Files
composable_kernel/example/51_avgpool3d_bwd/avgpool3d_bwd_common.hpp
emezh db2524be2d Verify HostTensorDescriptor when it is created (#2829)
* add proper GEMM layout verification

* Handle "auto" strides.

CalculateStrides only called when tensor's strides are empty or all of them are <=0 (auto strides).
CalculateStrides now supports GEMM::ColumnsMajor order. The assumption is still that it applies only to the inner two dims.
ValidateStrides throws if any of the tensor's strides is <=0.
profile_gemm_multiply_add updated to support "auto" strides for tensors.

Manual tests for profile_gemm_multiply_add (matrix B in Row and Col modes)
auto-strides
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 0 0 0 0 0
	bin/ckProfiler gemm_multiply_add 0 1 1 1 0 1 128 128 128 0 0 0 0 0
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 -1 -1 -1 -1 -1
Note, -1 should be deprecated (use 0 instead)

explicit strides (same as auto)
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 128 128 128 128 128
	bin/ckProfiler gemm_multiply_add 0 1 1 1 0 1 128 128 128 128 128 128 128 128

explicit strides (not the same as auto)
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 130 132 134 136 138
	bin/ckProfiler gemm_multiply_add 0 1 1 1 0 1 128 128 128 130 132 134 136 138

mix of explicit and auto strides
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 128 128 128 128 0

invalid stride
	bin/ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 0 0 0 0 64
	terminate called after throwing an instance of 'std::runtime_error'
	  what():  Invalid strides for RowMajor: mLens: 128 128 , mStrides: 64 1
	Aborted (core dumped)

* - add more names to ck::tensor_layout for easier namespace hierarchy checking
- updated convolutional layouts to use explicit ones or BaseConvolutionalLayout where it is not clear which layout to use (TBD) - see include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp

* added handling of partially initialized strides for GEMM. fixed more tests.

* clang-format and more fixes

* replace long dash by a simple hyphen - causes build failure in CK codegen.

* increase sizeof input, otherwise output size becomes zero or negative with large filter size

* select stride based on layout

* specify layout explicitly to avoid errors in HostTensorDescriptor creation

* add validation for higher GEMM tensor dimensions.; Add docstring to `HostTensorDescriptor`

* Not clear why permute test in test/permute_scale/test_permute_scale.cpp uses a lot of invalid strides. Setting layout to BypassLayoutVerification to avoid a lot of errors

* fix test (incl removing invalid config)

* fix moe examples:
- (in .cpp) add layout argument to non-2D tensors
- (in .hpp) fix asserts/failures that show up in Debug mode, specifically addressing 2D tensor by a single index (and 3D tensor by 2d index)

* fix moe_gemm2 example.

* fix profile and wmma examples

* clean-up early mods for ckprofile. verified with:
```
ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 0 0 0 0 0
ckProfiler gemm_multiply_add 0 1 1 1 0 1 128 128 128 0 0 0 0 0
ckProfiler gemm_multiply_add 0 0 1 1 0 1 128 128 128 130 132 134 136 138
ckProfiler gemm_multiply_add 0 1 1 1 0 1 128 128 128 130 132 134 136 138
#
ckProfiler gemm_fastgelu 1 0 1 2 0 1 128 128 128 0 0 0
ckProfiler gemm_fastgelu 1 1 1 2 0 1 128 128 128 0 0 0
ckProfiler gemm_fastgelu 1 2 1 2 0 1 128 128 128 0 0 0
ckProfiler gemm_fastgelu 1 3 1 2 0 1 128 128 128 0 0 0
ckProfiler gemm_fastgelu 1 0 1 2 0 1 128 128 128 128 128 128
#
ckProfiler gemm_add_relu 0 0 1 1 0 1 128 128 128 0 0 0 0
# ckProfiler gemm_add_relu 0 1 1 1 0 1 128 128 128 0 0 0 0    # not implemented
# ckProfiler gemm_add_relu 0 2 1 1 0 1 128 128 128 0 0 0 0    # not implemented
# ckProfiler gemm_add_relu 0 3 1 1 0 1 128 128 128 0 0 0 0    # not implemented
ckProfiler gemm_add_relu 0 0 1 1 0 1 128 128 128 128 128 128 128
#
ckProfiler gemm_add_relu_add_layernorm 1 0 1 1 0 0 128 128 128 0 0 0 0 0
ckProfiler gemm_add_relu_add_layernorm 1 1 1 1 0 0 128 128 128 0 0 0 0 0
ckProfiler gemm_add_relu_add_layernorm 1 2 1 1 0 0 128 128 128 0 0 0 0 0
ckProfiler gemm_add_relu_add_layernorm 1 3 1 1 0 0 128 128 128 0 0 0 0 0
ckProfiler gemm_add_relu_add_layernorm 1 0 1 1 0 0 128 128 128 130 132 134 136 138
#
example_gemm_add_multiply_dl_fp16
example_gemm_add_multiply_xdl_fp16
#
ckProfiler gemm_blockscale_wp 7 1 1 1 1 0 1 128 128 128 0 0 0
ckProfiler gemm_blockscale_wp 7 1 1 1 1 0 1 128 128 128 128 128 128
```

