Commit 7e8e54de authored by qinletao's avatar qinletao
Browse files

change rc matrix egister layout

parent ef77a1ca
...@@ -65,7 +65,7 @@ using ReferenceGemmInstance = ck::tensor_operation::host:: ...@@ -65,7 +65,7 @@ using ReferenceGemmInstance = ck::tensor_operation::host::
ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>; ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
template <typename DataType> template <typename DataType>
std::ostream& void show_2d_matrix(std::ostream& os, Tensor<DataType>& matrix) std::ostream& show_2d_matrix(std::ostream& os, Tensor<DataType>& matrix)
{ {
os << "[" << std::endl; os << "[" << std::endl;
for(int x = 0; x < matrix.mDesc.GetLengths()[0]; x++) for(int x = 0; x < matrix.mDesc.GetLengths()[0]; x++)
......
...@@ -387,8 +387,8 @@ struct mfma_type<MfmaInstr::mfma_i32_16x16x16i8> ...@@ -387,8 +387,8 @@ struct mfma_type<MfmaInstr::mfma_i32_16x16x16i8>
template <> template <>
struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64> struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
{ {
static constexpr index_t group_size = 4; static constexpr index_t group_size = 1;
static constexpr index_t num_groups_per_blk = 1; static constexpr index_t num_groups_per_blk = 4;
static constexpr index_t num_regs_per_blk = 4; // group_size * num_groups_per_blk; static constexpr index_t num_regs_per_blk = 4; // group_size * num_groups_per_blk;
static constexpr index_t num_threads_per_blk = 16; static constexpr index_t num_threads_per_blk = 16;
static constexpr index_t wave_size = 64; static constexpr index_t wave_size = 64;
......
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