Skip to content

GitLab

Commit ef2d6713 authored by carlushuang's avatar carlushuang
Browse files

Merge remote-tracking branch 'origin/develop' into stream-k-initial-impl

parents 1639689e a1e344b1
Show whitespace changes
Inline Side-by-side
Showing with 203 additions and 78 deletions
Jenkinsfile
View file @ ef2d6713
......@@ -19,7 +19,7 @@ def runShell(String command){
def getDockerImageName(){
def img
if (params.ROCMVERSION != "5.5" && params.ROCMVERSION != "5.6"){
if (params.ROCMVERSION != "5.6"){
if (params.COMPILER_VERSION == "") {
img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
}
......@@ -597,7 +597,7 @@ def process_results(Map conf=[:]){
//launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true
0 21 * * * % ROCMVERSION=5.4.3;COMPILER_VERSION=release;COMPILER_COMMIT=
0 21 * * * % ROCMVERSION=5.5;COMPILER_VERSION=release;COMPILER_COMMIT=
0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
pipeline {
......
cmake/EnableCompilerWarnings.cmake
View file @ ef2d6713
......@@ -92,6 +92,7 @@ else()
-Wno-unused-command-line-argument
-Wno-weak-vtables
-Wno-covered-switch-default
-Wno-unsafe-buffer-usage
)
else()
if (CMAKE_${COMPILER}_COMPILER_ID MATCHES "GNU" AND ${COMPILER} MATCHES "CXX")
......
example/27_layernorm/CMakeLists.txt
View file @ ef2d6713
add_example_executable(example_layernorm_blockwise layernorm_blockwise.cpp)
add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
example/27_layernorm/common.hpp 0 → 100644
View file @ ef2d6713
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#pragma once
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <getopt.h>
#include "ck/ck.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_common_util.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_layernorm.hpp"
example/27_layernorm/layernorm_fp16.cpp 0 → 100644
View file @ ef2d6713
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8>; // OutScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_groupnorm_example<DeviceInstance>(); }
example/27_layernorm/layernorm_splitk_fp16.cpp 0 → 100644
View file @ ef2d6713
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // XYVectorDim (0=M, 1=K)
8, // XScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8>; // YScalarPerVector
#include "run_layernorm_example.inc"
int main() { return run_groupnorm_example<DeviceInstance>(); }
example/27_layernorm/layernorm_blockwise.cpp → example/27_layernorm/run_layernorm_example.inc
View file @ ef2d6713
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <getopt.h>
#include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/check_err.hpp"
#include "ck/library/utility/device_memory.hpp"
#include "ck/library/utility/host_common_util.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_layernorm.hpp"
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using PassThrough = ck::tensor_operation::element_wise::PassThrough;
constexpr int Rank = 2;
constexpr int NumReduceDim = 1;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
PassThrough,
Rank,
NumReduceDim,
256, // BlockSize
8, // ClusterM
32, // ClusterK
1, // SliceM
8, // SliceK
1, // SrcVecDim (0=M, 1=K)
8, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
8, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
8, // BetaScalarPerVector
8>; // OutScalarPerVector
int main()
#pragma once
template <typename DeviceInstance>
int run_groupnorm_example()
{
bool time_kernel = false;
......@@ -111,6 +63,10 @@ int main()
return 1;
};
size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
DeviceMem workspace_dev(workspace_sz);
device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
auto invoker_ptr = device_instance.MakeInvokerPointer();
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
......@@ -133,7 +89,8 @@ int main()
ref_invoker.Run(ref_argument);
y_dev.FromDevice(y.mData.data());
pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results d1", 1e-3, 1e-3);
pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
}
return (pass ? 0 : 1);
}
example/31_batched_gemm_gemm/CMakeLists.txt
View file @ ef2d6713
add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
if(NOT GPU_TARGETS MATCHES "gfx940")
add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
......
example/41_grouped_conv_conv_fwd/CMakeLists.txt
View file @ ef2d6713
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
if(NOT GPU_TARGETS MATCHES "gfx940")
add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
endif()
if(USE_BITINT_EXTENSION_INT4)
add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
endif(USE_BITINT_EXTENSION_INT4)
example/42_groupnorm/CMakeLists.txt
View file @ ef2d6713
add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
example/42_groupnorm/common.hpp
View file @ ef2d6713
......@@ -12,6 +12,7 @@
#include "ck/ck.hpp"
#include "ck/utility/reduction_enums.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
#include "ck/library/utility/fill.hpp"
......
example/42_groupnorm/groupnorm_splitk_fp16.cpp 0 → 100644
View file @ ef2d6713
// SPDX-License-Identifier: MIT
// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
#include "common.hpp"
constexpr int Rank = 5;
constexpr int NumReduceDim = 3;
using XDataType = ck::half_t;
using GammaDataType = ck::half_t;
using BetaDataType = ck::half_t;
using YDataType = ck::half_t;
using ComputeDataType = float;
using YElementOp = ck::tensor_operation::element_wise::Swish;
using DeviceInstance =
ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
GammaDataType,
BetaDataType,
ComputeDataType,
YDataType,
YElementOp,
Rank,
NumReduceDim,
256, // BlockSize
1, // ClusterM
256, // ClusterK
1, // SliceM
16, // SliceK
1, // SrcVecDim (0=M, 1=K)
2, // SrcScalarPerVector
1, // GammaVecDim (0=M, 1=K)
2, // GammaScalarPerVector
1, // BetaVecDim (0=M, 1=K)
2, // BetaScalarPerVector
2>; // OutScalarPerVector
