tensor mA = tma.get_tma_tensor(shape(gmem_layout));
  Tensor mB = make_tensor(make_gmem_ptr<T>(g_out), gmem_layout);

  constexpr int R = rank_v<CTA_Tiler>;
  Tensor gA = flat_divide(mA, cta_tiler);               // (CTA_TILE_M,CTA_TILE_N,...REST_M,REST_N,...)
  Tensor gB = flat_divide(mB, cta_tiler);               // (CTA_TILE_M,CTA_TILE_N,...REST_M,REST_N,...)

  //
  // Prepare the TMA_LOAD
  //

  auto cta_tma = tma.get_slice(Int<0>{});                            // CTA slice
  Tensor tAgA_x = cta_tma.partition_S(gA);                           // (TMA,TMA_M,TMA_N,REST_M,REST_N)
  Tensor tAsA_x = cta_tma.partition_D(sA);                           // (TMA,TMA_M,TMA_N)

#if 0
  if (thread0()) {
    print(tma);
    print("TILE  :  "); print(cta_tiler); print("\n");
    print("  mA  :  "); print(  mA);   print("\n");
    print("  mB  :  "); print(  mB);   print("\n");
    print("  gA  :  "); print(  gA);   print("\n");
    print("  gB  :  "); print(  gB);   print("\n");
    print("  sA  :  "); print(  sA);   print("\n");
    print("tAgA_x:  "); print(tAgA_x); print("\n");
    print("tAsA_x:  "); print(tAsA_x); print("\n");
  }
#endif

  //
  // Perform the TMA_LOAD
  //

  // INPUT: Group the REST_X modes and the TMA_X modes to easily iterate through the tiles
  Tensor tAgA = group_modes<1,rank(tAgA_x)>(tAgA_x);                 // (TMA,REST)
  Tensor tAsA = group_modes<1,rank(tAsA_x)>(tAsA_x);                 // (TMA,REST)
  static_assert(size<1>(tAsA) == 1);

  // OUTPUT: Group the CTA_TILE_X modes and REST_X modes for output
  Tensor tBgB = group_modes<0,R>(group_modes<R,rank(gB)>(gB));       // (CTA_TILE, REST)

#if 0
  if (thread0()) {
    print("tAgA  :  "); print(tAgA); print("\n");
    print("tAsA  :  "); print(tAsA); print("\n");
    print("tBgB  :  "); print(tBgB); print("\n");
  }
#endif

  // Test L2 prefetch
  if (threadIdx.x == 0) {
    prefetch(tma, tAgA);
  }

  // Loop over the TMA stages, using smem as our buffer
  for (int stage = 0; stage < size<1>(tAgA); ++stage)
  {
    // Set the bytes transferred in this TMA transaction (may involve multiple issues)
    constexpr int kTmaTransactionBytes = sizeof(ArrayEngine<T, size(sA)>);

    if (threadIdx.x == 0)
    {
      /// Initialize shared memory barrier
      tma_load_mbar[0] = 0;
      cute::initialize_barrier(tma_load_mbar[0], 1 /*numThreads*/);
      cute::set_barrier_transaction_bytes(tma_load_mbar[0], kTmaTransactionBytes);

      copy(tma.with(tma_load_mbar[0]), tAgA(_,stage), tAsA(_,0));
    }
    __syncthreads();

    /// Wait on the shared memory barrier until the phase bit flips from kPhaseBit value
    constexpr int kPhaseBit = 0;
    cute::wait_barrier(tma_load_mbar[0], kPhaseBit);

    //
    // Write out trivially smem -> gmem
    //

    // Subbyte elements could cause race conditions, so be even more conservative
    if (thread0()) {
      copy(sA, tBgB(_,stage));
    }

    __syncthreads();
  }
}

template <class T, class TmaType = T, class CopyOp, class GMEM_Layout, class SMEM_Layout, class CTA_Tile>
auto
test_tma_load(CopyOp      const& copy_op,
              GMEM_Layout const& gmem_layout,
              SMEM_Layout const& smem_layout,
              CTA_Tile    const& cta_tile)
{
  using namespace cute;

  // Allocate and initialize host test data
  size_t N = ceil_div(cosize(gmem_layout) * sizeof_bits<T>::value, 8);
  thrust::host_vector<uint8_t> h_in(N);
  for (size_t i = 0; i < h_in.size(); ++i) {
    h_in[i] = uint8_t(i % 13);
  }
  Tensor hA_in  = make_tensor(recast_ptr<T>(h_in.data()), gmem_layout);

  // Allocate and initialize device test data
  thrust::device_vector<uint8_t> d_in = h_in;
  thrust::device_vector<uint8_t> d_out(h_in.size(), uint8_t(-1)); // overflow uint

  // Create TMA for this device Tensor
  Tensor gA = make_tensor(make_gmem_ptr<T>(raw_pointer_cast(d_in.data())), gmem_layout);
  auto tma = make_tma_copy<TmaType>(copy_op, gA, smem_layout, cta_tile, Int<1>{});
  //print(tma);

  // Launch
  int smem_size = int(sizeof(SharedStorage<T, decltype(smem_layout)>));
  tma_test_device_cute<<<1, 128, smem_size>>>(
    reinterpret_cast<T const*>(raw_pointer_cast(d_in.data())),
    reinterpret_cast<T*>      (raw_pointer_cast(d_out.data())),
    tma, cta_tile,
    gmem_layout,
    smem_layout);

  // Copy results back to host
  thrust::host_vector<uint8_t> h_out = d_out;
  Tensor hA_out = make_tensor(recast_ptr<T>(h_out.data()), gmem_layout);

  // Validate the results. Print only the first 3 errors.
  int count = 3;
  for (int i = 0; i < int(size(hA_out)) && count > 0; ++i) {
    EXPECT_EQ(hA_in(i), hA_out(i));
    if (hA_in(i) != hA_out(i)) {
      --count;
    }
  }

  return tma;
}

#endif

} // end namespace cutlass::test