Skip to content

GitLab

Commit f1d8217d authored by Jing Zhang's avatar Jing Zhang
Browse files

debuggging

parent fe728dc5
Hide whitespace changes
Inline Side-by-side
Showing with 48 additions and 45 deletions
example/01_gemm/gemm_wmma_fp16.cpp
View file @ f1d8217d
...@@ -19,48 +19,51 @@ using AElementOp = PassThrough; ...@@ -19,48 +19,51 @@ using AElementOp = PassThrough;
using BElementOp = PassThrough; using BElementOp = PassThrough;
using CElementOp = PassThrough; using CElementOp = PassThrough;
static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default; static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle<ALayout, // clang-format off
BLayout, using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmWmma_CShuffle
CLayout, < ALayout,
ADataType, BLayout,
BDataType, CLayout,
CDataType, ADataType,
AccDataType, BDataType,
CShuffleDataType, CDataType,
AElementOp, AccDataType,
BElementOp, CShuffleDataType,
CElementOp, AElementOp,
GemmDefault, BElementOp,
1, CElementOp,
32, GemmDefault,
16, 1, // Prefetch stage
32, 128, // BlockSize
64, 64, // MPerBlock
8, 128, // NPerBlock
16, 64, // KPerBlock
16, 8, // K1
1, 16, // MPerWmma
2, 16, // NPerWmma
S<2, 16, 1>, 2, // M-Repeat // M-PerWmma / M-Repeat = M-Wave
S<1, 0, 2>, 4, // N-Repeat // N-PerWmma / N-Repeat = N-Wave
S<1, 0, 2>, S<4, 32, 1>,
2, S<1, 0, 2>,
8, S<1, 0, 2>,
8, 2,
true, 8,
S<2, 16, 1>, 8,
S<1, 0, 2>, true,
S<1, 0, 2>, S<4, 32, 1>,
2, S<1, 0, 2>,
8, S<1, 0, 2>,
8, 2,
true, 8,
1, 8,
1, true,
S<1, 16, 1, 2>, 1, // C shuffle (M Repeat) Per store
8>; 1, // C shuffle (N Repeat) Per store
S<1, 32, 1, 4>,
8>;
// clang-format on
using ReferenceGemmInstance = ck::tensor_operation::host:: using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
......
example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
View file @ f1d8217d
...@@ -83,14 +83,14 @@ using DeviceOpInstanceKKNN = ...@@ -83,14 +83,14 @@ using DeviceOpInstanceKKNN =
2, 2,
4, 4,
4, 4,
true, false,
S<4, 32, 1>, S<4, 32, 1>,
S<1, 0, 2>, S<1, 0, 2>,
S<1, 0, 2>, S<1, 0, 2>,
2, 2,
4, 4,
4, 4,
true, false,
1, 1,
1, 1,
S<1, 64, 1, 2>, S<1, 64, 1, 2>,
......
include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
View file @ f1d8217d
...@@ -137,8 +137,8 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle ...@@ -137,8 +137,8 @@ struct DeviceBatchedContractionMultipleD_Wmma_CShuffle
static constexpr auto BEnableLds_auto = MWaves == 1 ? false : true; static constexpr auto BEnableLds_auto = MWaves == 1 ? false : true;
// If true, LDS is used unconditionally // If true, LDS is used unconditionally
static constexpr auto AEnableLds_manu = true; static constexpr auto AEnableLds_manu = false;
static constexpr auto BEnableLds_manu = true; static constexpr auto BEnableLds_manu = false;
static constexpr auto AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1); static constexpr auto AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1);
static constexpr auto BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1); static constexpr auto BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1);
......
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