Skip to content

GitLab

Commit 9dac9713 authored by mtgu0705's avatar mtgu0705
Browse files

enable blockwise pipelie v1 and v2. v1 is work for small K.

parent d58d55e5
Show whitespace changes
Inline Side-by-side
Showing with 115 additions and 42 deletions
example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_ab_scale.cpp
View file @ 9dac9713
...@@ -72,7 +72,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_ABScale_ ...@@ -72,7 +72,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_ABScale_
S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
1, 2, S<1, 32, 1, 8>, S<8>, 1, 2, S<1, 32, 1, 8>, S<8>,
ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, FP8>; ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, FP8>;
// clang-format on // clang-format on
int main(int argc, char* argv[]) int main(int argc, char* argv[])
......
include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v1_ab_scale.hpp
View file @ 9dac9713
...@@ -96,7 +96,8 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -96,7 +96,8 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
NPerXDL, NPerXDL,
MRepeat, MRepeat,
NRepeat, NRepeat,
KPack> KPack,
true>
{ {
using Base = BlockwiseGemmXdlops_pipeline_base<BlockSize, using Base = BlockwiseGemmXdlops_pipeline_base<BlockSize,
...@@ -117,7 +118,8 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -117,7 +118,8 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
NPerXDL, NPerXDL,
MRepeat, MRepeat,
NRepeat, NRepeat,
KPack>; KPack,
true>;
using Base::I0; using Base::I0;
using Base::KRepeat; using Base::KRepeat;
using Base::xdlops_gemm; using Base::xdlops_gemm;
...@@ -231,11 +233,26 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -231,11 +233,26 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step); a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step); b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
static_for<0, MRepeat, 1>{}([&](auto m0) {
a_scale_thread_copy.Run(a_scale_grid_desc, a_scale_thread_copy.Run(a_scale_grid_desc,
a_scale_grid_buf, a_scale_grid_buf,
a_scale_thread_desc, a_scale_thread_desc,
make_tuple(I0, I0), make_tuple(m0, I0),
a_scale_thread_buf); a_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<0>{}));
});
if(num_loop_per_scale == 1)
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<2>{}));
}
else
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<1>{}));
}
b_scale_thread_copy.Run(b_scale_grid_desc, b_scale_thread_copy.Run(b_scale_grid_desc,
b_scale_grid_buf, b_scale_grid_buf,
...@@ -243,7 +260,6 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -243,7 +260,6 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
make_tuple(I0, I0), make_tuple(I0, I0),
b_scale_thread_buf); b_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc, a_scale_thread_copy_step);
b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step); b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step);
// Local prefill 1 // Local prefill 1
...@@ -318,17 +334,32 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -318,17 +334,32 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
}); });
static_for<0, MRepeat, 1>{}([&](auto m0) {
a_scale_thread_copy.Run(a_scale_grid_desc, a_scale_thread_copy.Run(a_scale_grid_desc,
a_scale_grid_buf, a_scale_grid_buf,
a_scale_thread_desc, a_scale_thread_desc,
make_tuple(I0, I0), make_tuple(m0, I0),
a_scale_thread_buf); a_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<0>{}));
});
if(num_loop_per_scale == 1)
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<2>{}));
}
else
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<1>{}));
}
b_scale_thread_copy.Run(b_scale_grid_desc, b_scale_thread_copy.Run(b_scale_grid_desc,
b_scale_grid_buf, b_scale_grid_buf,
...@@ -336,7 +367,6 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -336,7 +367,6 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
make_tuple(I0, I0), make_tuple(I0, I0),
b_scale_thread_buf); b_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc, a_scale_thread_copy_step);
b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step); b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step);
block_sync_lds(); block_sync_lds();
...@@ -400,7 +430,7 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -400,7 +430,7 @@ struct BlockwiseGemmXdlops_pipeline_v1_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
......
include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v2_ab_scale.hpp
View file @ 9dac9713
...@@ -96,7 +96,8 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -96,7 +96,8 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
NPerXDL, NPerXDL,
MRepeat, MRepeat,
NRepeat, NRepeat,
KPack> KPack,
true>
{ {
using Base = BlockwiseGemmXdlops_pipeline_base<BlockSize, using Base = BlockwiseGemmXdlops_pipeline_base<BlockSize,
...@@ -117,7 +118,8 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -117,7 +118,8 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
