mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-05 22:22:27 +00:00
db2524be2d
* 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>
365 lines
16 KiB
C++
365 lines
16 KiB
C++
// SPDX-License-Identifier: MIT
|
|
// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
|
|
|
|
#pragma once
|
|
|
|
#include "ck/ck.hpp"
|
|
#include "ck/utility/reduction_operator.hpp"
|
|
#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
|
|
#include "ck/tensor_operation/gpu/device/device_gemm_reduce.hpp"
|
|
#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
|
|
|
|
#include "ck/library/utility/check_err.hpp"
|
|
#include "ck/library/utility/convolution_parameter.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_batched_gemm.hpp"
|
|
|
|
namespace ck {
|
|
namespace tensor_operation {
|
|
namespace device {
|
|
namespace instance {
|
|
|
|
using F32 = float;
|
|
using F16 = ck::half_t;
|
|
using ReducePtrsGlobal = ck::Tuple<F32*, F32*>;
|
|
using Identity = ck::tensor_operation::element_wise::PassThrough;
|
|
using Square = ck::tensor_operation::element_wise::UnarySquare;
|
|
using ReduceInElementOps = ck::Tuple<Identity, Square>;
|
|
using ReduceOutElementOps = ck::Tuple<Identity, Identity>;
|
|
|
|
using DeviceGemmReduceNoOpPtr =
|
|
ck::tensor_operation::device::DeviceGemmReducePtr<0, ReducePtrsGlobal::Size()>;
|
|
|
|
void add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instances(
|
|
std::vector<DeviceGemmReduceNoOpPtr>&);
|
|
|
|
void add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instances(
|
|
std::vector<DeviceGemmReduceNoOpPtr>&);
|
|
|
|
void add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instances(
|
|
std::vector<DeviceGemmReduceNoOpPtr>&);
|
|
|
|
void add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instances(
|
|
std::vector<DeviceGemmReduceNoOpPtr>&);
|
|
|
|
} // namespace instance
|
|
} // namespace device
|
|
} // namespace tensor_operation
|
|
} // namespace ck
|
|
|
|
namespace ck {
|
|
namespace profiler {
|
|
|
|
template <typename ADataType,
|
|
typename BDataType,
|
|
typename CDataType,
|
|
typename ReduceDataType,
|
|
typename ALayout,
|
|
typename BLayout,
|
|
typename CLayout>
|
|
bool profile_batched_gemm_reduce_impl(int do_verification,
|
|
int init_method,
|
|
bool do_log,
|
|
bool time_kernel,
|
|
int M,
|
|
int N,
|
|
int K,
|
|
int StrideA,
|
|
int StrideB,
|
|
int StrideC,
|
|
int BatchCount)
|
|
{
|
|
bool pass = true;
|
|
|
|
auto f_host_tensor_descriptor = [](std::size_t batch_count,
|
|
std::size_t row,
|
|
std::size_t col,
|
|
std::size_t stride,
|
|
auto layout) {
|
|
using namespace ck::literals;
|
|
|
|
if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
return HostTensorDescriptor(
|
|
{batch_count, row, col}, {row * stride, stride, 1_uz}, layout);
|
|
}
|
|
else
|
|
{
|
|
return HostTensorDescriptor(
|
|
{batch_count, row, col}, {col * stride, 1_uz, stride}, layout);
|
|
}
|
|
};
|
|
|
|
Tensor<ADataType> a_g_m_k(f_host_tensor_descriptor(BatchCount, M, K, StrideA, ALayout{}));
|
|
Tensor<BDataType> b_g_k_n(f_host_tensor_descriptor(BatchCount, K, N, StrideB, BLayout{}));
|
|
|
|
Tensor<CDataType> c_g_m_n_host_result(
|
|
f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{}));
|
|
Tensor<ReduceDataType> d0_g_m_host_result({BatchCount, M});
|
|
Tensor<ReduceDataType> d1_g_m_host_result({BatchCount, M});
|
|
|
|
Tensor<CDataType> c_g_m_n_device_result(
|
|
f_host_tensor_descriptor(BatchCount, M, N, StrideC, CLayout{}));
|
|
Tensor<ReduceDataType> d0_g_m_device_result({BatchCount, M});
|
|
Tensor<ReduceDataType> d1_g_m_device_result({BatchCount, M});
|
|
|
|
std::cout << "a_g_m_k: " << a_g_m_k.mDesc << std::endl;
|
|
std::cout << "b_g_k_n: " << b_g_k_n.mDesc << std::endl;
|
|
std::cout << "c_g_m_n: " << c_g_m_n_host_result.mDesc << std::endl;
|
|
std::cout << "d0_g_m: " << d0_g_m_host_result.mDesc << std::endl;
|
|
std::cout << "d1_g_m: " << d1_g_m_host_result.mDesc << std::endl;
|
|
|
|
std::size_t num_thread = std::thread::hardware_concurrency();
|
|
switch(init_method)
|
|
{
|
|
case 0: break;
|
|
case 1:
|
|
std::srand(0);
|
|
a_g_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5}, num_thread);
