ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 13:11:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Compile CK for all targets (#188)

Browse Source 
* compile ck for all targets

* update the target criteria

* change the target condition

* fixed some typos

* fixed missed file

* revert changes in README

* revert device_conv3d_fwd_xdl_...

* update device_conv3d_fwd_xdl_...

* update device_batched_gemm_reduce...

* test the unused arguments fix

* test the warning suppression

* try suppress warnings in device_batched_gemm_reduce_xdl...

* fix the last warnings

* replace UNUSED with std::ignore

* fix a typo

* replaced std::ignore with ignore

* add igonre header to common_header

* refactor atomicAdd

Co-authored-by: Chao Liu <chao.liu2@amd.com>
This commit is contained in:
Illia Silin
2022-04-15 12:17:28 -07:00
committed by GitHub
parent ac0d806650
commit 4221505d3e
17 changed files with 232 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
View File

@@ -48,6 +48,7 @@ __global__ void
const DGridDescriptor_MBlock_MPerBlock d_grid_desc_mblock_mperblock,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
@@ -66,6 +67,23 @@ __global__ void
c_grid_desc_mblock_mperblock_nblock_nperblock,
d_grid_desc_mblock_mperblock,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = p_d0_grid;
ignore = p_d1_grid;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = d0_reduce_op;
ignore = d1_reduce_op;
ignore = a_grid_desc_ak0_m_ak1;
ignore = b_grid_desc_bk0_n_bk1;
ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
ignore = d_grid_desc_mblock_mperblock;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <typename FloatAB,

13
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
View File

@@ -38,6 +38,7 @@ __global__ void
c_grid_desc_mblock_mperblock_nblock_nperblock,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
@@ -51,6 +52,18 @@ __global__ void
b_grid_desc_bk0_n_bk1,
c_grid_desc_mblock_mperblock_nblock_nperblock,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = a_grid_desc_ak0_m_ak1;
ignore = b_grid_desc_bk0_n_bk1;
ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <typename FloatAB,

21
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
View File

@@ -39,6 +39,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
@@ -52,6 +53,18 @@ __global__ void
b_element_op,
c_element_op,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = a_grid_desc_k0_m_k1;
ignore = b_grid_desc_k0_n_k1;
ignore = c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <typename GridwiseGemm,
@@ -74,6 +87,7 @@ __global__ void
const BElementwiseOperation b_element_op,
const CElementwiseOperation c_element_op)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const index_t block_id = get_block_1d_id();
@@ -126,6 +140,13 @@ __global__ void
gemm_desc_ptr[group_id].block_2_ctile_map_,
block_id_grp);
#endif
#else
ignore = gemm_desc_;
ignore = group_count;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <index_t BlockSize,

13
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
View File

@@ -37,6 +37,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const CBlockClusterAdaptor c_block_cluster_adaptor)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
constexpr index_t shared_block_size =
GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
@@ -53,6 +54,18 @@ __global__ void
b_element_op,
c_element_op,
c_block_cluster_adaptor);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = a_b_k0_m_k1_grid_desc;
ignore = b_b_k0_n_k1_grid_desc;
ignore = c_m0_n0_m1_n1_m2_m3_m4_n2_grid_desc;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = c_block_cluster_adaptor;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <index_t BlockSize,

13
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
View File

@@ -39,6 +39,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const CBlockClusterAdaptor c_block_cluster_adaptor)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
constexpr index_t shared_block_size =
GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
@@ -55,6 +56,18 @@ __global__ void
b_element_op,
c_element_op,
c_block_cluster_adaptor);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = a_b_k0_m_k1_grid_desc;
ignore = b_b_k0_n_k1_grid_desc;
ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = c_block_cluster_adaptor;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <index_t BlockSize,

13
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
View File

@@ -42,6 +42,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(
@@ -56,6 +57,18 @@ __global__ void
b_element_op,
c_element_op,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = a_grid_desc_ak0_m_ak1;
ignore = b_grid_desc_bk0_n_bk1;
ignore = c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <

15
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
View File

@@ -45,6 +45,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(
@@ -61,6 +62,20 @@ __global__ void
b_element_op,
c_element_op,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = p_c0_grid;
ignore = a_grid_desc_k0_m_k1;
ignore = b_grid_desc_k0_n_k1;
ignore = c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <

17
include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
View File

@@ -49,6 +49,7 @@ __global__ void
const CElementwiseOperation c_element_op,
const Block2CTileMap block_2_ctile_map)
{
#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
GridwiseGemm::template Run<HasMainK0BlockLoop>(
@@ -67,6 +68,22 @@ __global__ void
b_element_op,
c_element_op,
block_2_ctile_map);
#else
ignore = p_a_grid;
ignore = p_b_grid;
ignore = p_c_grid;
ignore = p_c0_grid;
ignore = p_c1_grid;
ignore = a_grid_desc_k0_m_k1;
ignore = b_grid_desc_k0_n_k1;
ignore = c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = c1_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl;
ignore = a_element_op;
ignore = b_element_op;
ignore = c_element_op;
ignore = block_2_ctile_map;
#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
}
template <

1
include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
View File

@@ -36,6 +36,7 @@ __global__ void kernel_buffer_set_value(const Grid1dBufferDescType grid_1d_buffe
DataType value)
{
using PassThroughOp = tensor_operation::element_wise::UnaryIdentic<DataType, DataType>;
constexpr auto I0 = Number<0>{};