NPerXDL, NPerXDL,
MRepeat, MRepeat,
NRepeat, NRepeat,
KPack>; KPack,
true>;
using Base::I0; using Base::I0;
using Base::KRepeat; using Base::KRepeat;
using Base::xdlops_gemm; using Base::xdlops_gemm;
...@@ -270,11 +272,26 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -270,11 +272,26 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step); a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step); b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
static_for<0, MRepeat, 1>{}([&](auto m0) {
a_scale_thread_copy.Run(a_scale_grid_desc, a_scale_thread_copy.Run(a_scale_grid_desc,
a_scale_grid_buf, a_scale_grid_buf,
a_scale_thread_desc, a_scale_thread_desc,
make_tuple(I0, I0), make_tuple(m0, I0),
a_scale_thread_buf); a_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<0>{}));
});
if(num_loop_per_scale == 1)
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<2>{}));
}
else
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<1>{}));
}
b_scale_thread_copy.Run(b_scale_grid_desc, b_scale_thread_copy.Run(b_scale_grid_desc,
b_scale_grid_buf, b_scale_grid_buf,
...@@ -282,7 +299,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -282,7 +299,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
make_tuple(I0, I0), make_tuple(I0, I0),
b_scale_thread_buf); b_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc, a_scale_thread_copy_step);
b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step); b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step);
// Local prefill 1 // Local prefill 1
...@@ -360,17 +376,32 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -360,17 +376,32 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
}); });
static_for<0, MRepeat, 1>{}([&](auto m0) {
a_scale_thread_copy.Run(a_scale_grid_desc, a_scale_thread_copy.Run(a_scale_grid_desc,
a_scale_grid_buf, a_scale_grid_buf,
a_scale_thread_desc, a_scale_thread_desc,
make_tuple(I0, I0), make_tuple(m0, I0),
a_scale_thread_buf); a_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<0>{}));
});
if(num_loop_per_scale == 1)
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<2>{}));
}
else
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<1>{}));
}
b_scale_thread_copy.Run(b_scale_grid_desc, b_scale_thread_copy.Run(b_scale_grid_desc,
b_scale_grid_buf, b_scale_grid_buf,
...@@ -378,8 +409,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -378,8 +409,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
make_tuple(I0, I0), make_tuple(I0, I0),
b_scale_thread_buf); b_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step);
b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc,
b_scale_thread_copy_step); b_scale_thread_copy_step);
...@@ -453,17 +482,32 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -453,17 +482,32 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
}); });
static_for<0, MRepeat, 1>{}([&](auto m0) {
a_scale_thread_copy.Run(a_scale_grid_desc, a_scale_thread_copy.Run(a_scale_grid_desc,
a_scale_grid_buf, a_scale_grid_buf,
a_scale_thread_desc, a_scale_thread_desc,
make_tuple(I0, I0), make_tuple(m0, I0),
a_scale_thread_buf); a_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<0>{}));
});
if(num_loop_per_scale == 1)
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<2>{}));
}
else
{
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc,
a_scale_thread_copy_step.At(Number<1>{}));
}
b_scale_thread_copy.Run(b_scale_grid_desc, b_scale_thread_copy.Run(b_scale_grid_desc,
b_scale_grid_buf, b_scale_grid_buf,
...@@ -471,7 +515,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -471,7 +515,6 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
make_tuple(I0, I0), make_tuple(I0, I0),
b_scale_thread_buf); b_scale_thread_buf);
a_scale_thread_copy.MoveSrcSliceWindow(a_scale_grid_desc, a_scale_thread_copy_step);
b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step); b_scale_thread_copy.MoveSrcSliceWindow(b_scale_grid_desc, b_scale_thread_copy_step);
block_sync_lds(); block_sync_lds();
...@@ -528,7 +571,7 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -528,7 +571,7 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
...@@ -586,7 +629,7 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr ...@@ -586,7 +629,7 @@ struct BlockwiseGemmXdlops_pipeline_v2_ab_scale<BlockGemmPipelineScheduler::Intr
c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t)); c_thread_desc_.CalculateOffset(make_tuple(m0, n0, t));
c_thread_buf(Number<c_offset>{}) += c_thread_buf(Number<c_offset>{}) +=
c_thread_buf_per_scale[Number<t>{}] * c_thread_buf_per_scale[Number<t>{}] *
type_convert<AccDataType>(a_scale_thread_buf[I0]) * type_convert<AccDataType>(a_scale_thread_buf[m0]) *
type_convert<AccDataType>(b_scale_thread_buf[I0]); type_convert<AccDataType>(b_scale_thread_buf[I0]);
}); });
}); });
......
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