|
|
b_g_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5}, num_thread);
|
|
break;
|
|
default:
|
|
std::srand(0);
|
|
a_g_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0}, num_thread);
|
|
b_g_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}, num_thread);
|
|
}
|
|
|
|
using AElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using BElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using CElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using ReduceOp0 = ck::reduce::Add;
|
|
using ReduceOp1 = ck::reduce::Add;
|
|
using UnaryIdenticElementOp = ck::tensor_operation::element_wise::PassThrough;
|
|
using UnarySquareElementOp = ck::tensor_operation::element_wise::UnarySquare;
|
|
|
|
auto a_element_op = AElementOp{};
|
|
auto b_element_op = BElementOp{};
|
|
auto c_element_op = CElementOp{};
|
|
std::array<void*, 3> gemm_element_ops = {&a_element_op, &b_element_op, &c_element_op};
|
|
|
|
const auto reduce0_op = ReduceOp0{};
|
|
const auto reduce1_op = ReduceOp1{};
|
|
|
|
auto passthrough = UnaryIdenticElementOp{};
|
|
auto square = UnarySquareElementOp{};
|
|
std::array<void*, 2> reduce_in_element_ops = {&passthrough, &square};
|
|
std::array<void*, 2> reduce_out_element_ops = {&passthrough, &passthrough};
|
|
|
|
if(do_verification)
|
|
{
|
|
using ReferenceBatchedGemmInstance =
|
|
ck::tensor_operation::host::ReferenceBatchedGemm<ADataType,
|
|
BDataType,
|
|
CDataType,
|
|
float,
|
|
AElementOp,
|
|
BElementOp,
|
|
CElementOp>;
|
|
|
|
using ReduceAccDataType = ReduceDataType;
|
|
|
|
auto ref_batched_gemm = ReferenceBatchedGemmInstance{};
|
|
auto ref_invoker = ref_batched_gemm.MakeInvoker();
|
|
|
|
auto ref_argument = ref_batched_gemm.MakeArgument(
|
|
a_g_m_k, b_g_k_n, c_g_m_n_host_result, a_element_op, b_element_op, c_element_op);
|
|
|
|
ref_invoker.Run(ref_argument);
|
|
|
|
for(int batch = 0; batch < BatchCount; ++batch)
|
|
{
|
|
for(int m = 0; m < M; ++m)
|
|
{
|
|
auto reduce0_acc = reduce0_op.GetIdentityValue<ReduceAccDataType>();
|
|
auto reduce1_acc = reduce1_op.GetIdentityValue<ReduceAccDataType>();
|
|
|
|
for(int n = 0; n < N; ++n)
|
|
{
|
|
ReduceAccDataType d0_val =
|
|
ck::type_convert<ReduceAccDataType>(c_g_m_n_host_result(batch, m, n));
|
|
ReduceAccDataType d1_val;
|
|
|
|
square(d1_val, d0_val);
|
|
reduce0_op(reduce0_acc, d0_val);
|
|
reduce1_op(reduce1_acc, d1_val);
|
|
}
|
|
|
|
d0_g_m_host_result(batch, m) = ck::type_convert<ReduceDataType>(reduce0_acc);
|
|
d1_g_m_host_result(batch, m) = ck::type_convert<ReduceDataType>(reduce1_acc);
|
|
}
|
|
}
|
|
}
|
|
|
|
DeviceMem a_device_buf(sizeof(ADataType) * a_g_m_k.mDesc.GetElementSpaceSize());
|
|
DeviceMem b_device_buf(sizeof(BDataType) * b_g_k_n.mDesc.GetElementSpaceSize());
|
|
DeviceMem c_device_buf(sizeof(CDataType) * c_g_m_n_device_result.mDesc.GetElementSpaceSize());
|
|
DeviceMem reduce0_device_buf(sizeof(ReduceDataType) *
|
|
d0_g_m_device_result.mDesc.GetElementSpaceSize());
|
|
DeviceMem reduce1_device_buf(sizeof(ReduceDataType) *
|
|
d1_g_m_device_result.mDesc.GetElementSpaceSize());
|
|
|
|
std::array<void*, 2> p_reduces = {reduce0_device_buf.GetDeviceBuffer(),
|
|
reduce1_device_buf.GetDeviceBuffer()};
|
|
|
|
a_device_buf.ToDevice(a_g_m_k.mData.data());
|
|
b_device_buf.ToDevice(b_g_k_n.mData.data());
|
|
|
|
// add device GEMM instances
|
|
std::vector<ck::tensor_operation::device::instance::DeviceGemmReduceNoOpPtr> gemm_ptrs;
|
|
|
|
if constexpr(is_same<ADataType, half_t>::value && is_same<BDataType, half_t>::value &&
|
|
is_same<CDataType, half_t>::value)
|
|
{
|
|
if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
|
|
is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
|
|
is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
ck::tensor_operation::device::instance::
|
|
add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instances(
|
|
gemm_ptrs);
|
|
}
|
|
else if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value &&
|
|
is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
|
|
is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
ck::tensor_operation::device::instance::
|
|
add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instances(
