Skip to content

GitLab

Commit 702c3379 authored by root's avatar root
Browse files

fixed clang format errors

parent 599497b0
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 432 additions and 448 deletions
include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
View file @ 702c3379
......@@ -39,7 +39,7 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
static constexpr auto I2 = Number<2>{};
static constexpr auto I3 = Number<3>{};
using ThisThreadBlock = ThisThreadBlock<BlockSize>;
using ThisThreadBlock = ThisThreadBlock<BlockSize>;
static constexpr index_t WaveSize = get_warp_size();
static constexpr index_t MPerBlock = AK0MK1BlockDesc{}.GetLength(I1);
......
include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
View file @ 702c3379
......@@ -57,7 +57,7 @@ struct ThreadGroupTensorSliceTransfer_v6r1
is_same<SliceLengths, decltype(thread_slice_lengths * ThreadClusterLengths{})>{},
"wrong! threads should be mapped to cover entire slicing window");
//static_assert(ThreadGroup::GetNumOfThread() >= thread_cluster_desc_.GetElementSize(),
// static_assert(ThreadGroup::GetNumOfThread() >= thread_cluster_desc_.GetElementSize(),
// "wrong! ThreadGroup::GetNumOfThread() too small");
if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
......
include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
View file @ 702c3379
......@@ -10,7 +10,6 @@
#include "gridwise_gemm_xdl_waveletmodel_cshuffle.hpp"
#include "gemm_specialization.hpp"
namespace ck {
namespace tensor_operation {
namespace device {
......@@ -438,7 +437,7 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle
{
using Argument = DeviceOp::Argument;
float Run(const Argument& arg, const StreamConfig& stream_config= StreamConfig{})
float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
{
#if 0
{
......@@ -485,11 +484,11 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle
typename GridwiseGemm::DefaultBlock2CTileMap,
true>;
ave_time =
launch_and_time_kernel(stream_config,
ave_time =
launch_and_time_kernel(stream_config,
kernel,
dim3(grid_size),
dim3(TileLoadThreadGroupSize + TileMathThreadGroupSize),
dim3(grid_size),
dim3(TileLoadThreadGroupSize + TileMathThreadGroupSize),
0,
arg.p_a_grid_,
arg.p_b_grid_,
......@@ -516,8 +515,8 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle
typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
typename GridwiseGemm::DefaultBlock2CTileMap,
false>;
ave_time =
launch_and_time_kernel(stream_config,
ave_time =
launch_and_time_kernel(stream_config,
kernel,
dim3(grid_size),
dim3(TileLoadThreadGroupSize + TileMathThreadGroupSize),
......@@ -539,7 +538,7 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle
// polymorphic
float Run(const BaseArgument* p_arg,
const StreamConfig& stream_config = StreamConfig{}) override
const StreamConfig& stream_config = StreamConfig{}) override
{
return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
}
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
View file @ 702c3379
......@@ -7,22 +7,22 @@ namespace ck {
template <typename TileLoadThreadGroup, index_t NumGemmKPrefetchStage>
struct GridwiseGemmLoadWave;
//1-stage prefetch
template<typename TileLoadThreadGroup>
// 1-stage prefetch
template <typename TileLoadThreadGroup>
struct GridwiseGemmLoadWave<TileLoadThreadGroup, 1>
{
__host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */)
{
// TODO: improve applicability
return true;
}
__host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */)
{
// TODO: improve applicability
return true;
}
__host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
{
return num_loop > 1;
}
__host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
{
return num_loop > 1;
}
template <bool HasMainLoop,
template <bool HasMainLoop,
typename AGridDesc,
typename ABlockDesc,
typename ABlockTransfer,
......@@ -36,43 +36,43 @@ struct GridwiseGemmLoadWave<TileLoadThreadGroup, 1>
typename BBlockBuffer,
typename BBlockTransferStep>
static __device__ void RunLoadWavePipeline(const AGridDesc& a_grid_desc,
const ABlockDesc& a_block_desc,
ABlockTransfer& a_blockwise_copy,
const AGridBuffer& a_grid_buf,
ABlockBuffer& a_block_buf,
const ABlockTransferStep& a_block_copy_step,
const BGridDesc& b_grid_desc,
const BBlockDesc& b_block_desc,
BBlockTransfer& b_blockwise_copy,
const BGridBuffer& b_grid_buf,
BBlockBuffer& b_block_buf,
const BBlockTransferStep& b_block_copy_step,
index_t num_loop)
const ABlockDesc& a_block_desc,
ABlockTransfer& a_blockwise_copy,
const AGridBuffer& a_grid_buf,
ABlockBuffer& a_block_buf,
const ABlockTransferStep& a_block_copy_step,
const BGridDesc& b_grid_desc,
const BBlockDesc& b_block_desc,
BBlockTransfer& b_blockwise_copy,
const BGridBuffer& b_grid_buf,
BBlockBuffer& b_block_buf,
const BBlockTransferStep& b_block_copy_step,
index_t num_loop)
{
// global read 0
a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
//move to 1
a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
// move to 1
a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
//LDS write 0
a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
// LDS write 0
a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
if constexpr(HasMainLoop)
{
index_t i=0;
if constexpr(HasMainLoop)
{
index_t i = 0;
do
do
{
//sync for Load threads()
block_sync_lds();
// sync for Load threads()
block_sync_lds();
// global read i + 1
a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
// move to i + 2
a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
......@@ -81,10 +81,9 @@ struct GridwiseGemmLoadWave<TileLoadThreadGroup, 1>
// sync with math threads()
block_sync_lds();
//LDS write i+1
a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
// LDS write i+1
a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
++i;
} while(i < (num_loop - 1));
......@@ -92,12 +91,10 @@ struct GridwiseGemmLoadWave<TileLoadThreadGroup, 1>
// tail
{
block_sync_lds();
// GEMM num_loop
}
block_sync_lds();
// GEMM num_loop
}
}
};
......@@ -105,29 +102,26 @@ template <typename TileMathThreadGroup, index_t NumGemmKPrefetchStage>
struct GridwiseGemmMathWave;
// 1- stage prefetch
template <typename TileMathThreadGroup>
struct GridwiseGemmMathWave<TileMathThreadGroup, 1>
struct GridwiseGemmMathWave<TileMathThreadGroup, 1>
{
__host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */)
{
return true;
}
__host__ __device__ static constexpr bool IsSupported(index_t /* num_loop */) { return true; }
__host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
{
return num_loop > 1;
}
__host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
{
return num_loop > 1;
}
template <bool HasMainLoop,
template <bool HasMainLoop,
typename ABlockBuffer,
typename BBlockBuffer,
typename BlockwiseGemm,
typename CThreadBuffer>
static __device__ void RunMathWavePipeline(ABlockBuffer& a_block_buf,
BBlockBuffer& b_block_buf,
const BlockwiseGemm& block_gemm,
CThreadBuffer& c_thread_buf,
index_t num_loop)
BBlockBuffer& b_block_buf,
const BlockwiseGemm& block_gemm,
CThreadBuffer& c_thread_buf,
index_t num_loop)
{
// Initialize C
c_thread_buf.Clear();
......@@ -155,7 +149,6 @@ struct GridwiseGemmMathWave<TileMathThreadGroup, 1>
// GEMM num_loop - 1
block_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
}
}
};
......
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
View file @ 702c3379
......@@ -249,8 +249,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4
}();
using BlockwiseGemm =
BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize
FloatAB,
BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize FloatAB,
FloatAcc,
decltype(a_k0_m_k1_block_desc),
decltype(b_k0_n_k1_block_desc),
......
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