* temporary skip first 8 test configs - they throw error

* temporary skip first 8 test configs in wmma too - they throw error

---------

Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
2025-09-25 18:22:13 -07:00

153 lines
6.5 KiB
C++

// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include "ck/ck.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_tensor.hpp"
#include "ck/library/utility/host_tensor_generator.hpp"
#include "ck/library/utility/literals.hpp"
#include "ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp"
template <typename TensorLayout>
std::vector<ck::index_t> f_tensor_strides_ncdhw(ck::index_t N_,
ck::index_t C_,
ck::index_t D,
ck::index_t H,
ck::index_t W,
TensorLayout layout)
{
using namespace ck::literals;
(void)N_;
if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
return {C_ * D * H * W, D * H * W, H * W, W, 1_uz};
else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
return {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_};
throw std::runtime_error("Avgpool3d_bwd: problem with layout. ");
return {0, 0, 0, 0, 0};
};
template <typename TensorLayout>
HostTensorDescriptor f_host_tensor_descriptor(std::size_t N_,
std::size_t C_,
std::size_t D,
std::size_t H,
std::size_t W,
TensorLayout layout)
{
using namespace ck::literals;
if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
{
return HostTensorDescriptor(
{N_, C_, D, H, W}, {C_ * D * H * W, D * H * W, H * W, W, 1_uz}, layout);
}
else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
{
return HostTensorDescriptor(
{N_, C_, D, H, W}, {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_}, layout);
}
throw std::runtime_error("Avgpool3d_bwd: problem with layout. ");
return HostTensorDescriptor({0, 0, 0, 0, 0}, {0, 0, 0, 0, 0}, layout);
};
template <typename DevicePoolBwdInstance,
typename DOutDataType,
typename DInDataType,
typename DOutLayout,
typename DInLayout>
bool pool3d_bwd_test(bool do_verification,
bool time_kernel,
ck::index_t N,
ck::index_t C,
ck::index_t Di,
ck::index_t Hi,
ck::index_t Wi,
std::vector<ck::index_t> window_lengths,
std::vector<ck::index_t> window_strides,
std::vector<ck::index_t> window_dilations,
std::vector<ck::index_t> dinput_left_pads,
std::vector<ck::index_t> dinput_right_pads)
{
auto OutSpatialLength = [&](auto InSpatialLength, int index) {
ck::index_t left_pad = dinput_left_pads[index];
ck::index_t right_pad = dinput_right_pads[index];
ck::index_t window_len = window_lengths[index];
ck::index_t stride = window_strides[index];
ck::index_t dilation = window_dilations[index];
ck::index_t eff = (window_len - 1) * dilation + 1;
return (InSpatialLength + left_pad + right_pad - eff) / stride + 1;
};
ck::index_t Do = OutSpatialLength(Di, 0);
ck::index_t Ho = OutSpatialLength(Hi, 1);
ck::index_t Wo = OutSpatialLength(Wi, 2);
Tensor<DOutDataType> dout(f_host_tensor_descriptor(N, C, Do, Ho, Wo, DOutLayout{}));
Tensor<DInDataType> din_dev(f_host_tensor_descriptor(N, C, Di, Hi, Wi, DInLayout{}));
Tensor<DInDataType> din_host(f_host_tensor_descriptor(N, C, Di, Hi, Wi, DInLayout{}));
std::cout << "dout: " << dout.mDesc << std::endl;
std::cout << "din_host: " << din_host.mDesc << std::endl;
dout.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{0.0, 1.0});
DeviceMem dout_device_buf(sizeof(DOutDataType) * dout.mDesc.GetElementSpaceSize());
DeviceMem din_device_buf(sizeof(DInDataType) * din_dev.mDesc.GetElementSpaceSize());
dout_device_buf.ToDevice(dout.mData.data());
din_device_buf.SetZero();
auto pool = DevicePoolBwdInstance{};
auto invoker_ptr = pool.MakeInvokerPointer();
auto argument_ptr =
pool.MakeArgumentPointer(static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
{N, C, Do, Ho, Wo},
{N, C, Di, Hi, Wi},
f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, DOutLayout{}),
f_tensor_strides_ncdhw(N, C, Di, Hi, Wi, DInLayout{}),
window_lengths,
window_strides,
window_dilations,
dinput_left_pads,
dinput_right_pads);
if(!pool.IsSupportedArgument(argument_ptr.get()))
{
throw std::runtime_error("wrong! device_op with the specified compilation parameters does "
"not support this problem");
}
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::cout << "Perf: " << ave_time << std::endl;
bool pass = true;
if(do_verification)
{
auto ref_pool =
ck::tensor_operation::host::ReferenceAvgPoolBwd<3, DInDataType, DOutDataType>();
auto ref_invoker = ref_pool.MakeInvoker();
auto ref_argument = ref_pool.MakeArgument(din_host,
dout,
window_lengths,
window_strides,
window_dilations,
dinput_left_pads,
dinput_right_pads);
ref_invoker.Run(ref_argument);
din_device_buf.FromDevice(din_dev.mData.data());
pass = ck::utils::check_err(din_dev, din_host);
}
return pass;
}