|
|
gemm_ptrs);
|
|
}
|
|
else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
|
|
is_same<BLayout, tensor_layout::gemm::RowMajor>::value &&
|
|
is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
ck::tensor_operation::device::instance::
|
|
add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instances(
|
|
gemm_ptrs);
|
|
}
|
|
else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value &&
|
|
is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value &&
|
|
is_same<CLayout, tensor_layout::gemm::RowMajor>::value)
|
|
{
|
|
ck::tensor_operation::device::instance::
|
|
add_device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instances(
|
|
gemm_ptrs);
|
|
}
|
|
}
|
|
|
|
if(gemm_ptrs.size() <= 0)
|
|
{
|
|
throw std::runtime_error("wrong! no device GEMM instance found");
|
|
}
|
|
|
|
std::string best_gemm_name;
|
|
float best_ave_time = 0;
|
|
float best_tflops = 0;
|
|
float best_gb_per_sec = 0;
|
|
|
|
// profile device GEMM instances
|
|
for(auto& gemm_ptr : gemm_ptrs)
|
|
{
|
|
auto argument_ptr = gemm_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
|
|
b_device_buf.GetDeviceBuffer(),
|
|
nullptr,
|
|
{},
|
|
c_device_buf.GetDeviceBuffer(),
|
|
p_reduces,
|
|
M,
|
|
N,
|
|
K,
|
|
StrideA,
|
|
StrideB,
|
|
StrideC,
|
|
{},
|
|
gemm_element_ops,
|
|
{},
|
|
reduce_in_element_ops,
|
|
reduce_out_element_ops,
|
|
BatchCount);
|
|
|
|
auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
|
|
|
|
if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
|
|
{
|
|
// init DO, D1 to 0
|
|
reduce0_device_buf.SetZero();
|
|
reduce1_device_buf.SetZero();
|
|
|
|
float ave_time =
|
|
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
|
|
|
|
std::string gemm_name = gemm_ptr->GetTypeString();
|
|
|
|
std::size_t flop = std::size_t(2) * BatchCount * M * N * K;
|
|
std::size_t num_btype = sizeof(ADataType) * BatchCount * M * K +
|
|
sizeof(BDataType) * BatchCount * K * N +
|
|
sizeof(CDataType) * BatchCount * M * N;
|
|
|
|
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
|
|
|
|
float gb_per_sec = num_btype / 1.E6 / ave_time;
|
|
|
|
std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
|
|
<< " GB/s, " << gemm_name << std::endl;
|
|
|
|
if(tflops > best_tflops)
|
|
{
|
|
best_gemm_name = gemm_name;
|
|
best_tflops = tflops;
|
|
best_ave_time = ave_time;
|
|
best_gb_per_sec = gb_per_sec;
|
|
}
|
|
|
|
if(do_verification)
|
|
{
|
|
c_device_buf.FromDevice(c_g_m_n_device_result.mData.data());
|
|
reduce0_device_buf.FromDevice(d0_g_m_device_result.mData.data());
|
|
reduce1_device_buf.FromDevice(d1_g_m_device_result.mData.data());
|
|
|
|
bool c_error = ck::utils::check_err(c_g_m_n_device_result, c_g_m_n_host_result);
|
|
bool d0_error = ck::utils::check_err(d0_g_m_device_result, d0_g_m_host_result);
|
|
bool d1_error = ck::utils::check_err(d1_g_m_device_result, d1_g_m_host_result);
|
|
|
|
pass = pass && (c_error == true);
|
|
pass = pass && (d0_error == true);
|
|
pass = pass && (d1_error == true);
|
|
|
|
if(do_log)
|
|
{
|
|
LogRangeAsType<float>(std::cout << "a : ", a_g_m_k.mData, ",") << std::endl;
|
|
LogRangeAsType<float>(std::cout << "b: ", b_g_k_n.mData, ",") << std::endl;
|
|
LogRangeAsType<float>(std::cout << "c_host: ", c_g_m_n_host_result.mData, ",")
|
|
<< std::endl;
|
|
LogRangeAsType<float>(
|
|
std::cout << "c_device: ", c_g_m_n_device_result.mData, ",")
|
|
<< std::endl;
|
|
LogRangeAsType<float>(std::cout << "d0_host: ", d0_g_m_host_result.mData, ",")
|
|
<< std::endl;
|
|
LogRangeAsType<float>(
|
|
std::cout << "d0_device: ", d0_g_m_device_result.mData, ",")
|
|
<< std::endl;
|
|
LogRangeAsType<float>(std::cout << "d1_host: ", d1_g_m_host_result.mData, ",")
|
|
<< std::endl;
|
|
LogRangeAsType<float>(
|
|
std::cout << "d1_device: ", d1_g_m_device_result.mData, ",")
|
|
<< std::endl;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
std::cout << "does not support this GEMM problem" << std::endl;
|
|
}
|
|
}
|
|
|
|
std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
|
|
<< best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
|
|
|
|
return pass;
|
|
}
|
|
|
|
} // namespace profiler
|
|
} // namespace ck
|