#include "run_groupnorm_example.inc"
int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
example/42_groupnorm/run_groupnorm_example.inc
View file @ ef2d6713
......@@ -73,6 +73,10 @@ int run_groupnorm_example(int argc, char* argv[])
return 1;
};
size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
DeviceMem workspace_dev(workspace_sz);
device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
auto invoker_ptr = device_instance.MakeInvokerPointer();
float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true, true});
......
include/ck/ck.hpp
View file @ ef2d6713
......@@ -31,7 +31,7 @@
#ifndef __HIP_DEVICE_COMPILE__ // for host code
#define CK_BUFFER_RESOURCE_3RD_DWORD -1
#elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx908__) || \
defined(__gfx90a__) // for GPU code
defined(__gfx90a__) || defined(__gfx940__) // for GPU code
#define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
#elif defined(__gfx1030__) // for GPU code
#define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
......@@ -43,8 +43,8 @@
#ifndef __HIP_DEVICE_COMPILE__ // for host code, define nothing
#elif defined(__gfx803__) || defined(__gfx900__) // for GPU code
#define CK_USE_AMD_V_MAC_F32
#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx1030__) // for GPU code
#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__) || \
defined(__gfx940__) // for GPU code
#define CK_USE_AMD_V_FMAC_F32
#define CK_USE_AMD_V_DOT2_F32_F16
#define CK_USE_AMD_V_DOT4_I32_I8
......@@ -53,14 +53,18 @@
// MFMA instruction
#ifndef __HIP_DEVICE_COMPILE__ // for host code
#define CK_USE_AMD_MFMA
#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
#define CK_USE_AMD_MFMA
#endif
#if defined(__gfx90a__)
#if(defined(__gfx90a__) || defined(__gfx940__))
#define CK_USE_AMD_MFMA_BF16_1K_OP
#endif
#if defined(__gfx940__)
#define CK_USE_AMD_MFMA_GFX940
#endif
// WMMA instruction
#ifndef __HIP_DEVICE_COMPILE__ // for host code
#define CK_USE_AMD_WMMA
......@@ -80,13 +84,13 @@
// buffer atomic add: floating point
#ifndef __HIP_DEVICE_COMPILE__ // for host code
#define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
#define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
#else // for GPU code
#define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 0
#endif
#if defined(__gfx90a__) // for GPU code
#if(defined(__gfx90a__) || defined(__gfx940__)) // for GPU code
#define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 1
#else
#define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 0
......@@ -171,7 +175,10 @@
// denorm test fix, required to work around dissue
#ifndef CK_WORKAROUND_DENORM_FIX
#define CK_WORKAROUND_DENORM_FIX 0
#endif
#elif
// enable only on MI200
#define CK_WORKAROUND_DENORM_FIX = CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
#endif // CK_WORKAROUND_DENORM_FIX
namespace ck {
......
include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
View file @ ef2d6713
......@@ -47,7 +47,8 @@ __global__ void
e_grid_desc_mblock_mperblock_nblock_nperblock,
const Block2ETileMap block_2_etile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
......@@ -416,7 +417,8 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
static bool IsSupportedArgument(const Argument& arg)
{
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
ck::get_device_name() == "gfx940"))
{
return false;
}
......
include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
View file @ ef2d6713
......@@ -135,7 +135,7 @@ __global__ void
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) || \
defined(__gfx90a__) || defined(__gfx908__))
defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__))
// offset base pointer for each work-group
const index_t num_blocks_per_batch =
__builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
......@@ -710,7 +710,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
// check device
if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908"))
ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908" ||
ck::get_device_name() == "gfx940"))
{
return false;
}
......
include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
View file @ ef2d6713
......@@ -43,7 +43,8 @@ __global__ void
const B1ElementwiseOperation b1_element_op,
const CElementwiseOperation c_element_op)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const index_t block_id = get_block_1d_id();
......@@ -678,7 +679,8 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
static bool IsSupportedArgument(const Argument& arg)
{
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
ck::get_device_name() == "gfx940"))
{
return false;
}
......
include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
View file @ ef2d6713
......@@ -56,7 +56,8 @@ __global__ void
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
const Block2ETileMap block_2_etile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const index_t num_blocks_per_batch =
......@@ -938,7 +939,8 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
static bool IsSupportedArgument(const Argument& arg)
{
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
ck::get_device_name() == "gfx940"))
{
return false;
}
......
include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
View file @ ef2d6713
......@@ -56,7 +56,8 @@ __global__ void
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
const Block2ETileMap block_2_etile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
defined(__gfx940__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const index_t num_blocks_per_batch =
......@@ -839,7 +840,8 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
static bool IsSupportedArgument(const Argument& arg)
{
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
ck::get_device_name() == "gfx940"))
{
return false;
}
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment